summaryrefslogtreecommitdiff
path: root/drivers/scsi/qla4xxx/ql4_bsg.c
diff options
context:
space:
mode:
Diffstat (limited to 'drivers/scsi/qla4xxx/ql4_bsg.c')
-rw-r--r--drivers/scsi/qla4xxx/ql4_bsg.c873
1 files changed, 873 insertions, 0 deletions
diff --git a/drivers/scsi/qla4xxx/ql4_bsg.c b/drivers/scsi/qla4xxx/ql4_bsg.c
new file mode 100644
index 000000000..9f92cbf96
--- /dev/null
+++ b/drivers/scsi/qla4xxx/ql4_bsg.c
@@ -0,0 +1,873 @@
+/*
+ * QLogic iSCSI HBA Driver
+ * Copyright (c) 2011-2013 QLogic Corporation
+ *
+ * See LICENSE.qla4xxx for copyright and licensing details.
+ */
+
+#include "ql4_def.h"
+#include "ql4_glbl.h"
+#include "ql4_bsg.h"
+
+static int
+qla4xxx_read_flash(struct bsg_job *bsg_job)
+{
+ struct Scsi_Host *host = iscsi_job_to_shost(bsg_job);
+ struct scsi_qla_host *ha = to_qla_host(host);
+ struct iscsi_bsg_reply *bsg_reply = bsg_job->reply;
+ struct iscsi_bsg_request *bsg_req = bsg_job->request;
+ uint32_t offset = 0;
+ uint32_t length = 0;
+ dma_addr_t flash_dma;
+ uint8_t *flash = NULL;
+ int rval = -EINVAL;
+
+ bsg_reply->reply_payload_rcv_len = 0;
+
+ if (unlikely(pci_channel_offline(ha->pdev)))
+ goto leave;
+
+ if (ql4xxx_reset_active(ha)) {
+ ql4_printk(KERN_ERR, ha, "%s: reset active\n", __func__);
+ rval = -EBUSY;
+ goto leave;
+ }
+
+ if (ha->flash_state != QLFLASH_WAITING) {
+ ql4_printk(KERN_ERR, ha, "%s: another flash operation "
+ "active\n", __func__);
+ rval = -EBUSY;
+ goto leave;
+ }
+
+ ha->flash_state = QLFLASH_READING;
+ offset = bsg_req->rqst_data.h_vendor.vendor_cmd[1];
+ length = bsg_job->reply_payload.payload_len;
+
+ flash = dma_alloc_coherent(&ha->pdev->dev, length, &flash_dma,
+ GFP_KERNEL);
+ if (!flash) {
+ ql4_printk(KERN_ERR, ha, "%s: dma alloc failed for flash "
+ "data\n", __func__);
+ rval = -ENOMEM;
+ goto leave;
+ }
+
+ rval = qla4xxx_get_flash(ha, flash_dma, offset, length);
+ if (rval) {
+ ql4_printk(KERN_ERR, ha, "%s: get flash failed\n", __func__);
+ bsg_reply->result = DID_ERROR << 16;
+ rval = -EIO;
+ } else {
+ bsg_reply->reply_payload_rcv_len =
+ sg_copy_from_buffer(bsg_job->reply_payload.sg_list,
+ bsg_job->reply_payload.sg_cnt,
+ flash, length);
+ bsg_reply->result = DID_OK << 16;
+ }
+
+ bsg_job_done(bsg_job, bsg_reply->result,
+ bsg_reply->reply_payload_rcv_len);
+ dma_free_coherent(&ha->pdev->dev, length, flash, flash_dma);
+leave:
+ ha->flash_state = QLFLASH_WAITING;
+ return rval;
+}
+
+static int
+qla4xxx_update_flash(struct bsg_job *bsg_job)
+{
+ struct Scsi_Host *host = iscsi_job_to_shost(bsg_job);
+ struct scsi_qla_host *ha = to_qla_host(host);
+ struct iscsi_bsg_reply *bsg_reply = bsg_job->reply;
+ struct iscsi_bsg_request *bsg_req = bsg_job->request;
+ uint32_t length = 0;
+ uint32_t offset = 0;
+ uint32_t options = 0;
+ dma_addr_t flash_dma;
+ uint8_t *flash = NULL;
+ int rval = -EINVAL;
+
