diff options
Diffstat (limited to 'drivers/infiniband/hw/amso1100')
23 files changed, 9079 insertions, 0 deletions
diff --git a/drivers/infiniband/hw/amso1100/Kbuild b/drivers/infiniband/hw/amso1100/Kbuild new file mode 100644 index 000000000..950dfabcd --- /dev/null +++ b/drivers/infiniband/hw/amso1100/Kbuild @@ -0,0 +1,6 @@ +ccflags-$(CONFIG_INFINIBAND_AMSO1100_DEBUG) := -DDEBUG + +obj-$(CONFIG_INFINIBAND_AMSO1100) += iw_c2.o + +iw_c2-y := c2.o c2_provider.o c2_rnic.o c2_alloc.o c2_mq.o c2_ae.o c2_vq.o \ + c2_intr.o c2_cq.o c2_qp.o c2_cm.o c2_mm.o c2_pd.o diff --git a/drivers/infiniband/hw/amso1100/Kconfig b/drivers/infiniband/hw/amso1100/Kconfig new file mode 100644 index 000000000..e6ce5f209 --- /dev/null +++ b/drivers/infiniband/hw/amso1100/Kconfig @@ -0,0 +1,15 @@ +config INFINIBAND_AMSO1100 + tristate "Ammasso 1100 HCA support" + depends on PCI && INET + ---help--- + This is a low-level driver for the Ammasso 1100 host + channel adapter (HCA). + +config INFINIBAND_AMSO1100_DEBUG + bool "Verbose debugging output" + depends on INFINIBAND_AMSO1100 + default n + ---help--- + This option causes the amso1100 driver to produce a bunch of + debug messages. Select this if you are developing the driver + or trying to diagnose a problem. diff --git a/drivers/infiniband/hw/amso1100/c2.c b/drivers/infiniband/hw/amso1100/c2.c new file mode 100644 index 000000000..766a71cce --- /dev/null +++ b/drivers/infiniband/hw/amso1100/c2.c @@ -0,0 +1,1241 @@ +/* + * Copyright (c) 2005 Ammasso, Inc. All rights reserved. + * Copyright (c) 2005 Open Grid Computing, Inc. All rights reserved. + * + * This software is available to you under a choice of one of two + * licenses. You may choose to be licensed under the terms of the GNU + * General Public License (GPL) Version 2, available from the file + * COPYING in the main directory of this source tree, or the + * OpenIB.org BSD license below: + * + * Redistribution and use in source and binary forms, with or + * without modification, are permitted provided that the following + * conditions are met: + * + * - Redistributions of source code must retain the above + * copyright notice, this list of conditions and the following + * disclaimer. + * + * - Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and/or other materials + * provided with the distribution. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS + * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN + * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ +#include <linux/module.h> +#include <linux/moduleparam.h> +#include <linux/pci.h> +#include <linux/netdevice.h> +#include <linux/etherdevice.h> +#include <linux/inetdevice.h> +#include <linux/interrupt.h> +#include <linux/delay.h> +#include <linux/ethtool.h> +#include <linux/mii.h> +#include <linux/if_vlan.h> +#include <linux/crc32.h> +#include <linux/in.h> +#include <linux/ip.h> +#include <linux/tcp.h> +#include <linux/init.h> +#include <linux/dma-mapping.h> +#include <linux/slab.h> +#include <linux/prefetch.h> + +#include <asm/io.h> +#include <asm/irq.h> +#include <asm/byteorder.h> + +#include <rdma/ib_smi.h> +#include "c2.h" +#include "c2_provider.h" + +MODULE_AUTHOR("Tom Tucker <tom@opengridcomputing.com>"); +MODULE_DESCRIPTION("Ammasso AMSO1100 Low-level iWARP Driver"); +MODULE_LICENSE("Dual BSD/GPL"); +MODULE_VERSION(DRV_VERSION); + +static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_LINK + | NETIF_MSG_IFUP | NETIF_MSG_IFDOWN; + +static int debug = -1; /* defaults above */ +module_param(debug, int, 0); +MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)"); + +static int c2_up(struct net_device *netdev); +static int c2_down(struct net_device *netdev); +static int c2_xmit_frame(struct sk_buff *skb, struct net_device *netdev); +static void c2_tx_interrupt(struct net_device *netdev); +static void c2_rx_interrupt(struct net_device *netdev); +static irqreturn_t c2_interrupt(int irq, void *dev_id); +static void c2_tx_timeout(struct net_device *netdev); +static int c2_change_mtu(struct net_device *netdev, int new_mtu); +static void c2_reset(struct c2_port *c2_port); + +static struct pci_device_id c2_pci_table[] = { + { PCI_DEVICE(0x18b8, 0xb001) }, + { 0 } +}; + +MODULE_DEVICE_TABLE(pci, c2_pci_table); + +static void c2_print_macaddr(struct net_device *netdev) +{ + pr_debug("%s: MAC %pM, IRQ %u\n", netdev->name, netdev->dev_addr, netdev->irq); +} + +static void c2_set_rxbufsize(struct c2_port *c2_port) +{ + struct net_device *netdev = c2_port->netdev; + + if (netdev->mtu > RX_BUF_SIZE) + c2_port->rx_buf_size = + netdev->mtu + ETH_HLEN + sizeof(struct c2_rxp_hdr) + + NET_IP_ALIGN; + else + c2_port->rx_buf_size = sizeof(struct c2_rxp_hdr) + RX_BUF_SIZE; +} + +/* + * Allocate TX ring elements and chain them together. + * One-to-one association of adapter descriptors with ring elements. + */ +static int c2_tx_ring_alloc(struct c2_ring *tx_ring, void *vaddr, + dma_addr_t base, void __iomem * mmio_txp_ring) +{ + struct c2_tx_desc *tx_desc; + struct c2_txp_desc __iomem *txp_desc; + struct c2_element *elem; + int i; + + tx_ring->start = kmalloc(sizeof(*elem) * tx_ring->count, GFP_KERNEL); + if (!tx_ring->start) + return -ENOMEM; + + elem = tx_ring->start; + tx_desc = vaddr; + txp_desc = mmio_txp_ring; + for (i = 0; i < tx_ring->count; i++, elem++, tx_desc++, txp_desc++) { + tx_desc->len = 0; + tx_desc->status = 0; + + /* Set TXP_HTXD_UNINIT */ + __raw_writeq((__force u64) cpu_to_be64(0x1122334455667788ULL), + (void __iomem *) txp_desc + C2_TXP_ADDR); + __raw_writew(0, (void __iomem *) txp_desc + C2_TXP_LEN); + __raw_writew((__force u16) cpu_to_be16(TXP_HTXD_UNINIT), + (void __iomem *) txp_desc + C2_TXP_FLAGS); + + elem->skb = NULL; + elem->ht_desc = tx_desc; + elem->hw_desc = txp_desc; + + if (i == tx_ring->count - 1) { + elem->next = tx_ring->start; + tx_desc->next_offset = base; + } else { + elem->next = elem + 1; + tx_desc->next_offset = + base + (i + 1) * sizeof(*tx_desc); + } + } + + tx_ring->to_use = tx_ring->to_clean = tx_ring->start; + + return 0; +} + +/* + * Allocate RX ring elements and chain them together. + * One-to-one association of adapter descriptors with ring elements. + */ +static int c2_rx_ring_alloc(struct c2_ring *rx_ring, void *vaddr, + dma_addr_t base, void __iomem * mmio_rxp_ring) +{ + struct c2_rx_desc *rx_desc; + struct c2_rxp_desc __iomem *rxp_desc; + struct c2_element *elem; + int i; + + rx_ring->start = kmalloc(sizeof(*elem) * rx_ring->count, GFP_KERNEL); + if (!rx_ring->start) + return -ENOMEM; + + elem = rx_ring->start; + rx_desc = vaddr; + rxp_desc = mmio_rxp_ring; + for (i = 0; i < rx_ring->count; i++, elem++, rx_desc++, rxp_desc++) { + rx_desc->len = 0; + rx_desc->status = 0; + + /* Set RXP_HRXD_UNINIT */ + __raw_writew((__force u16) cpu_to_be16(RXP_HRXD_OK), + (void __iomem *) rxp_desc + C2_RXP_STATUS); + __raw_writew(0, (void __iomem *) rxp_desc + C2_RXP_COUNT); + __raw_writew(0, (void __iomem *) rxp_desc + C2_RXP_LEN); + __raw_writeq((__force u64) cpu_to_be64(0x99aabbccddeeffULL), + (void __iomem *) rxp_desc + C2_RXP_ADDR); + __raw_writew((__force u16) cpu_to_be16(RXP_HRXD_UNINIT), + (void __iomem *) rxp_desc + C2_RXP_FLAGS); + + elem->skb = NULL; + elem->ht_desc = rx_desc; + elem->hw_desc = rxp_desc; + + if (i == rx_ring->count - 1) { + elem->next = rx_ring->start; + rx_desc->next_offset = base; + } else { + elem->next = elem + 1; + rx_desc->next_offset = + base + (i + 1) * sizeof(*rx_desc); + } + } + + rx_ring->to_use = rx_ring->to_clean = rx_ring->start; + + return 0; +} + +/* Setup buffer for receiving */ +static inline int c2_rx_alloc(struct c2_port *c2_port, struct c2_element *elem) +{ + struct c2_dev *c2dev = c2_port->c2dev; + struct c2_rx_desc *rx_desc = elem->ht_desc; + struct sk_buff *skb; + dma_addr_t mapaddr; + u32 maplen; + struct c2_rxp_hdr *rxp_hdr; + + skb = dev_alloc_skb(c2_port->rx_buf_size); + if (unlikely(!skb)) { + pr_debug("%s: out of memory for receive\n", + c2_port->netdev->name); + return -ENOMEM; + } + + /* Zero out the rxp hdr in the sk_buff */ + memset(skb->data, 0, sizeof(*rxp_hdr)); + + skb->dev = c2_port->netdev; + + maplen = c2_port->rx_buf_size; + mapaddr = + pci_map_single(c2dev->pcidev, skb->data, maplen, + PCI_DMA_FROMDEVICE); + + /* Set the sk_buff RXP_header to RXP_HRXD_READY */ + rxp_hdr = (struct c2_rxp_hdr *) skb->data; + rxp_hdr->flags = RXP_HRXD_READY; + + __raw_writew(0, elem->hw_desc + C2_RXP_STATUS); + __raw_writew((__force u16) cpu_to_be16((u16) maplen - sizeof(*rxp_hdr)), + elem->hw_desc + C2_RXP_LEN); + __raw_writeq((__force u64) cpu_to_be64(mapaddr), elem->hw_desc + C2_RXP_ADDR); + __raw_writew((__force u16) cpu_to_be16(RXP_HRXD_READY), + elem->hw_desc + C2_RXP_FLAGS); + + elem->skb = skb; + elem->mapaddr = mapaddr; + elem->maplen = maplen; + rx_desc->len = maplen; + + return 0; +} + +/* + * Allocate buffers for the Rx ring + * For receive: rx_ring.to_clean is next received frame + */ +static int c2_rx_fill(struct c2_port *c2_port) +{ + struct c2_ring *rx_ring = &c2_port->rx_ring; + struct c2_element *elem; + int ret = 0; + + elem = rx_ring->start; + do { + if (c2_rx_alloc(c2_port, elem)) { + ret = 1; + break; + } + } while ((elem = elem->next) != rx_ring->start); + + rx_ring->to_clean = rx_ring->start; + return ret; +} + +/* Free all buffers in RX ring, assumes receiver stopped */ +static void c2_rx_clean(struct c2_port *c2_port) +{ + struct c2_dev *c2dev = c2_port->c2dev; + struct c2_ring *rx_ring = &c2_port->rx_ring; + struct c2_element *elem; + struct c2_rx_desc *rx_desc; + + elem = rx_ring->start; + do { + rx_desc = elem->ht_desc; + rx_desc->len = 0; + + __raw_writew(0, elem->hw_desc + C2_RXP_STATUS); + __raw_writew(0, elem->hw_desc + C2_RXP_COUNT); + __raw_writew(0, elem->hw_desc + C2_RXP_LEN); + __raw_writeq((__force u64) cpu_to_be64(0x99aabbccddeeffULL), + elem->hw_desc + C2_RXP_ADDR); + __raw_writew((__force u16) cpu_to_be16(RXP_HRXD_UNINIT), + elem->hw_desc + C2_RXP_FLAGS); + + if (elem->skb) { + pci_unmap_single(c2dev->pcidev, elem->mapaddr, + elem->maplen, PCI_DMA_FROMDEVICE); + dev_kfree_skb(elem->skb); + elem->skb = NULL; + } + } while ((elem = elem->next) != rx_ring->start); +} + +static inline int c2_tx_free(struct c2_dev *c2dev, struct c2_element *elem) +{ + struct c2_tx_desc *tx_desc = elem->ht_desc; + + tx_desc->len = 0; + + pci_unmap_single(c2dev->pcidev, elem->mapaddr, elem->maplen, + PCI_DMA_TODEVICE); + + if (elem->skb) { + dev_kfree_skb_any(elem->skb); + elem->skb = NULL; + } + + return 0; +} + +/* Free all buffers in TX ring, assumes transmitter stopped */ +static void c2_tx_clean(struct c2_port *c2_port) +{ + struct c2_ring *tx_ring = &c2_port->tx_ring; + struct c2_element *elem; + struct c2_txp_desc txp_htxd; + int retry; + unsigned long flags; + + spin_lock_irqsave(&c2_port->tx_lock, flags); + + elem = tx_ring->start; + + do { + retry = 0; + do { + txp_htxd.flags = + readw(elem->hw_desc + C2_TXP_FLAGS); + + if (txp_htxd.flags == TXP_HTXD_READY) { + retry = 1; + __raw_writew(0, + elem->hw_desc + C2_TXP_LEN); + __raw_writeq(0, + elem->hw_desc + C2_TXP_ADDR); + __raw_writew((__force u16) cpu_to_be16(TXP_HTXD_DONE), + elem->hw_desc + C2_TXP_FLAGS); + c2_port->netdev->stats.tx_dropped++; + break; + } else { + __raw_writew(0, + elem->hw_desc + C2_TXP_LEN); + __raw_writeq((__force u64) cpu_to_be64(0x1122334455667788ULL), + elem->hw_desc + C2_TXP_ADDR); + __raw_writew((__force u16) cpu_to_be16(TXP_HTXD_UNINIT), + elem->hw_desc + C2_TXP_FLAGS); + } + + c2_tx_free(c2_port->c2dev, elem); + + } while ((elem = elem->next) != tx_ring->start); + } while (retry); + + c2_port->tx_avail = c2_port->tx_ring.count - 1; + c2_port->c2dev->cur_tx = tx_ring->to_use - tx_ring->start; + + if (c2_port->tx_avail > MAX_SKB_FRAGS + 1) + netif_wake_queue(c2_port->netdev); + + spin_unlock_irqrestore(&c2_port->tx_lock, flags); +} + +/* + * Process transmit descriptors marked 'DONE' by the firmware, + * freeing up their unneeded sk_buffs. + */ +static void c2_tx_interrupt(struct net_device *netdev) +{ + struct c2_port *c2_port = netdev_priv(netdev); + struct c2_dev *c2dev = c2_port->c2dev; + struct c2_ring *tx_ring = &c2_port->tx_ring; + struct c2_element *elem; + struct c2_txp_desc txp_htxd; + + spin_lock(&c2_port->tx_lock); + + for (elem = tx_ring->to_clean; elem != tx_ring->to_use; + elem = elem->next) { + txp_htxd.flags = + be16_to_cpu((__force __be16) readw(elem->hw_desc + C2_TXP_FLAGS)); + + if (txp_htxd.flags != TXP_HTXD_DONE) + break; + + if (netif_msg_tx_done(c2_port)) { + /* PCI reads are expensive in fast path */ + txp_htxd.len = + be16_to_cpu((__force __be16) readw(elem->hw_desc + C2_TXP_LEN)); + pr_debug("%s: tx done slot %3Zu status 0x%x len " + "%5u bytes\n", + netdev->name, elem - tx_ring->start, + txp_htxd.flags, txp_htxd.len); + } + + c2_tx_free(c2dev, elem); + ++(c2_port->tx_avail); + } + + tx_ring->to_clean = elem; + + if (netif_queue_stopped(netdev) + && c2_port->tx_avail > MAX_SKB_FRAGS + 1) + netif_wake_queue(netdev); + + spin_unlock(&c2_port->tx_lock); +} + +static void c2_rx_error(struct c2_port *c2_port, struct c2_element *elem) +{ + struct c2_rx_desc *rx_desc = elem->ht_desc; + struct c2_rxp_hdr *rxp_hdr = (struct c2_rxp_hdr *) elem->skb->data; + + if (rxp_hdr->status != RXP_HRXD_OK || + rxp_hdr->len > (rx_desc->len - sizeof(*rxp_hdr))) { + pr_debug("BAD RXP_HRXD\n"); + pr_debug(" rx_desc : %p\n", rx_desc); + pr_debug(" index : %Zu\n", + elem - c2_port->rx_ring.start); + pr_debug(" len : %u\n", rx_desc->len); + pr_debug(" rxp_hdr : %p [PA %p]\n", rxp_hdr, + (void *) __pa((unsigned long) rxp_hdr)); + pr_debug(" flags : 0x%x\n", rxp_hdr->flags); + pr_debug(" status: 0x%x\n", rxp_hdr->status); + pr_debug(" len : %u\n", rxp_hdr->len); + pr_debug(" rsvd : 0x%x\n", rxp_hdr->rsvd); + } + + /* Setup the skb for reuse since we're dropping this pkt */ + elem->skb->data = elem->skb->head; + skb_reset_tail_pointer(elem->skb); + + /* Zero out the rxp hdr in the sk_buff */ + memset(elem->skb->data, 0, sizeof(*rxp_hdr)); + + /* Write the descriptor to the adapter's rx ring */ + __raw_writew(0, elem->hw_desc + C2_RXP_STATUS); + __raw_writew(0, elem->hw_desc + C2_RXP_COUNT); + __raw_writew((__force u16) cpu_to_be16((u16) elem->maplen - sizeof(*rxp_hdr)), + elem->hw_desc + C2_RXP_LEN); + __raw_writeq((__force u64) cpu_to_be64(elem->mapaddr), + elem->hw_desc + C2_RXP_ADDR); + __raw_writew((__force u16) cpu_to_be16(RXP_HRXD_READY), + elem->hw_desc + C2_RXP_FLAGS); + + pr_debug("packet dropped\n"); + c2_port->netdev->stats.rx_dropped++; +} + +static void c2_rx_interrupt(struct net_device *netdev) +{ + struct c2_port *c2_port = netdev_priv(netdev); + struct c2_dev *c2dev = c2_port->c2dev; + struct c2_ring *rx_ring = &c2_port->rx_ring; + struct c2_element *elem; + struct c2_rx_desc *rx_desc; + struct c2_rxp_hdr *rxp_hdr; + struct sk_buff *skb; + dma_addr_t mapaddr; + u32 maplen, buflen; + unsigned long flags; + + spin_lock_irqsave(&c2dev->lock, flags); + + /* Begin where we left off */ + rx_ring->to_clean = rx_ring->start + c2dev->cur_rx; + + for (elem = rx_ring->to_clean; elem->next != rx_ring->to_clean; + elem = elem->next) { + rx_desc = elem->ht_desc; + mapaddr = elem->mapaddr; + maplen = elem->maplen; + skb = elem->skb; + rxp_hdr = (struct c2_rxp_hdr *) skb->data; + + if (rxp_hdr->flags != RXP_HRXD_DONE) + break; + buflen = rxp_hdr->len; + + /* Sanity check the RXP header */ + if (rxp_hdr->status != RXP_HRXD_OK || + buflen > (rx_desc->len - sizeof(*rxp_hdr))) { + c2_rx_error(c2_port, elem); + continue; + } + + /* + * Allocate and map a new skb for replenishing the host + * RX desc + */ + if (c2_rx_alloc(c2_port, elem)) { + c2_rx_error(c2_port, elem); + continue; + } + + /* Unmap the old skb */ + pci_unmap_single(c2dev->pcidev, mapaddr, maplen, + PCI_DMA_FROMDEVICE); + + prefetch(skb->data); + + /* + * Skip past the leading 8 bytes comprising of the + * "struct c2_rxp_hdr", prepended by the adapter + * to the usual Ethernet header ("struct ethhdr"), + * to the start of the raw Ethernet packet. + * + * Fix up the various fields in the sk_buff before + * passing it up to netif_rx(). The transfer size + * (in bytes) specified by the adapter len field of + * the "struct rxp_hdr_t" does NOT include the + * "sizeof(struct c2_rxp_hdr)". + */ + skb->data += sizeof(*rxp_hdr); + skb_set_tail_pointer(skb, buflen); + skb->len = buflen; + skb->protocol = eth_type_trans(skb, netdev); + + netif_rx(skb); + + netdev->stats.rx_packets++; + netdev->stats.rx_bytes += buflen; + } + + /* Save where we left off */ + rx_ring->to_clean = elem; + c2dev->cur_rx = elem - rx_ring->start; + C2_SET_CUR_RX(c2dev, c2dev->cur_rx); + + spin_unlock_irqrestore(&c2dev->lock, flags); +} + +/* + * Handle netisr0 TX & RX interrupts. + */ +static irqreturn_t c2_interrupt(int irq, void *dev_id) +{ + unsigned int netisr0, dmaisr; + int handled = 0; + struct c2_dev *c2dev = (struct c2_dev *) dev_id; + + /* Process CCILNET interrupts */ + netisr0 = readl(c2dev->regs + C2_NISR0); + if (netisr0) { + + /* + * There is an issue with the firmware that always + * provides the status of RX for both TX & RX + * interrupts. So process both queues here. + */ + c2_rx_interrupt(c2dev->netdev); + c2_tx_interrupt(c2dev->netdev); + + /* Clear the interrupt */ + writel(netisr0, c2dev->regs + C2_NISR0); + handled++; + } + + /* Process RNIC interrupts */ + dmaisr = readl(c2dev->regs + C2_DISR); + if (dmaisr) { + writel(dmaisr, c2dev->regs + C2_DISR); + c2_rnic_interrupt(c2dev); + handled++; + } + + if (handled) { + return IRQ_HANDLED; + } else { + return IRQ_NONE; + } +} + +static int c2_up(struct net_device *netdev) +{ + struct c2_port *c2_port = netdev_priv(netdev); + struct c2_dev *c2dev = c2_port->c2dev; + struct c2_element *elem; + struct c2_rxp_hdr *rxp_hdr; + struct in_device *in_dev; + size_t rx_size, tx_size; + int ret, i; + unsigned int netimr0; + + if (netif_msg_ifup(c2_port)) + pr_debug("%s: enabling interface\n", netdev->name); + + /* Set the Rx buffer size based on MTU */ + c2_set_rxbufsize(c2_port); + + /* Allocate DMA'able memory for Tx/Rx host descriptor rings */ + rx_size = c2_port->rx_ring.count * sizeof(struct c2_rx_desc); + tx_size = c2_port->tx_ring.count * sizeof(struct c2_tx_desc); + + c2_port->mem_size = tx_size + rx_size; + c2_port->mem = pci_zalloc_consistent(c2dev->pcidev, c2_port->mem_size, + &c2_port->dma); + if (c2_port->mem == NULL) { + pr_debug("Unable to allocate memory for " + "host descriptor rings\n"); + return -ENOMEM; + } + + /* Create the Rx host descriptor ring */ + if ((ret = + c2_rx_ring_alloc(&c2_port->rx_ring, c2_port->mem, c2_port->dma, + c2dev->mmio_rxp_ring))) { + pr_debug("Unable to create RX ring\n"); + goto bail0; + } + + /* Allocate Rx buffers for the host descriptor ring */ + if (c2_rx_fill(c2_port)) { + pr_debug("Unable to fill RX ring\n"); + goto bail1; + } + + /* Create the Tx host descriptor ring */ + if ((ret = c2_tx_ring_alloc(&c2_port->tx_ring, c2_port->mem + rx_size, + c2_port->dma + rx_size, + c2dev->mmio_txp_ring))) { + pr_debug("Unable to create TX ring\n"); + goto bail1; + } + + /* Set the TX pointer to where we left off */ + c2_port->tx_avail = c2_port->tx_ring.count - 1; + c2_port->tx_ring.to_use = c2_port->tx_ring.to_clean = + c2_port->tx_ring.start + c2dev->cur_tx; + + /* missing: Initialize MAC */ + + BUG_ON(c2_port->tx_ring.to_use != c2_port->tx_ring.to_clean); + + /* Reset the adapter, ensures the driver is in sync with the RXP */ + c2_reset(c2_port); + + /* Reset the READY bit in the sk_buff RXP headers & adapter HRXDQ */ + for (i = 0, elem = c2_port->rx_ring.start; i < c2_port->rx_ring.count; + i++, elem++) { + rxp_hdr = (struct c2_rxp_hdr *) elem->skb->data; + rxp_hdr->flags = 0; + __raw_writew((__force u16) cpu_to_be16(RXP_HRXD_READY), + elem->hw_desc + C2_RXP_FLAGS); + } + + /* Enable network packets */ + netif_start_queue(netdev); + + /* Enable IRQ */ + writel(0, c2dev->regs + C2_IDIS); + netimr0 = readl(c2dev->regs + C2_NIMR0); + netimr0 &= ~(C2_PCI_HTX_INT | C2_PCI_HRX_INT); + writel(netimr0, c2dev->regs + C2_NIMR0); + + /* Tell the stack to ignore arp requests for ipaddrs bound to + * other interfaces. This is needed to prevent the host stack + * from responding to arp requests to the ipaddr bound on the + * rdma interface. + */ + in_dev = in_dev_get(netdev); + IN_DEV_CONF_SET(in_dev, ARP_IGNORE, 1); + in_dev_put(in_dev); + + return 0; + + bail1: + c2_rx_clean(c2_port); + kfree(c2_port->rx_ring.start); + + bail0: + pci_free_consistent(c2dev->pcidev, c2_port->mem_size, c2_port->mem, + c2_port->dma); + + return ret; +} + +static int c2_down(struct net_device *netdev) +{ + struct c2_port *c2_port = netdev_priv(netdev); + struct c2_dev *c2dev = c2_port->c2dev; + + if (netif_msg_ifdown(c2_port)) + pr_debug("%s: disabling interface\n", + netdev->name); + + /* Wait for all the queued packets to get sent */ + c2_tx_interrupt(netdev); + + /* Disable network packets */ + netif_stop_queue(netdev); + + /* Disable IRQs by clearing the interrupt mask */ + writel(1, c2dev->regs + C2_IDIS); + writel(0, c2dev->regs + C2_NIMR0); + + /* missing: Stop transmitter */ + + /* missing: Stop receiver */ + + /* Reset the adapter, ensures the driver is in sync with the RXP */ + c2_reset(c2_port); + + /* missing: Turn off LEDs here */ + + /* Free all buffers in the host descriptor rings */ + c2_tx_clean(c2_port); + c2_rx_clean(c2_port); + + /* Free the host descriptor rings */ + kfree(c2_port->rx_ring.start); + kfree(c2_port->tx_ring.start); + pci_free_consistent(c2dev->pcidev, c2_port->mem_size, c2_port->mem, + c2_port->dma); + + return 0; +} + +static void c2_reset(struct c2_port *c2_port) +{ + struct c2_dev *c2dev = c2_port->c2dev; + unsigned int cur_rx = c2dev->cur_rx; + + /* Tell the hardware to quiesce */ + C2_SET_CUR_RX(c2dev, cur_rx | C2_PCI_HRX_QUI); + + /* + * The hardware will reset the C2_PCI_HRX_QUI bit once + * the RXP is quiesced. Wait 2 seconds for this. + */ + ssleep(2); + + cur_rx = C2_GET_CUR_RX(c2dev); + + if (cur_rx & C2_PCI_HRX_QUI) + pr_debug("c2_reset: failed to quiesce the hardware!\n"); + + cur_rx &= ~C2_PCI_HRX_QUI; + + c2dev->cur_rx = cur_rx; + + pr_debug("Current RX: %u\n", c2dev->cur_rx); +} + +static int c2_xmit_frame(struct sk_buff *skb, struct net_device *netdev) +{ + struct c2_port *c2_port = netdev_priv(netdev); + struct c2_dev *c2dev = c2_port->c2dev; + struct c2_ring *tx_ring = &c2_port->tx_ring; + struct c2_element *elem; + dma_addr_t mapaddr; + u32 maplen; + unsigned long flags; + unsigned int i; + + spin_lock_irqsave(&c2_port->tx_lock, flags); + + if (unlikely(c2_port->tx_avail < (skb_shinfo(skb)->nr_frags + 1))) { + netif_stop_queue(netdev); + spin_unlock_irqrestore(&c2_port->tx_lock, flags); + + pr_debug("%s: Tx ring full when queue awake!\n", + netdev->name); + return NETDEV_TX_BUSY; + } + + maplen = skb_headlen(skb); + mapaddr = + pci_map_single(c2dev->pcidev, skb->data, maplen, PCI_DMA_TODEVICE); + + elem = tx_ring->to_use; + elem->skb = skb; + elem->mapaddr = mapaddr; + elem->maplen = maplen; + + /* Tell HW to xmit */ + __raw_writeq((__force u64) cpu_to_be64(mapaddr), + elem->hw_desc + C2_TXP_ADDR); + __raw_writew((__force u16) cpu_to_be16(maplen), + elem->hw_desc + C2_TXP_LEN); + __raw_writew((__force u16) cpu_to_be16(TXP_HTXD_READY), + elem->hw_desc + C2_TXP_FLAGS); + + netdev->stats.tx_packets++; + netdev->stats.tx_bytes += maplen; + + /* Loop thru additional data fragments and queue them */ + if (skb_shinfo(skb)->nr_frags) { + for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { + const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; + maplen = skb_frag_size(frag); + mapaddr = skb_frag_dma_map(&c2dev->pcidev->dev, frag, + 0, maplen, DMA_TO_DEVICE); + elem = elem->next; + elem->skb = NULL; + elem->mapaddr = mapaddr; + elem->maplen = maplen; + + /* Tell HW to xmit */ + __raw_writeq((__force u64) cpu_to_be64(mapaddr), + elem->hw_desc + C2_TXP_ADDR); + __raw_writew((__force u16) cpu_to_be16(maplen), + elem->hw_desc + C2_TXP_LEN); + __raw_writew((__force u16) cpu_to_be16(TXP_HTXD_READY), + elem->hw_desc + C2_TXP_FLAGS); + + netdev->stats.tx_packets++; + netdev->stats.tx_bytes += maplen; + } + } + + tx_ring->to_use = elem->next; + c2_port->tx_avail -= (skb_shinfo(skb)->nr_frags + 1); + + if (c2_port->tx_avail <= MAX_SKB_FRAGS + 1) { + netif_stop_queue(netdev); + if (netif_msg_tx_queued(c2_port)) + pr_debug("%s: transmit queue full\n", + netdev->name); + } + + spin_unlock_irqrestore(&c2_port->tx_lock, flags); + + netdev->trans_start = jiffies; + + return NETDEV_TX_OK; +} + +static void c2_tx_timeout(struct net_device *netdev) +{ + struct c2_port *c2_port = netdev_priv(netdev); + + if (netif_msg_timer(c2_port)) + pr_debug("%s: tx timeout\n", netdev->name); + + c2_tx_clean(c2_port); +} + +static int c2_change_mtu(struct net_device *netdev, int new_mtu) +{ + int ret = 0; + + if (new_mtu < ETH_ZLEN || new_mtu > ETH_JUMBO_MTU) + return -EINVAL; + + netdev->mtu = new_mtu; + + if (netif_running(netdev)) { + c2_down(netdev); + + c2_up(netdev); + } + + return ret; +} + +static const struct net_device_ops c2_netdev = { + .ndo_open = c2_up, + .ndo_stop = c2_down, + .ndo_start_xmit = c2_xmit_frame, + .ndo_tx_timeout = c2_tx_timeout, + .ndo_change_mtu = c2_change_mtu, + .ndo_set_mac_address = eth_mac_addr, + .ndo_validate_addr = eth_validate_addr, +}; + +/* Initialize network device */ +static struct net_device *c2_devinit(struct c2_dev *c2dev, + void __iomem * mmio_addr) +{ + struct c2_port *c2_port = NULL; + struct net_device *netdev = alloc_etherdev(sizeof(*c2_port)); + + if (!netdev) { + pr_debug("c2_port etherdev alloc failed"); + return NULL; + } + + SET_NETDEV_DEV(netdev, &c2dev->pcidev->dev); + + netdev->netdev_ops = &c2_netdev; + netdev->watchdog_timeo = C2_TX_TIMEOUT; + netdev->irq = c2dev->pcidev->irq; + + c2_port = netdev_priv(netdev); + c2_port->netdev = netdev; + c2_port->c2dev = c2dev; + c2_port->msg_enable = netif_msg_init(debug, default_msg); + c2_port->tx_ring.count = C2_NUM_TX_DESC; + c2_port->rx_ring.count = C2_NUM_RX_DESC; + + spin_lock_init(&c2_port->tx_lock); + + /* Copy our 48-bit ethernet hardware address */ + memcpy_fromio(netdev->dev_addr, mmio_addr + C2_REGS_ENADDR, 6); + + /* Validate the MAC address */ + if (!is_valid_ether_addr(netdev->dev_addr)) { + pr_debug("Invalid MAC Address\n"); + c2_print_macaddr(netdev); + free_netdev(netdev); + return NULL; + } + + c2dev->netdev = netdev; + + return netdev; +} + +static int c2_probe(struct pci_dev *pcidev, const struct pci_device_id *ent) +{ + int ret = 0, i; + unsigned long reg0_start, reg0_flags, reg0_len; + unsigned long reg2_start, reg2_flags, reg2_len; + unsigned long reg4_start, reg4_flags, reg4_len; + unsigned kva_map_size; + struct net_device *netdev = NULL; + struct c2_dev *c2dev = NULL; + void __iomem *mmio_regs = NULL; + + printk(KERN_INFO PFX "AMSO1100 Gigabit Ethernet driver v%s loaded\n", + DRV_VERSION); + + /* Enable PCI device */ + ret = pci_enable_device(pcidev); + if (ret) { + printk(KERN_ERR PFX "%s: Unable to enable PCI device\n", + pci_name(pcidev)); + goto bail0; + } + + reg0_start = pci_resource_start(pcidev, BAR_0); + reg0_len = pci_resource_len(pcidev, BAR_0); + reg0_flags = pci_resource_flags(pcidev, BAR_0); + + reg2_start = pci_resource_start(pcidev, BAR_2); + reg2_len = pci_resource_len(pcidev, BAR_2); + reg2_flags = pci_resource_flags(pcidev, BAR_2); + + reg4_start = pci_resource_start(pcidev, BAR_4); + reg4_len = pci_resource_len(pcidev, BAR_4); + reg4_flags = pci_resource_flags(pcidev, BAR_4); + + pr_debug("BAR0 size = 0x%lX bytes\n", reg0_len); + pr_debug("BAR2 size = 0x%lX bytes\n", reg2_len); + pr_debug("BAR4 size = 0x%lX bytes\n", reg4_len); + + /* Make sure PCI base addr are MMIO */ + if (!(reg0_flags & IORESOURCE_MEM) || + !(reg2_flags & IORESOURCE_MEM) || !