+ bsg_reply->reply_payload_rcv_len = 0;
+
+ if (unlikely(pci_channel_offline(ha->pdev)))
+ goto leave;
+
+ if (ql4xxx_reset_active(ha)) {
+ ql4_printk(KERN_ERR, ha, "%s: reset active\n", __func__);
+ rval = -EBUSY;
+ goto leave;
+ }
+
+ if (ha->flash_state != QLFLASH_WAITING) {
+ ql4_printk(KERN_ERR, ha, "%s: another flash operation "
+ "active\n", __func__);
+ rval = -EBUSY;
+ goto leave;
+ }
+
+ ha->flash_state = QLFLASH_WRITING;
+ length = bsg_job->request_payload.payload_len;
+ offset = bsg_req->rqst_data.h_vendor.vendor_cmd[1];
+ options = bsg_req->rqst_data.h_vendor.vendor_cmd[2];
+
+ flash = dma_alloc_coherent(&ha->pdev->dev, length, &flash_dma,
+ GFP_KERNEL);
+ if (!flash) {
+ ql4_printk(KERN_ERR, ha, "%s: dma alloc failed for flash "
+ "data\n", __func__);
+ rval = -ENOMEM;
+ goto leave;
+ }
+
+ sg_copy_to_buffer(bsg_job->request_payload.sg_list,
+ bsg_job->request_payload.sg_cnt, flash, length);
+
+ rval = qla4xxx_set_flash(ha, flash_dma, offset, length, options);
+ if (rval) {
+ ql4_printk(KERN_ERR, ha, "%s: set flash failed\n", __func__);
+ bsg_reply->result = DID_ERROR << 16;
+ rval = -EIO;
+ } else
+ bsg_reply->result = DID_OK << 16;
+
+ bsg_job_done(bsg_job, bsg_reply->result,
+ bsg_reply->reply_payload_rcv_len);
+ dma_free_coherent(&ha->pdev->dev, length, flash, flash_dma);
+leave:
+ ha->flash_state = QLFLASH_WAITING;
+ return rval;
+}
+
+static int
+qla4xxx_get_acb_state(struct bsg_job *bsg_job)
+{
+ struct Scsi_Host *host = iscsi_job_to_shost(bsg_job);
+ struct scsi_qla_host *ha = to_qla_host(host);
+ struct iscsi_bsg_request *bsg_req = bsg_job->request;
+ struct iscsi_bsg_reply *bsg_reply = bsg_job->reply;
+ uint32_t status[MBOX_REG_COUNT];
+ uint32_t acb_idx;
+ uint32_t ip_idx;
+ int rval = -EINVAL;
+
+ bsg_reply->reply_payload_rcv_len = 0;
+
+ if (unlikely(pci_channel_offline(ha->pdev)))
+ goto leave;
+
+ /* Only 4022 and above adapters are supported */
+ if (is_qla4010(ha))
+ goto leave;
+
+ if (ql4xxx_reset_active(ha)) {
+ ql4_printk(KERN_ERR, ha, "%s: reset active\n", __func__);
+ rval = -EBUSY;
+ goto leave;
+ }
+
+ if (bsg_job->reply_payload.payload_len < sizeof(status)) {
+ ql4_printk(KERN_ERR, ha, "%s: invalid payload len %d\n",
+ __func__, bsg_job->reply_payload.payload_len);
+ rval = -EINVAL;
+ goto leave;
+ }
+
+ acb_idx = bsg_req->rqst_data.h_vendor.vendor_cmd[1];
+ ip_idx = bsg_req->rqst_data.h_vendor.vendor_cmd[2];
+
+ rval = qla4xxx_get_ip_state(ha, acb_idx, ip_idx, status);
+ if (rval) {
+ ql4_printk(KERN_ERR, ha, "%s: get ip state failed\n",
+ __func__);
+ bsg_reply->result = DID_ERROR << 16;
+ rval = -EIO;
+ } else {
+ bsg_reply->reply_payload_rcv_len =
+ sg_copy_from_buffer(bsg_job->reply_payload.sg_list,
+ bsg_job->reply_payload.sg_cnt,
+ status, sizeof(status));
+ bsg_reply->result = DID_OK << 16;
+ }
+
+ bsg_job_done(bsg_job, bsg_reply->result,
+ bsg_reply->reply_payload_rcv_len);
+leave:
+ return rval;
+}
+
+static int
+qla4xxx_read_nvram(struct bsg_job *bsg_job)
+{
+ struct Scsi_Host *host = iscsi_job_to_shost(bsg_job);
+ struct scsi_qla_host *ha = to_qla_host(host);
+ struct iscsi_bsg_request *bsg_req = bsg_job->request;
+ struct iscsi_bsg_reply *bsg_reply = bsg_job->reply;
+ uint32_t offset = 0;
+ uint32_t len = 0;
+ uint32_t total_len = 0;
+ dma_addr_t nvram_dma;
+ uint8_t *nvram = NULL;
+ int rval = -EINVAL;
+
+ bsg_reply->reply_payload_rcv_len = 0;
+
+ if (unlikely(pci_channel_offline(ha->pdev)))
+ goto leave;
+
+ /* Only 40xx adapters are supported */
+ if (!(is_qla4010(ha) || is_qla4022(ha) || is_qla4032(ha)))
+ goto leave;
+
+ if (ql4xxx_reset_active(ha)) {
+ ql4_printk(KERN_ERR, ha, "%s: reset active\n", __func__);
+ rval = -EBUSY;
+ goto leave;
+ }
+
+ offset = bsg_req->rqst_data.h_vendor.vendor_cmd[1];
+ len = bsg_job->reply_payload.payload_len;
+ total_len = offset + len;
+
+ /* total len should not be greater than max NVRAM size */
+ if ((is_qla4010(ha) && total_len > QL4010_NVRAM_SIZE) ||
+ ((is_qla4022(ha) || is_qla4032(ha)) &&
+ total_len > QL40X2_NVRAM_SIZE)) {
+ ql4_printk(KERN_ERR, ha, "%s: offset+len greater than max"
+ " nvram size, offset=%d len=%d\n",
+ __func__, offset, len);
+ goto leave;
+ }
+
+ nvram = dma_alloc_coherent(&ha->pdev->dev, len, &nvram_dma,
+ GFP_KERNEL);
+ if (!nvram) {
+ ql4_printk(KERN_ERR, ha, "%s: dma alloc failed for nvram "
+ "data\n", __func__);
+ rval = -ENOMEM;
+ goto leave;
+ }
+
+ rval = qla4xxx_get_nvram(ha, nvram_dma, offset, len);
+ if (rval) {
+ ql4_printk(KERN_ERR, ha, "%s: get nvram failed\n", __func__);
+ bsg_reply->result = DID_ERROR << 16;
+ rval = -EIO;
+ } else {
+ bsg_reply->reply_payload_rcv_len =
+ sg_copy_from_buffer(bsg_job->reply_payload.sg_list,
+ bsg_job->reply_payload.sg_cnt,
+ nvram, len);
+ bsg_reply->result = DID_OK << 16;
+ }
+
+ bsg_job_done(bsg_job, bsg_reply->result,
+ bsg_reply->reply_payload_rcv_len);
+ dma_free_coherent(&ha->pdev->dev, len, nvram, nvram_dma);
+leave:
+ return rval;
+}
+
+static int
+qla4xxx_update_nvram(struct bsg_job *bsg_job)
+{
+ struct Scsi_Host *host = iscsi_job_to_shost(bsg_job);
+ struct scsi_qla_host *ha = to_qla_host(host);
+ struct iscsi_bsg_request *bsg_req = bsg_job->request;
+ struct iscsi_bsg_reply *bsg_reply = bsg_job->reply;
+ uint32_t offset = 0;
+ uint32_t len = 0;
+ uint32_t total_len = 0;
+ dma_addr_t nvram_dma;
+ uint8_t *nvram = NULL;
+ int rval = -EINVAL;
+
+ bsg_reply->reply_payload_rcv_len = 0;
+
+ if (unlikely(pci_channel_offline(ha->pdev)))
+ goto leave;
+
+ if (!(is_qla4010(ha) || is_qla4022(ha) || is_qla4032(ha)))
+ goto leave;
+
+ if (ql4xxx_reset_active(ha)) {
+ ql4_printk(KERN_ERR, ha, "%s: reset active\n", __func__);
+ rval = -EBUSY;
+ goto leave;
+ }
+
+ offset = bsg_req->rqst_data.h_vendor.vendor_cmd[1];