(reg4_flags & IORESOURCE_MEM)) { + printk(KERN_ERR PFX "PCI regions not an MMIO resource\n"); + ret = -ENODEV; + goto bail1; + } + + /* Check for weird/broken PCI region reporting */ + if ((reg0_len < C2_REG0_SIZE) || + (reg2_len < C2_REG2_SIZE) || (reg4_len < C2_REG4_SIZE)) { + printk(KERN_ERR PFX "Invalid PCI region sizes\n"); + ret = -ENODEV; + goto bail1; + } + + /* Reserve PCI I/O and memory resources */ + ret = pci_request_regions(pcidev, DRV_NAME); + if (ret) { + printk(KERN_ERR PFX "%s: Unable to request regions\n", + pci_name(pcidev)); + goto bail1; + } + + if ((sizeof(dma_addr_t) > 4)) { + ret = pci_set_dma_mask(pcidev, DMA_BIT_MASK(64)); + if (ret < 0) { + printk(KERN_ERR PFX "64b DMA configuration failed\n"); + goto bail2; + } + } else { + ret = pci_set_dma_mask(pcidev, DMA_BIT_MASK(32)); + if (ret < 0) { + printk(KERN_ERR PFX "32b DMA configuration failed\n"); + goto bail2; + } + } + + /* Enables bus-mastering on the device */ + pci_set_master(pcidev); + + /* Remap the adapter PCI registers in BAR4 */ + mmio_regs = ioremap_nocache(reg4_start + C2_PCI_REGS_OFFSET, + sizeof(struct c2_adapter_pci_regs)); + if (!mmio_regs) { + printk(KERN_ERR PFX + "Unable to remap adapter PCI registers in BAR4\n"); + ret = -EIO; + goto bail2; + } + + /* Validate PCI regs magic */ + for (i = 0; i < sizeof(c2_magic); i++) { + if (c2_magic[i] != readb(mmio_regs + C2_REGS_MAGIC + i)) { + printk(KERN_ERR PFX "Downlevel Firmware boot loader " + "[%d/%Zd: got 0x%x, exp 0x%x]. Use the cc_flash " + "utility to update your boot loader\n", + i + 1, sizeof(c2_magic), + readb(mmio_regs + C2_REGS_MAGIC + i), + c2_magic[i]); + printk(KERN_ERR PFX "Adapter not claimed\n"); + iounmap(mmio_regs); + ret = -EIO; + goto bail2; + } + } + + /* Validate the adapter version */ + if (be32_to_cpu((__force __be32) readl(mmio_regs + C2_REGS_VERS)) != C2_VERSION) { + printk(KERN_ERR PFX "Version mismatch " + "[fw=%u, c2=%u], Adapter not claimed\n", + be32_to_cpu((__force __be32) readl(mmio_regs + C2_REGS_VERS)), + C2_VERSION); + ret = -EINVAL; + iounmap(mmio_regs); + goto bail2; + } + + /* Validate the adapter IVN */ + if (be32_to_cpu((__force __be32) readl(mmio_regs + C2_REGS_IVN)) != C2_IVN) { + printk(KERN_ERR PFX "Downlevel FIrmware level. You should be using " + "the OpenIB device support kit. " + "[fw=0x%x, c2=0x%x], Adapter not claimed\n", + be32_to_cpu((__force __be32) readl(mmio_regs + C2_REGS_IVN)), + C2_IVN); + ret = -EINVAL; + iounmap(mmio_regs); + goto bail2; + } + + /* Allocate hardware structure */ + c2dev = (struct c2_dev *) ib_alloc_device(sizeof(*c2dev)); + if (!c2dev) { + printk(KERN_ERR PFX "%s: Unable to alloc hardware struct\n", + pci_name(pcidev)); + ret = -ENOMEM; + iounmap(mmio_regs); + goto bail2; + } + + memset(c2dev, 0, sizeof(*c2dev)); + spin_lock_init(&c2dev->lock); + c2dev->pcidev = pcidev; + c2dev->cur_tx = 0; + + /* Get the last RX index */ + c2dev->cur_rx = + (be32_to_cpu((__force __be32) readl(mmio_regs + C2_REGS_HRX_CUR)) - + 0xffffc000) / sizeof(struct c2_rxp_desc); + + /* Request an interrupt line for the driver */ + ret = request_irq(pcidev->irq, c2_interrupt, IRQF_SHARED, DRV_NAME, c2dev); + if (ret) { + printk(KERN_ERR PFX "%s: requested IRQ %u is busy\n", + pci_name(pcidev), pcidev->irq); + iounmap(mmio_regs); + goto bail3; + } + + /* Set driver specific data */ + pci_set_drvdata(pcidev, c2dev); + + /* Initialize network device */ + if ((netdev = c2_devinit(c2dev, mmio_regs)) == NULL) { + ret = -ENOMEM; + iounmap(mmio_regs); + goto bail4; + } + + /* Save off the actual size prior to unmapping mmio_regs */ + kva_map_size = be32_to_cpu((__force __be32) readl(mmio_regs + C2_REGS_PCI_WINSIZE)); + + /* Unmap the adapter PCI registers in BAR4 */ + iounmap(mmio_regs); + + /* Register network device */ + ret = register_netdev(netdev); + if (ret) { + printk(KERN_ERR PFX "Unable to register netdev, ret = %d\n", + ret); + goto bail5; + } + + /* Disable network packets */ + netif_stop_queue(netdev); + + /* Remap the adapter HRXDQ PA space to kernel VA space */ + c2dev->mmio_rxp_ring = ioremap_nocache(reg4_start + C2_RXP_HRXDQ_OFFSET, + C2_RXP_HRXDQ_SIZE); + if (!c2dev->mmio_rxp_ring) { + printk(KERN_ERR PFX "Unable to remap MMIO HRXDQ region\n"); + ret = -EIO; + goto bail6; + } + + /* Remap the adapter HTXDQ PA space to kernel VA space */ + c2dev->mmio_txp_ring = ioremap_nocache(reg4_start + C2_TXP_HTXDQ_OFFSET, + C2_TXP_HTXDQ_SIZE); + if (!c2dev->mmio_txp_ring) { + printk(KERN_ERR PFX "Unable to remap MMIO HTXDQ region\n"); + ret = -EIO; + goto bail7; + } + + /* Save off the current RX index in the last 4 bytes of the TXP Ring */ + C2_SET_CUR_RX(c2dev, c2dev->cur_rx); + + /* Remap the PCI registers in adapter BAR0 to kernel VA space */ + c2dev->regs = ioremap_nocache(reg0_start, reg0_len); + if (!c2dev->regs) { + printk(KERN_ERR PFX "Unable to remap BAR0\n"); + ret = -EIO; + goto bail8; + } + + /* Remap the PCI registers in adapter BAR4 to kernel VA space */ + c2dev->pa = reg4_start + C2_PCI_REGS_OFFSET; + c2dev->kva = ioremap_nocache(reg4_start + C2_PCI_REGS_OFFSET, + kva_map_size); + if (!c2dev->kva) { + printk(KERN_ERR PFX "Unable to remap BAR4\n"); + ret = -EIO; + goto bail9; + } + + /* Print out the MAC address */ + c2_print_macaddr(netdev); + + ret = c2_rnic_init(c2dev); + if (ret) { + printk(KERN_ERR PFX "c2_rnic_init failed: %d\n", ret); + goto bail10; + } + + ret = c2_register_device(c2dev); + if (ret) + goto bail10; + + return 0; + + bail10: + iounmap(c2dev->kva); + + bail9: + iounmap(c2dev->regs); + + bail8: + iounmap(c2dev->mmio_txp_ring); + + bail7: + iounmap(c2dev->mmio_rxp_ring); + + bail6: + unregister_netdev(netdev); + + bail5: + free_netdev(netdev); + + bail4: + free_irq(pcidev->irq, c2dev); + + bail3: + ib_dealloc_device(&c2dev->ibdev); + + bail2: + pci_release_regions(pcidev); + + bail1: + pci_disable_device(pcidev); + + bail0: + return ret; +} + +static void c2_remove(struct pci_dev *pcidev) +{ + struct c2_dev *c2dev = pci_get_drvdata(pcidev); + struct net_device *netdev = c2dev->netdev; + + /* Unregister with OpenIB */ + c2_unregister_device(c2dev); + + /* Clean up the RNIC resources */ + c2_rnic_term(c2dev); + + /* Remove network device from the kernel */ + unregister_netdev(netdev); + + /* Free network device */ + free_netdev(netdev); + + /* Free the interrupt line */ + free_irq(pcidev->irq, c2dev); + + /* missing: Turn LEDs off here */ + + /* Unmap adapter PA space */ + iounmap(c2dev->kva); + iounmap(c2dev->regs); + iounmap(c2dev->mmio_txp_ring); + iounmap(c2dev->mmio_rxp_ring); + + /* Free the hardware structure */ + ib_dealloc_device(&c2dev->ibdev); + + /* Release reserved PCI I/O and memory resources */ + pci_release_regions(pcidev); + + /* Disable PCI device */ + pci_disable_device(pcidev); + + /* Clear driver specific data */ + pci_set_drvdata(pcidev, NULL); +} + +static struct pci_driver c2_pci_driver = { + .name = DRV_NAME, + .id_table = c2_pci_table, + .probe = c2_probe, + .remove = c2_remove, +}; + +module_pci_driver(c2_pci_driver); diff --git a/drivers/infiniband/hw/amso1100/c2.h b/drivers/infiniband/hw/amso1100/c2.h new file mode 100644 index 000000000..d619d7358 --- /dev/null +++ b/drivers/infiniband/hw/amso1100/c2.h @@ -0,0 +1,547 @@ +/* + * Copyright (c) 2005 Ammasso, Inc. All rights reserved. + * Copyright (c) 2005 Open Grid Computing, Inc. All rights reserved. + * + * This software is available to you under a choice of one of two + * licenses. You may choose to be licensed under the terms of the GNU + * General Public License (GPL) Version 2, available from the file + * COPYING in the main directory of this source tree, or the + * OpenIB.org BSD license below: + * + * Redistribution and use in source and binary forms, with or + * without modification, are permitted provided that the following + * conditions are met: + * + * - Redistributions of source code must retain the above + * copyright notice, this list of conditions and the following + * disclaimer. + * + * - Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and/or other materials + * provided with the distribution. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS + * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN + * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ + +#ifndef __C2_H +#define __C2_H + +#include <linux/netdevice.h> +#include <linux/spinlock.h> +#include <linux/kernel.h> +#include <linux/pci.h> +#include <linux/dma-mapping.h> +#include <linux/idr.h> + +#include "c2_provider.h" +#include "c2_mq.h" +#include "c2_status.h" + +#define DRV_NAME "c2" +#define DRV_VERSION "1.1" +#define PFX DRV_NAME ": " + +#define BAR_0 0 +#define BAR_2 2 +#define BAR_4 4 + +#define RX_BUF_SIZE (1536 + 8) +#define ETH_JUMBO_MTU 9000 +#define C2_MAGIC "CEPHEUS" +#define C2_VERSION 4 +#define C2_IVN (18 & 0x7fffffff) + +#define C2_REG0_SIZE (16 * 1024) +#define C2_REG2_SIZE (2 * 1024 * 1024) +#define C2_REG4_SIZE (256 * 1024 * 1024) +#define C2_NUM_TX_DESC 341 +#define C2_NUM_RX_DESC 256 +#define C2_PCI_REGS_OFFSET (0x10000) +#define C2_RXP_HRXDQ_OFFSET (((C2_REG4_SIZE)/2)) +#define C2_RXP_HRXDQ_SIZE (4096) +#define C2_TXP_HTXDQ_OFFSET (((C2_REG4_SIZE)/2) + C2_RXP_HRXDQ_SIZE) +#define C2_TXP_HTXDQ_SIZE (4096) +#define C2_TX_TIMEOUT (6*HZ) + +/* CEPHEUS */ +static const u8 c2_magic[] = { + 0x43, 0x45, 0x50, 0x48, 0x45, 0x55, 0x53 +}; + +enum adapter_pci_regs { + C2_REGS_MAGIC = 0x0000, + C2_REGS_VERS = 0x0008, + C2_REGS_IVN = 0x000C, + C2_REGS_PCI_WINSIZE = 0x0010, + C2_REGS_Q0_QSIZE = 0x0014, + C2_REGS_Q0_MSGSIZE = 0x0018, + C2_REGS_Q0_POOLSTART = 0x001C, + C2_REGS_Q0_SHARED = 0x0020, + C2_REGS_Q1_QSIZE = 0x0024, + C2_REGS_Q1_MSGSIZE = 0x0028, + C2_REGS_Q1_SHARED = 0x0030, + C2_REGS_Q2_QSIZE = 0x0034, + C2_REGS_Q2_MSGSIZE = 0x0038, + C2_REGS_Q2_SHARED = 0x0040, + C2_REGS_ENADDR = 0x004C, + C2_REGS_RDMA_ENADDR = 0x0054, + C2_REGS_HRX_CUR = 0x006C, +}; + +struct c2_adapter_pci_regs { + char reg_magic[8]; + u32 version; + u32 ivn; + u32 pci_window_size; + u32 q0_q_size; + u32 q0_msg_size; + u32 q0_pool_start; + u32 q0_shared; + u32 q1_q_size; + u32 q1_msg_size; + u32 q1_pool_start; + u32 q1_shared; + u32 q2_q_size; + u32 q2_msg_size; + u32 q2_pool_start; + u32 q2_shared; + u32 log_start; + u32 log_size; + u8 host_enaddr[8]; + u8 rdma_enaddr[8]; + u32 crash_entry; + u32 crash_ready[2]; + u32 fw_txd_cur; + u32 fw_hrxd_cur; + u32 fw_rxd_cur; +}; + +enum pci_regs { + C2_HISR = 0x0000, + C2_DISR = 0x0004, + C2_HIMR = 0x0008, + C2_DIMR = 0x000C, + C2_NISR0 = 0x0010, + C2_NISR1 = 0x0014, + C2_NIMR0 = 0x0018, + C2_NIMR1 = 0x001C, + C2_IDIS = 0x0020, +}; + +enum { + C2_PCI_HRX_INT = 1 << 8, + C2_PCI_HTX_INT = 1 << 17, + C2_PCI_HRX_QUI = 1 << 31, +}; + +/* + * Cepheus registers in BAR0. + */ +struct c2_pci_regs { + u32 hostisr; + u32 dmaisr; + u32 hostimr; + u32 dmaimr; + u32 netisr0; + u32 netisr1; + u32 netimr0; + u32 netimr1; + u32 int_disable; +}; + +/* TXP flags */ +enum c2_txp_flags { + TXP_HTXD_DONE = 0, + TXP_HTXD_READY = 1 << 0, + TXP_HTXD_UNINIT = 1 << 1, +}; + +/* RXP flags */ +enum c2_rxp_flags { + RXP_HRXD_UNINIT = 0, + RXP_HRXD_READY = 1 << 0, + RXP_HRXD_DONE = 1 << 1, +}; + +/* RXP status */ +enum c2_rxp_status { + RXP_HRXD_ZERO = 0, + RXP_HRXD_OK = 1 << 0, + RXP_HRXD_BUF_OV = 1 << 1, +}; + +/* TXP descriptor fields */ +enum txp_desc { + C2_TXP_FLAGS = 0x0000, + C2_TXP_LEN = 0x0002, + C2_TXP_ADDR = 0x0004, +}; + +/* RXP descriptor fields */ +enum rxp_desc { + C2_RXP_FLAGS = 0x0000, + C2_RXP_STATUS = 0x0002, + C2_RXP_COUNT = 0x0004, + C2_RXP_LEN = 0x0006, + C2_RXP_ADDR = 0x0008, +}; + +struct c2_txp_desc { + u16 flags; + u16 len; + u64 addr; +} __attribute__ ((packed)); + +struct c2_rxp_desc { + u16 flags; + u16 status; + u16 count; + u16 len; + u64 addr; +} __attribute__ ((packed)); + +struct c2_rxp_hdr { + u16 flags; + u16 status; + u16 len; + u16 rsvd; +} __attribute__ ((packed)); + +struct c2_tx_desc { + u32 len; + u32 status; + dma_addr_t next_offset; +}; + +struct c2_rx_desc { + u32 len; + u32 status; + dma_addr_t next_offset; +}; + +struct c2_alloc { + u32 last; + u32 max; + spinlock_t lock; + unsigned long *table; +}; + +struct c2_array { + struct { + void **page; + int used; + } *page_list; +}; + +/* + * The MQ shared pointer pool is organized as a linked list of + * chunks. Each chunk contains a linked list of free shared pointers + * that can be allocated to a given user mode client. + * + */ +struct sp_chunk { + struct sp_chunk *next; + dma_addr_t dma_addr; + DEFINE_DMA_UNMAP_ADDR(mapping); + u16 head; + u16 shared_ptr[0]; +}; + +struct c2_pd_table { + u32 last; + u32 max; + spinlock_t lock; + unsigned long *table; +}; + +struct c2_qp_table { + struct idr idr; + spinlock_t lock; +}; + +struct c2_element { + struct c2_element *next; + void *ht_desc; /* host descriptor */ + void __iomem *hw_desc; /* hardware descriptor */ + struct sk_buff *skb; + dma_addr_t mapaddr; + u32 maplen; +}; + +struct c2_ring { + struct c2_element *to_clean; + struct c2_element *to_use; + struct c2_element *start; + unsigned long count; +}; + +struct c2_dev { + struct ib_device ibdev; + void __iomem *regs; + void __iomem *mmio_txp_ring; /* remapped adapter memory for hw rings */ + void __iomem *mmio_rxp_ring; + spinlock_t lock; + struct pci_dev *pcidev; + struct net_device *netdev; + struct net_device *pseudo_netdev; + unsigned int cur_tx; + unsigned int cur_rx; + u32 adapter_handle; + int device_cap_flags; + void __iomem *kva; /* KVA device memory */ + unsigned long pa; /* PA device memory */ + void **qptr_array; + + struct kmem_cache *host_msg_cache; + + struct list_head cca_link; /* adapter list */ + struct list_head eh_wakeup_list; /* event wakeup list */ + wait_queue_head_t req_vq_wo; + + /* Cached RNIC properties */ + struct ib_device_attr props; + + struct c2_pd_table pd_table; + struct c2_qp_table qp_table; + int ports; /* num of GigE ports */ + int devnum; + spinlock_t vqlock; /* sync vbs req MQ */ + + /* Verbs Queues */ + struct c2_mq req_vq; /* Verbs Request MQ */ + struct c2_mq rep_vq; /* Verbs Reply MQ */ + struct c2_mq aeq; /* Async Events MQ */ + + /* Kernel client MQs */ + struct sp_chunk *kern_mqsp_pool; + + /* Device updates these values when posting messages to a host + * target queue */ + u16 req_vq_shared; + u16 rep_vq_shared; + u16 aeq_shared; + u16 irq_claimed; + + /* + * Shared host target pages for user-accessible MQs. + */ + int hthead; /* index of first free entry */ + void *htpages; /* kernel vaddr */ + int htlen; /* length of htpages memory */ + void *htuva; /* user mapped vaddr */ + spinlock_t htlock; /* serialize allocation */ + + u64 adapter_hint_uva; /* access to the activity FIFO */ + + // spinlock_t aeq_lock; + // spinlock_t rnic_lock; + + __be16 *hint_count; + dma_addr_t hint_count_dma; + u16 hints_read; + + int init; /* TRUE if it's ready */ + char ae_cache_name[16]; + char vq_cache_name[16]; +}; + +struct c2_port { + u32 msg_enable; + struct c2_dev *c2dev; + struct net_device *netdev; + + spinlock_t tx_lock; + u32 tx_avail; + struct c2_ring tx_ring; + struct c2_ring rx_ring; + + void *mem; /* PCI memory for host rings */ + dma_addr_t dma; + unsigned long mem_size; + + u32 rx_buf_size; +}; + +/* + * Activity FIFO registers in BAR0. + */ +#define PCI_BAR0_HOST_HINT 0x100 +#define PCI_BAR0_ADAPTER_HINT 0x2000 + +/* + * Ammasso PCI vendor id and Cepheus PCI device id. + */ +#define CQ_ARMED 0x01 +#define CQ_WAIT_FOR_DMA 0x80 + +/* + * The format of a hint is as follows: + * Lower 16 bits are the count of hints for the queue. + * Next 15 bits are the qp_index + * Upper most bit depends on who reads it: + * If read by producer, then it means Full (1) or Not-Full (0) + * If read by consumer, then it means Empty (1) or Not-Empty (0) + */ +#define C2_HINT_MAKE(q_index, hint_count) (((q_index) << 16) | hint_count) +#define C2_HINT_GET_INDEX(hint) (((hint) & 0x7FFF0000) >> 16) +#define C2_HINT_GET_COUNT(hint) ((hint) & 0x0000FFFF) + + +/* + * The following defines the offset in SDRAM for the c2_adapter_pci_regs_t + * struct. + */ +#define C2_ADAPTER_PCI_REGS_OFFSET 0x10000 + +#ifndef readq +static inline u64 readq(const void __iomem * addr) +{ + u64 ret = readl(addr + 4); + ret <<= 32; + ret |= readl(addr); + + return ret; +} +#endif + +#ifndef writeq +static inline void __raw_writeq(u64 val, void __iomem * addr) +{ + __raw_writel((u32) (val), addr); + __raw_writel((u32) (val >> 32), (addr + 4)); +} +#endif + +#define C2_SET_CUR_RX(c2dev, cur_rx) \ + __raw_writel((__force u32) cpu_to_be32(cur_rx), c2dev->mmio_txp_ring + 4092) + +#define C2_GET_CUR_RX(c2dev) \ + be32_to_cpu((__force __be32) readl(c2dev->mmio_txp_ring + 4092)) + +static inline struct c2_dev *to_c2dev(struct ib_device *ibdev) +{ + return container_of(ibdev, struct c2_dev, ibdev); +} + +static inline int c2_errno(void *reply) +{ + switch (c2_wr_get_result(reply)) { + case C2_OK: + return 0; + case CCERR_NO_BUFS: + case CCERR_INSUFFICIENT_RESOURCES: + case CCERR_ZERO_RDMA_READ_RESOURCES: + return -ENOMEM; + case CCERR_MR_IN_USE: + case CCERR_QP_IN_USE: + return -EBUSY; + case CCERR_ADDR_IN_USE: + return -EADDRINUSE; + case CCERR_ADDR_NOT_AVAIL: + return -EADDRNOTAVAIL; + case CCERR_CONN_RESET: + return -ECONNRESET; + case CCERR_NOT_IMPLEMENTED: + case CCERR_INVALID_WQE: + return -ENOSYS; + case CCERR_QP_NOT_PRIVILEGED: + return -EPERM; + case CCERR_STACK_ERROR: + return -EPROTO; + case CCERR_ACCESS_VIOLATION: + case CCERR_BASE_AND_BOUNDS_VIOLATION: + return -EFAULT; + case CCERR_STAG_STATE_NOT_INVALID: + case CCERR_INVALID_ADDRESS: + case CCERR_INVALID_CQ: + case CCERR_INVALID_EP: + case CCERR_INVALID_MODIFIER: + case CCERR_INVALID_MTU: + case CCERR_INVALID_PD_ID: + case CCERR_INVALID_QP: + case CCERR_INVALID_RNIC: + case CCERR_INVALID_STAG: + return -EINVAL; + default: + return -EAGAIN; + } +} + +/* Device */ +extern int c2_register_device(struct c2_dev *c2dev); +extern void c2_unregister_device(struct c2_dev *c2dev); +extern int c2_rnic_init(struct c2_dev *c2dev); +extern void c2_rnic_term(struct c2_dev *c2dev); +extern void c2_rnic_interrupt(struct c2_dev *c2dev); +extern int c2_del_addr(struct c2_dev *c2dev, __be32 inaddr, __be32 inmask); +extern int c2_add_addr(struct c2_dev *c2dev, __be32 inaddr, __be32 inmask); + +/* QPs */ +extern int c2_alloc_qp(struct c2_dev *c2dev, struct c2_pd *pd, + struct ib_qp_init_attr *qp_attrs, struct c2_qp *qp); +extern void c2_free_qp(struct c2_dev *c2dev, struct c2_qp *qp); +extern struct ib_qp *c2_get_qp(struct ib_device *device, int qpn); +extern int c2_qp_modify(struct c2_dev *c2dev, struct c2_qp *qp, + struct ib_qp_attr *attr, int attr_mask); +extern int c2_qp_set_read_limits(struct c2_dev *c2dev, struct c2_qp *qp, + int ord, int ird); +extern int c2_post_send(struct ib_qp *ibqp, struct ib_send_wr *ib_wr, + struct ib_send_wr **bad_wr); +extern int c2_post_receive(struct ib_qp *ibqp, struct ib_recv_wr *ib_wr, + struct ib_recv_wr **bad_wr); +extern void c2_init_qp_table(struct c2_dev *c2dev); +extern void c2_cleanup_qp_table(struct c2_dev *c2dev); +extern void c2_set_qp_state(struct c2_qp *, int); +extern struct c2_qp *c2_find_qpn(struct c2_dev *c2dev, int qpn); + +/* PDs */ +extern int c2_pd_alloc(struct c2_dev *c2dev, int privileged, struct c2_pd *pd); +extern void c2_pd_free(struct c2_dev *c2dev, struct c2_pd *pd); +extern int c2_init_pd_table(struct c2_dev *c2dev); +extern void c2_cleanup_pd_table(struct c2_dev *c2dev); + +/* CQs */ +extern int c2_init_cq(struct c2_dev *c2dev, int entries, + struct c2_ucontext *ctx, struct c2_cq *cq); +extern void c2_free_cq(struct c2_dev *c2dev, struct c2_cq *cq); +extern void c2_cq_event(struct c2_dev *c2dev, u32 mq_index); +extern void c2_cq_clean(struct c2_dev *c2dev, struct c2_qp *qp, u32 mq_index); +extern int c2_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *entry); +extern int c2_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags); + +/* CM */ +extern int c2_llp_connect(struct iw_cm_id *cm_id, + struct iw_cm_conn_param *iw_param); +extern int c2_llp_accept(struct iw_cm_id *cm_id, + struct iw_cm_conn_param *iw_param); +extern int c2_llp_reject(struct iw_cm_id *cm_id, const void *pdata, + u8 pdata_len); +extern int c2_llp_service_create(struct iw_cm_id *cm_id, int backlog); +extern int c2_llp_service_destroy(struct iw_cm_id *cm_id); + +/* MM */ +extern int c2_nsmr_register_phys_kern(struct c2_dev *c2dev, u64 *addr_list, + int page_size, int pbl_depth, u32 length, + u32 off, u64 *va, enum c2_acf acf, + struct c2_mr *mr); +extern int c2_stag_dealloc(struct c2_dev *c2dev, u32 stag_index); + +/* AE */ +extern void c2_ae_event(struct c2_dev *c2dev, u32 mq_index); + +/* MQSP Allocator */ +extern int c2_init_mqsp_pool(struct c2_dev *c2dev, gfp_t gfp_mask, + struct sp_chunk **root); +extern void c2_free_mqsp_pool(struct c2_dev *c2dev, struct sp_chunk *root); +extern __be16 *c2_alloc_mqsp(struct c2_dev *c2dev, struct sp_chunk *head, + dma_addr_t *dma_addr, gfp_t gfp_mask); +extern void c2_free_mqsp(__be16* mqsp); +#endif diff --git a/drivers/infiniband/hw/amso1100/c2_ae.c b/drivers/infiniband/hw/amso1100/c2_ae.c new file mode 100644 index 000000000..cedda2523 --- /dev/null +++ b/drivers/infiniband/hw/amso1100/c2_ae.c @@ -0,0 +1,327 @@ +/* + * Copyright (c) 2005 Ammasso, Inc. All rights reserved. + * Copyright (c) 2005 Open Grid Computing, Inc. All rights reserved. + * + * This software is available to you under a choice of one of two + * licenses. You may choose to be licensed under the terms of the GNU + * General Public License (GPL) Version 2, available from the file + * COPYING in the main directory of this source tree, or the + * OpenIB.org BSD license below: + * + * Redistribution and use in source and binary forms, with or + * without modification, are permitted provided that the following + * conditions are met: + * + * - Redistributions of source code must retain the above + * copyright notice, this list of conditions and the following + * disclaimer. + * + * - Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and/or other materials + * provided with the distribution. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS + * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN + * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ +#include "c2.h" +#include <rdma/iw_cm.h> +#include "c2_status.h" +#include "c2_ae.h" + +static int c2_convert_cm_status(u32 c2_status) +{ + switch (c2_status) { + case C2_CONN_STATUS_SUCCESS: + return 0; + case C2_CONN_STATUS_REJECTED: + return -ENETRESET; + case C2_CONN_STATUS_REFUSED: + return -ECONNREFUSED; + case C2_CONN_STATUS_TIMEDOUT: + return -ETIMEDOUT; + case C2_CONN_STATUS_NETUNREACH: + return -ENETUNREACH; + case C2_CONN_STATUS_HOSTUNREACH: + return -EHOSTUNREACH; + case C2_CONN_STATUS_INVALID_RNIC: + return -EINVAL; + case C2_CONN_STATUS_INVALID_QP: + return -EINVAL; + case C2_CONN_STATUS_INVALID_QP_STATE: + return -EINVAL; + case C2_CONN_STATUS_ADDR_NOT_AVAIL: + return -EADDRNOTAVAIL; + default: + printk(KERN_ERR PFX + "%s - Unable to convert CM status: %d\n", + __func__, c2_status); + return -EIO; + } +} + +static const char* to_event_str(int event) +{ + static const char* event_str[] = { + "CCAE_REMOTE_SHUTDOWN", + "CCAE_ACTIVE_CONNECT_RESULTS", + "CCAE_CONNECTION_REQUEST", + "CCAE_LLP_CLOSE_COMPLETE", + "CCAE_TERMINATE_MESSAGE_RECEIVED", + "CCAE_LLP_CONNECTION_RESET", + "CCAE_LLP_CONNECTION_LOST", + "CCAE_LLP_SEGMENT_SIZE_INVALID", + "CCAE_LLP_INVALID_CRC", + "CCAE_LLP_BAD_FPDU", + "CCAE_INVALID_DDP_VERSION", + "CCAE_INVALID_RDMA_VERSION", + "CCAE_UNEXPECTED_OPCODE", + "CCAE_INVALID_DDP_QUEUE_NUMBER", + "CCAE_RDMA_READ_NOT_ENABLED", + "CCAE_RDMA_WRITE_NOT_ENABLED", + "CCAE_RDMA_READ_TOO_SMALL", + "CCAE_NO_L_BIT", + "CCAE_TAGGED_INVALID_STAG", + "CCAE_TAGGED_BASE_BOUNDS_VIOLATION", + "CCAE_TAGGED_ACCESS_RIGHTS_VIOLATION", + "CCAE_TAGGED_INVALID_PD", + "CCAE_WRAP_ERROR", + "CCAE_BAD_CLOSE", + "CCAE_BAD_LLP_CLOSE", + "CCAE_INVALID_MSN_RANGE", + "CCAE_INVALID_MSN_GAP", + "CCAE_IRRQ_OVERFLOW", + "CCAE_IRRQ_MSN_GAP", + "CCAE_IRRQ_MSN_RANGE", + "CCAE_IRRQ_INVALID_STAG", + "CCAE_IRRQ_BASE_BOUNDS_VIOLATION", + "CCAE_IRRQ_ACCESS_RIGHTS_VIOLATION", + "CCAE_IRRQ_INVALID_PD", + "CCAE_IRRQ_WRAP_ERROR", + "CCAE_CQ_SQ_COMPLETION_OVERFLOW", + "CCAE_CQ_RQ_COMPLETION_ERROR", + "CCAE_QP_SRQ_WQE_ERROR", + "CCAE_QP_LOCAL_CATASTROPHIC_ERROR", + "CCAE_CQ_OVERFLOW", + "CCAE_CQ_OPERATION_ERROR", + "CCAE_SRQ_LIMIT_REACHED", + "CCAE_QP_RQ_LIMIT_REACHED", + "CCAE_SRQ_CATASTROPHIC_ERROR", + "CCAE_RNIC_CATASTROPHIC_ERROR" + }; + + if (event < CCAE_REMOTE_SHUTDOWN || + event > CCAE_RNIC_CATASTROPHIC_ERROR) + return "<invalid event>"; + + event -= CCAE_REMOTE_SHUTDOWN; + return event_str[event]; +} + +static const char *to_qp_state_str(int state) +{ + switch (state) { + case C2_QP_STATE_IDLE: + return "C2_QP_STATE_IDLE"; + case C2_QP_STATE_CONNECTING: + return "C2_QP_STATE_CONNECTING"; + case C2_QP_STATE_RTS: + return "C2_QP_STATE_RTS"; + case C2_QP_STATE_CLOSING: + return "C2_QP_STATE_CLOSING"; + case C2_QP_STATE_TERMINATE: + return "C2_QP_STATE_TERMINATE"; + case C2_QP_STATE_ERROR: + return "C2_QP_STATE_ERROR"; + default: + return "<invalid QP state>"; + } +} + +void c2_ae_event(struct c2_dev *c2dev, u32 mq_index) +{ + struct c2_mq *mq = c2dev->qptr_array[mq_index]; + union c2wr *wr; + void *resource_user_context; + struct iw_cm_event cm_event; + struct ib_event ib_event; + enum c2_resource_indicator resource_indicator; + enum c2_event_id event_id; + unsigned long flags; + int status; + struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_event.local_addr; + struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_event.remote_addr; + + /* + * retrieve the message + */ + wr = c2_mq_consume(mq); + if (!wr) + return; + + memset(&ib_event, 0, sizeof(ib_event)); + memset(&cm_event, 0, sizeof(cm_event)); + + event_id = c2_wr_get_id(wr); + resource_indicator = be32_to_cpu(wr->ae.ae_generic.resource_type); + resource_user_context = + (void *) (unsigned long) wr->ae.ae_generic.user_context; + + status = cm_event.status = c2_convert_cm_status(c2_wr_get_result(wr)); + + pr_debug("event received c2_dev=%p, event_id=%d, " + "resource_indicator=%d, user_context=%p, status = %d\n", + c2dev, event_id, resource_indicator, resource_user_context, + status); + + switch (resource_indicator) { + case C2_RES_IND_QP:{ + + struct c2_qp *qp = (struct c2_qp *)resource_user_context; + struct iw_cm_id *cm_id = qp->cm_id; + struct c2wr_ae_active_connect_results *res; + + if (!cm_id) { + pr_debug("event received, but cm_id is <nul>, qp=%p!\n", + qp); + goto ignore_it; + } + pr_debug("%s: event = %s, user_context=%llx, " + "resource_type=%x, " + "resource=%x, qp_state=%s\n", + __func__, + to_event_str(event_id), + (unsigned long long) wr->ae.ae_generic.user_context, + be32_to_cpu(wr->ae.ae_generic.resource_type), + be32_to_cpu(wr->ae.ae_generic.resource), + to_qp_state_str(be32_to_cpu(wr->ae.ae_generic.qp_state))); + + c2_set_qp_state(qp, be32_to_cpu(wr->ae.ae_generic.qp_state)); + + switch (event_id) { + case CCAE_ACTIVE_CONNECT_RESULTS: + res = &wr->ae.ae_active_connect_results; + cm_event.event = IW_CM_EVENT_CONNECT_REPLY; + laddr->sin_addr.s_addr = res->laddr; + raddr->sin_addr.s_addr = res->raddr; + laddr->sin_port = res->lport; + raddr->sin_port = res->rport; + if (status == 0) { + cm_event.private_data_len = + be32_to_cpu(res->private_data_length); + cm_event.private_data = res->private_data; + } else { + spin_lock_irqsave(&qp->lock, flags); + if (qp->cm_id) { + qp->cm_id->rem_ref(qp->cm_id); + qp->cm_id = NULL; + } + spin_unlock_irqrestore(&qp->lock, flags); + cm_event.private_data_len = 0; + cm_event.private_data = NULL; + } + if (cm_id->event_handler) + cm_id->event_handler(cm_id, &cm_event); + break; + case CCAE_TERMINATE_MESSAGE_RECEIVED: + case CCAE_CQ_SQ_COMPLETION_OVERFLOW: + ib_event.device = &c2dev->ibdev; + ib_event.element.qp = &qp->ibqp; + ib_event.event = IB_EVENT_QP_REQ_ERR; + + if (qp->ibqp.event_handler) + qp->ibqp.event_handler(&ib_event, + qp->ibqp. + qp_context); + break; + case CCAE_BAD_CLOSE: + case CCAE_LLP_CLOSE_COMPLETE: + case CCAE_LLP_CONNECTION_RESET: + case CCAE_LLP_CONNECTION_LOST: + BUG_ON(cm_id->event_handler==(void*)0x6b6b6b6b); + + spin_lock_irqsave(&qp->lock, flags); + if (qp->cm_id) { + qp->cm_id->rem_ref(qp->cm_id); + qp->cm_id = NULL; + } + spin_unlock_irqrestore(&qp->lock, flags); + cm_event.event = IW_CM_EVENT_CLOSE; + cm_event.status = 0; + if (cm_id->event_handler) + cm_id->event_handler(cm_id, &cm_event); + break; + default: + BUG_ON(1); + pr_debug("%s:%d Unexpected event_id=%d on QP=%p, " + "CM_ID=%p\n", + __func__, __LINE__, + event_id, qp, cm_id); + break; + } + break; + } + + case C2_RES_IND_EP:{ + + struct c2wr_ae_connection_request *req = + &wr->ae.ae_connection_request; + struct iw_cm_id *cm_id = + (struct iw_cm_id *)resource_user_context; + + pr_debug("C2_RES_IND_EP event_id=%d\n", event_id); + if (event_id != CCAE_CONNECTION_REQUEST) { + pr_debug("%s: Invalid event_id: %d\n", + __func__, event_id); + break; + } + cm_event.event = IW_CM_EVENT_CONNECT_REQUEST; + cm_event.provider_data = (void*)(unsigned long)req->cr_handle; + laddr->sin_addr.s_addr = req->laddr; + raddr->sin_addr.s_addr = req->raddr; + laddr->sin_port = req->lport; + raddr->sin_port = req->rport; + cm_event.private_data_len = + be32_to_cpu(req->private_data_length); + cm_event.private_data = req->private_data; + /* + * Until ird/ord negotiation via MPAv2 support is added, send + * max supported values + */ + cm_event.ird = cm_event.ord = 128; + + if (cm_id->event_handler) + cm_id->event_handler(cm_id, &cm_event); + break; + } + + case C2_RES_IND_CQ:{ + struct c2_cq *cq = + (struct c2_cq *) resource_user_context; + + pr_debug("IB_EVENT_CQ_ERR\n"); + ib_event.device = &c2dev->ibdev; + ib_event.element.cq = &cq->ibcq; + ib_event.event = IB_EVENT_CQ_ERR; + + if (cq->ibcq.event_handler) + cq->ibcq.event_handler(&ib_event, + cq->ibcq.cq_context); + break; + } + + default: + printk("Bad resource indicator = %d\n", + resource_indicator); + break; + } + + ignore_it: + c2_mq_free(mq); +} diff --git a/drivers/infiniband/hw/amso1100/c2_ae.h b/drivers/infiniband/hw/amso1100/c2_ae.h new file mode 100644 index 000000000..3a065c33b --- /dev/null +++ b/drivers/infiniband/hw/amso1100/c2_ae.h @@ -0,0 +1,108 @@ +/* + * Copyright (c) 2005 Ammasso, Inc. All rights reserved. + * Copyright (c) 2005 Open Grid Computing, Inc. All rights reserved. + * + * This software is available to you under a choice of one of two + * licenses. You may choose to be licensed under the terms of the GNU + * General Public License (GPL) Version 2, available from the file + * COPYING in the main directory of this source tree, or the + * OpenIB.org BSD license below: + * + * Redistribution and use in source and binary forms, with or + * without modification, are permitted provided that the following + * conditions are met: + * + * - Redistributions of source code must retain the above + * copyright notice, this list of conditions and the following + * disclaimer. + * + * - Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and/or other materials + * provided with the distribution. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS + * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN + * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ +#ifndef _C2_AE_H_ +#define _C2_AE_H_ + +/* + * WARNING: If you change this file, also bump C2_IVN_BASE + * in common/include/clustercore/c2_ivn.h. + */ + +/* + * Asynchronous Event Identifiers + * + * These start at 0x80 only so it's obvious from inspection that + * they are not work-request statuses. This isn't critical. + * + * NOTE: these event id's must fit in eight bits. + */ +enum c2_event_id { + CCAE_REMOTE_SHUTDOWN = 0x80, + CCAE_ACTIVE_CONNECT_RESULTS, + CCAE_CONNECTION_REQUEST, + CCAE_LLP_CLOSE_COMPLETE, + CCAE_TERMINATE_MESSAGE_RECEIVED, + CCAE_LLP_CONNECTION_RESET, + CCAE_LLP_CONNECTION_LOST, + CCAE_LLP_SEGMENT_SIZE_INVALID, + CCAE_LLP_INVALID_CRC, + CCAE_LLP_BAD_FPDU, + CCAE_INVALID_DDP_VERSION, + CCAE_INVALID_RDMA_VERSION, + CCAE_UNEXPECTED_OPCODE, + CCAE_INVALID_DDP_QUEUE_NUMBER, + CCAE_RDMA_READ_NOT_ENABLED, + CCAE_RDMA_WRITE_NOT_ENABLED, + CCAE_RDMA_READ_TOO_SMALL, + CCAE_NO_L_BIT, + CCAE_TAGGED_INVALID_STAG, + CCAE_TAGGED_BASE_BOUNDS_VIOLATION, + CCAE_TAGGED_ACCESS_RIGHTS_VIOLATION, + CCAE_TAGGED_INVALID_PD, + CCAE_WRAP_ERROR, + CCAE_BAD_CLOSE, + CCAE_BAD_LLP_CLOSE, + CCAE_INVALID_MSN_RANGE, + CCAE_INVALID_MSN_GAP, + CCAE_IRRQ_OVERFLOW, + CCAE_IRRQ_MSN_GAP, + CCAE_IRRQ_MSN_RANGE, + CCAE_IRRQ_INVALID_STAG, + CCAE_IRRQ_BASE_BOUNDS_VIOLATION, + CCAE_IRRQ_ACCESS_RIGHTS_VIOLATION, + CCAE_IRRQ_INVALID_PD, + CCAE_IRRQ_WRAP_ERROR, + CCAE_CQ_SQ_COMPLETION_OVERFLOW, + CCAE_CQ_RQ_COMPLETION_ERROR, + CCAE_QP_SRQ_WQE_ERROR, + CCAE_QP_LOCAL_CATASTROPHIC_ERROR, + CCAE_CQ_OVERFLOW, + CCAE_CQ_OPERATION_ERROR, + CCAE_SRQ_LIMIT_REACHED, + CCAE_QP_RQ_LIMIT_REACHED, + CCAE_SRQ_CATASTROPHIC_ERROR, + CCAE_RNIC_CATASTROPHIC_ERROR +/* WARNING If you add more id's, make sure their values fit in eight bits. */ +}; + +/* + * Resource Indicators and Identifiers + */ +enum c2_resource_indicator { + C2_RES_IND_QP = 1, + C2_RES_IND_EP, + C2_RES_IND_CQ, + C2_RES_IND_SRQ, +}; + +#endif /* _C2_AE_H_ */ diff --git a/drivers/infiniband/hw/amso1100/c2_alloc.c b/drivers/infiniband/hw/amso1100/c2_alloc.c new file mode 100644 index 000000000..78d247ec6 --- /dev/null +++ b/drivers/infiniband/hw/amso1100/c2_alloc.c @@ -0,0 +1,142 @@ +/* + * Copyright (c) 2004 Topspin Communications. All rights reserved. + * Copyright (c) 2005 Open Grid Computing, Inc. All rights reserved. + * + * This software is available to you under a choice of one of two + * licenses. You may choose to be licensed under the terms of the GNU + * General Public License (GPL) Version 2, available from the file + * COPYING in the main directory of this source tree, or the + * OpenIB.org BSD license below: + * + * Redistribution and use in source and binary forms, with or + * without modification, are permitted provided that the following + * conditions are met: + * + * - Redistributions of source code must retain the above + * copyright notice, this list of conditions and the following + * disclaimer. + * + * - Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and/or other materials + * provided with the distribution. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS + * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN + * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ + +#include <linux/errno.h> +#include <linux/bitmap.h> + +#include "c2.h" + +static int c2_alloc_mqsp_chunk(struct c2_dev *c2dev, gfp_t gfp_mask, + struct sp_chunk **head) +{ + int i; + struct sp_chunk *new_head; + dma_addr_t dma_addr; + + new_head = dma_alloc_coherent(&c2dev->pcidev->dev, PAGE_SIZE, + &dma_addr, gfp_mask); + if (new_head == NULL) + return -ENOMEM; + + new_head->dma_addr = dma_addr; + dma_unmap_addr_set(new_head, mapping, new_head->dma_addr); + + new_head->next = NULL; + new_head->head = 0; + + /* build list where each index is the next free slot */ + for (i = 0; + i < (PAGE_SIZE - sizeof(struct sp_chunk) - + sizeof(u16)) / sizeof(u16) - 1; + i++) { + new_head->shared_ptr[i] = i + 1; + } + /* terminate list */ + new_head->shared_ptr[i] = 0xFFFF; + + *head = new_head; + return 0; +} + +int c2_init_mqsp_pool(struct c2_dev *c2dev, gfp_t gfp_mask, + struct sp_chunk **root) +{ + return c2_alloc_mqsp_chunk(c2dev, gfp_mask, root); +} + +void c2_free_mqsp_pool(struct c2_dev *c2dev, struct sp_chunk *root) +{ + struct sp_chunk *next; + + while (root) { + next = root->next; + dma_free_coherent(&c2dev->pcidev->dev, PAGE_SIZE, root, + dma_unmap_addr(root, mapping)); + root = next; + } +} + +__be16 *c2_alloc_mqsp(struct c2_dev *c2dev, struct sp_chunk *head, + dma_addr_t *dma_addr, gfp_t gfp_mask) +{ + u16 mqsp; + + while (head) { + mqsp = head->head; + if (mqsp != 0xFFFF) { + head->head = head->shared_ptr[mqsp]; + break; + } else if (head->next == NULL) { + if (c2_alloc_mqsp_chunk(c2dev, gfp_mask, &head->next) == + 0) { + head = head->next; + mqsp = head->head; + head->head = head->shared_ptr[mqsp]; + break; + } else + return NULL; + } else + head = head->next; + } + if (head) { + *dma_addr = head->dma_addr + + ((unsigned long) &(head->shared_ptr[mqsp]) - + (unsigned long) head); + pr_debug("%s addr %p dma_addr %llx\n", __func__, + &(head->shared_ptr[mqsp]), (unsigned long long) *dma_addr); + return (__force __be16 *) &(head->shared_ptr[mqsp]); + } + return NULL; +} + +void c2_free_mqsp(__be16 *mqsp) +{ + struct sp_chunk *head; + u16 idx; + + /* The chunk containing this ptr begins at the page boundary */ + head = (struct sp_chunk *) ((unsigned long) mqsp & PAGE_MASK); + + /* Link head to new mqsp */ + *mqsp = (__force __be16) head->head; + + /* Compute the shared_ptr index */ + idx = ((unsigned long) mqsp & ~PAGE_MASK) >> 1; + idx -= (unsigned long) &(((struct sp_chunk *) 0)->shared_ptr[0]) >> 1; + + /* Point this index at the head */ + head->shared_ptr[idx] = head->head; + + /* Point head at this index */ + head->head = idx; +} diff --git a/drivers/infiniband/hw/amso1100/c2_cm.c b/drivers/infiniband/hw/amso1100/c2_cm.c new file mode 100644 index 000000000..23bfa94fb --- /dev/null +++ b/drivers/infiniband/hw/amso1100/c2_cm.c @@ -0,0 +1,461 @@ +/* + * Copyright (c) 2005 Ammasso, Inc. All rights reserved. + * Copyright (c) 2005 Open Grid Computing, Inc. All rights reserved. + * + * This software is available to you under a choice of one of two + * licenses. You may choose to be licensed under the terms of the GNU + * General Public License (GPL) Version 2, available from the file + * COPYING in the main directory of this source tree, or the + * OpenIB.org BSD license below: + * + * Redistribution and use in source and binary forms, with or + * without modification, are permitted provided that the following + * conditions are met: + * + * - Redistributions of source code must retain the above + * copyright notice, this list of conditions and the following + * disclaimer. + * + * - Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and/or other materials + * provided with the distribution. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS + * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN + * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + * + */ +#include <linux/slab.h> + +#include "c2.h" +#include "c2_wr.h" +#include "c2_vq.h" +#include <rdma/iw_cm.h> + +int c2_llp_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *iw_param) +{ + struct c2_dev *c2dev = to_c2dev(cm_id->device); + struct ib_qp *ibqp; + struct c2_qp *qp; + struct c2wr_qp_connect_req *wr; /* variable size needs a malloc. */ + struct c2_vq_req *vq_req; + int err; + struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->remote_addr; + + if (cm_id->remote_addr.ss_family != AF_INET) + return -ENOSYS; + + ibqp = c2_get_qp(cm_id->device, iw_param->qpn); + if (!ibqp) + return -EINVAL; + qp = to_c2qp(ibqp); + + /* Associate QP <--> CM_ID */ + cm_id->provider_data = qp; + cm_id->add_ref(cm_id); + qp->cm_id = cm_id; + + /* + * only support the max private_data length + */ + if (iw_param->private_data_len > C2_MAX_PRIVATE_DATA_SIZE) { + err = -EINVAL; + goto bail0; + } + /* + * Set the rdma read limits + */ + err = c2_qp_set_read_limits(c2dev, qp, iw_param->ord, iw_param->ird); + if (err) + goto bail0; + + /* + * Create and send a WR_QP_CONNECT... + */ + wr = kmalloc(c2dev->req_vq.msg_size, GFP_KERNEL); + if (!wr) { + err = -ENOMEM; + goto bail0; + } + + vq_req = vq_req_alloc(c2dev); + if (!vq_req) { + err = -ENOMEM; + goto bail1; + } + + c2_wr_set_id(wr, CCWR_QP_CONNECT); + wr->hdr.context = 0; + wr->rnic_handle = c2dev->adapter_handle; + wr->qp_handle = qp->adapter_handle; + + wr->remote_addr = raddr->sin_addr.s_addr; + wr->remote_port = raddr->sin_port; + + /* + * Move any private data from the callers's buf into + * the WR. + */ + if (iw_param->private_data) { + wr->private_data_length = + cpu_to_be32(iw_param->private_data_len); + memcpy(&wr->private_data[0], iw_param->private_data, + iw_param->private_data_len); + } else + wr->private_data_length = 0; + + /* + * Send WR to adapter. NOTE: There is no synch reply from + * the adapter. + */ + err = vq_send_wr(c2dev, (union c2wr *) wr); + vq_req_free(c2dev, vq_req); + + bail1: + kfree(wr); + bail0: + if (err) { + /* + * If we fail, release reference on QP and + * disassociate QP from CM_ID + */ + cm_id->provider_data = NULL; + qp->cm_id = NULL; + cm_id->rem_ref(cm_id); + } + return err; +} + +int c2_llp_service_create(struct iw_cm_id *cm_id, int backlog) +{ + struct c2_dev *c2dev; + struct c2wr_ep_listen_create_req wr; + struct c2wr_ep_listen_create_rep *reply; + struct c2_vq_req *vq_req; + int err; + struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->local_addr; + + if (cm_id->local_addr.ss_family != AF_INET) + return -ENOSYS; + + c2dev = to_c2dev(cm_id->device); + if (c2dev == NULL) + return -EINVAL; + + /* + * Allocate verbs request. + */ + vq_req = vq_req_alloc(c2dev); + if (!vq_req) + return -ENOMEM; + + /* + * Build the WR + */ + c2_wr_set_id(&wr, CCWR_EP_LISTEN_CREATE); + wr.hdr.context = (u64) (unsigned long) vq_req; + wr.rnic_handle = c2dev->adapter_handle; + wr.local_addr = laddr->sin_addr.s_addr; + wr.local_port = laddr->sin_port; + wr.backlog = cpu_to_be32(backlog); + wr.user_context = (u64) (unsigned long) cm_id; + + /* + * Reference the request struct. Dereferenced in the int handler. + */ + vq_req_get(c2dev, vq_req); + + /* + * Send WR to adapter + */ + err = vq_send_wr(c2dev, (union c2wr *) & wr); + if (err) { + vq_req_put(c2dev, vq_req); + goto bail0; + } + + /* + * Wait for reply from adapter + */ + err = vq_wait_for_reply(c2dev, vq_req); + if (err) + goto bail0; + + /* + * Process reply + */ + reply = + (struct c2wr_ep_listen_create_rep *) (unsigned long) vq_req->reply_msg; + if (!reply) { + err = -ENOMEM; + goto bail1; + } + + if ((err = c2_errno(reply)) != 0) + goto bail1; + + /* + * Keep the adapter handle. Used in subsequent destroy + */ + cm_id->provider_data = (void*)(unsigned long) reply->ep_handle; + + /* + * free vq stuff + */ + vq_repbuf_free(c2dev, reply); + vq_req_free(c2dev, vq_req); + + return 0; + + bail1: + vq_repbuf_free(c2dev, reply); + bail0: + vq_req_free(c2dev, vq_req); + return err; +} + + +int c2_llp_service_destroy(struct iw_cm_id *cm_id) +{ + + struct c2_dev *c2dev; + struct c2wr_ep_listen_destroy_req wr; + struct c2wr_ep_listen_destroy_rep *reply; + struct c2_vq_req *vq_req; + int err; + + c2dev = to_c2dev(cm_id->device); + if (c2dev == NULL) + return -EINVAL; + + /* + * Allocate verbs request. + */ + vq_req = vq_req_alloc(c2dev); + if (!vq_req) + return -ENOMEM; + + /* + * Build the WR + */ + c2_wr_set_id(&wr, CCWR_EP_LISTEN_DESTROY); + wr.hdr.context = (unsigned long) vq_req; + wr.rnic_handle = c2dev->adapter_handle; + wr.ep_handle = (u32)(unsigned long)cm_id->provider_data; + + /* + * reference the request struct. dereferenced in the int handler. + */ + vq_req_get(c2dev, vq_req); + + /* + * Send WR to adapter + */ + err = vq_send_wr(c2dev, (union c2wr *) & wr); + if (err) { + vq_req_put(c2dev, vq_req); + goto bail0; + } + + /* + * Wait for reply from adapter + */ + err = vq_wait_for_reply(c2dev, vq_req); + if (err) + goto bail0; + + /* + * Process reply + */ + reply=(struct c2wr_ep_listen_destroy_rep *)(unsigned long)vq_req->reply_msg; + if (!reply) { + err = -ENOMEM; + goto bail0; + } + if ((err = c2_errno(reply)) != 0) + goto bail1; + + bail1: + vq_repbuf_free(c2dev, reply); + bail0: + vq_req_free(c2dev, vq_req); + return err; +} + +int c2_llp_accept(struct iw_cm_id *cm_id, struct iw_cm_conn_param *iw_param) +{ + struct c2_dev *c2dev = to_c2dev(cm_id->device); + struct c2_qp *qp; + struct ib_qp *ibqp; + struct c2wr_cr_accept_req *wr; /* variable length WR */ + struct c2_vq_req *vq_req; + struct c2wr_cr_accept_rep *reply; /* VQ Reply msg ptr. */ + int err; + + ibqp = c2_get_qp(cm_id->device, iw_param->qpn); + if (!ibqp) + return -EINVAL; + qp = to_c2qp(ibqp); + + /* Set the RDMA read limits */ + err = c2_qp_set_read_limits(c2dev, qp, iw_param->ord, iw_param->ird); + if (err) + goto bail0; + + /* Allocate verbs request. */ + vq_req = vq_req_alloc(c2dev); + if (!vq_req) { + err = -ENOMEM; + goto bail0; + } + vq_req->qp = qp; + vq_req->cm_id = cm_id; + vq_req->event = IW_CM_EVENT_ESTABLISHED; + + wr = kmalloc(c2dev->req_vq.msg_size, GFP_KERNEL); + if (!wr) { + err = -ENOMEM; + goto bail1; + } + + /* Build the WR */ + c2_wr_set_id(wr, CCWR_CR_ACCEPT); + wr->hdr.context = (unsigned long) vq_req; + wr->rnic_handle = c2dev->adapter_handle; + wr->ep_handle = (u32) (unsigned long) cm_id->provider_data; + wr->qp_handle = qp->adapter_handle; + + /* Replace the cr_handle with the QP after accept */ + cm_id->provider_data = qp; + cm_id->add_ref(cm_id); + qp->cm_id = cm_id; + + cm_id->provider_data = qp; + + /* Validate private_data length */ + if (iw_param->private_data_len > C2_MAX_PRIVATE_DATA_SIZE) { + err = -EINVAL; + goto bail1; + } + + if (iw_param->private_data) { + wr->private_data_length = cpu_to_be32(iw_param->private_data_len); + memcpy(&wr->private_data[0], + iw_param->private_data, iw_param->private_data_len); + } else + wr->private_data_length = 0; + + /* Reference the request struct. Dereferenced in the int handler. */ + vq_req_get(c2dev, vq_req); + + /* Send WR to adapter */ + err = vq_send_wr(c2dev, (union c2wr *) wr); + if (err) { + vq_req_put(c2dev, vq_req); + goto bail1; + } + + /* Wait for reply from adapter */ + err = vq_wait_for_reply(c2dev, vq_req); + if (err) + goto bail1; + + /* Check that reply is present */ + reply = (struct c2wr_cr_accept_rep *) (unsigned long) vq_req->reply_msg; + if (!reply) { + err = -ENOMEM; + goto bail1; + } + + err = c2_errno(reply); + vq_repbuf_free(c2dev, reply); + + if (!err) + c2_set_qp_state(qp, C2_QP_STATE_RTS); + bail1: + kfree(wr); + vq_req_free(c2dev, vq_req); + bail0: + if (err) { + /* + * If we fail, release reference on QP and + * disassociate QP from CM_ID + */ + cm_id->provider_data = NULL; + qp->cm_id = NULL; + cm_id->rem_ref(cm_id); + } + return err; +} + +int c2_llp_reject(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len) +{ + struct c2_dev *c2dev; + struct c2wr_cr_reject_req wr; + struct c2_vq_req *vq_req; + struct c2wr_cr_reject_rep *reply; + int err; + + c2dev = to_c2dev(cm_id->device); + + /* + * Allocate verbs request. + */ + vq_req = vq_req_alloc(c2dev); + if (!vq_req) + return -ENOMEM; + + /* + * Build the WR + */ + c2_wr_set_id(&wr, CCWR_CR_REJECT); + wr.hdr.context = (unsigned long) vq_req; + wr.rnic_handle = c2dev->adapter_handle; + wr.ep_handle = (u32) (unsigned long) cm_id->provider_data; + + /* + * reference the request struct. dereferenced in the int handler. + */ + vq_req_get(c2dev, vq_req); + + /* + * Send WR to adapter + */ + err = vq_send_wr(c2dev, (union c2wr *) & wr); + if (err) { + vq_req_put(c2dev, vq_req); + goto bail0; + } + + /* + * Wait for reply from adapter + */ + err = vq_wait_for_reply(c2dev, vq_req); + if (err) + goto bail0; + + /* + * Process reply + */ + reply = (struct c2wr_cr_reject_rep *) (unsigned long) + vq_req->reply_msg; + if (!reply) { + err = -ENOMEM; + goto bail0; + } + err = c2_errno(reply); + /* + * free vq stuff + */ + vq_repbuf_free(c2dev, reply); + + bail0: + vq_req_free(c2dev, vq_req); + return err; +} diff --git a/drivers/infiniband/hw/amso1100/c2_cq.c b/drivers/infiniband/hw/amso1100/c2_cq.c new file mode 100644 index 000000000..1b63185b4 --- /dev/null +++ b/drivers/infiniband/hw/amso1100/c2_cq.c @@ -0,0 +1,440 @@ +/* + * Copyright (c) 2004, 2005 Topspin Communications. All rights reserved. + * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved. + * Copyright (c) 2005 Cisco Systems, Inc. All rights reserved. + * Copyright (c) 2005 Mellanox Technologies. All rights reserved. + * Copyright (c) 2004 Voltaire, Inc. All rights reserved. + * Copyright (c) 2005 Open Grid Computing, Inc. All rights reserved. + * + * This software is available to you under a choice of one of two + * licenses. You may choose to be licensed under the terms of the GNU + * General Public License (GPL) Version 2, available from the file + * COPYING in the main directory of this source tree, or the + * OpenIB.org BSD license below: + * + * Redistribution and use in source and binary forms, with or + * without modification, are permitted provided that the following + * conditions are met: + * + * - Redistributions of source code must retain the above + * copyright notice, this list of conditions and the following + * disclaimer. + * + * - Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and/or other materials + * provided with the distribution. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS + * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN + * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + * + */ +#include <linux/gfp.h> + +#include "c2.h" +#include "c2_vq.h" +#include "c2_status.h" + +#define C2_CQ_MSG_SIZE ((sizeof(struct c2wr_ce) + 32-1) & ~(32-1)) + +static struct c2_cq *c2_cq_get(struct c2_dev *c2dev, int cqn) +{ + struct c2_cq *cq; + unsigned long flags; + + spin_lock_irqsave(&c2dev->lock, flags); + cq = c2dev->qptr_array[cqn]; + if (!cq) { + spin_unlock_irqrestore(&c2dev->lock, flags); + return NULL; + } + atomic_inc(&cq->refcount); + spin_unlock_irqrestore(&c2dev->lock, flags); + return cq; +} + +static void c2_cq_put(struct c2_cq *cq) +{ + if (atomic_dec_and_test(&cq->refcount)) + wake_up(&cq->wait); +} + +void c2_cq_event(struct c2_dev *c2dev, u32 mq_index) +{ + struct c2_cq *cq; + + cq = c2_cq_get(c2dev, mq_index); + if (!cq) { + printk("discarding events on destroyed CQN=%d\n", mq_index); + return; + } + + (*cq->ibcq.comp_handler) (&cq->ibcq, cq->ibcq.cq_context); + c2_cq_put(cq); +} + +void c2_cq_clean(struct c2_dev *c2dev, struct c2_qp *qp, u32 mq_index) +{ + struct c2_cq *cq; + struct c2_mq *q; + + cq = c2_cq_get(c2dev, mq_index); + if (!cq) + return; + + spin_lock_irq(&cq->lock); + q = &cq->mq; + if (q && !c2_mq_empty(q)) { + u16 priv = q->priv; + struct c2wr_ce *msg; + + while (priv != be16_to_cpu(*q->shared)) { + msg = (struct c2wr_ce *) + (q->msg_pool.host + priv * q->msg_size); + if (msg->qp_user_context == (u64) (unsigned long) qp) { + msg->qp_user_context = (u64) 0; + } + priv = (priv + 1) % q->q_size; + } + } + spin_unlock_irq(&cq->lock); + c2_cq_put(cq); +} + +static inline enum ib_wc_status c2_cqe_status_to_openib(u8 status) +{ + switch (status) { + case C2_OK: + return IB_WC_SUCCESS; + case CCERR_FLUSHED: + return IB_WC_WR_FLUSH_ERR; + case CCERR_BASE_AND_BOUNDS_VIOLATION: + return IB_WC_LOC_PROT_ERR; + case CCERR_ACCESS_VIOLATION: + return IB_WC_LOC_ACCESS_ERR; + case CCERR_TOTAL_LENGTH_TOO_BIG: + return IB_WC_LOC_LEN_ERR; + case CCERR_INVALID_WINDOW: + return IB_WC_MW_BIND_ERR; + default: + return IB_WC_GENERAL_ERR; + } +} + + +static inline int c2_poll_one(struct c2_dev *c2dev, + struct c2_cq *cq, struct ib_wc *entry) +{ + struct c2wr_ce *ce; + struct c2_qp *qp; + int is_recv = 0; + + ce = c2_mq_consume(&cq->mq); + if (!ce) { + return -EAGAIN; + } + + /* + * if the qp returned is null then this qp has already + * been freed and we are unable process the completion. + * try pulling the next message + */ + while ((qp = + (struct c2_qp *) (unsigned long) ce->qp_user_context) == NULL) { + c2_mq_free(&cq->mq); + ce = c2_mq_consume(&cq->mq); + if (!ce) + return -EAGAIN; + } + + entry->status = c2_cqe_status_to_openib(c2_wr_get_result(ce)); + entry->wr_id = ce->hdr.context; + entry->qp = &qp->ibqp; + entry->wc_flags = 0; + entry->slid = 0; + entry->sl = 0; + entry->src_qp = 0; + entry->dlid_path_bits = 0; + entry->pkey_index = 0; + + switch (c2_wr_get_id(ce)) { + case C2_WR_TYPE_SEND: + entry->opcode = IB_WC_SEND; + break; + case C2_WR_TYPE_RDMA_WRITE: + entry->opcode = IB_WC_RDMA_WRITE; + break; + case C2_WR_TYPE_RDMA_READ: + entry->opcode = IB_WC_RDMA_READ; + break; + case C2_WR_TYPE_BIND_MW: + entry->opcode = IB_WC_BIND_MW; + break; + case C2_WR_TYPE_RECV: + entry->byte_len = be32_to_cpu(ce->bytes_rcvd); + entry->opcode = IB_WC_RECV; + is_recv = 1; + break; + default: + break; + } + + /* consume the WQEs */ + if (is_recv) + c2_mq_lconsume(&qp->rq_mq, 1); + else + c2_mq_lconsume(&qp->sq_mq, + be32_to_cpu(c2_wr_get_wqe_count(ce)) + 1); + + /* free the message */ + c2_mq_free(&cq->mq); + + return 0; +} + +int c2_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *entry) +{ + struct c2_dev *c2dev = to_c2dev(ibcq->device); + struct c2_cq *cq = to_c2cq(ibcq); + unsigned long flags; + int npolled, err; + + spin_lock_irqsave(&cq->lock, flags); + + for (npolled = 0; npolled < num_entries; ++npolled) { + + err = c2_poll_one(c2dev, cq, entry + npolled); + if (err) + break; + } + + spin_unlock_irqrestore(&cq->lock, flags); + + return npolled; +} + +int c2_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags notify_flags) +{ + struct c2_mq_shared __iomem *shared; + struct c2_cq *cq; + unsigned long flags; + int ret = 0; + + cq = to_c2cq(ibcq); + shared = cq->mq.peer; + + if ((notify_flags & IB_CQ_SOLICITED_MASK) == IB_CQ_NEXT_COMP) + writeb(C2_CQ_NOTIFICATION_TYPE_NEXT, &shared->notification_type); + else if ((notify_flags & IB_CQ_SOLICITED_MASK) == IB_CQ_SOLICITED) + writeb(C2_CQ_NOTIFICATION_TYPE_NEXT_SE, &shared->notification_type); + else + return -EINVAL; + + writeb(CQ_WAIT_FOR_DMA | CQ_ARMED, &shared->armed); + + /* + * Now read back shared->armed to make the PCI + * write synchronous. This is necessary for + * correct cq notification semantics. + */ + readb(&shared->armed); + + if (notify_flags & IB_CQ_REPORT_MISSED_EVENTS) { + spin_lock_irqsave(&cq->lock, flags); + ret = !c2_mq_empty(&cq->mq); + spin_unlock_irqrestore(&cq->lock, flags); + } + + return ret; +} + +static void c2_free_cq_buf(struct c2_dev *c2dev, struct c2_mq *mq) +{ + dma_free_coherent(&c2dev->pcidev->dev, mq->q_size * mq->msg_size, + mq->msg_pool.host, dma_unmap_addr(mq, mapping)); +} + +static int c2_alloc_cq_buf(struct c2_dev *c2dev, struct c2_mq *mq, + size_t q_size, size_t msg_size) +{ + u8 *pool_start; + + if (q_size > SIZE_MAX / msg_size) + return -EINVAL; + + pool_start = dma_alloc_coherent(&c2dev->pcidev->dev, q_size * msg_size, + &mq->host_dma, GFP_KERNEL); + if (!pool_start) + return -ENOMEM; + + c2_mq_rep_init(mq, + 0, /* index (currently unknown) */ + q_size, + msg_size, + pool_start, + NULL, /* peer (currently unknown) */ + C2_MQ_HOST_TARGET); + + dma_unmap_addr_set(mq, mapping, mq->host_dma); + + return 0; +} + +int c2_init_cq(struct c2_dev *c2dev, int entries, + struct c2_ucontext *ctx, struct c2_cq *cq) +{ + struct c2wr_cq_create_req wr; + struct c2wr_cq_create_rep *reply; + unsigned long peer_pa; + struct c2_vq_req *vq_req; + int err; + + might_sleep(); + + cq->ibcq.cqe = entries - 1; + cq->is_kernel = !ctx; + + /* Allocate a shared pointer */ + cq->mq.shared = c2_alloc_mqsp(c2dev, c2dev->kern_mqsp_pool, + &cq->mq.shared_dma, GFP_KERNEL); + if (!cq->mq.shared) + return -ENOMEM; + + /* Allocate pages for the message pool */ + err = c2_alloc_cq_buf(c2dev, &cq->mq, entries + 1, C2_CQ_MSG_SIZE); + if (err) + goto bail0; + + vq_req = vq_req_alloc(c2dev); + if (!vq_req) { + err = -ENOMEM; + goto bail1; + } + + memset(&wr, 0, sizeof(wr)); + c2_wr_set_id(&wr, CCWR_CQ_CREATE); + wr.hdr.context = (unsigned long) vq_req; + wr.rnic_handle = c2dev->adapter_handle; + wr.msg_size = cpu_to_be32(cq->mq.msg_size); + wr.depth = cpu_to_be32(cq->mq.q_size); + wr.shared_ht = cpu_to_be64(cq->mq.shared_dma); + wr.msg_pool = cpu_to_be64(cq->mq.host_dma); + wr.user_context = (u64) (unsigned long) (cq); + + vq_req_get(c2dev, vq_req); + + err = vq_send_wr(c2dev, (union c2wr *) & wr); + if (err) { + vq_req_put(c2dev, vq_req); + goto bail2; + } + + err = vq_wait_for_reply(c2dev, vq_req); + if (err) + goto bail2; + + reply = (struct c2wr_cq_create_rep *) (unsigned long) (vq_req->reply_msg); + if (!reply) { + err = -ENOMEM; + goto bail2; + } + + if ((err = c2_errno(reply)) != 0) + goto bail3; + + cq->adapter_handle = reply->cq_handle; + cq->mq.index = be32_to_cpu(reply->mq_index); + + peer_pa = c2dev->pa + be32_to_cpu(reply->adapter_shared); + cq->mq.peer = ioremap_nocache(peer_pa, PAGE_SIZE); + if (!cq->mq.peer) { + err = -ENOMEM; + goto bail3; + } + + vq_repbuf_free(c2dev, reply); + vq_req_free(c2dev, vq_req); + + spin_lock_init(&cq->lock); + atomic_set(&cq->refcount, 1); + init_waitqueue_head(&cq->wait); + + /* + * Use the MQ index allocated by the adapter to + * store the CQ in the qptr_array + */ + cq->cqn = cq->mq.index; + c2dev->qptr_array[cq->cqn] = cq; + + return 0; + + bail3: + vq_repbuf_free(c2dev, reply); + bail2: + vq_req_free(c2dev, vq_req); + bail1: + c2_free_cq_buf(c2dev, &cq->mq); + bail0: + c2_free_mqsp(cq->mq.shared); + + return err; +} + +void c2_free_cq(struct c2_dev *c2dev, struct c2_cq *cq) +{ + int err; + struct c2_vq_req *vq_req; + struct c2wr_cq_destroy_req wr; + struct c2wr_cq_destroy_rep *reply; + + might_sleep(); + + /* Clear CQ from the qptr array */ + spin_lock_irq(&c2dev->lock); + c2dev->qptr_array[cq->mq.index] = NULL; + atomic_dec(&cq->refcount); + spin_unlock_irq(&c2dev->lock); + + wait_event(cq->wait, !atomic_read(&cq->refcount)); + + vq_req = vq_req_alloc(c2dev); + if (!vq_req) { + goto bail0; + } + + memset(&wr, 0, sizeof(wr)); + c2_wr_set_id(&wr, CCWR_CQ_DESTROY); + wr.hdr.context = (unsigned long) vq_req; + wr.rnic_handle = c2dev->adapter_handle; + wr.cq_handle = cq->adapter_handle; + + vq_req_get(c2dev, vq_req); + + err = vq_send_wr(c2dev, (union c2wr *) & wr); + if (err) { + vq_req_put(c2dev, vq_req); + goto bail1; + } + + err = vq_wait_for_reply(c2dev, vq_req); + if (err) + goto bail1; + + reply = (struct c2wr_cq_destroy_rep *) (unsigned long) (vq_req->reply_msg); + if (reply) + vq_repbuf_free(c2dev, reply); + bail1: + vq_req_free(c2dev, vq_req); + bail0: + if (cq->is_kernel) { + c2_free_cq_buf(c2dev, &cq->mq); + } + + return; +} diff --git a/drivers/infiniband/hw/amso1100/c2_intr.c b/drivers/infiniband/hw/amso1100/c2_intr.c new file mode 100644 index 000000000..3a17d9b36 --- /dev/null +++ b/drivers/infiniband/hw/amso1100/c2_intr.c @@ -0,0 +1,219 @@ +/* + * Copyright (c) 2005 Ammasso, Inc. All rights reserved. + * Copyright (c) 2005 Open Grid Computing, Inc. All rights reserved. + * + * This software is available to you under a choice of one of two + * licenses. You may choose to be licensed under the terms of the GNU + * General Public License (GPL) Version 2, available from the file + * COPYING in the main directory of this source tree, or the + * OpenIB.org BSD license below: + * + * Redistribution and use in source and binary forms, with or + * without modification, are permitted provided that the following + * conditions are met: + * + * - Redistributions of source code must retain the above + * copyright notice, this list of conditions and the following + * disclaimer. + * + * - Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and/or other materials + * provided with the distribution. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS + * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN + * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ +#include "c2.h" +#include <rdma/iw_cm.h> +#include "c2_vq.h" + +static void handle_mq(struct c2_dev *c2dev, u32 index); +static void handle_vq(struct c2_dev *c2dev, u32 mq_index); + +/* + * Handle RNIC interrupts + */ +void c2_rnic_interrupt(struct c2_dev *c2dev) +{ + unsigned int mq_index; + + while (c2dev->hints_read != be16_to_cpu(*c2dev->hint_count)) { + mq_index = readl(c2dev->regs + PCI_BAR0_HOST_HINT); + if (mq_index & 0x80000000) { + break; + } + + c2dev->hints_read++; + handle_mq(c2dev, mq_index); + } + +} + +/* + * Top level MQ handler + */ +static void handle_mq(struct c2_dev *c2dev, u32 mq_index) +{ + if (c2dev->qptr_array[mq_index] == NULL) { + pr_debug("handle_mq: stray activity for mq_index=%d\n", + mq_index); + return; + } + + switch (mq_index) { + case (0): + /* + * An index of 0 in the activity queue + * indicates the req vq now has messages + * available... + * + * Wake up any waiters waiting on req VQ + * message availability. + */ + wake_up(&c2dev->req_vq_wo); + break; + case (1): + handle_vq(c2dev, mq_index); + break; + case (2): + /* We have to purge the VQ in case there are pending + * accept reply requests that would result in the + * generation of an ESTABLISHED event. If we don't + * generate these first, a CLOSE event could end up + * being delivered before the ESTABLISHED event. + */ + handle_vq(c2dev, 1); + + c2_ae_event(c2dev, mq_index); + break; + default: + /* There is no event synchronization between CQ events + * and AE or CM events. In fact, CQE could be + * delivered for all of the I/O up to and including the + * FLUSH for a peer disconenct prior to the ESTABLISHED + * event being delivered to the app. The reason for this + * is that CM events are delivered on a thread, while AE + * and CM events are delivered on interrupt context. + */ + c2_cq_event(c2dev, mq_index); + break; + } + + return; +} + +/* + * Handles verbs WR replies. + */ +static void handle_vq(struct c2_dev *c2dev, u32 mq_index) +{ + void *adapter_msg, *reply_msg; + struct c2wr_hdr *host_msg; + struct c2wr_hdr tmp; + struct c2_mq *reply_vq; + struct c2_vq_req *req; + struct iw_cm_event cm_event; + int err; + + reply_vq = (struct c2_mq *) c2dev->qptr_array[mq_index]; + + /* + * get next msg from mq_index into adapter_msg. + * don't free it yet. + */ + adapter_msg = c2_mq_consume(reply_vq); + if (adapter_msg == NULL) { + return; + } + + host_msg = vq_repbuf_alloc(c2dev); + + /* + * If we can't get a host buffer, then we'll still + * wakeup the waiter, we just won't give him the msg. + * It is assumed the waiter will deal with this... + */ + if (!host_msg) { + pr_debug("handle_vq: no repbufs!\n"); + + /* + * just copy the WR header into a local variable. + * this allows us to still demux on the context + */ + host_msg = &tmp; + memcpy(host_msg, adapter_msg, sizeof(tmp)); + reply_msg = NULL; + } else { + memcpy(host_msg, adapter_msg, reply_vq->msg_size); + reply_msg = host_msg; + } + + /* + * consume the msg from the MQ + */ + c2_mq_free(reply_vq); + + /* + * wakeup the waiter. + */ + req = (struct c2_vq_req *) (unsigned long) host_msg->context; + if (req == NULL) { + /* + * We should never get here, as the adapter should + * never send us a reply that we're not expecting. + */ + if (reply_msg != NULL) + vq_repbuf_free(c2dev, host_msg); + pr_debug("handle_vq: UNEXPECTEDLY got NULL req\n"); + return; + } + + if (reply_msg) + err = c2_errno(reply_msg); + else + err = -ENOMEM; + + if (!err) switch (req->event) { + case IW_CM_EVENT_ESTABLISHED: + c2_set_qp_state(req->qp, + C2_QP_STATE_RTS); + /* + * Until ird/ord negotiation via MPAv2 support is added, send + * max supported values + */ + cm_event.ird = cm_event.ord = 128; + case IW_CM_EVENT_CLOSE: + + /* + * Move the QP to RTS if this is + * the established event + */ + cm_event.event = req->event; + cm_event.status = 0; + cm_event.local_addr = req->cm_id->local_addr; + cm_event.remote_addr = req->cm_id->remote_addr; + cm_event.private_data = NULL; + cm_event.private_data_len = 0; + req->cm_id->event_handler(req->cm_id, &cm_event); + break; + default: + break; + } + + req->reply_msg = (u64) (unsigned long) (reply_msg); + atomic_set(&req->reply_ready, 1); + wake_up(&req->wait_object); + + /* + * If the request was cancelled, then this put will + * free the vq_req memory...and reply_msg!!! + */ + vq_req_put(c2dev, req); +} diff --git a/drivers/infiniband/hw/amso1100/c2_mm.c b/drivers/infiniband/hw/amso1100/c2_mm.c new file mode 100644 index 000000000..119c4f3d9 --- /dev/null +++ b/drivers/infiniband/hw/amso1100/c2_mm.c @@ -0,0 +1,377 @@ +/* + * Copyright (c) 2005 Ammasso, Inc. All rights reserved. + * Copyright (c) 2005 Open Grid Computing, Inc. All rights reserved. + * + * This software is available to you under a choice of one of two + * licenses. You may choose to be licensed under the terms of the GNU + * General Public License (GPL) Version 2, available from the file + * COPYING in the main directory of this source tree, or the + * OpenIB.org BSD license below: + * + * Redistribution and use in source and binary forms, with or + * without modification, are permitted provided that the following + * conditions are met: + * + * - Redistributions of source code must retain the above + * copyright notice, this list of conditions and the following + * disclaimer. + * + * - Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and/or other materials + * provided with the distribution. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS + * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN + * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ +#include <linux/slab.h> + +#include "c2.h" +#include "c2_vq.h" + +#define PBL_VIRT 1 +#define PBL_PHYS 2 + +/* + * Send all the PBL messages to convey the remainder of the PBL + * Wait for the adapter's reply on the last one. + * This is indicated by setting the MEM_PBL_COMPLETE in the flags. + * + * NOTE: vq_req is _not_ freed by this function. The VQ Host + * Reply buffer _is_ freed by this function. + */ +static int +send_pbl_messages(struct c2_dev *c2dev, __be32 stag_index, + unsigned long va, u32 pbl_depth, + struct c2_vq_req *vq_req, int pbl_type) +{ + u32 pbe_count; /* amt that fits in a PBL msg */ + u32 count; /* amt in this PBL MSG. */ + struct c2wr_nsmr_pbl_req *wr; /* PBL WR ptr */ + struct c2wr_nsmr_pbl_rep *reply; /* reply ptr */ + int err, pbl_virt, pbl_index, i; + + switch (pbl_type) { + case PBL_VIRT: + pbl_virt = 1; + break; + case PBL_PHYS: + pbl_virt = 0; + break; + default: + return -EINVAL; + break; + } + + pbe_count = (c2dev->req_vq.msg_size - + sizeof(struct c2wr_nsmr_pbl_req)) / sizeof(u64); + wr = kmalloc(c2dev->req_vq.msg_size, GFP_KERNEL); + if (!wr) { + return -ENOMEM; + } + c2_wr_set_id(wr, CCWR_NSMR_PBL); + + /* + * Only the last PBL message will generate a reply from the verbs, + * so we set the context to 0 indicating there is no kernel verbs + * handler blocked awaiting this reply. + */ + wr->hdr.context = 0; + wr->rnic_handle = c2dev->adapter_handle; + wr->stag_index = stag_index; /* already swapped */ + wr->flags = 0; + pbl_index = 0; + while (pbl_depth) { + count = min(pbe_count, pbl_depth); + wr->addrs_length = cpu_to_be32(count); + + /* + * If this is the last message, then reference the + * vq request struct cuz we're gonna wait for a reply. + * also make this PBL msg as the last one. + */ + if (count == pbl_depth) { + /* + * reference the request struct. dereferenced in the + * int handler. + */ + vq_req_get(c2dev, vq_req); + wr->flags = cpu_to_be32(MEM_PBL_COMPLETE); + + /* + * This is the last PBL message. + * Set the context to our VQ Request Object so we can + * wait for the reply. + */ + wr->hdr.context = (unsigned long) vq_req; + } + + /* + * If pbl_virt is set then va is a virtual address + * that describes a virtually contiguous memory + * allocation. The wr needs the start of each virtual page + * to be converted to the corresponding physical address + * of the page. If pbl_virt is not set then va is an array + * of physical addresses and there is no conversion to do. + * Just fill in the wr with what is in the array. + */ + for (i = 0; i < count; i++) { + if (pbl_virt) { + va += PAGE_SIZE; + } else { + wr->paddrs[i] = + cpu_to_be64(((u64 *)va)[pbl_index + i]); + } + } + + /* + * Send WR to adapter + */ + err = vq_send_wr(c2dev, (union c2wr *) wr); + if (err) { + if (count <= pbe_count) { + vq_req_put(c2dev, vq_req); + } + goto bail0; + } + pbl_depth -= count; + pbl_index += count; + } + + /* + * Now wait for the reply... + */ + err = vq_wait_for_reply(c2dev, vq_req); + if (err) { + goto bail0; + } + + /* + * Process reply + */ + reply = (struct c2wr_nsmr_pbl_rep *) (unsigned long) vq_req->reply_msg; + if (!reply) { + err = -ENOMEM; + goto bail0; + } + + err = c2_errno(reply); + + vq_repbuf_free(c2dev, reply); + bail0: + kfree(wr); + return err; +} + +#define C2_PBL_MAX_DEPTH 131072 +int +c2_nsmr_register_phys_kern(struct c2_dev *c2dev, u64 *addr_list, + int page_size, int pbl_depth, u32 length, + u32 offset, u64 *va, enum c2_acf acf, + struct c2_mr *mr) +{ + struct c2_vq_req *vq_req; + struct c2wr_nsmr_register_req *wr; + struct c2wr_nsmr_register_rep *reply; + u16 flags; + int i, pbe_count, count; + int err; + + if (!va || !length || !addr_list || !pbl_depth) + return -EINTR; + + /* + * Verify PBL depth is within rnic max + */ + if (pbl_depth > C2_PBL_MAX_DEPTH) { + return -EINTR; + } + + /* + * allocate verbs request object + */ + vq_req = vq_req_alloc(c2dev); + if (!vq_req) + return -ENOMEM; + + wr = kmalloc(c2dev->req_vq.msg_size, GFP_KERNEL); + if (!wr) { + err = -ENOMEM; + goto bail0; + } + + /* + * build the WR + */ + c2_wr_set_id(wr, CCWR_NSMR_REGISTER); + wr->hdr.context = (unsigned long) vq_req; + wr->rnic_handle = c2dev->adapter_handle; + + flags = (acf | MEM_VA_BASED | MEM_REMOTE); + + /* + * compute how many pbes can fit in the message + */ + pbe_count = (c2dev->req_vq.msg_size - + sizeof(struct c2wr_nsmr_register_req)) / sizeof(u64); + + if (pbl_depth <= pbe_count) { + flags |= MEM_PBL_COMPLETE; + } + wr->flags = cpu_to_be16(flags); + wr->stag_key = 0; //stag_key; + wr->va = cpu_to_be64(*va); + wr->pd_id = mr->pd->pd_id; + wr->pbe_size = cpu_to_be32(page_size); + wr->length = cpu_to_be32(length); + wr->pbl_depth = cpu_to_be32(pbl_depth); + wr->fbo = cpu_to_be32(offset); + count = min(pbl_depth, pbe_count); + wr->addrs_length = cpu_to_be32(count); + + /* + * fill out the PBL for this message + */ + for (i = 0; i < count; i++) { + wr->paddrs[i] = cpu_to_be64(addr_list[i]); + } + + /* + * regerence the request struct + */ + vq_req_get(c2dev, vq_req); + + /* + * send the WR to the adapter + */ + err = vq_send_wr(c2dev, (union c2wr *) wr); + if (err) { + vq_req_put(c2dev, vq_req); + goto bail1; + } + + /* + * wait for reply from adapter + */ + err = vq_wait_for_reply(c2dev, vq_req); + if (err) { + goto bail1; + } + + /* + * process reply + */ + reply = + (struct c2wr_nsmr_register_rep *) (unsigned long) (vq_req->reply_msg); + if (!reply) { + err = -ENOMEM; + goto bail1; + } + if ((err = c2_errno(reply))) { + goto bail2; + } + //*p_pb_entries = be32_to_cpu(reply->pbl_depth); + mr->ibmr.lkey = mr->ibmr.rkey = be32_to_cpu(reply->stag_index); + vq_repbuf_free(c2dev, reply); + + /* + * if there are still more PBEs we need to send them to + * the adapter and wait for a reply on the final one. + * reuse vq_req for this purpose. + */ + pbl_depth -= count; + if (pbl_depth) { + + vq_req->reply_msg = (unsigned long) NULL; + atomic_set(&vq_req->reply_ready, 0); + err = send_pbl_messages(c2dev, + cpu_to_be32(mr->ibmr.lkey), + (unsigned long) &addr_list[i], + pbl_depth, vq_req, PBL_PHYS); + if (err) { + goto bail1; + } + } + + vq_req_free(c2dev, vq_req); + kfree(wr); + + return err; + + bail2: + vq_repbuf_free(c2dev, reply); + bail1: + kfree(wr); + bail0: + vq_req_free(c2dev, vq_req); + return err; +} + +int c2_stag_dealloc(struct c2_dev *c2dev, u32 stag_index) +{ + struct c2_vq_req *vq_req; /* verbs request object */ + struct c2wr_stag_dealloc_req wr; /* work request */ + struct c2wr_stag_dealloc_rep *reply; /* WR reply */ + int err; + + + /* + * allocate verbs request object + */ + vq_req = vq_req_alloc(c2dev); + if (!vq_req) { + return -ENOMEM; + } + + /* + * Build the WR + */ + c2_wr_set_id(&wr, CCWR_STAG_DEALLOC); + wr.hdr.context = (u64) (unsigned long) vq_req; + wr.rnic_handle = c2dev->adapter_handle; + wr.stag_index = cpu_to_be32(stag_index); + + /* + * reference the request struct. dereferenced in the int handler. + */ + vq_req_get(c2dev, vq_req); + + /* + * Send WR to adapter + */ + err = vq_send_wr(c2dev, (union c2wr *) & wr); + if (err) { + vq_req_put(c2dev, vq_req); + goto bail0; + } + + /* + * Wait for reply from adapter + */ + err = vq_wait_for_reply(c2dev, vq_req); + if (err) { + goto bail0; + } + + /* + * Process reply + */ + reply = (struct c2wr_stag_dealloc_rep *) (unsigned long) vq_req->reply_msg; + if (!reply) { + err = -ENOMEM; + goto bail0; + } + + err = c2_errno(reply); + + vq_repbuf_free(c2dev, reply); + bail0: + vq_req_free(c2dev, vq_req); + return err; +} diff --git a/drivers/infiniband/hw/amso1100/c2_mq.c b/drivers/infiniband/hw/amso1100/c2_mq.c new file mode 100644 index 000000000..0cddc49be --- /dev/null +++ b/drivers/infiniband/hw/amso1100/c2_mq.c @@ -0,0 +1,174 @@ +/* + * Copyright (c) 2005 Ammasso, Inc. All rights reserved. + * Copyright (c) 2005 Open Grid Computing, Inc. All rights reserved. + * + * This software is available to you under a choice of one of two + * licenses. You may choose to be licensed under the terms of the GNU + * General Public License (GPL) Version 2, available from the file + * COPYING in the main directory of this source tree, or the + * OpenIB.org BSD license below: + * + * Redistribution and use in source and binary forms, with or + * without modification, are permitted provided that the following + * conditions are met: + * + * - Redistributions of source code must retain the above + * copyright notice, this list of conditions and the following + * disclaimer. + * + * - Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and/or other materials + * provided with the distribution. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS + * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN + * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ +#include "c2.h" +#include "c2_mq.h" + +void *c2_mq_alloc(struct c2_mq *q) +{ + BUG_ON(q->magic != C2_MQ_MAGIC); + BUG_ON(q->type != C2_MQ_ADAPTER_TARGET); + + if (c2_mq_full(q)) { + return NULL; + } else { +#ifdef DEBUG + struct c2wr_hdr *m = + (struct c2wr_hdr *) (q->msg_pool.host + q->priv * q->msg_size); +#ifdef CCMSGMAGIC + BUG_ON(m->magic != be32_to_cpu(~CCWR_MAGIC)); + m->magic = cpu_to_be32(CCWR_MAGIC); +#endif + return m; +#else + return q->msg_pool.host + q->priv * q->msg_size; +#endif + } +} + +void c2_mq_produce(struct c2_mq *q) +{ + BUG_ON(q->magic != C2_MQ_MAGIC); + BUG_ON(q->type != C2_MQ_ADAPTER_TARGET); + + if (!c2_mq_full(q)) { + q->priv = (q->priv + 1) % q->q_size; + q->hint_count++; + /* Update peer's offset. */ + __raw_writew((__force u16) cpu_to_be16(q->priv), &q->peer->shared); + } +} + +void *c2_mq_consume(struct c2_mq *q) +{ + BUG_ON(q->magic != C2_MQ_MAGIC); + BUG_ON(q->type != C2_MQ_HOST_TARGET); + + if (c2_mq_empty(q)) { + return NULL; + } else { +#ifdef DEBUG + struct c2wr_hdr *m = (struct c2wr_hdr *) + (q->msg_pool.host + q->priv * q->msg_size); +#ifdef CCMSGMAGIC + BUG_ON(m->magic != be32_to_cpu(CCWR_MAGIC)); +#endif + return m; +#else + return q->msg_pool.host + q->priv * q->msg_size; +#endif + } +} + +void c2_mq_free(struct c2_mq *q) +{ + BUG_ON(q->magic != C2_MQ_MAGIC); + BUG_ON(q->type != C2_MQ_HOST_TARGET); + + if (!c2_mq_empty(q)) { + +#ifdef CCMSGMAGIC + { + struct c2wr_hdr __iomem *m = (struct c2wr_hdr __iomem *) + (q->msg_pool.adapter + q->priv * q->msg_size); + __raw_writel(cpu_to_be32(~CCWR_MAGIC), &m->magic); + } +#endif + q->priv = (q->priv + 1) % q->q_size; + /* Update peer's offset. */ + __raw_writew((__force u16) cpu_to_be16(q->priv), &q->peer->shared); + } +} + + +void c2_mq_lconsume(struct c2_mq *q, u32 wqe_count) +{ + BUG_ON(q->magic != C2_MQ_MAGIC); + BUG_ON(q->type != C2_MQ_ADAPTER_TARGET); + + while (wqe_count--) { + BUG_ON(c2_mq_empty(q)); + *q->shared = cpu_to_be16((be16_to_cpu(*q->shared)+1) % q->q_size); + } +} + +#if 0 +u32 c2_mq_count(struct c2_mq *q) +{ + s32 count; + + if (q->type == C2_MQ_HOST_TARGET) + count = be16_to_cpu(*q->shared) - q->priv; + else + count = q->priv - be16_to_cpu(*q->shared); + + if (count < 0) + count += q->q_size; + + return (u32) count; +} +#endif /* 0 */ + +void c2_mq_req_init(struct c2_mq *q, u32 index, u32 q_size, u32 msg_size, + u8 __iomem *pool_start, u16 __iomem *peer, u32 type) +{ + BUG_ON(!q->shared); + + /* This code assumes the byte swapping has already been done! */ + q->index = index; + q->q_size = q_size; + q->msg_size = msg_size; + q->msg_pool.adapter = pool_start; + q->peer = (struct c2_mq_shared __iomem *) peer; + q->magic = C2_MQ_MAGIC; + q->type = type; + q->priv = 0; + q->hint_count = 0; + return; +} +void c2_mq_rep_init(struct c2_mq *q, u32 index, u32 q_size, u32 msg_size, + u8 *pool_start, u16 __iomem *peer, u32 type) +{ + BUG_ON(!q->shared); + + /* This code assumes the byte swapping has already been done! */ + q->index = index; + q->q_size = q_size; + q->msg_size = msg_size; + q->msg_pool.host = pool_start; + q->peer = (struct c2_mq_shared __iomem *) peer; + q->magic = C2_MQ_MAGIC; + q->type = type; + q->priv = 0; + q->hint_count = 0; + return; +} diff --git a/drivers/infiniband/hw/amso1100/c2_mq.h b/drivers/infiniband/hw/amso1100/c2_mq.h new file mode 100644 index 000000000..fc1b9a7ce --- /dev/null +++ b/drivers/infiniband/hw/amso1100/c2_mq.h @@ -0,0 +1,106 @@ +/* + * Copyright (c) 2005 Ammasso, Inc. All rights reserved. + * Copyright (c) 2005 Open Grid Computing, Inc. All rights reserved. + * + * This software is available to you under a choice of one of two + * licenses. You may choose to be licensed under the terms of the GNU + * General Public License (GPL) Version 2, available from the file + * COPYING in the main directory of this source tree, or the + * OpenIB.org BSD license below: + * + * Redistribution and use in source and binary forms, with or + * without modification, are permitted provided that the following + * conditions are met: + * + * - Redistributions of source code must retain the above + * copyright notice, this list of conditions and the following + * disclaimer. + * + * - Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and/or other materials + * provided with the distribution. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS + * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN + * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ + +#ifndef _C2_MQ_H_ +#define _C2_MQ_H_ +#include <linux/kernel.h> +#include <linux/dma-mapping.h> +#include "c2_wr.h" + +enum c2_shared_regs { + + C2_SHARED_ARMED = 0x10, + C2_SHARED_NOTIFY = 0x18, + C2_SHARED_SHARED = 0x40, +}; + +struct c2_mq_shared { + u16 unused1; + u8 armed; + u8 notification_type; + u32 unused2; + u16 shared; + /* Pad to 64 bytes. */ + u8 pad[64 - sizeof(u16) - 2 * sizeof(u8) - sizeof(u32) - sizeof(u16)]; +}; + +enum c2_mq_type { + C2_MQ_HOST_TARGET = 1, + C2_MQ_ADAPTER_TARGET = 2, +}; + +/* + * c2_mq_t is for kernel-mode MQs like the VQs Cand the AEQ. + * c2_user_mq_t (which is the same format) is for user-mode MQs... + */ +#define C2_MQ_MAGIC 0x4d512020 /* 'MQ ' */ +struct c2_mq { + u32 magic; + union { + u8 *host; + u8 __iomem *adapter; + } msg_pool; + dma_addr_t host_dma; + DEFINE_DMA_UNMAP_ADDR(mapping); + u16 hint_count; + u16 priv; + struct c2_mq_shared __iomem *peer; + __be16 *shared; + dma_addr_t shared_dma; + u32 q_size; + u32 msg_size; + u32 index; + enum c2_mq_type type; +}; + +static __inline__ int c2_mq_empty(struct c2_mq *q) +{ + return q->priv == be16_to_cpu(*q->shared); +} + +static __inline__ int c2_mq_full(struct c2_mq *q) +{ + return q->priv == (be16_to_cpu(*q->shared) + q->q_size - 1) % q->q_size; +} + +extern void c2_mq_lconsume(struct c2_mq *q, u32 wqe_count); +extern void *c2_mq_alloc(struct c2_mq *q); +extern void c2_mq_produce(struct c2_mq *q); +extern void *c2_mq_consume(struct c2_mq *q); +extern void c2_mq_free(struct c2_mq *q); +extern void c2_mq_req_init(struct c2_mq *q, u32 index, u32 q_size, u32 msg_size, + u8 __iomem *pool_start, u16 __iomem *peer, u32 type); +extern void c2_mq_rep_init(struct c2_mq *q, u32 index, u32 q_size, u32 msg_size, + u8 *pool_start, u16 __iomem *peer, u32 type); + +#endif /* _C2_MQ_H_ */ diff --git a/drivers/infiniband/hw/amso1100/c2_pd.c b/drivers/infiniband/hw/amso1100/c2_pd.c new file mode 100644 index 000000000..f3e81dc35 --- /dev/null +++ b/drivers/infiniband/hw/amso1100/c2_pd.c @@ -0,0 +1,90 @@ +/* + * Copyright (c) 2004 Topspin Communications. All rights reserved. + * Copyright (c) 2005 Cisco Systems. All rights reserved. + * Copyright (c) 2005 Mellanox Technologies. All rights reserved. + * Copyright (c) 2005 Open Grid Computing, Inc. All rights reserved. + * + * This software is available to you under a choice of one of two + * licenses. You may choose to be licensed under the terms of the GNU + * General Public License (GPL) Version 2, available from the file + * COPYING in the main directory of this source tree, or the + * OpenIB.org BSD license below: + * + * Redistribution and use in source and binary forms, with or + * without modification, are permitted provided that the following + * conditions are met: + * + * - Redistributions of source code must retain the above + * copyright notice, this list of conditions and the following + * disclaimer. + * + * - Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and/or other materials + * provided with the distribution. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS + * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN + * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ + +#include <linux/init.h> +#include <linux/slab.h> +#include <linux/errno.h> + +#include "c2.h" +#include "c2_provider.h" + +int c2_pd_alloc(struct c2_dev *c2dev, int privileged, struct c2_pd *pd) +{ + u32 obj; + int ret = 0; + + spin_lock(&c2dev->pd_table.lock); + obj = find_next_zero_bit(c2dev->pd_table.table, c2dev->pd_table.max, + c2dev->pd_table.last); + if (obj >= c2dev->pd_table.max) + obj = find_first_zero_bit(c2dev->pd_table.table, + c2dev->pd_table.max); + if (obj < c2dev->pd_table.max) { + pd->pd_id = obj; + __set_bit(obj, c2dev->pd_table.table); + c2dev->pd_table.last = obj+1; + if (c2dev->pd_table.last >= c2dev->pd_table.max) + c2dev->pd_table.last = 0; + } else + ret = -ENOMEM; + spin_unlock(&c2dev->pd_table.lock); + return ret; +} + +void c2_pd_free(struct c2_dev *c2dev, struct c2_pd *pd) +{ + spin_lock(&c2dev->pd_table.lock); + __clear_bit(pd->pd_id, c2dev->pd_table.table); + spin_unlock(&c2dev->pd_table.lock); +} + +int c2_init_pd_table(struct c2_dev *c2dev) +{ + + c2dev->pd_table.last = 0; + c2dev->pd_table.max = c2dev->props.max_pd; + spin_lock_init(&c2dev->pd_table.lock); + c2dev->pd_table.table = kmalloc(BITS_TO_LONGS(c2dev->props.max_pd) * + sizeof(long), GFP_KERNEL); + if (!c2dev->pd_table.table) + return -ENOMEM; + bitmap_zero(c2dev->pd_table.table, c2dev->props.max_pd); + return 0; +} + +void c2_cleanup_pd_table(struct c2_dev *c2dev) +{ + kfree(c2dev->pd_table.table); +} diff --git a/drivers/infiniband/hw/amso1100/c2_provider.c b/drivers/infiniband/hw/amso1100/c2_provider.c new file mode 100644 index 000000000..bdf350781 --- /dev/null +++ b/drivers/infiniband/hw/amso1100/c2_provider.c @@ -0,0 +1,882 @@ +/* + * Copyright (c) 2005 Ammasso, Inc. All rights reserved. + * Copyright (c) 2005 Open Grid Computing, Inc. All rights reserved. + * + * This software is available to you under a choice of one of two + * licenses. You may choose to be licensed under the terms of the GNU + * General Public License (GPL) Version 2, available from the file + * COPYING in the main directory of this source tree, or the + * OpenIB.org BSD license below: + * + * Redistribution and use in source and binary forms, with or + * without modification, are permitted provided that the following + * conditions are met: + * + * - Redistributions of source code must retain the above + * copyright notice, this list of conditions and the following + * disclaimer. + * + * - Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and/or other materials + * provided with the distribution. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS + * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN + * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + * + */ + +#include <linux/module.h> +#include <linux/moduleparam.h> +#include <linux/pci.h> +#include <linux/netdevice.h> +#include <linux/etherdevice.h> +#include <linux/inetdevice.h> +#include <linux/delay.h> +#include <linux/ethtool.h> +#include <linux/mii.h> +#include <linux/if_vlan.h> +#include <linux/crc32.h> +#include <linux/in.h> +#include <linux/ip.h> +#include <linux/tcp.h> +#include <linux/init.h> +#include <linux/dma-mapping.h> +#include <linux/if_arp.h> +#include <linux/vmalloc.h> +#include <linux/slab.h> + +#include <asm/io.h> +#include <asm/irq.h> +#include <asm/byteorder.h> + +#include <rdma/ib_smi.h> +#include <rdma/ib_umem.h> +#include <rdma/ib_user_verbs.h> +#include "c2.h" +#include "c2_provider.h" +#include "c2_user.h" + +static int c2_query_device(struct ib_device *ibdev, + struct ib_device_attr *props) +{ + struct c2_dev *c2dev = to_c2dev(ibdev); + + pr_debug("%s:%u\n", __func__, __LINE__); + + *props = c2dev->props; + return 0; +} + +static int c2_query_port(struct ib_device *ibdev, + u8 port, struct ib_port_attr *props) +{ + pr_debug("%s:%u\n", __func__, __LINE__); + + props->max_mtu = IB_MTU_4096; + props->lid = 0; + props->lmc = 0; + props->sm_lid = 0; + props->sm_sl = 0; + props->state = IB_PORT_ACTIVE; + props->phys_state = 0; + props->port_cap_flags = + IB_PORT_CM_SUP | + IB_PORT_REINIT_SUP | + IB_PORT_VENDOR_CLASS_SUP | IB_PORT_BOOT_MGMT_SUP; + props->gid_tbl_len = 1; + props->pkey_tbl_len = 1; + props->qkey_viol_cntr = 0; + props->active_width = 1; + props->active_speed = IB_SPEED_SDR; + + return 0; +} + +static int c2_query_pkey(struct ib_device *ibdev, + u8 port, u16 index, u16 * pkey) +{ + pr_debug("%s:%u\n", __func__, __LINE__); + *pkey = 0; + return 0; +} + +static int c2_query_gid(struct ib_device *ibdev, u8 port, + int index, union ib_gid *gid) +{ + struct c2_dev *c2dev = to_c2dev(ibdev); + + pr_debug("%s:%u\n", __func__, __LINE__); + memset(&(gid->raw[0]), 0, sizeof(gid->raw)); + memcpy(&(gid->raw[0]), c2dev->pseudo_netdev->dev_addr, 6); + + return 0; +} + +/* Allocate the user context data structure. This keeps track + * of all objects associated with a particular user-mode client. + */ +static struct ib_ucontext *c2_alloc_ucontext(struct ib_device *ibdev, + struct ib_udata *udata) +{ + struct c2_ucontext *context; + + pr_debug("%s:%u\n", __func__, __LINE__); + context = kmalloc(sizeof(*context), GFP_KERNEL); + if (!context) + return ERR_PTR(-ENOMEM); + + return &context->ibucontext; +} + +static int c2_dealloc_ucontext(struct ib_ucontext *context) +{ + pr_debug("%s:%u\n", __func__, __LINE__); + kfree(context); + return 0; +} + +static int c2_mmap_uar(struct ib_ucontext *context, struct vm_area_struct *vma) +{ + pr_debug("%s:%u\n", __func__, __LINE__); + return -ENOSYS; +} + +static struct ib_pd *c2_alloc_pd(struct ib_device *ibdev, + struct ib_ucontext *context, + struct ib_udata *udata) +{ + struct c2_pd *pd; + int err; + + pr_debug("%s:%u\n", __func__, __LINE__); + + pd = kmalloc(sizeof(*pd), GFP_KERNEL); + if (!pd) + return ERR_PTR(-ENOMEM); + + err = c2_pd_alloc(to_c2dev(ibdev), !context, pd); + if (err) { + kfree(pd); + return ERR_PTR(err); + } + + if (context) { + if (ib_copy_to_udata(udata, &pd->pd_id, sizeof(__u32))) { + c2_pd_free(to_c2dev(ibdev), pd); + kfree(pd); + return ERR_PTR(-EFAULT); + } + } + + return &pd->ibpd; +} + +static int c2_dealloc_pd(struct ib_pd *pd) +{ + pr_debug("%s:%u\n", __func__, __LINE__); + c2_pd_free(to_c2dev(pd->device), to_c2pd(pd)); + kfree(pd); + + return 0; +} + +static struct ib_ah *c2_ah_create(struct ib_pd *pd, struct ib_ah_attr *ah_attr) +{ + pr_debug("%s:%u\n", __func__, __LINE__); + return ERR_PTR(-ENOSYS); +} + +static int c2_ah_destroy(struct ib_ah *ah) +{ + pr_debug("%s:%u\n", __func__, __LINE__); + return -ENOSYS; +} + +static void c2_add_ref(struct ib_qp *ibqp) +{ + struct c2_qp *qp; + BUG_ON(!ibqp); + qp = to_c2qp(ibqp); + atomic_inc(&qp->refcount); +} + +static void c2_rem_ref(struct ib_qp *ibqp) +{ + struct c2_qp *qp; + BUG_ON(!ibqp); + qp = to_c2qp(ibqp); + if (atomic_dec_and_test(&qp->refcount)) + wake_up(&qp->wait); +} + +struct ib_qp *c2_get_qp(struct ib_device *device, int qpn) +{ + struct c2_dev* c2dev = to_c2dev(device); + struct c2_qp *qp; + + qp = c2_find_qpn(c2dev, qpn); + pr_debug("%s Returning QP=%p for QPN=%d, device=%p, refcount=%d\n", + __func__, qp, qpn, device, + (qp?atomic_read(&qp->refcount):0)); + + return (qp?