+ len = bsg_job->request_payload.payload_len;
+ total_len = offset + len;
+
+ /* total len should not be greater than max NVRAM size */
+ if ((is_qla4010(ha) && total_len > QL4010_NVRAM_SIZE) ||
+ ((is_qla4022(ha) || is_qla4032(ha)) &&
+ total_len > QL40X2_NVRAM_SIZE)) {
+ ql4_printk(KERN_ERR, ha, "%s: offset+len greater than max"
+ " nvram size, offset=%d len=%d\n",
+ __func__, offset, len);
+ goto leave;
+ }
+
+ nvram = dma_alloc_coherent(&ha->pdev->dev, len, &nvram_dma,
+ GFP_KERNEL);
+ if (!nvram) {
+ ql4_printk(KERN_ERR, ha, "%s: dma alloc failed for flash "
+ "data\n", __func__);
+ rval = -ENOMEM;
+ goto leave;
+ }
+
+ sg_copy_to_buffer(bsg_job->request_payload.sg_list,
+ bsg_job->request_payload.sg_cnt, nvram, len);
+
+ rval = qla4xxx_set_nvram(ha, nvram_dma, offset, len);
+ if (rval) {
+ ql4_printk(KERN_ERR, ha, "%s: set nvram failed\n", __func__);
+ bsg_reply->result = DID_ERROR << 16;
+ rval = -EIO;
+ } else
+ bsg_reply->result = DID_OK << 16;
+
+ bsg_job_done(bsg_job, bsg_reply->result,
+ bsg_reply->reply_payload_rcv_len);
+ dma_free_coherent(&ha->pdev->dev, len, nvram, nvram_dma);
+leave:
+ return rval;
+}
+
+static int
+qla4xxx_restore_defaults(struct bsg_job *bsg_job)
+{
+ struct Scsi_Host *host = iscsi_job_to_shost(bsg_job);
+ struct scsi_qla_host *ha = to_qla_host(host);
+ struct iscsi_bsg_request *bsg_req = bsg_job->request;
+ struct iscsi_bsg_reply *bsg_reply = bsg_job->reply;
+ uint32_t region = 0;
+ uint32_t field0 = 0;
+ uint32_t field1 = 0;
+ int rval = -EINVAL;
+
+ bsg_reply->reply_payload_rcv_len = 0;
+
+ if (unlikely(pci_channel_offline(ha->pdev)))
+ goto leave;
+
+ if (is_qla4010(ha))
+ goto leave;
+
+ if (ql4xxx_reset_active(ha)) {
+ ql4_printk(KERN_ERR, ha, "%s: reset active\n", __func__);
+ rval = -EBUSY;
+ goto leave;
+ }
+
+ region = bsg_req->rqst_data.h_vendor.vendor_cmd[1];
+ field0 = bsg_req->rqst_data.h_vendor.vendor_cmd[2];
+ field1 = bsg_req->rqst_data.h_vendor.vendor_cmd[3];
+
+ rval = qla4xxx_restore_factory_defaults(ha, region, field0, field1);
+ if (rval) {
+ ql4_printk(KERN_ERR, ha, "%s: set nvram failed\n", __func__);
+ bsg_reply->result = DID_ERROR << 16;
+ rval = -EIO;
+ } else
+ bsg_reply->result = DID_OK << 16;
+
+ bsg_job_done(bsg_job, bsg_reply->result,
+ bsg_reply->reply_payload_rcv_len);
+leave:
+ return rval;
+}
+
+static int
+qla4xxx_bsg_get_acb(struct bsg_job *bsg_job)
+{
+ struct Scsi_Host *host = iscsi_job_to_shost(bsg_job);
+ struct scsi_qla_host *ha = to_qla_host(host);
+ struct iscsi_bsg_request *bsg_req = bsg_job->request;
+ struct iscsi_bsg_reply *bsg_reply = bsg_job->reply;
+ uint32_t acb_type = 0;
+ uint32_t len = 0;
+ dma_addr_t acb_dma;
+ uint8_t *acb = NULL;
+ int rval = -EINVAL;
+
+ bsg_reply->reply_payload_rcv_len = 0;
+
+ if (unlikely(pci_channel_offline(ha->pdev)))
+ goto leave;
+
+ /* Only 4022 and above adapters are supported */
+ if (is_qla4010(ha))
+ goto leave;
+
+ if (ql4xxx_reset_active(ha)) {
+ ql4_printk(KERN_ERR, ha, "%s: reset active\n", __func__);
+ rval = -EBUSY;
+ goto leave;
+ }
+
+ acb_type = bsg_req->rqst_data.h_vendor.vendor_cmd[1];
+ len = bsg_job->reply_payload.payload_len;
+ if (len < sizeof(struct addr_ctrl_blk)) {
+ ql4_printk(KERN_ERR, ha, "%s: invalid acb len %d\n",
+ __func__, len);
+ rval = -EINVAL;
+ goto leave;
+ }
+
+ acb = dma_alloc_coherent(&ha->pdev->dev, len, &acb_dma, GFP_KERNEL);
+ if (!acb) {
+ ql4_printk(KERN_ERR, ha, "%s: dma alloc failed for acb "
+ "data\n", __func__);
+ rval = -ENOMEM;
+ goto leave;
+ }
+
+ rval = qla4xxx_get_acb(ha, acb_dma, acb_type, len);
+ if (rval) {
+ ql4_printk(KERN_ERR, ha, "%s: get acb failed\n", __func__);
+ bsg_reply->result = DID_ERROR << 16;
+ rval = -EIO;
+ } else {
+ bsg_reply->reply_payload_rcv_len =
+ sg_copy_from_buffer(bsg_job->reply_payload.sg_list,
+ bsg_job->reply_payload.sg_cnt,
+ acb, len);
+ bsg_reply->result = DID_OK << 16;
+ }
+
+ bsg_job_done(bsg_job, bsg_reply->result,
+ bsg_reply->reply_payload_rcv_len);
+ dma_free_coherent(&ha->pdev->dev, len, acb, acb_dma);
+leave:
+ return rval;
+}
+
+static void ql4xxx_execute_diag_cmd(struct bsg_job *bsg_job)
+{
+ struct Scsi_Host *host = iscsi_job_to_shost(bsg_job);
+ struct scsi_qla_host *ha = to_qla_host(host);
+ struct iscsi_bsg_request *bsg_req = bsg_job->request;
+ struct iscsi_bsg_reply *bsg_reply = bsg_job->reply;
+ uint8_t *rsp_ptr = NULL;
+ uint32_t mbox_cmd[MBOX_REG_COUNT];
+ uint32_t mbox_sts[MBOX_REG_COUNT];
+ int status = QLA_ERROR;
+
+ DEBUG2(ql4_printk(KERN_INFO, ha, "%s: in\n", __func__));
+
+ if (test_bit(DPC_RESET_HA, &ha->dpc_flags)) {
+ ql4_printk(KERN_INFO, ha, "%s: Adapter reset in progress. Invalid Request\n",
+ __func__);
+ bsg_reply->result = DID_ERROR << 16;
+ goto exit_diag_mem_test;
+ }
+
+ bsg_reply->reply_payload_rcv_len = 0;
+ memcpy(mbox_cmd, &bsg_req->rqst_data.h_vendor.vendor_cmd[1],
+ sizeof(uint32_t) * MBOX_REG_COUNT);
+
+ DEBUG2(ql4_printk(KERN_INFO, ha,
+ "%s: mbox_cmd: %08X %08X %08X %08X %08X %08X %08X %08X\n",
+ __func__, mbox_cmd[0], mbox_cmd[1], mbox_cmd[2],
+ mbox_cmd[3], mbox_cmd[4], mbox_cmd[5], mbox_cmd[6],
+ mbox_cmd[7]));
+
+ status = qla4xxx_mailbox_command(ha, MBOX_REG_COUNT, 8, &mbox_cmd[0],
+ &mbox_sts[0]);
+
+ DEBUG2(ql4_printk(KERN_INFO, ha,
+ "%s: mbox_sts: %08X %08X %08X %08X %08X %08X %08X %08X\n",
+ __func__, mbox_sts[0], mbox_sts[1], mbox_sts[2],
+ mbox_sts[3], mbox_sts[4], mbox_sts[5], mbox_sts[6],
+ mbox_sts[7]));
+
+ if (status == QLA_SUCCESS)
+ bsg_reply->result = DID_OK << 16;
+ else
+ bsg_reply->result = DID_ERROR << 16;
+
+ /* Send mbox_sts to application */
+ bsg_job->reply_len = sizeof(struct iscsi_bsg_reply) + sizeof(mbox_sts);
+ rsp_ptr = ((uint8_t *)bsg_reply) + sizeof(struct iscsi_bsg_reply);