&qp->ibqp:NULL); +} + +static struct ib_qp *c2_create_qp(struct ib_pd *pd, + struct ib_qp_init_attr *init_attr, + struct ib_udata *udata) +{ + struct c2_qp *qp; + int err; + + pr_debug("%s:%u\n", __func__, __LINE__); + + if (init_attr->create_flags) + return ERR_PTR(-EINVAL); + + switch (init_attr->qp_type) { + case IB_QPT_RC: + qp = kzalloc(sizeof(*qp), GFP_KERNEL); + if (!qp) { + pr_debug("%s: Unable to allocate QP\n", __func__); + return ERR_PTR(-ENOMEM); + } + spin_lock_init(&qp->lock); + if (pd->uobject) { + /* userspace specific */ + } + + err = c2_alloc_qp(to_c2dev(pd->device), + to_c2pd(pd), init_attr, qp); + + if (err && pd->uobject) { + /* userspace specific */ + } + + break; + default: + pr_debug("%s: Invalid QP type: %d\n", __func__, + init_attr->qp_type); + return ERR_PTR(-EINVAL); + } + + if (err) { + kfree(qp); + return ERR_PTR(err); + } + + return &qp->ibqp; +} + +static int c2_destroy_qp(struct ib_qp *ib_qp) +{ + struct c2_qp *qp = to_c2qp(ib_qp); + + pr_debug("%s:%u qp=%p,qp->state=%d\n", + __func__, __LINE__, ib_qp, qp->state); + c2_free_qp(to_c2dev(ib_qp->device), qp); + kfree(qp); + return 0; +} + +static struct ib_cq *c2_create_cq(struct ib_device *ibdev, int entries, int vector, + struct ib_ucontext *context, + struct ib_udata *udata) +{ + struct c2_cq *cq; + int err; + + cq = kmalloc(sizeof(*cq), GFP_KERNEL); + if (!cq) { + pr_debug("%s: Unable to allocate CQ\n", __func__); + return ERR_PTR(-ENOMEM); + } + + err = c2_init_cq(to_c2dev(ibdev), entries, NULL, cq); + if (err) { + pr_debug("%s: error initializing CQ\n", __func__); + kfree(cq); + return ERR_PTR(err); + } + + return &cq->ibcq; +} + +static int c2_destroy_cq(struct ib_cq *ib_cq) +{ + struct c2_cq *cq = to_c2cq(ib_cq); + + pr_debug("%s:%u\n", __func__, __LINE__); + + c2_free_cq(to_c2dev(ib_cq->device), cq); + kfree(cq); + + return 0; +} + +static inline u32 c2_convert_access(int acc) +{ + return (acc & IB_ACCESS_REMOTE_WRITE ? C2_ACF_REMOTE_WRITE : 0) | + (acc & IB_ACCESS_REMOTE_READ ? C2_ACF_REMOTE_READ : 0) | + (acc & IB_ACCESS_LOCAL_WRITE ? C2_ACF_LOCAL_WRITE : 0) | + C2_ACF_LOCAL_READ | C2_ACF_WINDOW_BIND; +} + +static struct ib_mr *c2_reg_phys_mr(struct ib_pd *ib_pd, + struct ib_phys_buf *buffer_list, + int num_phys_buf, int acc, u64 * iova_start) +{ + struct c2_mr *mr; + u64 *page_list; + u32 total_len; + int err, i, j, k, page_shift, pbl_depth; + + pbl_depth = 0; + total_len = 0; + + page_shift = PAGE_SHIFT; + /* + * If there is only 1 buffer we assume this could + * be a map of all phy mem...use a 32k page_shift. + */ + if (num_phys_buf == 1) + page_shift += 3; + + for (i = 0; i < num_phys_buf; i++) { + + if (buffer_list[i].addr & ~PAGE_MASK) { + pr_debug("Unaligned Memory Buffer: 0x%x\n", + (unsigned int) buffer_list[i].addr); + return ERR_PTR(-EINVAL); + } + + if (!buffer_list[i].size) { + pr_debug("Invalid Buffer Size\n"); + return ERR_PTR(-EINVAL); + } + + total_len += buffer_list[i].size; + pbl_depth += ALIGN(buffer_list[i].size, + (1 << page_shift)) >> page_shift; + } + + page_list = vmalloc(sizeof(u64) * pbl_depth); + if (!page_list) { + pr_debug("couldn't vmalloc page_list of size %zd\n", + (sizeof(u64) * pbl_depth)); + return ERR_PTR(-ENOMEM); + } + + for (i = 0, j = 0; i < num_phys_buf; i++) { + + int naddrs; + + naddrs = ALIGN(buffer_list[i].size, + (1 << page_shift)) >> page_shift; + for (k = 0; k < naddrs; k++) + page_list[j++] = (buffer_list[i].addr + + (k << page_shift)); + } + + mr = kmalloc(sizeof(*mr), GFP_KERNEL); + if (!mr) { + vfree(page_list); + return ERR_PTR(-ENOMEM); + } + + mr->pd = to_c2pd(ib_pd); + mr->umem = NULL; + pr_debug("%s - page shift %d, pbl_depth %d, total_len %u, " + "*iova_start %llx, first pa %llx, last pa %llx\n", + __func__, page_shift, pbl_depth, total_len, + (unsigned long long) *iova_start, + (unsigned long long) page_list[0], + (unsigned long long) page_list[pbl_depth-1]); + err = c2_nsmr_register_phys_kern(to_c2dev(ib_pd->device), page_list, + (1 << page_shift), pbl_depth, + total_len, 0, iova_start, + c2_convert_access(acc), mr); + vfree(page_list); + if (err) { + kfree(mr); + return ERR_PTR(err); + } + + return &mr->ibmr; +} + +static struct ib_mr *c2_get_dma_mr(struct ib_pd *pd, int acc) +{ + struct ib_phys_buf bl; + u64 kva = 0; + + pr_debug("%s:%u\n", __func__, __LINE__); + + /* AMSO1100 limit */ + bl.size = 0xffffffff; + bl.addr = 0; + return c2_reg_phys_mr(pd, &bl, 1, acc, &kva); +} + +static struct ib_mr *c2_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, + u64 virt, int acc, struct ib_udata *udata) +{ + u64 *pages; + u64 kva = 0; + int shift, n, len; + int i, k, entry; + int err = 0; + struct scatterlist *sg; + struct c2_pd *c2pd = to_c2pd(pd); + struct c2_mr *c2mr; + + pr_debug("%s:%u\n", __func__, __LINE__); + + c2mr = kmalloc(sizeof(*c2mr), GFP_KERNEL); + if (!c2mr) + return ERR_PTR(-ENOMEM); + c2mr->pd = c2pd; + + c2mr->umem = ib_umem_get(pd->uobject->context, start, length, acc, 0); + if (IS_ERR(c2mr->umem)) { + err = PTR_ERR(c2mr->umem); + kfree(c2mr); + return ERR_PTR(err); + } + + shift = ffs(c2mr->umem->page_size) - 1; + n = c2mr->umem->nmap; + + pages = kmalloc(n * sizeof(u64), GFP_KERNEL); + if (!pages) { + err = -ENOMEM; + goto err; + } + + i = 0; + for_each_sg(c2mr->umem->sg_head.sgl, sg, c2mr->umem->nmap, entry) { + len = sg_dma_len(sg) >> shift; + for (k = 0; k < len; ++k) { + pages[i++] = + sg_dma_address(sg) + + (c2mr->umem->page_size * k); + } + } + + kva = virt; + err = c2_nsmr_register_phys_kern(to_c2dev(pd->device), + pages, + c2mr->umem->page_size, + i, + length, + ib_umem_offset(c2mr->umem), + &kva, + c2_convert_access(acc), + c2mr); + kfree(pages); + if (err) + goto err; + return &c2mr->ibmr; + +err: + ib_umem_release(c2mr->umem); + kfree(c2mr); + return ERR_PTR(err); +} + +static int c2_dereg_mr(struct ib_mr *ib_mr) +{ + struct c2_mr *mr = to_c2mr(ib_mr); + int err; + + pr_debug("%s:%u\n", __func__, __LINE__); + + err = c2_stag_dealloc(to_c2dev(ib_mr->device), ib_mr->lkey); + if (err) + pr_debug("c2_stag_dealloc failed: %d\n", err); + else { + if (mr->umem) + ib_umem_release(mr->umem); + kfree(mr); + } + + return err; +} + +static ssize_t show_rev(struct device *dev, struct device_attribute *attr, + char *buf) +{ + struct c2_dev *c2dev = container_of(dev, struct c2_dev, ibdev.dev); + pr_debug("%s:%u\n", __func__, __LINE__); + return sprintf(buf, "%x\n", c2dev->props.hw_ver); +} + +static ssize_t show_fw_ver(struct device *dev, struct device_attribute *attr, + char *buf) +{ + struct c2_dev *c2dev = container_of(dev, struct c2_dev, ibdev.dev); + pr_debug("%s:%u\n", __func__, __LINE__); + return sprintf(buf, "%x.%x.%x\n", + (int) (c2dev->props.fw_ver >> 32), + (int) (c2dev->props.fw_ver >> 16) & 0xffff, + (int) (c2dev->props.fw_ver & 0xffff)); +} + +static ssize_t show_hca(struct device *dev, struct device_attribute *attr, + char *buf) +{ + pr_debug("%s:%u\n", __func__, __LINE__); + return sprintf(buf, "AMSO1100\n"); +} + +static ssize_t show_board(struct device *dev, struct device_attribute *attr, + char *buf) +{ + pr_debug("%s:%u\n", __func__, __LINE__); + return sprintf(buf, "%.*s\n", 32, "AMSO1100 Board ID"); +} + +static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL); +static DEVICE_ATTR(fw_ver, S_IRUGO, show_fw_ver, NULL); +static DEVICE_ATTR(hca_type, S_IRUGO, show_hca, NULL); +static DEVICE_ATTR(board_id, S_IRUGO, show_board, NULL); + +static struct device_attribute *c2_dev_attributes[] = { + &dev_attr_hw_rev, + &dev_attr_fw_ver, + &dev_attr_hca_type, + &dev_attr_board_id +}; + +static int c2_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, + int attr_mask, struct ib_udata *udata) +{ + int err; + + err = + c2_qp_modify(to_c2dev(ibqp->device), to_c2qp(ibqp), attr, + attr_mask); + + return err; +} + +static int c2_multicast_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid) +{ + pr_debug("%s:%u\n", __func__, __LINE__); + return -ENOSYS; +} + +static int c2_multicast_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid) +{ + pr_debug("%s:%u\n", __func__, __LINE__); + return -ENOSYS; +} + +static int c2_process_mad(struct ib_device *ibdev, + int mad_flags, + u8 port_num, + struct ib_wc *in_wc, + struct ib_grh *in_grh, + struct ib_mad *in_mad, struct ib_mad *out_mad) +{ + pr_debug("%s:%u\n", __func__, __LINE__); + return -ENOSYS; +} + +static int c2_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *iw_param) +{ + pr_debug("%s:%u\n", __func__, __LINE__); + + /* Request a connection */ + return c2_llp_connect(cm_id, iw_param); +} + +static int c2_accept(struct iw_cm_id *cm_id, struct iw_cm_conn_param *iw_param) +{ + pr_debug("%s:%u\n", __func__, __LINE__); + + /* Accept the new connection */ + return c2_llp_accept(cm_id, iw_param); +} + +static int c2_reject(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len) +{ + int err; + + pr_debug("%s:%u\n", __func__, __LINE__); + + err = c2_llp_reject(cm_id, pdata, pdata_len); + return err; +} + +static int c2_service_create(struct iw_cm_id *cm_id, int backlog) +{ + int err; + + pr_debug("%s:%u\n", __func__, __LINE__); + err = c2_llp_service_create(cm_id, backlog); + pr_debug("%s:%u err=%d\n", + __func__, __LINE__, + err); + return err; +} + +static int c2_service_destroy(struct iw_cm_id *cm_id) +{ + int err; + pr_debug("%s:%u\n", __func__, __LINE__); + + err = c2_llp_service_destroy(cm_id); + + return err; +} + +static int c2_pseudo_up(struct net_device *netdev) +{ + struct in_device *ind; + struct c2_dev *c2dev = netdev->ml_priv; + + ind = in_dev_get(netdev); + if (!ind) + return 0; + + pr_debug("adding...\n"); + for_ifa(ind) { +#ifdef DEBUG + u8 *ip = (u8 *) & ifa->ifa_address; + + pr_debug("%s: %d.%d.%d.%d\n", + ifa->ifa_label, ip[0], ip[1], ip[2], ip[3]); +#endif + c2_add_addr(c2dev, ifa->ifa_address, ifa->ifa_mask); + } + endfor_ifa(ind); + in_dev_put(ind); + + return 0; +} + +static int c2_pseudo_down(struct net_device *netdev) +{ + struct in_device *ind; + struct c2_dev *c2dev = netdev->ml_priv; + + ind = in_dev_get(netdev); + if (!ind) + return 0; + + pr_debug("deleting...\n"); + for_ifa(ind) { +#ifdef DEBUG + u8 *ip = (u8 *) & ifa->ifa_address; + + pr_debug("%s: %d.%d.%d.%d\n", + ifa->ifa_label, ip[0], ip[1], ip[2], ip[3]); +#endif + c2_del_addr(c2dev, ifa->ifa_address, ifa->ifa_mask); + } + endfor_ifa(ind); + in_dev_put(ind); + + return 0; +} + +static int c2_pseudo_xmit_frame(struct sk_buff *skb, struct net_device *netdev) +{ + kfree_skb(skb); + return NETDEV_TX_OK; +} + +static int c2_pseudo_change_mtu(struct net_device *netdev, int new_mtu) +{ + if (new_mtu < ETH_ZLEN || new_mtu > ETH_JUMBO_MTU) + return -EINVAL; + + netdev->mtu = new_mtu; + + /* TODO: Tell rnic about new rmda interface mtu */ + return 0; +} + +static const struct net_device_ops c2_pseudo_netdev_ops = { + .ndo_open = c2_pseudo_up, + .ndo_stop = c2_pseudo_down, + .ndo_start_xmit = c2_pseudo_xmit_frame, + .ndo_change_mtu = c2_pseudo_change_mtu, + .ndo_validate_addr = eth_validate_addr, +}; + +static void setup(struct net_device *netdev) +{ + netdev->netdev_ops = &c2_pseudo_netdev_ops; + + netdev->watchdog_timeo = 0; + netdev->type = ARPHRD_ETHER; + netdev->mtu = 1500; + netdev->hard_header_len = ETH_HLEN; + netdev->addr_len = ETH_ALEN; + netdev->tx_queue_len = 0; + netdev->flags |= IFF_NOARP; +} + +static struct net_device *c2_pseudo_netdev_init(struct c2_dev *c2dev) +{ + char name[IFNAMSIZ]; + struct net_device *netdev; + + /* change ethxxx to iwxxx */ + strcpy(name, "iw"); + strcat(name, &c2dev->netdev->name[3]); + netdev = alloc_netdev(0, name, NET_NAME_UNKNOWN, setup); + if (!netdev) { + printk(KERN_ERR PFX "%s - etherdev alloc failed", + __func__); + return NULL; + } + + netdev->ml_priv = c2dev; + + SET_NETDEV_DEV(netdev, &c2dev->pcidev->dev); + + memcpy_fromio(netdev->dev_addr, c2dev->kva + C2_REGS_RDMA_ENADDR, 6); + + /* Print out the MAC address */ + pr_debug("%s: MAC %pM\n", netdev->name, netdev->dev_addr); + +#if 0 + /* Disable network packets */ + netif_stop_queue(netdev); +#endif + return netdev; +} + +int c2_register_device(struct c2_dev *dev) +{ + int ret = -ENOMEM; + int i; + + /* Register pseudo network device */ + dev->pseudo_netdev = c2_pseudo_netdev_init(dev); + if (!dev->pseudo_netdev) + goto out; + + ret = register_netdev(dev->pseudo_netdev); + if (ret) + goto out_free_netdev; + + pr_debug("%s:%u\n", __func__, __LINE__); + strlcpy(dev->ibdev.name, "amso%d", IB_DEVICE_NAME_MAX); + dev->ibdev.owner = THIS_MODULE; + dev->ibdev.uverbs_cmd_mask = + (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) | + (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE) | + (1ull << IB_USER_VERBS_CMD_QUERY_PORT) | + (1ull << IB_USER_VERBS_CMD_ALLOC_PD) | + (1ull << IB_USER_VERBS_CMD_DEALLOC_PD) | + (1ull << IB_USER_VERBS_CMD_REG_MR) | + (1ull << IB_USER_VERBS_CMD_DEREG_MR) | + (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) | + (1ull << IB_USER_VERBS_CMD_CREATE_CQ) | + (1ull << IB_USER_VERBS_CMD_DESTROY_CQ) | + (1ull << IB_USER_VERBS_CMD_REQ_NOTIFY_CQ) | + (1ull << IB_USER_VERBS_CMD_CREATE_QP) | + (1ull << IB_USER_VERBS_CMD_MODIFY_QP) | + (1ull << IB_USER_VERBS_CMD_POLL_CQ) | + (1ull << IB_USER_VERBS_CMD_DESTROY_QP) | + (1ull << IB_USER_VERBS_CMD_POST_SEND) | + (1ull << IB_USER_VERBS_CMD_POST_RECV); + + dev->ibdev.node_type = RDMA_NODE_RNIC; + memset(&dev->ibdev.node_guid, 0, sizeof(dev->ibdev.node_guid)); + memcpy(&dev->ibdev.node_guid, dev->pseudo_netdev->dev_addr, 6); + dev->ibdev.phys_port_cnt = 1; + dev->ibdev.num_comp_vectors = 1; + dev->ibdev.dma_device = &dev->pcidev->dev; + dev->ibdev.query_device = c2_query_device; + dev->ibdev.query_port = c2_query_port; + dev->ibdev.query_pkey = c2_query_pkey; + dev->ibdev.query_gid = c2_query_gid; + dev->ibdev.alloc_ucontext = c2_alloc_ucontext; + dev->ibdev.dealloc_ucontext = c2_dealloc_ucontext; + dev->ibdev.mmap = c2_mmap_uar; + dev->ibdev.alloc_pd = c2_alloc_pd; + dev->ibdev.dealloc_pd = c2_dealloc_pd; + dev->ibdev.create_ah = c2_ah_create; + dev->ibdev.destroy_ah = c2_ah_destroy; + dev->ibdev.create_qp = c2_create_qp; + dev->ibdev.modify_qp = c2_modify_qp; + dev->ibdev.destroy_qp = c2_destroy_qp; + dev->ibdev.create_cq = c2_create_cq; + dev->ibdev.destroy_cq = c2_destroy_cq; + dev->ibdev.poll_cq = c2_poll_cq; + dev->ibdev.get_dma_mr = c2_get_dma_mr; + dev->ibdev.reg_phys_mr = c2_reg_phys_mr; + dev->ibdev.reg_user_mr = c2_reg_user_mr; + dev->ibdev.dereg_mr = c2_dereg_mr; + + dev->ibdev.alloc_fmr = NULL; + dev->ibdev.unmap_fmr = NULL; + dev->ibdev.dealloc_fmr = NULL; + dev->ibdev.map_phys_fmr = NULL; + + dev->ibdev.attach_mcast = c2_multicast_attach; + dev->ibdev.detach_mcast = c2_multicast_detach; + dev->ibdev.process_mad = c2_process_mad; + + dev->ibdev.req_notify_cq = c2_arm_cq; + dev->ibdev.post_send = c2_post_send; + dev->ibdev.post_recv = c2_post_receive; + + dev->ibdev.iwcm = kmalloc(sizeof(*dev->ibdev.iwcm), GFP_KERNEL); + if (dev->ibdev.iwcm == NULL) { + ret = -ENOMEM; + goto out_unregister_netdev; + } + dev->ibdev.iwcm->add_ref = c2_add_ref; + dev->ibdev.iwcm->rem_ref = c2_rem_ref; + dev->ibdev.iwcm->get_qp = c2_get_qp; + dev->ibdev.iwcm->connect = c2_connect; + dev->ibdev.iwcm->accept = c2_accept; + dev->ibdev.iwcm->reject = c2_reject; + dev->ibdev.iwcm->create_listen = c2_service_create; + dev->ibdev.iwcm->destroy_listen = c2_service_destroy; + + ret = ib_register_device(&dev->ibdev, NULL); + if (ret) + goto out_free_iwcm; + + for (i = 0; i < ARRAY_SIZE(c2_dev_attributes); ++i) { + ret = device_create_file(&dev->ibdev.dev, + c2_dev_attributes[i]); + if (ret) + goto out_unregister_ibdev; + } + goto out; + +out_unregister_ibdev: + ib_unregister_device(&dev->ibdev); +out_free_iwcm: + kfree(dev->ibdev.iwcm); +out_unregister_netdev: + unregister_netdev(dev->pseudo_netdev); +out_free_netdev: + free_netdev(dev->pseudo_netdev); +out: + pr_debug("%s:%u ret=%d\n", __func__, __LINE__, ret); + return ret; +} + +void c2_unregister_device(struct c2_dev *dev) +{ + pr_debug("%s:%u\n", __func__, __LINE__); + unregister_netdev(dev->pseudo_netdev); + free_netdev(dev->pseudo_netdev); + ib_unregister_device(&dev->ibdev); +} diff --git a/drivers/infiniband/hw/amso1100/c2_provider.h b/drivers/infiniband/hw/amso1100/c2_provider.h new file mode 100644 index 000000000..bf1899877 --- /dev/null +++ b/drivers/infiniband/hw/amso1100/c2_provider.h @@ -0,0 +1,182 @@ +/* + * Copyright (c) 2005 Ammasso, Inc. All rights reserved. + * Copyright (c) 2005 Open Grid Computing, Inc. All rights reserved. + * + * This software is available to you under a choice of one of two + * licenses. You may choose to be licensed under the terms of the GNU + * General Public License (GPL) Version 2, available from the file + * COPYING in the main directory of this source tree, or the + * OpenIB.org BSD license below: + * + * Redistribution and use in source and binary forms, with or + * without modification, are permitted provided that the following + * conditions are met: + * + * - Redistributions of source code must retain the above + * copyright notice, this list of conditions and the following + * disclaimer. + * + * - Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and/or other materials + * provided with the distribution. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS + * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN + * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + * + */ + +#ifndef C2_PROVIDER_H +#define C2_PROVIDER_H +#include <linux/inetdevice.h> + +#include <rdma/ib_verbs.h> +#include <rdma/ib_pack.h> + +#include "c2_mq.h" +#include <rdma/iw_cm.h> + +#define C2_MPT_FLAG_ATOMIC (1 << 14) +#define C2_MPT_FLAG_REMOTE_WRITE (1 << 13) +#define C2_MPT_FLAG_REMOTE_READ (1 << 12) +#define C2_MPT_FLAG_LOCAL_WRITE (1 << 11) +#define C2_MPT_FLAG_LOCAL_READ (1 << 10) + +struct c2_buf_list { + void *buf; + DEFINE_DMA_UNMAP_ADDR(mapping); +}; + + +/* The user context keeps track of objects allocated for a + * particular user-mode client. */ +struct c2_ucontext { + struct ib_ucontext ibucontext; +}; + +struct c2_mtt; + +/* All objects associated with a PD are kept in the + * associated user context if present. + */ +struct c2_pd { + struct ib_pd ibpd; + u32 pd_id; +}; + +struct c2_mr { + struct ib_mr ibmr; + struct c2_pd *pd; + struct ib_umem *umem; +}; + +struct c2_av; + +enum c2_ah_type { + C2_AH_ON_HCA, + C2_AH_PCI_POOL, + C2_AH_KMALLOC +}; + +struct c2_ah { + struct ib_ah ibah; +}; + +struct c2_cq { + struct ib_cq ibcq; + spinlock_t lock; + atomic_t refcount; + int cqn; + int is_kernel; + wait_queue_head_t wait; + + u32 adapter_handle; + struct c2_mq mq; +}; + +struct c2_wq { + spinlock_t lock; +}; +struct iw_cm_id; +struct c2_qp { + struct ib_qp ibqp; + struct iw_cm_id *cm_id; + spinlock_t lock; + atomic_t refcount; + wait_queue_head_t wait; + int qpn; + + u32 adapter_handle; + u32 send_sgl_depth; + u32 recv_sgl_depth; + u32 rdma_write_sgl_depth; + u8 state; + + struct c2_mq sq_mq; + struct c2_mq rq_mq; +}; + +struct c2_cr_query_attrs { + u32 local_addr; + u32 remote_addr; + u16 local_port; + u16 remote_port; +}; + +static inline struct c2_pd *to_c2pd(struct ib_pd *ibpd) +{ + return container_of(ibpd, struct c2_pd, ibpd); +} + +static inline struct c2_ucontext *to_c2ucontext(struct ib_ucontext *ibucontext) +{ + return container_of(ibucontext, struct c2_ucontext, ibucontext); +} + +static inline struct c2_mr *to_c2mr(struct ib_mr *ibmr) +{ + return container_of(ibmr, struct c2_mr, ibmr); +} + + +static inline struct c2_ah *to_c2ah(struct ib_ah *ibah) +{ + return container_of(ibah, struct c2_ah, ibah); +} + +static inline struct c2_cq *to_c2cq(struct ib_cq *ibcq) +{ + return container_of(ibcq, struct c2_cq, ibcq); +} + +static inline struct c2_qp *to_c2qp(struct ib_qp *ibqp) +{ + return container_of(ibqp, struct c2_qp, ibqp); +} + +static inline int is_rnic_addr(struct net_device *netdev, u32 addr) +{ + struct in_device *ind; + int ret = 0; + + ind = in_dev_get(netdev); + if (!ind) + return 0; + + for_ifa(ind) { + if (ifa->ifa_address == addr) { + ret = 1; + break; + } + } + endfor_ifa(ind); + in_dev_put(ind); + return ret; +} +#endif /* C2_PROVIDER_H */ diff --git a/drivers/infiniband/hw/amso1100/c2_qp.c b/drivers/infiniband/hw/amso1100/c2_qp.c new file mode 100644 index 000000000..86708dee5 --- /dev/null +++ b/drivers/infiniband/hw/amso1100/c2_qp.c @@ -0,0 +1,1024 @@ +/* + * Copyright (c) 2004 Topspin Communications. All rights reserved. + * Copyright (c) 2005 Cisco Systems. All rights reserved. + * Copyright (c) 2005 Mellanox Technologies. All rights reserved. + * Copyright (c) 2004 Voltaire, Inc. All rights reserved. + * Copyright (c) 2005 Open Grid Computing, Inc. All rights reserved. + * + * This software is available to you under a choice of one of two + * licenses. You may choose to be licensed under the terms of the GNU + * General Public License (GPL) Version 2, available from the file + * COPYING in the main directory of this source tree, or the + * OpenIB.org BSD license below: + * + * Redistribution and use in source and binary forms, with or + * without modification, are permitted provided that the following + * conditions are met: + * + * - Redistributions of source code must retain the above + * copyright notice, this list of conditions and the following + * disclaimer. + * + * - Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and/or other materials + * provided with the distribution. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS + * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN + * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + * + */ + +#include <linux/delay.h> +#include <linux/gfp.h> + +#include "c2.h" +#include "c2_vq.h" +#include "c2_status.h" + +#define C2_MAX_ORD_PER_QP 128 +#define C2_MAX_IRD_PER_QP 128 + +#define C2_HINT_MAKE(q_index, hint_count) (((q_index) << 16) | hint_count) +#define C2_HINT_GET_INDEX(hint) (((hint) & 0x7FFF0000) >> 16) +#define C2_HINT_GET_COUNT(hint) ((hint) & 0x0000FFFF) + +#define NO_SUPPORT -1 +static const u8 c2_opcode[] = { + [IB_WR_SEND] = C2_WR_TYPE_SEND, + [IB_WR_SEND_WITH_IMM] = NO_SUPPORT, + [IB_WR_RDMA_WRITE] = C2_WR_TYPE_RDMA_WRITE, + [IB_WR_RDMA_WRITE_WITH_IMM] = NO_SUPPORT, + [IB_WR_RDMA_READ] = C2_WR_TYPE_RDMA_READ, + [IB_WR_ATOMIC_CMP_AND_SWP] = NO_SUPPORT, + [IB_WR_ATOMIC_FETCH_AND_ADD] = NO_SUPPORT, +}; + +static int to_c2_state(enum ib_qp_state ib_state) +{ + switch (ib_state) { + case IB_QPS_RESET: + return C2_QP_STATE_IDLE; + case IB_QPS_RTS: + return C2_QP_STATE_RTS; + case IB_QPS_SQD: + return C2_QP_STATE_CLOSING; + case IB_QPS_SQE: + return C2_QP_STATE_CLOSING; + case IB_QPS_ERR: + return C2_QP_STATE_ERROR; + default: + return -1; + } +} + +static int to_ib_state(enum c2_qp_state c2_state) +{ + switch (c2_state) { + case C2_QP_STATE_IDLE: + return IB_QPS_RESET; + case C2_QP_STATE_CONNECTING: + return IB_QPS_RTR; + case C2_QP_STATE_RTS: + return IB_QPS_RTS; + case C2_QP_STATE_CLOSING: + return IB_QPS_SQD; + case C2_QP_STATE_ERROR: + return IB_QPS_ERR; + case C2_QP_STATE_TERMINATE: + return IB_QPS_SQE; + default: + return -1; + } +} + +static const char *to_ib_state_str(int ib_state) +{ + static const char *state_str[] = { + "IB_QPS_RESET", + "IB_QPS_INIT", + "IB_QPS_RTR", + "IB_QPS_RTS", + "IB_QPS_SQD", + "IB_QPS_SQE", + "IB_QPS_ERR" + }; + if (ib_state < IB_QPS_RESET || + ib_state > IB_QPS_ERR) + return "<invalid IB QP state>"; + + ib_state -= IB_QPS_RESET; + return state_str[ib_state]; +} + +void c2_set_qp_state(struct c2_qp *qp, int c2_state) +{ + int new_state = to_ib_state(c2_state); + + pr_debug("%s: qp[%p] state modify %s --> %s\n", + __func__, + qp, + to_ib_state_str(qp->state), + to_ib_state_str(new_state)); + qp->state = new_state; +} + +#define C2_QP_NO_ATTR_CHANGE 0xFFFFFFFF + +int c2_qp_modify(struct c2_dev *c2dev, struct c2_qp *qp, + struct ib_qp_attr *attr, int attr_mask) +{ + struct c2wr_qp_modify_req wr; + struct c2wr_qp_modify_rep *reply; + struct c2_vq_req *vq_req; + unsigned long flags; + u8 next_state; + int err; + + pr_debug("%s:%d qp=%p, %s --> %s\n", + __func__, __LINE__, + qp, + to_ib_state_str(qp->state), + to_ib_state_str(attr->qp_state)); + + vq_req = vq_req_alloc(c2dev); + if (!vq_req) + return -ENOMEM; + + c2_wr_set_id(&wr, CCWR_QP_MODIFY); + wr.hdr.context = (unsigned long) vq_req; + wr.rnic_handle = c2dev->adapter_handle; + wr.qp_handle = qp->adapter_handle; + wr.ord = cpu_to_be32(C2_QP_NO_ATTR_CHANGE); + wr.ird = cpu_to_be32(C2_QP_NO_ATTR_CHANGE); + wr.sq_depth = cpu_to_be32(C2_QP_NO_ATTR_CHANGE); + wr.rq_depth = cpu_to_be32(C2_QP_NO_ATTR_CHANGE); + + if (attr_mask & IB_QP_STATE) { + /* Ensure the state is valid */ + if (attr->qp_state < 0 || attr->qp_state > IB_QPS_ERR) { + err = -EINVAL; + goto bail0; + } + + wr.next_qp_state = cpu_to_be32(to_c2_state(attr->qp_state)); + + if (attr->qp_state == IB_QPS_ERR) { + spin_lock_irqsave(&qp->lock, flags); + if (qp->cm_id && qp->state == IB_QPS_RTS) { + pr_debug("Generating CLOSE event for QP-->ERR, " + "qp=%p, cm_id=%p\n",qp,qp->cm_id); + /* Generate an CLOSE event */ + vq_req->cm_id = qp->cm_id; + vq_req->event = IW_CM_EVENT_CLOSE; + } + spin_unlock_irqrestore(&qp->lock, flags); + } + next_state = attr->qp_state; + + } else if (attr_mask & IB_QP_CUR_STATE) { + + if (attr->cur_qp_state != IB_QPS_RTR && + attr->cur_qp_state != IB_QPS_RTS && + attr->cur_qp_state != IB_QPS_SQD && + attr->cur_qp_state != IB_QPS_SQE) { + err = -EINVAL; + goto bail0; + } else + wr.next_qp_state = + cpu_to_be32(to_c2_state(attr->cur_qp_state)); + + next_state = attr->cur_qp_state; + + } else { + err = 0; + goto bail0; + } + + /* reference the request struct */ + vq_req_get(c2dev, vq_req); + + err = vq_send_wr(c2dev, (union c2wr *) & wr); + if (err) { + vq_req_put(c2dev, vq_req); + goto bail0; + } + + err = vq_wait_for_reply(c2dev, vq_req); + if (err) + goto bail0; + + reply = (struct c2wr_qp_modify_rep *) (unsigned long) vq_req->reply_msg; + if (!reply) { + err = -ENOMEM; + goto bail0; + } + + err = c2_errno(reply); + if (!err) + qp->state = next_state; +#ifdef DEBUG + else + pr_debug("%s: c2_errno=%d\n", __func__, err); +#endif + /* + * If we're going to error and generating the event here, then + * we need to remove the reference because there will be no + * close event generated by the adapter + */ + spin_lock_irqsave(&qp->lock, flags); + if (vq_req->event==IW_CM_EVENT_CLOSE && qp->cm_id) { + qp->cm_id->rem_ref(qp->cm_id); + qp->cm_id = NULL; + } + spin_unlock_irqrestore(&qp->lock, flags); + + vq_repbuf_free(c2dev, reply); + bail0: + vq_req_free(c2dev, vq_req); + + pr_debug("%s:%d qp=%p, cur_state=%s\n", + __func__, __LINE__, + qp, + to_ib_state_str(qp->state)); + return err; +} + +int c2_qp_set_read_limits(struct c2_dev *c2dev, struct c2_qp *qp, + int ord, int ird) +{ + struct c2wr_qp_modify_req wr; + struct c2wr_qp_modify_rep *reply; + struct c2_vq_req *vq_req; + int err; + + vq_req = vq_req_alloc(c2dev); + if (!vq_req) + return -ENOMEM; + + c2_wr_set_id(&wr, CCWR_QP_MODIFY); + wr.hdr.context = (unsigned long) vq_req; + wr.rnic_handle = c2dev->adapter_handle; + wr.qp_handle = qp->adapter_handle; + wr.ord = cpu_to_be32(ord); + wr.ird = cpu_to_be32(ird); + wr.sq_depth = cpu_to_be32(C2_QP_NO_ATTR_CHANGE); + wr.rq_depth = cpu_to_be32(C2_QP_NO_ATTR_CHANGE); + wr.next_qp_state = cpu_to_be32(C2_QP_NO_ATTR_CHANGE); + + /* reference the request struct */ + vq_req_get(c2dev, vq_req); + + err = vq_send_wr(c2dev, (union c2wr *) & wr); + if (err) { + vq_req_put(c2dev, vq_req); + goto bail0; + } + + err = vq_wait_for_reply(c2dev, vq_req); + if (err) + goto bail0; + + reply = (struct c2wr_qp_modify_rep *) (unsigned long) + vq_req->reply_msg; + if (!reply) { + err = -ENOMEM; + goto bail0; + } + + err = c2_errno(reply); + vq_repbuf_free(c2dev, reply); + bail0: + vq_req_free(c2dev, vq_req); + return err; +} + +static int destroy_qp(struct c2_dev *c2dev, struct c2_qp *qp) +{ + struct c2_vq_req *vq_req; + struct c2wr_qp_destroy_req wr; + struct c2wr_qp_destroy_rep *reply; + unsigned long flags; + int err; + + /* + * Allocate a verb request message + */ + vq_req = vq_req_alloc(c2dev); + if (!vq_req) { + return -ENOMEM; + } + + /* + * Initialize the WR + */ + c2_wr_set_id(&wr, CCWR_QP_DESTROY); + wr.hdr.context = (unsigned long) vq_req; + wr.rnic_handle = c2dev->adapter_handle; + wr.qp_handle = qp->adapter_handle; + + /* + * reference the request struct. dereferenced in the int handler. + */ + vq_req_get(c2dev, vq_req); + + spin_lock_irqsave(&qp->lock, flags); + if (qp->cm_id && qp->state == IB_QPS_RTS) { + pr_debug("destroy_qp: generating CLOSE event for QP-->ERR, " + "qp=%p, cm_id=%p\n",qp,qp->cm_id); + /* Generate