+ memcpy(rsp_ptr, mbox_sts, sizeof(mbox_sts));
+
+exit_diag_mem_test:
+ DEBUG2(ql4_printk(KERN_INFO, ha,
+ "%s: bsg_reply->result = x%x, status = %s\n",
+ __func__, bsg_reply->result, STATUS(status)));
+
+ bsg_job_done(bsg_job, bsg_reply->result,
+ bsg_reply->reply_payload_rcv_len);
+}
+
+static int qla4_83xx_wait_for_loopback_config_comp(struct scsi_qla_host *ha,
+ int wait_for_link)
+{
+ int status = QLA_SUCCESS;
+
+ if (!wait_for_completion_timeout(&ha->idc_comp, (IDC_COMP_TOV * HZ))) {
+ ql4_printk(KERN_INFO, ha, "%s: IDC Complete notification not received, Waiting for another %d timeout",
+ __func__, ha->idc_extend_tmo);
+ if (ha->idc_extend_tmo) {
+ if (!wait_for_completion_timeout(&ha->idc_comp,
+ (ha->idc_extend_tmo * HZ))) {
+ ha->notify_idc_comp = 0;
+ ha->notify_link_up_comp = 0;
+ ql4_printk(KERN_WARNING, ha, "%s: Aborting: IDC Complete notification not received",
+ __func__);
+ status = QLA_ERROR;
+ goto exit_wait;
+ } else {
+ DEBUG2(ql4_printk(KERN_INFO, ha,
+ "%s: IDC Complete notification received\n",
+ __func__));
+ }
+ }
+ } else {
+ DEBUG2(ql4_printk(KERN_INFO, ha,
+ "%s: IDC Complete notification received\n",
+ __func__));
+ }
+ ha->notify_idc_comp = 0;
+
+ if (wait_for_link) {
+ if (!wait_for_completion_timeout(&ha->link_up_comp,
+ (IDC_COMP_TOV * HZ))) {
+ ha->notify_link_up_comp = 0;
+ ql4_printk(KERN_WARNING, ha, "%s: Aborting: LINK UP notification not received",
+ __func__);
+ status = QLA_ERROR;
+ goto exit_wait;
+ } else {
+ DEBUG2(ql4_printk(KERN_INFO, ha,
+ "%s: LINK UP notification received\n",
+ __func__));
+ }
+ ha->notify_link_up_comp = 0;
+ }
+
+exit_wait:
+ return status;
+}
+
+static int qla4_83xx_pre_loopback_config(struct scsi_qla_host *ha,
+ uint32_t *mbox_cmd)
+{
+ uint32_t config = 0;
+ int status = QLA_SUCCESS;
+
+ DEBUG2(ql4_printk(KERN_INFO, ha, "%s: in\n", __func__));
+
+ status = qla4_83xx_get_port_config(ha, &config);
+ if (status != QLA_SUCCESS)
+ goto exit_pre_loopback_config;
+
+ DEBUG2(ql4_printk(KERN_INFO, ha, "%s: Default port config=%08X\n",
+ __func__, config));
+
+ if ((config & ENABLE_INTERNAL_LOOPBACK) ||
+ (config & ENABLE_EXTERNAL_LOOPBACK)) {
+ ql4_printk(KERN_INFO, ha, "%s: Loopback diagnostics already in progress. Invalid requiest\n",
+ __func__);
+ goto exit_pre_loopback_config;
+ }
+
+ if (mbox_cmd[1] == QL_DIAG_CMD_TEST_INT_LOOPBACK)
+ config |= ENABLE_INTERNAL_LOOPBACK;
+
+ if (mbox_cmd[1] == QL_DIAG_CMD_TEST_EXT_LOOPBACK)
+ config |= ENABLE_EXTERNAL_LOOPBACK;
+
+ config &= ~ENABLE_DCBX;
+
+ DEBUG2(ql4_printk(KERN_INFO, ha, "%s: New port config=%08X\n",
+ __func__, config));
+
+ ha->notify_idc_comp = 1;
+ ha->notify_link_up_comp = 1;
+
+ /* get the link state */
+ qla4xxx_get_firmware_state(ha);
+
+ status = qla4_83xx_set_port_config(ha, &config);
+ if (status != QLA_SUCCESS) {
+ ha->notify_idc_comp = 0;
+ ha->notify_link_up_comp = 0;
+ goto exit_pre_loopback_config;
+ }
+exit_pre_loopback_config:
+ DEBUG2(ql4_printk(KERN_INFO, ha, "%s: status = %s\n", __func__,
+ STATUS(status)));
+ return status;
+}
+
+static int qla4_83xx_post_loopback_config(struct scsi_qla_host *ha,
+ uint32_t *mbox_cmd)
+{
+ int status = QLA_SUCCESS;
+ uint32_t config = 0;
+
+ DEBUG2(ql4_printk(KERN_INFO, ha, "%s: in\n", __func__));
+
+ status = qla4_83xx_get_port_config(ha, &config);
+ if (status != QLA_SUCCESS)
+ goto exit_post_loopback_config;
+
+ DEBUG2(ql4_printk(KERN_INFO, ha, "%s: port config=%08X\n", __func__,
+ config));
+
+ if (mbox_cmd[1] == QL_DIAG_CMD_TEST_INT_LOOPBACK)
+ config &= ~ENABLE_INTERNAL_LOOPBACK;
+ else if (mbox_cmd[1] == QL_DIAG_CMD_TEST_EXT_LOOPBACK)
+ config &= ~ENABLE_EXTERNAL_LOOPBACK;
+
+ config |= ENABLE_DCBX;
+
+ DEBUG2(ql4_printk(KERN_INFO, ha,
+ "%s: Restore default port config=%08X\n", __func__,
+ config));
+
+ ha->notify_idc_comp = 1;
+ if (ha->addl_fw_state & FW_ADDSTATE_LINK_UP)
+ ha->notify_link_up_comp = 1;
+
+ status = qla4_83xx_set_port_config(ha, &config);
+ if (status != QLA_SUCCESS) {
+ ql4_printk(KERN_INFO, ha, "%s: Scheduling adapter reset\n",
+ __func__);
+ set_bit(DPC_RESET_HA, &ha->dpc_flags);
+ clear_bit(AF_LOOPBACK, &ha->flags);
+ goto exit_post_loopback_config;
+ }
+
+exit_post_loopback_config:
+ DEBUG2(ql4_printk(KERN_INFO, ha, "%s: status = %s\n", __func__,
+ STATUS(status)));
+ return status;
+}
+
+static void qla4xxx_execute_diag_loopback_cmd(struct bsg_job *bsg_job)
+{
+ struct Scsi_Host *host = iscsi_job_to_shost(bsg_job);
+ struct scsi_qla_host *ha = to_qla_host(host);
+ struct iscsi_bsg_request *bsg_req = bsg_job->request;
+ struct iscsi_bsg_reply *bsg_reply = bsg_job->reply;
+ uint8_t *rsp_ptr = NULL;
+ uint32_t mbox_cmd[MBOX_REG_COUNT];
+ uint32_t mbox_sts[MBOX_REG_COUNT];
+ int wait_for_link = 1;
+ int status = QLA_ERROR;
+
+ DEBUG2(ql4_printk(KERN_INFO, ha, "%s: in\n", __func__));
+
+ bsg_reply->reply_payload_rcv_len = 0;
+
+ if (test_bit(AF_LOOPBACK, &ha->flags)) {
+ ql4_printk(KERN_INFO, ha, "%s: Loopback Diagnostics already in progress. Invalid Request\n",
+ __func__);
+ bsg_reply->result = DID_ERROR << 16;
+ goto exit_loopback_cmd;
+ }
+
+ if (test_bit(DPC_RESET_HA, &ha->dpc_flags)) {
+ ql4_printk(KERN_INFO, ha, "%s: Adapter reset in progress. Invalid Request\n",
+ __func__);
+ bsg_reply->result = DID_ERROR << 16;
+ goto exit_loopback_cmd;
+ }
+
+ memcpy(mbox_cmd, &bsg_req->rqst_data.h_vendor.vendor_cmd[1],
+ sizeof(uint32_t) * MBOX_REG_COUNT);
+
+ if (is_qla8032(ha) || is_qla8042(ha)) {
+ status = qla4_83xx_pre_loopback_config(ha, mbox_cmd);
+ if (status != QLA_SUCCESS) {
+ bsg_reply->result = DID_ERROR << 16;
+ goto exit_loopback_cmd;
+ }
+
+ status = qla4_83xx_wait_for_loopback_config_comp(ha,
+ wait_for_link);
+ if (status != QLA_SUCCESS) {