an CLOSE event */ + vq_req->qp = qp; + vq_req->cm_id = qp->cm_id; + vq_req->event = IW_CM_EVENT_CLOSE; + } + spin_unlock_irqrestore(&qp->lock, flags); + + /* + * Send WR to adapter + */ + err = vq_send_wr(c2dev, (union c2wr *) & wr); + if (err) { + vq_req_put(c2dev, vq_req); + goto bail0; + } + + /* + * Wait for reply from adapter + */ + err = vq_wait_for_reply(c2dev, vq_req); + if (err) { + goto bail0; + } + + /* + * Process reply + */ + reply = (struct c2wr_qp_destroy_rep *) (unsigned long) (vq_req->reply_msg); + if (!reply) { + err = -ENOMEM; + goto bail0; + } + + spin_lock_irqsave(&qp->lock, flags); + if (qp->cm_id) { + qp->cm_id->rem_ref(qp->cm_id); + qp->cm_id = NULL; + } + spin_unlock_irqrestore(&qp->lock, flags); + + vq_repbuf_free(c2dev, reply); + bail0: + vq_req_free(c2dev, vq_req); + return err; +} + +static int c2_alloc_qpn(struct c2_dev *c2dev, struct c2_qp *qp) +{ + int ret; + + idr_preload(GFP_KERNEL); + spin_lock_irq(&c2dev->qp_table.lock); + + ret = idr_alloc_cyclic(&c2dev->qp_table.idr, qp, 0, 0, GFP_NOWAIT); + if (ret >= 0) + qp->qpn = ret; + + spin_unlock_irq(&c2dev->qp_table.lock); + idr_preload_end(); + return ret < 0 ? ret : 0; +} + +static void c2_free_qpn(struct c2_dev *c2dev, int qpn) +{ + spin_lock_irq(&c2dev->qp_table.lock); + idr_remove(&c2dev->qp_table.idr, qpn); + spin_unlock_irq(&c2dev->qp_table.lock); +} + +struct c2_qp *c2_find_qpn(struct c2_dev *c2dev, int qpn) +{ + unsigned long flags; + struct c2_qp *qp; + + spin_lock_irqsave(&c2dev->qp_table.lock, flags); + qp = idr_find(&c2dev->qp_table.idr, qpn); + spin_unlock_irqrestore(&c2dev->qp_table.lock, flags); + return qp; +} + +int c2_alloc_qp(struct c2_dev *c2dev, + struct c2_pd *pd, + struct ib_qp_init_attr *qp_attrs, struct c2_qp *qp) +{ + struct c2wr_qp_create_req wr; + struct c2wr_qp_create_rep *reply; + struct c2_vq_req *vq_req; + struct c2_cq *send_cq = to_c2cq(qp_attrs->send_cq); + struct c2_cq *recv_cq = to_c2cq(qp_attrs->recv_cq); + unsigned long peer_pa; + u32 q_size, msg_size, mmap_size; + void __iomem *mmap; + int err; + + err = c2_alloc_qpn(c2dev, qp); + if (err) + return err; + qp->ibqp.qp_num = qp->qpn; + qp->ibqp.qp_type = IB_QPT_RC; + + /* Allocate the SQ and RQ shared pointers */ + qp->sq_mq.shared = c2_alloc_mqsp(c2dev, c2dev->kern_mqsp_pool, + &qp->sq_mq.shared_dma, GFP_KERNEL); + if (!qp->sq_mq.shared) { + err = -ENOMEM; + goto bail0; + } + + qp->rq_mq.shared = c2_alloc_mqsp(c2dev, c2dev->kern_mqsp_pool, + &qp->rq_mq.shared_dma, GFP_KERNEL); + if (!qp->rq_mq.shared) { + err = -ENOMEM; + goto bail1; + } + + /* Allocate the verbs request */ + vq_req = vq_req_alloc(c2dev); + if (vq_req == NULL) { + err = -ENOMEM; + goto bail2; + } + + /* Initialize the work request */ + memset(&wr, 0, sizeof(wr)); + c2_wr_set_id(&wr, CCWR_QP_CREATE); + wr.hdr.context = (unsigned long) vq_req; + wr.rnic_handle = c2dev->adapter_handle; + wr.sq_cq_handle = send_cq->adapter_handle; + wr.rq_cq_handle = recv_cq->adapter_handle; + wr.sq_depth = cpu_to_be32(qp_attrs->cap.max_send_wr + 1); + wr.rq_depth = cpu_to_be32(qp_attrs->cap.max_recv_wr + 1); + wr.srq_handle = 0; + wr.flags = cpu_to_be32(QP_RDMA_READ | QP_RDMA_WRITE | QP_MW_BIND | + QP_ZERO_STAG | QP_RDMA_READ_RESPONSE); + wr.send_sgl_depth = cpu_to_be32(qp_attrs->cap.max_send_sge); + wr.recv_sgl_depth = cpu_to_be32(qp_attrs->cap.max_recv_sge); + wr.rdma_write_sgl_depth = cpu_to_be32(qp_attrs->cap.max_send_sge); + wr.shared_sq_ht = cpu_to_be64(qp->sq_mq.shared_dma); + wr.shared_rq_ht = cpu_to_be64(qp->rq_mq.shared_dma); + wr.ord = cpu_to_be32(C2_MAX_ORD_PER_QP); + wr.ird = cpu_to_be32(C2_MAX_IRD_PER_QP); + wr.pd_id = pd->pd_id; + wr.user_context = (unsigned long) qp; + + vq_req_get(c2dev, vq_req); + + /* Send the WR to the adapter */ + err = vq_send_wr(c2dev, (union c2wr *) & wr); + if (err) { + vq_req_put(c2dev, vq_req); + goto bail3; + } + + /* Wait for the verb reply */ + err = vq_wait_for_reply(c2dev, vq_req); + if (err) { + goto bail3; + } + + /* Process the reply */ + reply = (struct c2wr_qp_create_rep *) (unsigned long) (vq_req->reply_msg); + if (!reply) { + err = -ENOMEM; + goto bail3; + } + + if ((err = c2_wr_get_result(reply)) != 0) { + goto bail4; + } + + /* Fill in the kernel QP struct */ + atomic_set(&qp->refcount, 1); + qp->adapter_handle = reply->qp_handle; + qp->state = IB_QPS_RESET; + qp->send_sgl_depth = qp_attrs->cap.max_send_sge; + qp->rdma_write_sgl_depth = qp_attrs->cap.max_send_sge; + qp->recv_sgl_depth = qp_attrs->cap.max_recv_sge; + init_waitqueue_head(&qp->wait); + + /* Initialize the SQ MQ */ + q_size = be32_to_cpu(reply->sq_depth); + msg_size = be32_to_cpu(reply->sq_msg_size); + peer_pa = c2dev->pa + be32_to_cpu(reply->sq_mq_start); + mmap_size = PAGE_ALIGN(sizeof(struct c2_mq_shared) + msg_size * q_size); + mmap = ioremap_nocache(peer_pa, mmap_size); + if (!mmap) { + err = -ENOMEM; + goto bail5; + } + + c2_mq_req_init(&qp->sq_mq, + be32_to_cpu(reply->sq_mq_index), + q_size, + msg_size, + mmap + sizeof(struct c2_mq_shared), /* pool start */ + mmap, /* peer */ + C2_MQ_ADAPTER_TARGET); + + /* Initialize the RQ mq */ + q_size = be32_to_cpu(reply->rq_depth); + msg_size = be32_to_cpu(reply->rq_msg_size); + peer_pa = c2dev->pa + be32_to_cpu(reply->rq_mq_start); + mmap_size = PAGE_ALIGN(sizeof(struct c2_mq_shared) + msg_size * q_size); + mmap = ioremap_nocache(peer_pa, mmap_size); + if (!mmap) { + err = -ENOMEM; + goto bail6; + } + + c2_mq_req_init(&qp->rq_mq, + be32_to_cpu(reply->rq_mq_index), + q_size, + msg_size, + mmap + sizeof(struct c2_mq_shared), /* pool start */ + mmap, /* peer */ + C2_MQ_ADAPTER_TARGET); + + vq_repbuf_free(c2dev, reply); + vq_req_free(c2dev, vq_req); + + return 0; + + bail6: + iounmap(qp->sq_mq.peer); + bail5: + destroy_qp(c2dev, qp); + bail4: + vq_repbuf_free(c2dev, reply); + bail3: + vq_req_free(c2dev, vq_req); + bail2: + c2_free_mqsp(qp->rq_mq.shared); + bail1: + c2_free_mqsp(qp->sq_mq.shared); + bail0: + c2_free_qpn(c2dev, qp->qpn); + return err; +} + +static inline void c2_lock_cqs(struct c2_cq *send_cq, struct c2_cq *recv_cq) +{ + if (send_cq == recv_cq) + spin_lock_irq(&send_cq->lock); + else if (send_cq > recv_cq) { + spin_lock_irq(&send_cq->lock); + spin_lock_nested(&recv_cq->lock, SINGLE_DEPTH_NESTING); + } else { + spin_lock_irq(&recv_cq->lock); + spin_lock_nested(&send_cq->lock, SINGLE_DEPTH_NESTING); + } +} + +static inline void c2_unlock_cqs(struct c2_cq *send_cq, struct c2_cq *recv_cq) +{ + if (send_cq == recv_cq) + spin_unlock_irq(&send_cq->lock); + else if (send_cq > recv_cq) { + spin_unlock(&recv_cq->lock); + spin_unlock_irq(&send_cq->lock); + } else { + spin_unlock(&send_cq->lock); + spin_unlock_irq(&recv_cq->lock); + } +} + +void c2_free_qp(struct c2_dev *c2dev, struct c2_qp *qp) +{ + struct c2_cq *send_cq; + struct c2_cq *recv_cq; + + send_cq = to_c2cq(qp->ibqp.send_cq); + recv_cq = to_c2cq(qp->ibqp.recv_cq); + + /* + * Lock CQs here, so that CQ polling code can do QP lookup + * without taking a lock. + */ + c2_lock_cqs(send_cq, recv_cq); + c2_free_qpn(c2dev, qp->qpn); + c2_unlock_cqs(send_cq, recv_cq); + + /* + * Destroy qp in the rnic... + */ + destroy_qp(c2dev, qp); + + /* + * Mark any unreaped CQEs as null and void. + */ + c2_cq_clean(c2dev, qp, send_cq->cqn); + if (send_cq != recv_cq) + c2_cq_clean(c2dev, qp, recv_cq->cqn); + /* + * Unmap the MQs and return the shared pointers + * to the message pool. + */ + iounmap(qp->sq_mq.peer); + iounmap(qp->rq_mq.peer); + c2_free_mqsp(qp->sq_mq.shared); + c2_free_mqsp(qp->rq_mq.shared); + + atomic_dec(&qp->refcount); + wait_event(qp->wait, !atomic_read(&qp->refcount)); +} + +/* + * Function: move_sgl + * + * Description: + * Move an SGL from the user's work request struct into a CCIL Work Request + * message, swapping to WR byte order and ensure the total length doesn't + * overflow. + * + * IN: + * dst - ptr to CCIL Work Request message SGL memory. + * src - ptr to the consumers SGL memory. + * + * OUT: none + * + * Return: + * CCIL status codes. + */ +static int +move_sgl(struct c2_data_addr * dst, struct ib_sge *src, int count, u32 * p_len, + u8 * actual_count) +{ + u32 tot = 0; /* running total */ + u8 acount = 0; /* running total non-0 len sge's */ + + while (count > 0) { + /* + * If the addition of this SGE causes the + * total SGL length to exceed 2^32-1, then + * fail-n-bail. + * + * If the current total plus the next element length + * wraps, then it will go negative and be less than the + * current total... + */ + if ((tot + src->length) < tot) { + return -EINVAL; + } + /* + * Bug: 1456 (as well as 1498 & 1643) + * Skip over any sge's supplied with len=0 + */ + if (src->length) { + tot += src->length; + dst->stag = cpu_to_be32(src->lkey); + dst->to = cpu_to_be64(src->addr); + dst->length = cpu_to_be32(src->length); + dst++; + acount++; + } + src++; + count--; + } + + if (acount == 0) { + /* + * Bug: 1476 (as well as 1498, 1456 and 1643) + * Setup the SGL in the WR to make it easier for the RNIC. + * This way, the FW doesn't have to deal with special cases. + * Setting length=0 should be sufficient. + */ + dst->stag = 0; + dst->to = 0; + dst->length = 0; + } + + *p_len = tot; + *actual_count = acount; + return 0; +} + +/* + * Function: c2_activity (private function) + * + * Description: + * Post an mq index to the host->adapter activity fifo. + * + * IN: + * c2dev - ptr to c2dev structure + * mq_index - mq index to post + * shared - value most recently written to shared + * + * OUT: + * + * Return: + * none + */ +static inline void c2_activity(struct c2_dev *c2dev, u32 mq_index, u16 shared) +{ + /* + * First read the register to see if the FIFO is full, and if so, + * spin until it's not. This isn't perfect -- there is no + * synchronization among the clients of the register, but in + * practice it prevents multiple CPU from hammering the bus + * with PCI RETRY. Note that when this does happen, the card + * cannot get on the bus and the card and system hang in a + * deadlock -- thus the need for this code. [TOT] + */ + while (readl(c2dev->regs + PCI_BAR0_ADAPTER_HINT) & 0x80000000) + udelay(10); + + __raw_writel(C2_HINT_MAKE(mq_index, shared), + c2dev->regs + PCI_BAR0_ADAPTER_HINT); +} + +/* + * Function: qp_wr_post + * + * Description: + * This in-line function allocates a MQ msg, then moves the host-copy of + * the completed WR into msg. Then it posts the message. + * + * IN: + * q - ptr to user MQ. + * wr - ptr to host-copy of the WR. + * qp - ptr to user qp + * size - Number of bytes to post. Assumed to be divisible by 4. + * + * OUT: none + * + * Return: + * CCIL status codes. + */ +static int qp_wr_post(struct c2_mq *q, union c2wr * wr, struct c2_qp *qp, u32 size) +{ + union c2wr *msg; + + msg = c2_mq_alloc(q); + if (msg == NULL) { + return -EINVAL; + } +#ifdef CCMSGMAGIC + ((c2wr_hdr_t *) wr)->magic = cpu_to_be32(CCWR_MAGIC); +#endif + + /* + * Since all header fields in the WR are the same as the + * CQE, set the following so the adapter need not. + */ + c2_wr_set_result(wr, CCERR_PENDING); + + /* + * Copy the wr down to the adapter + */ + memcpy((void *) msg, (void *) wr, size); + + c2_mq_produce(q); + return 0; +} + + +int c2_post_send(struct ib_qp *ibqp, struct ib_send_wr *ib_wr, + struct ib_send_wr **bad_wr) +{ + struct c2_dev *c2dev = to_c2dev(ibqp->device); + struct c2_qp *qp = to_c2qp(ibqp); + union c2wr wr; + unsigned long lock_flags; + int err = 0; + + u32 flags; + u32 tot_len; + u8 actual_sge_count; + u32 msg_size; + + if (qp->state > IB_QPS_RTS) { + err = -EINVAL; + goto out; + } + + while (ib_wr) { + + flags = 0; + wr.sqwr.sq_hdr.user_hdr.hdr.context = ib_wr->wr_id; + if (ib_wr->send_flags & IB_SEND_SIGNALED) { + flags |= SQ_SIGNALED; + } + + switch (ib_wr->opcode) { + case IB_WR_SEND: + case IB_WR_SEND_WITH_INV: + if (ib_wr->opcode == IB_WR_SEND) { + if (ib_wr->send_flags & IB_SEND_SOLICITED) + c2_wr_set_id(&wr, C2_WR_TYPE_SEND_SE); + else + c2_wr_set_id(&wr, C2_WR_TYPE_SEND); + wr.sqwr.send.remote_stag = 0; + } else { + if (ib_wr->send_flags & IB_SEND_SOLICITED) + c2_wr_set_id(&wr, C2_WR_TYPE_SEND_SE_INV); + else + c2_wr_set_id(&wr, C2_WR_TYPE_SEND_INV); + wr.sqwr.send.remote_stag = + cpu_to_be32(ib_wr->ex.invalidate_rkey); + } + + msg_size = sizeof(struct c2wr_send_req) + + sizeof(struct c2_data_addr) * ib_wr->num_sge; + if (ib_wr->num_sge > qp->send_sgl_depth) { + err = -EINVAL; + break; + } + if (ib_wr->send_flags & IB_SEND_FENCE) { + flags |= SQ_READ_FENCE; + } + err = move_sgl((struct c2_data_addr *) & (wr.sqwr.send.data), + ib_wr->sg_list, + ib_wr->num_sge, + &tot_len, &actual_sge_count); + wr.sqwr.send.sge_len = cpu_to_be32(tot_len); + c2_wr_set_sge_count(&wr, actual_sge_count); + break; + case IB_WR_RDMA_WRITE: + c2_wr_set_id(&wr, C2_WR_TYPE_RDMA_WRITE); + msg_size = sizeof(struct c2wr_rdma_write_req) + + (sizeof(struct c2_data_addr) * ib_wr->num_sge); + if (ib_wr->num_sge > qp->rdma_write_sgl_depth) { + err = -EINVAL; + break; + } + if (ib_wr->send_flags & IB_SEND_FENCE) { + flags |= SQ_READ_FENCE; + } + wr.sqwr.rdma_write.remote_stag = + cpu_to_be32(ib_wr->wr.rdma.rkey); + wr.sqwr.rdma_write.remote_to = + cpu_to_be64(ib_wr->wr.rdma.remote_addr); + err = move_sgl((struct c2_data_addr *) + & (wr.sqwr.rdma_write.data), + ib_wr->sg_list, + ib_wr->num_sge, + &tot_len, &actual_sge_count); + wr.sqwr.rdma_write.sge_len = cpu_to_be32(tot_len); + c2_wr_set_sge_count(&wr, actual_sge_count); + break; + case IB_WR_RDMA_READ: + c2_wr_set_id(&wr, C2_WR_TYPE_RDMA_READ); + msg_size = sizeof(struct c2wr_rdma_read_req); + + /* IWarp only suppots 1 sge for RDMA reads */ + if (ib_wr->num_sge > 1) { + err = -EINVAL; + break; + } + + /* + * Move the local and remote stag/to/len into the WR. + */ + wr.sqwr.rdma_read.local_stag = + cpu_to_be32(ib_wr->sg_list->lkey); + wr.sqwr.rdma_read.local_to = + cpu_to_be64(ib_wr->sg_list->addr); + wr.sqwr.rdma_read.remote_stag = + cpu_to_be32(ib_wr->wr.rdma.rkey); + wr.sqwr.rdma_read.remote_to = + cpu_to_be64(ib_wr->wr.rdma.remote_addr); + wr.sqwr.rdma_read.length = + cpu_to_be32(ib_wr->sg_list->length); + break; + default: + /* error */ + msg_size = 0; + err = -EINVAL; + break; + } + + /* + * If we had an error on the last wr build, then + * break out. Possible errors include bogus WR + * type, and a bogus SGL length... + */ + if (err) { + break; + } + + /* + * Store flags + */ + c2_wr_set_flags(&wr, flags); + + /* + * Post the puppy! + */ + spin_lock_irqsave(&qp->lock, lock_flags); + err = qp_wr_post(&qp->sq_mq, &wr, qp, msg_size); + if (err) { + spin_unlock_irqrestore(&qp->lock, lock_flags); + break; + } + + /* + * Enqueue mq index to activity FIFO. + */ + c2_activity(c2dev, qp->sq_mq.index, qp->sq_mq.hint_count); + spin_unlock_irqrestore(&qp->lock, lock_flags); + + ib_wr = ib_wr->next; + } + +out: + if (err) + *bad_wr = ib_wr; + return err; +} + +int c2_post_receive(struct ib_qp *ibqp, struct ib_recv_wr *ib_wr, + struct ib_recv_wr **bad_wr) +{ + struct c2_dev *c2dev = to_c2dev(ibqp->device); + struct c2_qp *qp = to_c2qp(ibqp); + union c2wr wr; + unsigned long lock_flags; + int err = 0; + + if (qp->state > IB_QPS_RTS) { + err = -EINVAL; + goto out; + } + + /* + * Try and post each work request + */ + while (ib_wr) { + u32 tot_len; + u8 actual_sge_count; + + if (ib_wr->num_sge > qp->recv_sgl_depth) { + err = -EINVAL; + break; + } + + /* + * Create local host-copy of the WR + */ + wr.rqwr.rq_hdr.user_hdr.hdr.context = ib_wr->wr_id; + c2_wr_set_id(&wr, CCWR_RECV); + c2_wr_set_flags(&wr, 0); + + /* sge_count is limited to eight bits. */ + BUG_ON(ib_wr->num_sge >= 256); + err = move_sgl((struct c2_data_addr *) & (wr.rqwr.data), + ib_wr->sg_list, + ib_wr->num_sge, &tot_len, &actual_sge_count); + c2_wr_set_sge_count(&wr, actual_sge_count); + + /* + * If we had an error on the last wr build, then + * break out. Possible errors include bogus WR + * type, and a bogus SGL length... + */ + if (err) { + break; + } + + spin_lock_irqsave(&qp->lock, lock_flags); + err = qp_wr_post(&qp->rq_mq, &wr, qp, qp->rq_mq.msg_size); + if (err) { + spin_unlock_irqrestore(&qp->lock, lock_flags); + break; + } + + /* + * Enqueue mq index to activity FIFO + */ + c2_activity(c2dev, qp->rq_mq.index, qp->rq_mq.hint_count); + spin_unlock_irqrestore(&qp->lock, lock_flags); + + ib_wr = ib_wr->next; + } + +out: + if (err) + *bad_wr = ib_wr; + return err; +} + +void c2_init_qp_table(struct c2_dev *c2dev) +{ + spin_lock_init(&c2dev->qp_table.lock); + idr_init(&c2dev->qp_table.idr); +} + +void c2_cleanup_qp_table(struct c2_dev *c2dev) +{ + idr_destroy(&c2dev->qp_table.idr); +} diff --git a/drivers/infiniband/hw/amso1100/c2_rnic.c b/drivers/infiniband/hw/amso1100/c2_rnic.c new file mode 100644 index 000000000..d2a6d9613 --- /dev/null +++ b/drivers/infiniband/hw/amso1100/c2_rnic.c @@ -0,0 +1,655 @@ +/* + * Copyright (c) 2005 Ammasso, Inc. All rights reserved. + * Copyright (c) 2005 Open Grid Computing, Inc. All rights reserved. + * + * This software is available to you under a choice of one of two + * licenses. You may choose to be licensed under the terms of the GNU + * General Public License (GPL) Version 2, available from the file + * COPYING in the main directory of this source tree, or the + * OpenIB.org BSD license below: + * + * Redistribution and use in source and binary forms, with or + * without modification, are permitted provided that the following + * conditions are met: + * + * - Redistributions of source code must retain the above + * copyright notice, this list of conditions and the following + * disclaimer. + * + * - Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and/or other materials + * provided with the distribution. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS + * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN + * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + * + */ + + +#include <linux/module.h> +#include <linux/moduleparam.h> +#include <linux/pci.h> +#include <linux/netdevice.h> +#include <linux/etherdevice.h> +#include <linux/delay.h> +#include <linux/ethtool.h> +#include <linux/mii.h> +#include <linux/if_vlan.h> +#include <linux/crc32.h> +#include <linux/in.h> +#include <linux/ip.h> +#include <linux/tcp.h> +#include <linux/init.h> +#include <linux/dma-mapping.h> +#include <linux/mm.h> +#include <linux/inet.h> +#include <linux/vmalloc.h> +#include <linux/slab.h> + +#include <linux/route.h> + +#include <asm/io.h> +#include <asm/irq.h> +#include <asm/byteorder.h> +#include <rdma/ib_smi.h> +#include "c2.h" +#include "c2_vq.h" + +/* Device capabilities */ +#define C2_MIN_PAGESIZE 1024 + +#define C2_MAX_MRS 32768 +#define C2_MAX_QPS 16000 +#define C2_MAX_WQE_SZ 256 +#define C2_MAX_QP_WR ((128*1024)/C2_MAX_WQE_SZ) +#define C2_MAX_SGES 4 +#define C2_MAX_SGE_RD 1 +#define C2_MAX_CQS 32768 +#define C2_MAX_CQES 4096 +#define C2_MAX_PDS 16384 + +/* + * Send the adapter INIT message to the amso1100 + */ +static int c2_adapter_init(struct c2_dev *c2dev) +{ + struct c2wr_init_req wr; + int err; + + memset(&wr, 0, sizeof(wr)); + c2_wr_set_id(&wr, CCWR_INIT); + wr.hdr.context = 0; + wr.hint_count = cpu_to_be64(c2dev->hint_count_dma); + wr.q0_host_shared = cpu_to_be64(c2dev->req_vq.shared_dma); + wr.q1_host_shared = cpu_to_be64(c2dev->rep_vq.shared_dma); + wr.q1_host_msg_pool = cpu_to_be64(c2dev->rep_vq.host_dma); + wr.q2_host_shared = cpu_to_be64(c2dev->aeq.shared_dma); + wr.q2_host_msg_pool = cpu_to_be64(c2dev->aeq.host_dma); + + /* Post the init message */ + err = vq_send_wr(c2dev, (union c2wr *) & wr); + + return err; +} + +/* + * Send the adapter TERM message to the amso1100 + */ +static void c2_adapter_term(struct c2_dev *c2dev) +{ + struct c2wr_init_req wr; + + memset(&wr, 0, sizeof(wr)); + c2_wr_set_id(&wr, CCWR_TERM); + wr.hdr.context = 0; + + /* Post the init message */ + vq_send_wr(c2dev, (union c2wr *) & wr); + c2dev->init = 0; + + return; +} + +/* + * Query the adapter + */ +static int c2_rnic_query(struct c2_dev *c2dev, struct ib_device_attr *props) +{ + struct c2_vq_req *vq_req; + struct c2wr_rnic_query_req wr; + struct c2wr_rnic_query_rep *reply; + int err; + + vq_req = vq_req_alloc(c2dev); + if (!vq_req) + return -ENOMEM; + + c2_wr_set_id(&wr, CCWR_RNIC_QUERY); + wr.hdr.context = (unsigned long) vq_req; + wr.rnic_handle = c2dev->adapter_handle; + + vq_req_get(c2dev, vq_req); + + err = vq_send_wr(c2dev, (union c2wr *) &wr); + if (err) { + vq_req_put(c2dev, vq_req); + goto bail1; + } + + err = vq_wait_for_reply(c2dev, vq_req); + if (err) + goto bail1; + + reply = + (struct c2wr_rnic_query_rep *) (unsigned long) (vq_req->reply_msg); + if (!reply) + err = -ENOMEM; + else + err = c2_errno(reply); + if (err) + goto bail2; + + props->fw_ver = + ((u64)be32_to_cpu(reply->fw_ver_major) << 32) | + ((be32_to_cpu(reply->fw_ver_minor) & 0xFFFF) << 16) | + (be32_to_cpu(reply->fw_ver_patch) & 0xFFFF); + memcpy(&props->sys_image_guid, c2dev->netdev->dev_addr, 6); + props->max_mr_size = 0xFFFFFFFF; + props->page_size_cap = ~(C2_MIN_PAGESIZE-1); + props->vendor_id = be32_to_cpu(reply->vendor_id); + props->vendor_part_id = be32_to_cpu(reply->part_number); + props->hw_ver = be32_to_cpu(reply->hw_version); + props->max_qp = be32_to_cpu(reply->max_qps); + props->max_qp_wr = be32_to_cpu(reply->max_qp_depth); + props->device_cap_flags = c2dev->device_cap_flags; + props->max_sge = C2_MAX_SGES; + props->max_sge_rd = C2_MAX_SGE_RD; + props->max_cq = be32_to_cpu(reply->max_cqs); + props->max_cqe = be32_to_cpu(reply->max_cq_depth); + props->max_mr = be32_to_cpu(reply->max_mrs); + props->max_pd = be32_to_cpu(reply->max_pds); + props->max_qp_rd_atom = be32_to_cpu(reply->max_qp_ird); + props->max_ee_rd_atom = 0; + props->max_res_rd_atom = be32_to_cpu(reply->max_global_ird); + props->max_qp_init_rd_atom = be32_to_cpu(reply->max_qp_ord); + props->max_ee_init_rd_atom = 0; + props->atomic_cap = IB_ATOMIC_NONE; + props->max_ee = 0; + props->max_rdd = 0; + props->max_mw = be32_to_cpu(reply->max_mws); + props->max_raw_ipv6_qp = 0; + props->max_raw_ethy_qp = 0; + props->max_mcast_grp = 0; + props->max_mcast_qp_attach = 0; + props->max_total_mcast_qp_attach = 0; + props->max_ah = 0; + props->max_fmr = 0; + props->max_map_per_fmr = 0; + props->max_srq = 0; + props->max_srq_wr = 0; + props->max_srq_sge = 0; + props->max_pkeys = 0; + props->local_ca_ack_delay = 0; + + bail2: + vq_repbuf_free(c2dev, reply); + + bail1: + vq_req_free(c2dev, vq_req); + return err; +} + +/* + * Add an IP address to the RNIC interface + */ +int c2_add_addr(struct c2_dev *c2dev, __be32 inaddr, __be32 inmask) +{ + struct c2_vq_req *vq_req; + struct c2wr_rnic_setconfig_req *wr; + struct c2wr_rnic_setconfig_rep *reply; + struct c2_netaddr netaddr; + int err, len; + + vq_req = vq_req_alloc(c2dev); + if (!vq_req) + return -ENOMEM; + + len = sizeof(struct c2_netaddr); + wr = kmalloc(c2dev->req_vq.msg_size, GFP_KERNEL); + if (!wr) { + err = -ENOMEM; + goto bail0; + } + + c2_wr_set_id(wr, CCWR_RNIC_SETCONFIG); + wr->hdr.context = (unsigned long) vq_req; + wr->rnic_handle = c2dev->adapter_handle; + wr->option = cpu_to_be32(C2_CFG_ADD_ADDR); + + netaddr.ip_addr = inaddr; + netaddr.netmask = inmask; + netaddr.mtu = 0; + + memcpy(wr->data, &netaddr, len); + + vq_req_get(c2dev, vq_req); + + err = vq_send_wr(c2dev, (union c2wr *) wr); + if (err) { + vq_req_put(c2dev, vq_req); + goto bail1; + } + + err = vq_wait_for_reply(c2dev, vq_req); + if (err) + goto bail1; + + reply = + (struct c2wr_rnic_setconfig_rep *) (unsigned long) (vq_req->reply_msg); + if (!reply) { + err = -ENOMEM; + goto bail1; + } + + err = c2_errno(reply); + vq_repbuf_free(c2dev, reply); + + bail1: + kfree(wr); + bail0: + vq_req_free(c2dev, vq_req); + return err; +} + +/* + * Delete an IP address from the RNIC interface + */ +int c2_del_addr(struct c2_dev *c2dev, __be32 inaddr, __be32 inmask) +{ + struct c2_vq_req *vq_req; + struct c2wr_rnic_setconfig_req *wr; + struct c2wr_rnic_setconfig_rep *reply; + struct c2_netaddr netaddr; + int err, len; + + vq_req = vq_req_alloc(c2dev); + if (!vq_req) + return -ENOMEM; + + len = sizeof(struct c2_netaddr); + wr = kmalloc(c2dev->req_vq.msg_size, GFP_KERNEL); + if (!wr) { + err = -ENOMEM; + goto bail0; + } + + c2_wr_set_id(wr, CCWR_RNIC_SETCONFIG); + wr->hdr.context = (unsigned long) vq_req; + wr->rnic_handle = c2dev->adapter_handle; + wr->option = cpu_to_be32(C2_CFG_DEL_ADDR); + + netaddr.ip_addr = inaddr; + netaddr.netmask = inmask; + netaddr.mtu = 0; + + memcpy(wr->data, &netaddr, len); + + vq_req_get(c2dev, vq_req); + + err = vq_send_wr(c2dev, (union c2wr *) wr); + if (err) { + vq_req_put(c2dev, vq_req); + goto bail1; + } + + err = vq_wait_for_reply(c2dev, vq_req); + if (err) + goto bail1; + + reply = + (struct c2wr_rnic_setconfig_rep *) (unsigned long) (vq_req->reply_msg); + if (!reply) { + err = -ENOMEM; + goto bail1; + } + + err = c2_errno(reply); + vq_repbuf_free(c2dev, reply); + + bail1: + kfree(wr); + bail0: + vq_req_free(c2dev, vq_req); + return err; +} + +/* + * Open a single RNIC instance to use with all + * low level openib calls + */ +static int c2_rnic_open(struct c2_dev *c2dev) +{ + struct c2_vq_req *vq_req; + union c2wr wr; + struct c2wr_rnic_open_rep *reply; + int err; + + vq_req = vq_req_alloc(c2dev); + if (vq_req == NULL) { + return -ENOMEM; + } + + memset(&wr, 0, sizeof(wr)); + c2_wr_set_id(&wr, CCWR_RNIC_OPEN); + wr.rnic_open.req.hdr.context = (unsigned long) (vq_req); + wr.rnic_open.req.flags = cpu_to_be16(RNIC_PRIV_MODE); + wr.rnic_open.req.port_num = cpu_to_be16(0); + wr.rnic_open.req.user_context = (unsigned long) c2dev; + + vq_req_get(c2dev, vq_req); + + err = vq_send_wr(c2dev, &wr); + if (err) { + vq_req_put(c2dev, vq_req); + goto bail0; + } + + err = vq_wait_for_reply(c2dev, vq_req); + if (err) { + goto bail0; + } + + reply = (struct c2wr_rnic_open_rep *) (unsigned long) (vq_req->reply_msg); + if (!reply) { + err = -ENOMEM; + goto bail0; + } + + if ((err = c2_errno(reply)) != 0) { + goto bail1; + } + + c2dev->adapter_handle = reply->rnic_handle; + + bail1: + vq_repbuf_free(c2dev, reply); + bail0: + vq_req_free(c2dev, vq_req); + return err; +} + +/* + * Close the RNIC instance + */ +static int c2_rnic_close(struct c2_dev *c2dev) +{ + struct c2_vq_req *vq_req; + union c2wr wr; + struct c2wr_rnic_close_rep *reply; + int err; + + vq_req = vq_req_alloc(c2dev); + if (vq_req == NULL) { + return -ENOMEM; + } + + memset(&wr, 0, sizeof(wr)); + c2_wr_set_id(&wr, CCWR_RNIC_CLOSE); + wr.rnic_close.req.hdr.context = (unsigned long) vq_req; + wr.rnic_close.req.rnic_handle = c2dev->adapter_handle; + + vq_req_get(c2dev, vq_req); + + err = vq_send_wr(c2dev, &wr); + if (err) { + vq_req_put(c2dev, vq_req); + goto bail0; + } + + err = vq_wait_for_reply(c2dev, vq_req); + if (err) { + goto bail0; + } + + reply = (struct c2wr_rnic_close_rep *) (unsigned long) (vq_req->reply_msg); + if (!reply) { + err = -ENOMEM; + goto bail0; + } + + if ((err = c2_errno(reply)) != 0) { + goto bail1; + } + + c2dev->adapter_handle = 0; + + bail1: + vq_repbuf_free(c2dev, reply); + bail0: + vq_req_free(c2dev, vq_req); + return err; +} + +/* + * Called by c2_probe to initialize the RNIC. This principally + * involves initializing the various limits and resource pools that + * comprise the RNIC instance. + */ +int c2_rnic_init(struct c2_dev *c2dev) +{ + int err; + u32 qsize, msgsize; + void *q1_pages; + void *q2_pages; + void __iomem *mmio_regs; + + /* Device capabilities */ + c2dev->device_cap_flags = + (IB_DEVICE_RESIZE_MAX_WR | + IB_DEVICE_CURR_QP_STATE_MOD | + IB_DEVICE_SYS_IMAGE_GUID | + IB_DEVICE_LOCAL_DMA_LKEY | + IB_DEVICE_MEM_WINDOW); + + /* Allocate the qptr_array */ + c2dev->qptr_array = vzalloc(C2_MAX_CQS * sizeof(void *)); + if (!c2dev->qptr_array) { + return -ENOMEM; + } + + /* Initialize the qptr_array */ + c2dev->qptr_array[0] = (void *) &c2dev->req_vq; + c2dev->qptr_array[1] = (void *) &c2dev->rep_vq; + c2dev->qptr_array[2] = (void *) &c2dev->aeq; + + /* Initialize data structures */ + init_waitqueue_head(&c2dev->req_vq_wo); + spin_lock_init(&c2dev->vqlock); + spin_lock_init(&c2dev->lock); + + /* Allocate MQ shared pointer pool for kernel clients. User + * mode client pools are hung off the user context + */ + err = c2_init_mqsp_pool(c2dev, GFP_KERNEL, &c2dev->kern_mqsp_pool); + if (err) { + goto bail0; + } + + /* Allocate shared pointers for Q0, Q1, and Q2 from + * the shared pointer pool. + */ + + c2dev->hint_count = c2_alloc_mqsp(c2dev, c2dev->kern_mqsp_pool, + &c2dev->hint_count_dma, + GFP_KERNEL); + c2dev->req_vq.shared = c2_alloc_mqsp(c2dev, c2dev->kern_mqsp_pool, + &c2dev->req_vq.shared_dma, + GFP_KERNEL); + c2dev->rep_vq.shared = c2_alloc_mqsp(c2dev, c2dev->kern_mqsp_pool, + &c2dev->rep_vq.shared_dma, + GFP_KERNEL); + c2dev->aeq.shared = c2_alloc_mqsp(c2dev, c2dev->kern_mqsp_pool, + &c2dev->aeq.shared_dma, GFP_KERNEL); + if (!c2dev->hint_count || !c2dev->req_vq.shared || + !c2dev->rep_vq.shared || !c2dev->aeq.shared) { + err = -ENOMEM; + goto bail1; + } + + mmio_regs = c2dev->kva; + /* Initialize the Verbs Request Queue */ + c2_mq_req_init(&c2dev->req_vq, 0, + be32_to_cpu((__force __be32) readl(mmio_regs + C2_REGS_Q0_QSIZE)), + be32_to_cpu((__force __be32) readl(mmio_regs + C2_REGS_Q0_MSGSIZE)), + mmio_regs + + be32_to_cpu((__force __be32) readl(mmio_regs + C2_REGS_Q0_POOLSTART)), + mmio_regs + + be32_to_cpu((__force __be32) readl(mmio_regs + C2_REGS_Q0_SHARED)), + C2_MQ_ADAPTER_TARGET); + + /* Initialize the Verbs Reply Queue */ + qsize = be32_to_cpu((__force __be32) readl(mmio_regs + C2_REGS_Q1_QSIZE)); + msgsize = be32_to_cpu((__force __be32) readl(mmio_regs + C2_REGS_Q1_MSGSIZE)); + q1_pages = dma_alloc_coherent(&c2dev->pcidev->dev, qsize * msgsize, + &c2dev->rep_vq.host_dma, GFP_KERNEL); + if (!q1_pages) { + err = -ENOMEM; + goto bail1; + } + dma_unmap_addr_set(&c2dev->rep_vq, mapping, c2dev->rep_vq.host_dma); + pr_debug("%s rep_vq va %p dma %llx\n", __func__, q1_pages, + (unsigned long long) c2dev->rep_vq.host_dma); + c2_mq_rep_init(&c2dev->rep_vq, + 1, + qsize, + msgsize, + q1_pages, + mmio_regs + + be32_to_cpu((__force __be32) readl(mmio_regs + C2_REGS_Q1_SHARED)), + C2_MQ_HOST_TARGET); + + /* Initialize the Asynchronus Event Queue */ + qsize = be32_to_cpu((__force __be32) readl(mmio_regs + C2_REGS_Q2_QSIZE)); + msgsize = be32_to_cpu((__force __be32) readl(mmio_regs + C2_REGS_Q2_MSGSIZE)); + q2_pages = dma_alloc_coherent(&c2dev->pcidev->dev, qsize * msgsize, + &c2dev->aeq.host_dma, GFP_KERNEL); + if (!q2_pages) { + err = -ENOMEM; + goto bail2; + } + dma_unmap_addr_set(&c2dev->aeq, mapping, c2dev->aeq.host_dma); + pr_debug("%s aeq va %p dma %llx\n", __func__, q2_pages, + (unsigned long long) c2dev->aeq.host_dma); + c2_mq_rep_init(&c2dev->aeq, + 2, + qsize, + msgsize, + q2_pages, + mmio_regs + + be32_to_cpu((__force __be32) readl(mmio_regs + C2_REGS_Q2_SHARED)), + C2_MQ_HOST_TARGET); + + /* Initialize the verbs request allocator */ + err = vq_init(c2dev); + if (err) + goto bail3; + + /* Enable interrupts on the adapter */ + writel(0, c2dev->regs + C2_IDIS); + + /* create the WR init message */ + err = c2_adapter_init(c2dev); + if (err) + goto bail4; + c2dev->init++; + + /* open an adapter instance */ + err = c2_rnic_open(c2dev); + if (err) + goto bail4; + + /* Initialize cached the adapter limits */ + err = c2_rnic_query(c2dev, &c2dev->props); + if (err) + goto bail5; + + /* Initialize the PD pool */ + err = c2_init_pd_table(c2dev); + if (err) + goto bail5; + + /* Initialize the QP pool */ + c2_init_qp_table(c2dev); + return 0; + + bail5: + c2_rnic_close(c2dev); + bail4: + vq_term(c2dev); + bail3: + dma_free_coherent(&c2dev->pcidev->dev, + c2dev->aeq.q_size * c2dev->aeq.msg_size, + q2_pages, dma_unmap_addr(&c2dev->aeq, mapping)); + bail2: + dma_free_coherent(&c2dev->pcidev->dev, + c2dev->rep_vq.q_size * c2dev->rep_vq.msg_size, + q1_pages, dma_unmap_addr(&c2dev->rep_vq, mapping)); + bail1: + c2_free_mqsp_pool(c2dev, c2dev->kern_mqsp_pool); + bail0: + vfree(c2dev->qptr_array); + + return err; +} + +/* + * Called by c2_remove to cleanup the RNIC resources. + */ +void c2_rnic_term(struct c2_dev *c2dev) +{ + + /* Close the open adapter instance */ + c2_rnic_close(c2dev); + + /* Send the TERM message to the adapter */ + c2_adapter_term(c2dev); + + /* Disable interrupts on the adapter */ + writel(1, c2dev->regs + C2_IDIS); + + /* Free the QP pool */ + c2_cleanup_qp_table(c2dev); + + /* Free the PD pool */ + c2_cleanup_pd_table(c2dev); + + /* Free the verbs request allocator */ + vq_term(c2dev); + + /* Free the asynchronus event queue */ + dma_free_coherent(&c2dev->pcidev->dev, + c2dev->aeq.q_size * c2dev->aeq.msg_size, + c2dev->aeq.msg_pool.host, + dma_unmap_addr(&c2dev->aeq, mapping)); + + /* Free the verbs reply queue */ + dma_free_coherent(&c2dev->pcidev->dev, + c2dev->rep_vq.q_size * c2dev->rep_vq.msg_size, + c2dev->rep_vq.msg_pool.host, + dma_unmap_addr(&c2dev->rep_vq, mapping)); + + /* Free the MQ shared pointer pool */ + c2_free_mqsp_pool(c2dev, c2dev->kern_mqsp_pool); + + /* Free the qptr_array */ + vfree(c2dev->qptr_array); + + return; +} diff --git a/drivers/infiniband/hw/amso1100/c2_status.h b/drivers/infiniband/hw/amso1100/c2_status.h new file mode 100644 index 000000000..6ee4aa92d --- /dev/null +++ b/drivers/infiniband/hw/amso1100/c2_status.h @@ -0,0 +1,158 @@ +/* + * Copyright (c) 2005 Ammasso, Inc. All rights reserved. + * Copyright (c) 2005 Open Grid Computing, Inc. All rights reserved. + * + * This software is available to you under a choice of one of two + * licenses. You may choose to be licensed under the terms of the GNU + * General Public License (GPL) Version 2, available from the file + * COPYING in the main directory of this source tree, or the + * OpenIB.org BSD license below: + * + * Redistribution and use in source and binary forms, with or + * without modification, are permitted provided that the following + * conditions are met: + * + * - Redistributions of source code must retain the above + * copyright notice, this list of conditions and the following + * disclaimer. + * + * - Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and/or other materials + * provided with the distribution. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS + * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN + * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ +#ifndef _C2_STATUS_H_ +#define _C2_STATUS_H_ + +/* + * Verbs Status Codes + */ +enum c2_status { + C2_OK = 0, /* This must be zero */ + CCERR_INSUFFICIENT_RESOURCES = 1, + CCERR_INVALID_MODIFIER = 2, + CCERR_INVALID_MODE = 3, + CCERR_IN_USE = 4, + CCERR_INVALID_RNIC = 5, + CCERR_INTERRUPTED_OPERATION = 6, + CCERR_INVALID_EH = 7, + CCERR_INVALID_CQ = 8, + CCERR_CQ_EMPTY = 9, + CCERR_NOT_IMPLEMENTED = 10, + CCERR_CQ_DEPTH_TOO_SMALL = 11, + CCERR_PD_IN_USE = 12, + CCERR_INVALID_PD = 13, + CCERR_INVALID_SRQ = 14, + CCERR_INVALID_ADDRESS = 15, + CCERR_INVALID_NETMASK = 16, + CCERR_INVALID_QP = 17, + CCERR_INVALID_QP_STATE = 18, + CCERR_TOO_MANY_WRS_POSTED = 19, + CCERR_INVALID_WR_TYPE = 20, + CCERR_INVALID_SGL_LENGTH = 21, + CCERR_INVALID_SQ_DEPTH = 22, + CCERR_INVALID_RQ_DEPTH = 23, + CCERR_INVALID_ORD = 24, + CCERR_INVALID_IRD = 25, + CCERR_QP_ATTR_CANNOT_CHANGE = 26, + CCERR_INVALID_STAG = 27, + CCERR_QP_IN_USE = 28, + CCERR_OUTSTANDING_WRS = 29, + CCERR_STAG_IN_USE = 30, + CCERR_INVALID_STAG_INDEX = 31, + CCERR_INVALID_SGL_FORMAT = 32, + CCERR_ADAPTER_TIMEOUT = 33, + CCERR_INVALID_CQ_DEPTH = 34, + CCERR_INVALID_PRIVATE_DATA_LENGTH = 35, + CCERR_INVALID_EP = 36, + CCERR_MR_IN_USE = CCERR_STAG_IN_USE, + CCERR_FLUSHED = 38, + CCERR_INVALID_WQE = 39, + CCERR_LOCAL_QP_CATASTROPHIC_ERROR = 40, + CCERR_REMOTE_TERMINATION_ERROR = 41, + CCERR_BASE_AND_BOUNDS_VIOLATION = 42, + CCERR_ACCESS_VIOLATION = 43, + CCERR_INVALID_PD_ID = 44, + CCERR_WRAP_ERROR = 45, + CCERR_INV_STAG_ACCESS_ERROR = 46, + CCERR_ZERO_RDMA_READ_RESOURCES = 47, + CCERR_QP_NOT_PRIVILEGED = 48, + CCERR_STAG_STATE_NOT_INVALID = 49, + CCERR_INVALID_PAGE_SIZE = 50, + CCERR_INVALID_BUFFER_SIZE = 51, + CCERR_INVALID_PBE = 52, + CCERR_INVALID_FBO = 53, + CCERR_INVALID_LENGTH = 54, + CCERR_INVALID_ACCESS_RIGHTS = 55, + CCERR_PBL_TOO_BIG = 56, + CCERR_INVALID_VA = 57, + CCERR_INVALID_REGION = 58, + CCERR_INVALID_WINDOW = 59, + CCERR_TOTAL_LENGTH_TOO_BIG = 60, + CCERR_INVALID_QP_ID = 61, + CCERR_ADDR_IN_USE = 62, + CCERR_ADDR_NOT_AVAIL = 63, + CCERR_NET_DOWN = 64, + CCERR_NET_UNREACHABLE = 65, + CCERR_CONN_ABORTED = 66, + CCERR_CONN_RESET = 67, + CCERR_NO_BUFS = 68, + CCERR_CONN_TIMEDOUT = 69, + CCERR_CONN_REFUSED = 70, + CCERR_HOST_UNREACHABLE = 71, + CCERR_INVALID_SEND_SGL_DEPTH = 72, + CCERR_INVALID_RECV_SGL_DEPTH = 73, + CCERR_INVALID_RDMA_WRITE_SGL_DEPTH = 74, + CCERR_INSUFFICIENT_PRIVILEGES = 75, + CCERR_STACK_ERROR = 76, + CCERR_INVALID_VERSION = 77, + CCERR_INVALID_MTU = 78, + CCERR_INVALID_IMAGE = 79, + CCERR_PENDING = 98, /* not an error; user internally by adapter */ + CCERR_DEFER = 99, /* not an error; used internally by adapter */ + CCERR_FAILED_WRITE = 100, + CCERR_FAILED_ERASE = 101, + CCERR_FAILED_VERIFICATION = 102, + CCERR_NOT_FOUND = 103, + +}; + +/* + * CCAE_ACTIVE_CONNECT_RESULTS status result codes. + */ +enum c2_connect_status { + C2_CONN_STATUS_SUCCESS = C2_OK, + C2_CONN_STATUS_NO_MEM = CCERR_INSUFFICIENT_RESOURCES, + C2_CONN_STATUS_TIMEDOUT = CCERR_CONN_TIMEDOUT, + C2_CONN_STATUS_REFUSED = CCERR_CONN_REFUSED, + C2_CONN_STATUS_NETUNREACH = CCERR_NET_UNREACHABLE, + C2_CONN_STATUS_HOSTUNREACH = CCERR_HOST_UNREACHABLE, + C2_CONN_STATUS_INVALID_RNIC = CCERR_INVALID_RNIC, + C2_CONN_STATUS_INVALID_QP = CCERR_INVALID_QP, + C2_CONN_STATUS_INVALID_QP_STATE = CCERR_INVALID_QP_STATE, + C2_CONN_STATUS_REJECTED = CCERR_CONN_RESET, + C2_CONN_STATUS_ADDR_NOT_AVAIL = CCERR_ADDR_NOT_AVAIL, +}; + +/* + * Flash programming status codes. + */ +enum c2_flash_status { + C2_FLASH_STATUS_SUCCESS = 0x0000, + C2_FLASH_STATUS_VERIFY_ERR = 0x0002, + C2_FLASH_STATUS_IMAGE_ERR = 0x0004, + C2_FLASH_STATUS_ECLBS = 0x0400, + C2_FLASH_STATUS_PSLBS = 0x0800, + C2_FLASH_STATUS_VPENS = 0x1000, +}; + +#endif /* _C2_STATUS_H_ */ diff --git a/drivers/infiniband/hw/amso1100/c2_user.h b/drivers/infiniband/hw/amso1100/c2_user.h new file mode 100644 index 000000000..7e9e7ad65 --- /dev/null +++ b/drivers/infiniband/hw/amso1100/c2_user.h @@ -0,0 +1,82 @@ +/* + * Copyright (c) 2005 Topspin Communications. All rights reserved. + * Copyright (c) 2005 Cisco Systems. All rights reserved. + * Copyright (c) 2005 Open Grid Computing, Inc. All rights reserved. + * + * This software is available to you under a choice of one of two + * licenses. You may choose to be licensed under the terms of the GNU + * General Public License (GPL) Version 2, available from the file + * COPYING in the main directory of this source tree, or the + * OpenIB.org BSD license below: + * + * Redistribution and use in source and binary forms, with or + * without modification, are permitted provided that the following + * conditions are met: + * + * - Redistributions of source code must retain the above + * copyright notice, this list of conditions and the following + * disclaimer. + * + * - Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and/or other materials + * provided with the distribution. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS + * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN + * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + * + */ + +#ifndef C2_USER_H +#define C2_USER_H + +#include <linux/types.h> + +/* + * Make sure that all structs defined in this file remain laid out so + * that they pack the same way on 32-bit and 64-bit architectures (to + * avoid incompatibility between 32-bit userspace and 64-bit kernels). + * In particular do not use pointer types -- pass pointers in __u64 + * instead. + */ + +struct c2_alloc_ucontext_resp { + __u32 qp_tab_size; + __u32 uarc_size; +}; + +struct c2_alloc_pd_resp { + __u32 pdn; + __u32 reserved; +}; + +struct c2_create_cq { + __u32 lkey; + __u32 pdn; + __u64 arm_db_page; + __u64 set_db_page; + __u32 arm_db_index; + __u32 set_db_index; +}; + +struct c2_create_cq_resp { + __u32 cqn; + __u32 reserved; +}; + +struct c2_create_qp { + __u32 lkey; + __u32 reserved; + __u64 sq_db_page; + __u64 rq_db_page; + __u32 sq_db_index; + __u32 rq_db_index; +}; + +#endif /* C2_USER_H */ diff --git a/drivers/infiniband/hw/amso1100/c2_vq.c b/drivers/infiniband/hw/amso1100/c2_vq.c new file mode 100644 index 000000000..2ec716fb2 --- /dev/null +++ b/drivers/infiniband/hw/amso1100/c2_vq.c @@ -0,0 +1,260 @@ +/* + * Copyright (c) 2005 Ammasso, Inc. All rights reserved. + * Copyright (c) 2005 Open Grid Computing, Inc. All rights reserved. + * + * This software is available to you under a choice of one of two + * licenses. You may choose to be licensed under the terms of the GNU + * General Public License (GPL) Version 2, available from the file + * COPYING in the main directory of this source tree, or the + * OpenIB.org BSD license below: + * + * Redistribution and use in source and binary forms, with or + * without modification, are permitted provided that the following + * conditions are met: + * + * - Redistributions of source code must retain the above + * copyright notice, this list of conditions and the following + * disclaimer. + * + * - Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and/or other materials + * provided with the distribution. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS + * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN + * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ +#include <linux/slab.h> +#include <linux/spinlock.h> + +#include "c2_vq.h" +#include "c2_provider.h" + +/* + * Verbs Request Objects: + * + * VQ Request Objects are allocated by the kernel verbs handlers. + * They contain a wait object, a refcnt, an atomic bool indicating that the + * adapter has replied, and a copy of the verb reply work request. + * A pointer to the VQ Request Object is passed down in the context + * field of the work request message, and reflected back by the adapter + * in the verbs reply message. The function handle_vq() in the interrupt + * path will use this pointer to: + * 1) append a copy of the verbs reply message + * 2) mark that the reply is ready + * 3) wake up the kernel verbs handler blocked awaiting the reply. + * + * + * The kernel verbs handlers do a "get" to put a 2nd reference on the + * VQ Request object. If the kernel verbs handler exits before the adapter + * can respond, this extra reference will keep the VQ Request object around + * until the adapter's reply can be processed. The reason we need this is + * because a pointer to this object is stuffed into the context field of + * the verbs work request message, and reflected back in the reply message. + * It is used in the interrupt handler (handle_vq()) to wake up the appropriate + * kernel verb handler that is blocked awaiting the verb reply. + * So handle_vq() will do a "put" on the object when it's done accessing it. + * NOTE: If we guarantee that the kernel verb handler will never bail before + * getting the reply, then we don't need these refcnts. + * + * + * VQ Request objects are freed by the kernel verbs handlers only + * after the verb has been processed, or when the adapter fails and + * does not reply. + * + * + * Verbs Reply Buffers: + * + * VQ Reply bufs are local host memory copies of a + * outstanding Verb Request reply + * message. The are always allocated by the kernel verbs handlers, and _may_ be + * freed by either the kernel verbs handler -or- the interrupt handler. The + * kernel verbs handler _must_ free the repbuf, then free the vq request object + * in that order. + */ + +int vq_init(struct c2_dev *c2dev) +{ + sprintf(c2dev->vq_cache_name, "c2-vq:dev%c", + (char) ('0' + c2dev->devnum)); + c2dev->host_msg_cache = + kmem_cache_create(c2dev->vq_cache_name, c2dev->rep_vq.msg_size, 0, + SLAB_HWCACHE_ALIGN, NULL); + if (c2dev->host_msg_cache == NULL) { + return -ENOMEM; + } + return 0; +} + +void vq_term(struct c2_dev *c2dev) +{ + kmem_cache_destroy(c2dev->host_msg_cache); +} + +/* vq_req_alloc - allocate a VQ Request Object and initialize it. + * The refcnt is set to 1. + */ +struct c2_vq_req *vq_req_alloc(struct c2_dev *c2dev) +{ + struct c2_vq_req *r; + + r = kmalloc(sizeof(struct c2_vq_req), GFP_KERNEL); + if (r) { + init_waitqueue_head(&r->wait_object); + r->reply_msg = 0; + r->event = 0; + r->cm_id = NULL; + r->qp = NULL; + atomic_set(&r->refcnt, 1); + atomic_set(&r->reply_ready, 0); + } + return r; +} + + +/* vq_req_free - free the VQ Request Object. It is assumed the verbs handler + * has already free the VQ Reply Buffer if it existed. + */ +void vq_req_free(struct c2_dev *c2dev, struct c2_vq_req *r) +{ + r->reply_msg = 0; + if (atomic_dec_and_test(&r->refcnt)) { + kfree(r); + } +} + +/* vq_req_get - reference a VQ Request Object. Done + * only in the kernel verbs handlers. + */ +void vq_req_get(struct c2_dev *c2dev, struct c2_vq_req *r) +{ + atomic_inc(&r->refcnt); +} + + +/* vq_req_put - dereference and potentially free a VQ Request Object. + * + * This is only called by handle_vq() on the + * interrupt when it is done processing + * a verb reply message. If the associated + * kernel verbs handler has already bailed, + * then this put will actually free the VQ + * Request object _and_ the VQ Reply Buffer + * if it exists. + */ +void vq_req_put(struct c2_dev *c2dev, struct c2_vq_req *r) +{ + if (atomic_dec_and_test(&r->refcnt)) { + if (r->reply_msg != 0) + vq_repbuf_free(c2dev, + (void *) (unsigned long) r->reply_msg); + kfree(r); + } +} + + +/* + * vq_repbuf_alloc - allocate a VQ Reply Buffer. + */ +void *vq_repbuf_alloc(struct c2_dev *c2dev) +{ + return kmem_cache_alloc(c2dev->host_msg_cache, GFP_ATOMIC); +} + +/* + * vq_send_wr - post a verbs request message to the Verbs Request Queue. + * If a message is not available in the MQ, then block until one is available. + * NOTE: handle_mq() on the interrupt context will wake up threads blocked here. + * When the adapter drains the Verbs Request Queue, + * it inserts MQ index 0 in to the + * adapter->host activity fifo and interrupts the host. + */ +int vq_send_wr(struct c2_dev *c2dev, union c2wr *wr) +{ + void *msg; + wait_queue_t __wait; + + /* + * grab adapter vq lock + */ + spin_lock(&c2dev->vqlock); + + /* + * allocate msg + */ + msg = c2_mq_alloc(&c2dev->req_vq); + + /* + * If we cannot get a msg, then we'll wait + * When a messages are available, the int handler will wake_up() + * any waiters. + */ + while (msg == NULL) { + pr_debug("%s:%d no available msg in VQ, waiting...\n", + __func__, __LINE__); + init_waitqueue_entry(&__wait, current); + add_wait_queue(&c2dev->req_vq_wo, &__wait); + spin_unlock(&c2dev->vqlock); + for (;;) { + set_current_state(TASK_INTERRUPTIBLE); + if (!c2_mq_full(&c2dev->req_vq)) { + break; + } + if (!signal_pending(current)) { + schedule_timeout(1 * HZ); /* 1 second... */ + continue; + } + set_current_state(TASK_RUNNING); + remove_wait_queue(&c2dev->req_vq_wo, &__wait); + return -EINTR; + } + set_current_state(TASK_RUNNING); + remove_wait_queue(&c2dev->req_vq_wo, &__wait); + spin_lock(&c2dev->vqlock); + msg = c2_mq_alloc(&c2dev->req_vq); + } + + /* + * copy wr into adapter msg + */ + memcpy(msg, wr, c2dev->req_vq.msg_size); + + /* + * post msg + */ + c2_mq_produce(&c2dev->req_vq); + + /* + * release adapter vq lock + */ + spin_unlock(&c2dev->vqlock); + return 0; +} + + +/* + * vq_wait_for_reply - block until the adapter posts a Verb Reply Message. + */ +int vq_wait_for_reply(struct c2_dev *c2dev, struct c2_vq_req *req) +{ + if (!wait_event_timeout(req->wait_object, + atomic_read(&req->reply_ready), + 60*HZ)) + return -ETIMEDOUT; + + return 0; +} + +/* + * vq_repbuf_free - Free a Verbs Reply Buffer. + */ +void vq_repbuf_free(struct c2_dev *c2dev, void *reply) +{ + kmem_cache_free(c2dev->host_msg_cache, reply); +} diff --git a/drivers/infiniband/hw/amso1100/c2_vq.h b/drivers/infiniband/hw/amso1100/c2_vq.h new file mode 100644 index 000000000..33805627a --- /dev/null +++ b/drivers/infiniband/hw/amso1100/c2_vq.h @@ -0,0 +1,63 @@ +/* + * Copyright (c) 2005 Ammasso, Inc. All rights reserved. + * Copyright (c) 2005 Open Grid Computing, Inc. All rights reserved. + * + * This software is available to you under a choice of one of two + * licenses. You may choose to be licensed under the terms of the GNU + * General Public License (GPL) Version 2, available from the file + * COPYING in the main directory of this source tree, or the + * OpenIB.org BSD license below: + * + * Redistribution and use in source and binary forms, with or + * without modification, are permitted provided that the following + * conditions are met: + * + * - Redistributions of source code must retain the above + * copyright notice, this list of conditions and the following + * disclaimer. + * + * - Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and/or other materials + * provided with the distribution. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS + * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN + * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ +#ifndef _C2_VQ_H_ +#define _C2_VQ_H_ +#include <linux/sched.h> +#include "c2.h" +#include "c2_wr.h" +#include "c2_provider.h" + +struct c2_vq_req { + u64 reply_msg; /* ptr to reply msg */ + wait_queue_head_t wait_object; /* wait object for vq reqs */ + atomic_t reply_ready; /* set when reply is ready */ + atomic_t refcnt; /* used to cancel WRs... */ + int event; + struct iw_cm_id *cm_id; + struct c2_qp *qp; +}; + +extern int vq_init(struct c2_dev *c2dev); +extern void vq_term(struct c2_dev *c2dev); + +extern struct c2_vq_req *vq_req_alloc(struct c2_dev *c2dev); +extern void vq_req_free(struct c2_dev *c2dev, struct c2_vq_req *req); +extern void vq_req_get(struct c2_dev *c2dev, struct c2_vq_req *req); +extern void vq_req_put(struct c2_dev *c2dev, struct c2_vq_req *req); +extern int vq_send_wr(struct c2_dev *c2dev, union c2wr * wr); + +extern void *vq_repbuf_alloc(struct c2_dev *c2dev); +extern void vq_repbuf_free(struct c2_dev *c2dev, void *reply); + +extern int vq_wait_for_reply(struct c2_dev *c2dev, struct c2_vq_req *req); +#endif /* _C2_VQ_H_ */ diff --git a/drivers/infiniband/hw/amso1100/c2_wr.h b/drivers/infiniband/hw/amso1100/c2_wr.h new file mode 100644 index 000000000..8d4b4ca46 --- /dev/null +++ b/drivers/infiniband/hw/amso1100/c2_wr.h @@ -0,0 +1,1520 @@ +/* + * Copyright (c) 2005 Ammasso, Inc. All rights reserved. + * Copyright (c) 2005 Open Grid Computing, Inc. All rights reserved. + * + * This software is available to you under a choice of one of two + * licenses. You may choose to be licensed under the terms of the GNU + * General Public License (GPL) Version 2, available from the file + * COPYING in the main directory of this source tree, or the + * OpenIB.org BSD license below: + * + * Redistribution and use in source and binary forms, with or + * without modification, are permitted provided that the following + * conditions are met: + * + * - Redistributions of source code must retain the above + * copyright notice, this list of conditions and the following + * disclaimer. + * + * - Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and/or other materials + * provided with the distribution. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS + * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN + * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ +#ifndef _C2_WR_H_ +#define _C2_WR_H_ + +#ifdef CCDEBUG +#define CCWR_MAGIC 0xb07700b0 +#endif + +#define C2_QP_NO_ATTR_CHANGE 0xFFFFFFFF + +/* Maximum allowed size in bytes of private_data exchange + * on connect. + */ +#define C2_MAX_PRIVATE_DATA_SIZE 200 + +/* + * These types are shared among the adapter, host, and CCIL consumer. + */ +enum c2_cq_notification_type { + C2_CQ_NOTIFICATION_TYPE_NONE = 1, + C2_CQ_NOTIFICATION_TYPE_NEXT, + C2_CQ_NOTIFICATION_TYPE_NEXT_SE +}; + +enum c2_setconfig_cmd { + C2_CFG_ADD_ADDR = 1, + C2_CFG_DEL_ADDR = 2, + C2_CFG_ADD_ROUTE = 3, + C2_CFG_DEL_ROUTE = 4 +}; + +enum c2_getconfig_cmd { + C2_GETCONFIG_ROUTES = 1, + C2_GETCONFIG_ADDRS +}; + +/* + * CCIL Work Request Identifiers + */ +enum c2wr_ids { + CCWR_RNIC_OPEN = 1, + CCWR_RNIC_QUERY, + CCWR_RNIC_SETCONFIG, + CCWR_RNIC_GETCONFIG, + CCWR_RNIC_CLOSE, + CCWR_CQ_CREATE, + CCWR_CQ_QUERY, + CCWR_CQ_MODIFY, + CCWR_CQ_DESTROY, + CCWR_QP_CONNECT, + CCWR_PD_ALLOC, + CCWR_PD_DEALLOC, + CCWR_SRQ_CREATE, + CCWR_SRQ_QUERY, + CCWR_SRQ_MODIFY, + CCWR_SRQ_DESTROY, + CCWR_QP_CREATE, + CCWR_QP_QUERY, + CCWR_QP_MODIFY, + CCWR_QP_DESTROY, + CCWR_NSMR_STAG_ALLOC, + CCWR_NSMR_REGISTER, + CCWR_NSMR_PBL, + CCWR_STAG_DEALLOC, + CCWR_NSMR_REREGISTER, + CCWR_SMR_REGISTER, + CCWR_MR_QUERY, + CCWR_MW_ALLOC, + CCWR_MW_QUERY, + CCWR_EP_CREATE, + CCWR_EP_GETOPT, + CCWR_EP_SETOPT, + CCWR_EP_DESTROY, + CCWR_EP_BIND, + CCWR_EP_CONNECT, + CCWR_EP_LISTEN, + CCWR_EP_SHUTDOWN, + CCWR_EP_LISTEN_CREATE, + CCWR_EP_LISTEN_DESTROY, + CCWR_EP_QUERY, + CCWR_CR_ACCEPT, + CCWR_CR_REJECT, + CCWR_CONSOLE, + CCWR_TERM, + CCWR_FLASH_INIT, + CCWR_FLASH, + CCWR_BUF_ALLOC, + CCWR_BUF_FREE, + CCWR_FLASH_WRITE, + CCWR_INIT, /* WARNING: Don't move this ever again! */ + + + + /* Add new IDs here */ + + + + /* + * WARNING: CCWR_LAST must always be the last verbs id defined! + * All the preceding IDs are fixed, and must not change. + * You can add new IDs, but must not remove or reorder + * any IDs. If you do, YOU will ruin any hope of + * compatibility between versions. + */ + CCWR_LAST, + + /* + * Start over at 1 so that arrays indexed by user wr id's + * begin at 1. This is OK since the verbs and user wr id's + * are always used on disjoint sets of queues. + */ + /* + * The order of the CCWR_SEND_XX verbs must + * match the order of the RDMA_OPs + */ + CCWR_SEND = 1, + CCWR_SEND_INV, + CCWR_SEND_SE, + CCWR_SEND_SE_INV, + CCWR_RDMA_WRITE, + CCWR_RDMA_READ, + CCWR_RDMA_READ_INV, + CCWR_MW_BIND, + CCWR_NSMR_FASTREG, + CCWR_STAG_INVALIDATE, + CCWR_RECV, + CCWR_NOP, + CCWR_UNIMPL, +/* WARNING: This must always be the last user wr id defined! */ +}; +#define RDMA_SEND_OPCODE_FROM_WR_ID(x) (x+2) + +/* + * SQ/RQ Work Request Types + */ +enum c2_wr_type { + C2_WR_TYPE_SEND = CCWR_SEND, + C2_WR_TYPE_SEND_SE = CCWR_SEND_SE, + C2_WR_TYPE_SEND_INV = CCWR_SEND_INV, + C2_WR_TYPE_SEND_SE_INV = CCWR_SEND_SE_INV, + C2_WR_TYPE_RDMA_WRITE = CCWR_RDMA_WRITE, + C2_WR_TYPE_RDMA_READ = CCWR_RDMA_READ, + C2_WR_TYPE_RDMA_READ_INV_STAG = CCWR_RDMA_READ_INV, + C2_WR_TYPE_BIND_MW = CCWR_MW_BIND, + C2_WR_TYPE_FASTREG_NSMR = CCWR_NSMR_FASTREG, + C2_WR_TYPE_INV_STAG = CCWR_STAG_INVALIDATE, + C2_WR_TYPE_RECV = CCWR_RECV, + C2_WR_TYPE_NOP = CCWR_NOP, +}; + +struct c2_netaddr { + __be32 ip_addr; + __be32 netmask; + u32 mtu; +}; + +struct c2_route { + u32 ip_addr; /* 0 indicates the default route */ + u32 netmask; /* netmask associated with dst */ + u32 flags; + union { + u32 ipaddr; /* address of the nexthop interface */ + u8 enaddr[6]; + } nexthop; +}; + +/* + * A Scatter Gather Entry. + */ +struct c2_data_addr { + __be32 stag; + __be32 length; + __be64 to; +}; + +/* + * MR and MW flags used by the consumer, RI, and RNIC. + */ +enum c2_mm_flags { + MEM_REMOTE = 0x0001, /* allow mw binds with remote access. */ + MEM_VA_BASED = 0x0002, /* Not Zero-based */ + MEM_PBL_COMPLETE = 0x0004, /* PBL array is complete in this msg */ + MEM_LOCAL_READ = 0x0008, /* allow local reads */ + MEM_LOCAL_WRITE = 0x0010, /* allow local writes */ + MEM_REMOTE_READ = 0x0020, /* allow remote reads */ + MEM_REMOTE_WRITE = 0x0040, /* allow remote writes */ + MEM_WINDOW_BIND = 0x0080, /* binds allowed */ + MEM_SHARED = 0x0100, /* set if MR is shared */ + MEM_STAG_VALID = 0x0200 /* set if STAG is in valid state */ +}; + +/* + * CCIL API ACF flags defined in terms of the low level mem flags. + * This minimizes translation needed in the user API + */ +enum c2_acf { + C2_ACF_LOCAL_READ = MEM_LOCAL_READ, + C2_ACF_LOCAL_WRITE = MEM_LOCAL_WRITE, + C2_ACF_REMOTE_READ = MEM_REMOTE_READ, + C2_ACF_REMOTE_WRITE = MEM_REMOTE_WRITE, + C2_ACF_WINDOW_BIND = MEM_WINDOW_BIND +}; + +/* + * Image types of objects written to flash + */ +#define C2_FLASH_IMG_BITFILE 1 +#define C2_FLASH_IMG_OPTION_ROM 2 +#define C2_FLASH_IMG_VPD 3 + +/* + * to fix bug 1815 we define the max size allowable of the + * terminate message (per the IETF spec).Refer to the IETF + * protocol specification, section 12.1.6, page 64) + * The message is prefixed by 20 types of DDP info. + * + * Then the message has 6 bytes for the terminate control + * and DDP segment length info plus a DDP header (either + * 14 or 18 byts) plus 28 bytes for the RDMA header. + * Thus the max size in: + * 20 + (6 + 18 + 28) = 72 + */ +#define C2_MAX_TERMINATE_MESSAGE_SIZE (72) + +/* + * Build String Length. It must be the same as C2_BUILD_STR_LEN in ccil_api.h + */ +#define WR_BUILD_STR_LEN 64 + +/* + * WARNING: All of these structs need to align any 64bit types on + * 64 bit boundaries! 64bit types include u64 and u64. + */ + +/* + * Clustercore Work Request Header. Be sensitive to field layout + * and alignment. + */ +struct c2wr_hdr { + /* wqe_count is part of the cqe. It is put here so the + * adapter can write to it while the wr is pending without + * clobbering part of the wr. This word need not be dma'd + * from the host to adapter by libccil, but we copy it anyway + * to make the memcpy to the adapter better aligned. + */ + __be32 wqe_count; + + /* Put these fields next so that later 32- and 64-bit + * quantities are naturally aligned. + */ + u8 id; + u8 result; /* adapter -> host */ + u8 sge_count; /* host -> adapter */ + u8 flags; /* host -> adapter */ + + u64 context; +#ifdef CCMSGMAGIC + u32 magic; + u32 pad; +#endif +} __attribute__((packed)); + +/* + *------------------------ RNIC ------------------------ + */ + +/* + * WR_RNIC_OPEN + */ + +/* + * Flags for the RNIC WRs + */ +enum c2_rnic_flags { + RNIC_IRD_STATIC = 0x0001, + RNIC_ORD_STATIC = 0x0002, + RNIC_QP_STATIC = 0x0004, + RNIC_SRQ_SUPPORTED = 0x0008, + RNIC_PBL_BLOCK_MODE = 0x0010, + RNIC_SRQ_MODEL_ARRIVAL = 0x0020, + RNIC_CQ_OVF_DETECTED = 0x0040, + RNIC_PRIV_MODE = 0x0080 +}; + +struct c2wr_rnic_open_req { + struct c2wr_hdr hdr; + u64 user_context; + __be16 flags; /* See enum c2_rnic_flags */ + __be16 port_num; +} __attribute__((packed)); + +struct c2wr_rnic_open_rep { + struct c2wr_hdr hdr; + u32 rnic_handle; +} __attribute__((packed)); + +union c2wr_rnic_open { + struct c2wr_rnic_open_req req; + struct c2wr_rnic_open_rep rep; +} __attribute__((packed)); + +struct c2wr_rnic_query_req { + struct c2wr_hdr hdr; + u32 rnic_handle; +} __attribute__((packed)); + +/* + * WR_RNIC_QUERY + */ +struct c2wr_rnic_query_rep { + struct c2wr_hdr hdr; + u64 user_context; + __be32 vendor_id; + __be32 part_number; + __be32 hw_version; + __be32 fw_ver_major; + __be32 fw_ver_minor; + __be32 fw_ver_patch; + char fw_ver_build_str[WR_BUILD_STR_LEN]; + __be32 max_qps; + __be32 max_qp_depth; + u32 max_srq_depth; + u32 max_send_sgl_depth; + u32 max_rdma_sgl_depth; + __be32 max_cqs; + __be32 max_cq_depth; + u32 max_cq_event_handlers; + __be32 max_mrs; + u32 max_pbl_depth; + __be32 max_pds; + __be32 max_global_ird; + u32 max_global_ord; + __be32 max_qp_ird; + __be32 max_qp_ord; + u32 flags; + __be32 max_mws; + u32 pbe_range_low; + u32 pbe_range_high; + u32 max_srqs; + u32 page_size; +} __attribute__((packed)); + +union c2wr_rnic_query { + struct c2wr_rnic_query_req req; + struct c2wr_rnic_query_rep rep; +} __attribute__((packed)); + +/* + * WR_RNIC_GETCONFIG + */ + +struct c2wr_rnic_getconfig_req { + struct c2wr_hdr hdr; + u32 rnic_handle; + u32 option; /* see c2_getconfig_cmd_t */ + u64 reply_buf; + u32 reply_buf_len; +} __attribute__((packed)) ; + +struct c2wr_rnic_getconfig_rep { + struct c2wr_hdr hdr; + u32 option; /* see c2_getconfig_cmd_t */ + u32 count_len; /* length of the number of addresses configured */ +} __attribute__((packed)) ; + +union c2wr_rnic_getconfig { + struct c2wr_rnic_getconfig_req req; + struct c2wr_rnic_getconfig_rep rep; +} __attribute__((packed)) ; + +/* + * WR_RNIC_SETCONFIG + */ +struct c2wr_rnic_setconfig_req { + struct c2wr_hdr hdr; + u32 rnic_handle; + __be32 option; /* See c2_setconfig_cmd_t */ + /* variable data and pad. See c2_netaddr and c2_route */ + u8 data[0]; +} __attribute__((packed)) ; + +struct c2wr_rnic_setconfig_rep { + struct c2wr_hdr hdr; +} __attribute__((packed)) ; + +union c2wr_rnic_setconfig { + struct c2wr_rnic_setconfig_req req; + struct c2wr_rnic_setconfig_rep rep; +} __attribute__((packed)) ; + +/* + * WR_RNIC_CLOSE + */ +struct c2wr_rnic_close_req { + struct c2wr_hdr hdr; + u32 rnic_handle; +} __attribute__((packed)) ; + +struct c2wr_rnic_close_rep { + struct c2wr_hdr hdr; +} __attribute__((packed)) ; + +union c2wr_rnic_close { + struct c2wr_rnic_close_req req; + struct c2wr_rnic_close_rep rep; +} __attribute__((packed)) ; + +/* + *------------------------ CQ ------------------------ + */ +struct c2wr_cq_create_req { + struct c2wr_hdr hdr; + __be64 shared_ht; + u64 user_context; + __be64 msg_pool; + u32 rnic_handle; + __be32 msg_size; + __be32 depth; +} __attribute__((packed)) ; + +struct c2wr_cq_create_rep { + struct c2wr_hdr hdr; + __be32 mq_index; + __be32 adapter_shared; + u32 cq_handle; +} __attribute__((packed)) ; + +union c2wr_cq_create { + struct c2wr_cq_create_req req; + struct c2wr_cq_create_rep rep; +} __attribute__((packed)) ; + +struct c2wr_cq_modify_req { + struct c2wr_hdr hdr; + u32 rnic_handle; + u32 cq_handle; + u32 new_depth; + u64 new_msg_pool; +} __attribute__((packed)) ; + +struct c2wr_cq_modify_rep { + struct c2wr_hdr hdr; +} __attribute__((packed)) ; + +union c2wr_cq_modify { + struct c2wr_cq_modify_req req; + struct c2wr_cq_modify_rep rep; +} __attribute__((packed)) ; + +struct c2wr_cq_destroy_req { + struct c2wr_hdr hdr; + u32 rnic_handle; + u32 cq_handle; +} __attribute__((packed)) ; + +struct c2wr_cq_destroy_rep { + struct c2wr_hdr hdr; +} __attribute__((packed)) ; + +union c2wr_cq_destroy { + struct c2wr_cq_destroy_req req; + struct c2wr_cq_destroy_rep rep; +} __attribute__((packed)) ; + +/* + *------------------------ PD ------------------------ + */ +struct c2wr_pd_alloc_req { + struct c2wr_hdr hdr; + u32 rnic_handle; + u32 pd_id; +} __attribute__((packed)) ; + +struct c2wr_pd_alloc_rep { + struct c2wr_hdr hdr; +} __attribute__((packed)) ; + +union c2wr_pd_alloc { + struct c2wr_pd_alloc_req req; + struct c2wr_pd_alloc_rep rep; +} __attribute__((packed)) ; + +struct c2wr_pd_dealloc_req { + struct c2wr_hdr hdr; + u32 rnic_handle; + u32 pd_id; +} __attribute__((packed)) ; + +struct c2wr_pd_dealloc_rep { + struct c2wr_hdr hdr; +} __attribute__((packed)) ; + +union c2wr_pd_dealloc { + struct c2wr_pd_dealloc_req req; + struct c2wr_pd_dealloc_rep rep; +} __attribute__((packed)) ; + +/* + *------------------------ SRQ ------------------------ + */ +struct c2wr_srq_create_req { + struct c2wr_hdr hdr; + u64 shared_ht; + u64 user_context; + u32 rnic_handle; + u32 srq_depth; + u32 srq_limit; + u32 sgl_depth; + u32 pd_id; +} __attribute__((packed)) ; + +struct c2wr_srq_create_rep { + struct c2wr_hdr hdr; + u32 srq_depth; + u32 sgl_depth; + u32 msg_size; + u32 mq_index; + u32 mq_start; + u32 srq_handle; +} __attribute__((packed)) ; + +union c2wr_srq_create { + struct c2wr_srq_create_req req; + struct c2wr_srq_create_rep rep; +} __attribute__((packed)) ; + +struct c2wr_srq_destroy_req { + struct c2wr_hdr hdr; + u32 rnic_handle; + u32 srq_handle; +} __attribute__((packed)) ; + +struct c2wr_srq_destroy_rep { + struct c2wr_hdr hdr; +} __attribute__((packed)) ; + +union c2wr_srq_destroy { + struct c2wr_srq_destroy_req req; + struct c2wr_srq_destroy_rep rep; +} __attribute__((packed)) ; + +/* + *------------------------ QP ------------------------ + */ +enum c2wr_qp_flags { + QP_RDMA_READ = 0x00000001, /* RDMA read enabled? */ + QP_RDMA_WRITE = 0x00000002, /* RDMA write enabled? */ + QP_MW_BIND = 0x00000004, /* MWs enabled */ + QP_ZERO_STAG = 0x00000008, /* enabled? */ + QP_REMOTE_TERMINATION = 0x00000010, /* remote end terminated */ + QP_RDMA_READ_RESPONSE = 0x00000020 /* Remote RDMA read */ + /* enabled? */ +}; + +struct c2wr_qp_create_req { + struct c2wr_hdr hdr; + __be64 shared_sq_ht; + __be64 shared_rq_ht; + u64 user_context; + u32 rnic_handle; + u32 sq_cq_handle; + u32 rq_cq_handle; + __be32 sq_depth; + __be32 rq_depth; + u32 srq_handle; + u32 srq_limit; + __be32 flags; /* see enum c2wr_qp_flags */ + __be32 send_sgl_depth; + __be32 recv_sgl_depth; + __be32 rdma_write_sgl_depth; + __be32 ord; + __be32 ird; + u32 pd_id; +} __attribute__((packed)) ; + +struct c2wr_qp_create_rep { + struct c2wr_hdr hdr; + __be32 sq_depth; + __be32 rq_depth; + u32 send_sgl_depth; + u32 recv_sgl_depth; + u32 rdma_write_sgl_depth; + u32 ord; + u32 ird; + __be32 sq_msg_size; + __be32 sq_mq_index; + __be32 sq_mq_start; + __be32 rq_msg_size; + __be32 rq_mq_index; + __be32 rq_mq_start; + u32 qp_handle; +} __attribute__((packed)) ; + +union c2wr_qp_create { + struct c2wr_qp_create_req req; + struct c2wr_qp_create_rep rep; +} __attribute__((packed)) ; + +struct c2wr_qp_query_req { + struct c2wr_hdr hdr; + u32 rnic_handle; + u32 qp_handle; +} __attribute__((packed)) ; + +struct c2wr_qp_query_rep { + struct c2wr_hdr hdr; + u64 user_context; + u32 rnic_handle; + u32 sq_depth; + u32 rq_depth; + u32 send_sgl_depth; + u32 rdma_write_sgl_depth; + u32 recv_sgl_depth; + u32 ord; + u32 ird; + u16 qp_state; + u16 flags; /* see c2wr_qp_flags_t */ + u32 qp_id; + u32 local_addr; + u32 remote_addr; + u16 local_port; + u16 remote_port; + u32 terminate_msg_length; /* 0 if not present */ + u8 data[0]; + /* Terminate Message in-line here. */ +} __attribute__((packed)) ; + +union c2wr_qp_query { + struct c2wr_qp_query_req req; + struct c2wr_qp_query_rep rep; +} __attribute__((packed)) ; + +struct c2wr_qp_modify_req { + struct c2wr_hdr hdr; + u64 stream_msg; + u32 stream_msg_length; + u32 rnic_handle; + u32 qp_handle; + __be32 next_qp_state; + __be32 ord; + __be32 ird; + __be32 sq_depth; + __be32 rq_depth; + u32 llp_ep_handle; +} __attribute__((packed)) ; + +struct c2wr_qp_modify_rep { + struct c2wr_hdr hdr; + u32 ord; + u32 ird; + u32 sq_depth; + u32 rq_depth; + u32 sq_msg_size; + u32 sq_mq_index; + u32 sq_mq_start; + u32 rq_msg_size; + u32 rq_mq_index; + u32 rq_mq_start; +} __attribute__((packed)) ; + +union c2wr_qp_modify { + struct c2wr_qp_modify_req req; + struct c2wr_qp_modify_rep rep; +} __attribute__((packed)) ; + +struct c2wr_qp_destroy_req { + struct c2wr_hdr hdr; + u32 rnic_handle; + u32 qp_handle; +} __attribute__((packed)) ; + +struct c2wr_qp_destroy_rep { + struct c2wr_hdr hdr; +} __attribute__((packed)) ; + +union c2wr_qp_destroy { + struct c2wr_qp_destroy_req req; + struct c2wr_qp_destroy_rep rep; +} __attribute__((packed)) ; + +/* + * The CCWR_QP_CONNECT msg is posted on the verbs request queue. It can + * only be posted when a QP is in IDLE state. After the connect request is + * submitted to the LLP, the adapter moves the QP to CONNECT_PENDING state. + * No synchronous reply from adapter to this WR. The results of + * connection are passed back in an async event CCAE_ACTIVE_CONNECT_RESULTS + * See c2wr_ae_active_connect_results_t + */ +struct c2wr_qp_connect_req { + struct c2wr_hdr hdr; + u32 rnic_handle; + u32 qp_handle; + __be32 remote_addr; + __be16 remote_port; + u16 pad; + __be32 private_data_length; + u8 private_data[0]; /* Private data in-line. */ +} __attribute__((packed)) ; + +struct c2wr_qp_connect { + struct c2wr_qp_connect_req req; + /* no synchronous reply. */ +} __attribute__((packed)) ; + + +/* + *------------------------ MM ------------------------ + */ + +struct c2wr_nsmr_stag_alloc_req { + struct c2wr_hdr hdr; + u32 rnic_handle; + u32 pbl_depth; + u32 pd_id; + u32 flags; +} __attribute__((packed)) ; + +struct c2wr_nsmr_stag_alloc_rep { + struct c2wr_hdr hdr; + u32 pbl_depth; + u32 stag_index; +} __attribute__((packed)) ; + +union c2wr_nsmr_stag_alloc { + struct c2wr_nsmr_stag_alloc_req req; + struct c2wr_nsmr_stag_alloc_rep rep; +} __attribute__((packed)) ; + +struct c2wr_nsmr_register_req { + struct c2wr_hdr hdr; + __be64 va; + u32 rnic_handle; + __be16 flags; + u8 stag_key; + u8 pad; + u32 pd_id; + __be32 pbl_depth; + __be32 pbe_size; + __be32 fbo; + __be32 length; + __be32 addrs_length; + /* array of paddrs (must be aligned on a 64bit boundary) */ + __be64 paddrs[0]; +} __attribute__((packed)) ; + +struct c2wr_nsmr_register_rep { + struct c2wr_hdr hdr; + u32 pbl_depth; + __be32 stag_index; +} __attribute__((packed)) ; + +union c2wr_nsmr_register { + struct c2wr_nsmr_register_req req; + struct c2wr_nsmr_register_rep rep; +} __attribute__((packed)) ; + +struct c2wr_nsmr_pbl_req { + struct c2wr_hdr hdr; + u32 rnic_handle; + __be32 flags; + __be32 stag_index; + __be32 addrs_length; + /* array of paddrs (must be aligned on a 64bit boundary) */ + __be64 paddrs[0]; +} __attribute__((packed)) ; + +struct c2wr_nsmr_pbl_rep { + struct c2wr_hdr hdr; +} __attribute__((packed)) ; + +union c2wr_nsmr_pbl { + struct c2wr_nsmr_pbl_req req; + struct c2wr_nsmr_pbl_rep rep; +} __attribute__((packed)) ; + +struct c2wr_mr_query_req { + struct c2wr_hdr hdr; + u32 rnic_handle; + u32 stag_index; +} __attribute__((packed)) ; + +struct c2wr_mr_query_rep { + struct c2wr_hdr hdr; + u8 stag_key; + u8 pad[3]; + u32 pd_id; + u32 flags; + u32 pbl_depth; +} __attribute__((packed)) ; + +union c2wr_mr_query { + struct c2wr_mr_query_req req; + struct c2wr_mr_query_rep rep; +} __attribute__((packed)) ; + +struct c2wr_mw_query_req { + struct c2wr_hdr hdr; + u32 rnic_handle; + u32 stag_index; +} __attribute__((packed)) ; + +struct c2wr_mw_query_rep { + struct c2wr_hdr hdr; + u8 stag_key; + u8 pad[3]; + u32 pd_id; + u32 flags; +} __attribute__((packed)) ; + +union c2wr_mw_query { + struct c2wr_mw_query_req req; + struct c2wr_mw_query_rep rep; +} __attribute__((packed)) ; + + +struct c2wr_stag_dealloc_req { + struct c2wr_hdr hdr; + u32 rnic_handle; + __be32 stag_index; +} __attribute__((packed)) ; + +struct c2wr_stag_dealloc_rep { + struct c2wr_hdr hdr; +} __attribute__((packed)) ; + +union c2wr_stag_dealloc { + struct c2wr_stag_dealloc_req req; + struct c2wr_stag_dealloc_rep rep; +} __attribute__((packed)) ; + +struct c2wr_nsmr_reregister_req { + struct c2wr_hdr hdr; + u64 va; + u32 rnic_handle; + u16 flags; + u8 stag_key; + u8 pad; + u32 stag_index; + u32 pd_id; + u32 pbl_depth; + u32 pbe_size; + u32 fbo; + u32 length; + u32 addrs_length; + u32 pad1; + /* array of paddrs (must be aligned on a 64bit boundary) */ + u64 paddrs[0]; +} __attribute__((packed)) ; + +struct c2wr_nsmr_reregister_rep { + struct c2wr_hdr hdr; + u32 pbl_depth; + u32 stag_index; +} __attribute__((packed)) ; + +union c2wr_nsmr_reregister { + struct c2wr_nsmr_reregister_req req; + struct c2wr_nsmr_reregister_rep rep; +} __attribute__((packed)) ; + +struct c2wr_smr_register_req { + struct c2wr_hdr hdr; + u64 va; + u32 rnic_handle; + u16 flags; + u8 stag_key; + u8 pad; + u32 stag_index; + u32 pd_id; +} __attribute__((packed)) ; + +struct c2wr_smr_register_rep { + struct c2wr_hdr hdr; + u32 stag_index; +} __attribute__((packed)) ; + +union c2wr_smr_register { + struct c2wr_smr_register_req req; + struct c2wr_smr_register_rep rep; +} __attribute__((packed)) ; + +struct c2wr_mw_alloc_req { + struct c2wr_hdr hdr; + u32 rnic_handle; + u32 pd_id; +} __attribute__((packed)) ; + +struct c2wr_mw_alloc_rep { + struct c2wr_hdr hdr; + u32 stag_index; +} __attribute__((packed)) ; + +union c2wr_mw_alloc { + struct c2wr_mw_alloc_req req; + struct c2wr_mw_alloc_rep rep; +} __attribute__((packed)) ; + +/* + *------------------------ WRs ----------------------- + */ + +struct c2wr_user_hdr { + struct c2wr_hdr hdr; /* Has status and WR Type */ +} __attribute__((packed)) ; + +enum c2_qp_state { + C2_QP_STATE_IDLE = 0x01, + C2_QP_STATE_CONNECTING = 0x02, + C2_QP_STATE_RTS = 0x04, + C2_QP_STATE_CLOSING = 0x08, + C2_QP_STATE_TERMINATE = 0x10, + C2_QP_STATE_ERROR = 0x20, +}; + +/* Completion queue entry. */ +struct c2wr_ce { + struct c2wr_hdr hdr; /* Has status and WR Type */ + u64 qp_user_context; /* c2_user_qp_t * */ + u32 qp_state; /* Current QP State */ + u32 handle; /* QPID or EP Handle */ + __be32 bytes_rcvd; /* valid for RECV WCs */ + u32 stag; +} __attribute__((packed)) ; + + +/* + * Flags used for all post-sq WRs. These must fit in the flags + * field of the struct c2wr_hdr (eight bits). + */ +enum { + SQ_SIGNALED = 0x01, + SQ_READ_FENCE = 0x02, + SQ_FENCE = 0x04, +}; + +/* + * Common fields for all post-sq WRs. Namely the standard header and a + * secondary header with fields common to all post-sq WRs. + */ +struct c2_sq_hdr { + struct c2wr_user_hdr user_hdr; +} __attribute__((packed)); + +/* + * Same as above but for post-rq WRs. + */ +struct c2_rq_hdr { + struct c2wr_user_hdr user_hdr; +} __attribute__((packed)); + +/* + * use the same struct for all sends. + */ +struct c2wr_send_req { + struct c2_sq_hdr sq_hdr; + __be32 sge_len; + __be32 remote_stag; + u8 data[0]; /* SGE array */ +} __attribute__((packed)); + +union c2wr_send { + struct c2wr_send_req req; + struct c2wr_ce rep; +} __attribute__((packed)); + +struct c2wr_rdma_write_req { + struct c2_sq_hdr sq_hdr; + __be64 remote_to; + __be32 remote_stag; + __be32 sge_len; + u8 data[0]; /* SGE array */ +} __attribute__((packed)); + +union c2wr_rdma_write { + struct c2wr_rdma_write_req req; + struct c2wr_ce rep; +} __attribute__((packed)); + +struct c2wr_rdma_read_req { + struct c2_sq_hdr sq_hdr; + __be64 local_to; + __be64 remote_to; + __be32 local_stag; + __be32 remote_stag; + __be32 length; +} __attribute__((packed)); + +union c2wr_rdma_read { + struct c2wr_rdma_read_req req; + struct c2wr_ce rep; +} __attribute__((packed)); + +struct c2wr_mw_bind_req { + struct c2_sq_hdr sq_hdr; + u64 va; + u8 stag_key; + u8 pad[3]; + u32 mw_stag_index; + u32 mr_stag_index; + u32 length; + u32 flags; +} __attribute__((packed)); + +union c2wr_mw_bind { + struct c2wr_mw_bind_req req; + struct c2wr_ce rep; +} __attribute__((packed)); + +struct c2wr_nsmr_fastreg_req { + struct c2_sq_hdr sq_hdr; + u64 va; + u8 stag_key; + u8 pad[3]; + u32 stag_index; + u32 pbe_size; + u32 fbo; + u32 length; + u32 addrs_length; + /* array of paddrs (must be aligned on a 64bit boundary) */ + u64 paddrs[0]; +} __attribute__((packed)); + +union c2wr_nsmr_fastreg { + struct c2wr_nsmr_fastreg_req req; + struct c2wr_ce rep; +} __attribute__((packed)); + +struct c2wr_stag_invalidate_req { + struct c2_sq_hdr sq_hdr; + u8 stag_key; + u8 pad[3]; + u32 stag_index; +} __attribute__((packed)); + +union c2wr_stag_invalidate { + struct c2wr_stag_invalidate_req req; + struct c2wr_ce rep; +} __attribute__((packed)); + +union c2wr_sqwr { + struct c2_sq_hdr sq_hdr; + struct c2wr_send_req send; + struct c2wr_send_req send_se; + struct c2wr_send_req send_inv; + struct c2wr_send_req send_se_inv; + struct c2wr_rdma_write_req rdma_write; + struct c2wr_rdma_read_req rdma_read; + struct c2wr_mw_bind_req mw_bind; + struct c2wr_nsmr_fastreg_req nsmr_fastreg; + struct c2wr_stag_invalidate_req stag_inv; +} __attribute__((packed)); + + +/* + * RQ WRs + */ +struct c2wr_rqwr { + struct c2_rq_hdr rq_hdr; + u8 data[0]; /* array of SGEs */ +} __attribute__((packed)); + +union c2wr_recv { + struct c2wr_rqwr req; + struct c2wr_ce rep; +} __attribute__((packed)); + +/* + * All AEs start with this header. Most AEs only need to convey the + * information in the header. Some, like LLP connection events, need + * more info. The union typdef c2wr_ae_t has all the possible AEs. + * + * hdr.context is the user_context from the rnic_open WR. NULL If this + * is not affiliated with an rnic + * + * hdr.id is the AE identifier (eg; CCAE_REMOTE_SHUTDOWN, + * CCAE_LLP_CLOSE_COMPLETE) + * + * resource_type is one of: C2_RES_IND_QP, C2_RES_IND_CQ, C2_RES_IND_SRQ + * + * user_context is the context passed down when the host created the resource. + */ +struct c2wr_ae_hdr { + struct c2wr_hdr hdr; + u64 user_context; /* user context for this res. */ + __be32 resource_type; /* see enum c2_resource_indicator */ + __be32 resource; /* handle for resource */ + __be32 qp_state; /* current QP State */ +} __attribute__((packed)); + +/* + * After submitting the CCAE_ACTIVE_CONNECT_RESULTS message on the AEQ, + * the adapter moves the QP into RTS state + */ +struct c2wr_ae_active_connect_results { + struct c2wr_ae_hdr ae_hdr; + __be32 laddr; + __be32 raddr; + __be16 lport; + __be16 rport; + __be32 private_data_length; + u8 private_data[0]; /* data is in-line in the msg. */ +} __attribute__((packed)); + +/* + * When connections are established by the stack (and the private data + * MPA frame is received), the adapter will generate an event to the host. + * The details of the connection, any private data, and the new connection + * request handle is passed up via the CCAE_CONNECTION_REQUEST msg on the + * AE queue: + */ +struct c2wr_ae_connection_request { + struct c2wr_ae_hdr ae_hdr; + u32 cr_handle; /* connreq handle (sock ptr) */ + __be32 laddr; + __be32 raddr; + __be16 lport; + __be16 rport; + __be32 private_data_length; + u8 private_data[0]; /* data is in-line in the msg. */ +} __attribute__((packed)); + +union c2wr_ae { + struct c2wr_ae_hdr ae_generic; + struct c2wr_ae_active_connect_results ae_active_connect_results; + struct c2wr_ae_connection_request ae_connection_request; +} __attribute__((packed)); + +struct c2wr_init_req { + struct c2wr_hdr hdr; + __be64 hint_count; + __be64 q0_host_shared; + __be64 q1_host_shared; + __be64 q1_host_msg_pool; + __be64 q2_host_shared; + __be64 q2_host_msg_pool; +} __attribute__((packed)); + +struct c2wr_init_rep { + struct c2wr_hdr hdr; +} __attribute__((packed)); + +union c2wr_init { + struct c2wr_init_req req; + struct c2wr_init_rep rep; +} __attribute__((packed)); + +/* + * For upgrading flash. + */ + +struct c2wr_flash_init_req { + struct c2wr_hdr hdr; + u32 rnic_handle; +} __attribute__((packed)); + +struct c2wr_flash_init_rep { + struct c2wr_hdr hdr; + u32 adapter_flash_buf_offset; + u32 adapter_flash_len; +} __attribute__((packed)); + +union c2wr_flash_init { + struct c2wr_flash_init_req req; + struct c2wr_flash_init_rep rep; +} __attribute__((packed)); + +struct c2wr_flash_req { + struct c2wr_hdr hdr; + u32 rnic_handle; + u32 len; +} __attribute__((packed)); + +struct c2wr_flash_rep { + struct c2wr_hdr hdr; + u32 status; +} __attribute__((packed)); + +union c2wr_flash { + struct c2wr_flash_req req; + struct c2wr_flash_rep rep; +} __attribute__((packed)); + +struct c2wr_buf_alloc_req { + struct c2wr_hdr hdr; + u32 rnic_handle; + u32 size; +} __attribute__((packed)); + +struct c2wr_buf_alloc_rep { + struct c2wr_hdr hdr; + u32 offset; /* 0 if mem not available */ + u32 size; /* 0 if mem not available */ +} __attribute__((packed)); + +union c2wr_buf_alloc { + struct c2wr_buf_alloc_req req; + struct c2wr_buf_alloc_rep rep; +} __attribute__((packed)); + +struct c2wr_buf_free_req { + struct c2wr_hdr hdr; + u32 rnic_handle; + u32 offset; /* Must match value from alloc */ + u32 size; /* Must match value from alloc */ +} __attribute__((packed)); + +struct c2wr_buf_free_rep { + struct c2wr_hdr hdr; +} __attribute__((packed)); + +union c2wr_buf_free { + struct c2wr_buf_free_req req; + struct c2wr_ce rep; +} __attribute__((packed)); + +struct c2wr_flash_write_req { + struct c2wr_hdr hdr; + u32 rnic_handle; + u32 offset; + u32 size; + u32 type; + u32 flags; +} __attribute__((packed)); + +struct c2wr_flash_write_rep { + struct c2wr_hdr hdr; + u32 status; +} __attribute__((packed)); + +union c2wr_flash_write { + struct c2wr_flash_write_req req; + struct c2wr_flash_write_rep rep; +} __attribute__((packed)); + +/* + * Messages for LLP connection setup. + */ + +/* + * Listen Request. This allocates a listening endpoint to allow passive + * connection setup. Newly established LLP connections are passed up + * via an AE. See c2wr_ae_connection_request_t + */ +struct c2wr_ep_listen_create_req { + struct c2wr_hdr hdr; + u64 user_context; /* returned in AEs. */ + u32 rnic_handle; + __be32 local_addr; /* local addr, or 0 */ + __be16 local_port; /* 0 means "pick one" */ + u16 pad; + __be32 backlog; /* tradional tcp listen bl */ +} __attribute__((packed)); + +struct c2wr_ep_listen_create_rep { + struct c2wr_hdr hdr; + u32 ep_handle; /* handle to new listening ep */ + u16 local_port; /* resulting port... */ + u16 pad; +} __attribute__((packed)); + +union c2wr_ep_listen_create { + struct c2wr_ep_listen_create_req req; + struct c2wr_ep_listen_create_rep rep; +} __attribute__((packed)); + +struct c2wr_ep_listen_destroy_req { + struct c2wr_hdr hdr; + u32 rnic_handle; + u32 ep_handle; +} __attribute__((packed)); + +struct c2wr_ep_listen_destroy_rep { + struct c2wr_hdr hdr; +} __attribute__((packed)); + +union c2wr_ep_listen_destroy { + struct c2wr_ep_listen_destroy_req req; + struct c2wr_ep_listen_destroy_rep rep; +} __attribute__((packed)); + +struct c2wr_ep_query_req { + struct c2wr_hdr hdr; + u32 rnic_handle; + u32 ep_handle; +} __attribute__((packed)); + +struct c2wr_ep_query_rep { + struct c2wr_hdr hdr; + u32 rnic_handle; + u32 local_addr; + u32 remote_addr; + u16 local_port; + u16 remote_port; +} __attribute__((packed)); + +union c2wr_ep_query { + struct c2wr_ep_query_req req; + struct c2wr_ep_query_rep rep; +} __attribute__((packed)); + + +/* + * The host passes this down to indicate acceptance of a pending iWARP + * connection. The cr_handle was obtained from the CONNECTION_REQUEST + * AE passed up by the adapter. See c2wr_ae_connection_request_t. + */ +struct c2wr_cr_accept_req { + struct c2wr_hdr hdr; + u32 rnic_handle; + u32 qp_handle; /* QP to bind to this LLP conn */ + u32 ep_handle; /* LLP handle to accept */ + __be32 private_data_length; + u8 private_data[0]; /* data in-line in msg. */ +} __attribute__((packed)); + +/* + * adapter sends reply when private data is successfully submitted to + * the LLP. + */ +struct c2wr_cr_accept_rep { + struct c2wr_hdr hdr; +} __attribute__((packed)); + +union c2wr_cr_accept { + struct c2wr_cr_accept_req req; + struct c2wr_cr_accept_rep rep; +} __attribute__((packed)); + +/* + * The host sends this down if a given iWARP connection request was + * rejected by the consumer. The cr_handle was obtained from a + * previous c2wr_ae_connection_request_t AE sent by the adapter. + */ +struct c2wr_cr_reject_req { + struct c2wr_hdr hdr; + u32 rnic_handle; + u32 ep_handle; /* LLP handle to reject */ +} __attribute__((packed)); + +/* + * Dunno if this is needed, but we'll add it for now. The adapter will + * send the reject_reply after the LLP endpoint has been destroyed. + */ +struct c2wr_cr_reject_rep { + struct c2wr_hdr hdr; +} __attribute__((packed)); + +union c2wr_cr_reject { + struct c2wr_cr_reject_req req; + struct c2wr_cr_reject_rep rep; +} __attribute__((packed)); + +/* + * console command. Used to implement a debug console over the verbs + * request and reply queues. + */ + +/* + * Console request message. It contains: + * - message hdr with id = CCWR_CONSOLE + * - the physaddr/len of host memory to be used for the reply. + * - the command string. eg: "netstat -s" or "zoneinfo" + */ +struct c2wr_console_req { + struct c2wr_hdr hdr; /* id = CCWR_CONSOLE */ + u64 reply_buf; /* pinned host buf for reply */ + u32 reply_buf_len; /* length of reply buffer */ + u8 command[0]; /* NUL terminated ascii string */ + /* containing the command req */ +} __attribute__((packed)); + +/* + * flags used in the console reply. + */ +enum c2_console_flags { + CONS_REPLY_TRUNCATED = 0x00000001 /* reply was truncated */ +} __attribute__((packed)); + +/* + * Console reply message. + * hdr.result contains the c2_status_t error if the reply was _not_ generated, + * or C2_OK if the reply was generated. + */ +struct c2wr_console_rep { + struct c2wr_hdr hdr; /* id = CCWR_CONSOLE */ + u32 flags; +} __attribute__((packed)); + +union c2wr_console { + struct c2wr_console_req req; + struct c2wr_console_rep rep; +} __attribute__((packed)); + + +/* + * Giant union with all WRs. Makes life easier... + */ +union c2wr { + struct c2wr_hdr hdr; + struct c2wr_user_hdr user_hdr; + union c2wr_rnic_open rnic_open; + union c2wr_rnic_query rnic_query; + union c2wr_rnic_getconfig rnic_getconfig; + union c2wr_rnic_setconfig rnic_setconfig; + union c2wr_rnic_close rnic_close; + union c2wr_cq_create cq_create; + union c2wr_cq_modify cq_modify; + union c2wr_cq_destroy cq_destroy; + union c2wr_pd_alloc pd_alloc; + union c2wr_pd_dealloc pd_dealloc; + union c2wr_srq_create srq_create; + union c2wr_srq_destroy srq_destroy; + union c2wr_qp_create qp_create; + union c2wr_qp_query qp_query; + union c2wr_qp_modify qp_modify; + union c2wr_qp_destroy qp_destroy; + struct c2wr_qp_connect qp_connect; + union c2wr_nsmr_stag_alloc nsmr_stag_alloc; + union c2wr_nsmr_register nsmr_register; + union c2wr_nsmr_pbl nsmr_pbl; + union c2wr_mr_query mr_query; + union c2wr_mw_query mw_query; + union c2wr_stag_dealloc stag_dealloc; + union c2wr_sqwr sqwr; + struct c2wr_rqwr rqwr; + struct c2wr_ce ce; + union c2wr_ae ae; + union c2wr_init init; + union c2wr_ep_listen_create ep_listen_create; + union c2wr_ep_listen_destroy ep_listen_destroy; + union c2wr_cr_accept cr_accept; + union c2wr_cr_reject cr_reject; + union c2wr_console console; + union c2wr_flash_init flash_init; + union c2wr_flash flash; + union c2wr_buf_alloc buf_alloc; + union c2wr_buf_free buf_free; + union c2wr_flash_write flash_write; +} __attribute__((packed)); + + +/* + * Accessors for the wr fields that are packed together tightly to + * reduce the wr message size. The wr arguments are void* so that + * either a struct c2wr*, a struct c2wr_hdr*, or a pointer to any of the types + * in the struct c2wr union can be passed in. + */ +static __inline__ u8 c2_wr_get_id(void *wr) +{ + return ((struct c2wr_hdr *) wr)->id; +} +static __inline__ void c2_wr_set_id(void *wr, u8 id) +{ + ((struct c2wr_hdr *) wr)->id = id; +} +static __inline__ u8 c2_wr_get_result(void *wr) +{ + return ((struct c2wr_hdr *) wr)->result; +} +static __inline__ void c2_wr_set_result(void *wr, u8 result) +{ + ((struct c2wr_hdr *) wr)->result = result; +} +static __inline__ u8 c2_wr_get_flags(void *wr) +{ + return ((struct c2wr_hdr *) wr)->flags; +} +static __inline__ void c2_wr_set_flags(void *wr, u8 flags) +{ + ((struct c2wr_hdr *) wr)->flags = flags; +} +static __inline__ u8 c2_wr_get_sge_count(void *wr) +{ + return ((struct c2wr_hdr *) wr)->sge_count; +} +static __inline__ void c2_wr_set_sge_count(void *wr, u8 sge_count) +{ + ((struct c2wr_hdr *) wr)->sge_count = sge_count; +} +static __inline__ __be32 c2_wr_get_wqe_count(void *wr) +{ + return ((struct c2wr_hdr *) wr)->wqe_count; +} +static __inline__ void c2_wr_set_wqe_count(void *wr, u32 wqe_count) +{ + ((struct c2wr_hdr *) wr)->wqe_count = wqe_count; +} + +#endif /* _C2_WR_H_ */ |