+ bsg_reply->result = DID_TIME_OUT << 16;
+ goto restore;
+ }
+ }
+
+ DEBUG2(ql4_printk(KERN_INFO, ha,
+ "%s: mbox_cmd: %08X %08X %08X %08X %08X %08X %08X %08X\n",
+ __func__, mbox_cmd[0], mbox_cmd[1], mbox_cmd[2],
+ mbox_cmd[3], mbox_cmd[4], mbox_cmd[5], mbox_cmd[6],
+ mbox_cmd[7]));
+
+ status = qla4xxx_mailbox_command(ha, MBOX_REG_COUNT, 8, &mbox_cmd[0],
+ &mbox_sts[0]);
+
+ if (status == QLA_SUCCESS)
+ bsg_reply->result = DID_OK << 16;
+ else
+ bsg_reply->result = DID_ERROR << 16;
+
+ DEBUG2(ql4_printk(KERN_INFO, ha,
+ "%s: mbox_sts: %08X %08X %08X %08X %08X %08X %08X %08X\n",
+ __func__, mbox_sts[0], mbox_sts[1], mbox_sts[2],
+ mbox_sts[3], mbox_sts[4], mbox_sts[5], mbox_sts[6],
+ mbox_sts[7]));
+
+ /* Send mbox_sts to application */
+ bsg_job->reply_len = sizeof(struct iscsi_bsg_reply) + sizeof(mbox_sts);
+ rsp_ptr = ((uint8_t *)bsg_reply) + sizeof(struct iscsi_bsg_reply);
+ memcpy(rsp_ptr, mbox_sts, sizeof(mbox_sts));
+restore:
+ if (is_qla8032(ha) || is_qla8042(ha)) {
+ status = qla4_83xx_post_loopback_config(ha, mbox_cmd);
+ if (status != QLA_SUCCESS) {
+ bsg_reply->result = DID_ERROR << 16;
+ goto exit_loopback_cmd;
+ }
+
+ /* for pre_loopback_config() wait for LINK UP only
+ * if PHY LINK is UP */
+ if (!(ha->addl_fw_state & FW_ADDSTATE_LINK_UP))
+ wait_for_link = 0;
+
+ status = qla4_83xx_wait_for_loopback_config_comp(ha,
+ wait_for_link);
+ if (status != QLA_SUCCESS) {
+ bsg_reply->result = DID_TIME_OUT << 16;
+ goto exit_loopback_cmd;
+ }
+ }
+exit_loopback_cmd:
+ DEBUG2(ql4_printk(KERN_INFO, ha,
+ "%s: bsg_reply->result = x%x, status = %s\n",
+ __func__, bsg_reply->result, STATUS(status)));
+ bsg_job_done(bsg_job, bsg_reply->result,
+ bsg_reply->reply_payload_rcv_len);
+}
+
+static int qla4xxx_execute_diag_test(struct bsg_job *bsg_job)
+{
+ struct Scsi_Host *host = iscsi_job_to_shost(bsg_job);
+ struct scsi_qla_host *ha = to_qla_host(host);
+ struct iscsi_bsg_request *bsg_req = bsg_job->request;
+ uint32_t diag_cmd;
+ int rval = -EINVAL;
+
+ DEBUG2(ql4_printk(KERN_INFO, ha, "%s: in\n", __func__));
+
+ diag_cmd = bsg_req->rqst_data.h_vendor.vendor_cmd[1];
+ if (diag_cmd == MBOX_CMD_DIAG_TEST) {
+ switch (bsg_req->rqst_data.h_vendor.vendor_cmd[2]) {
+ case QL_DIAG_CMD_TEST_DDR_SIZE:
+ case QL_DIAG_CMD_TEST_DDR_RW:
+ case QL_DIAG_CMD_TEST_ONCHIP_MEM_RW:
+ case QL_DIAG_CMD_TEST_NVRAM:
+ case QL_DIAG_CMD_TEST_FLASH_ROM:
+ case QL_DIAG_CMD_TEST_DMA_XFER:
+ case QL_DIAG_CMD_SELF_DDR_RW:
+ case QL_DIAG_CMD_SELF_ONCHIP_MEM_RW:
+ /* Execute diag test for adapter RAM/FLASH */
+ ql4xxx_execute_diag_cmd(bsg_job);
+ /* Always return success as we want to sent bsg_reply
+ * to Application */
+ rval = QLA_SUCCESS;
+ break;
+
+ case QL_DIAG_CMD_TEST_INT_LOOPBACK:
+ case QL_DIAG_CMD_TEST_EXT_LOOPBACK:
+ /* Execute diag test for Network */
+ qla4xxx_execute_diag_loopback_cmd(bsg_job);
+ /* Always return success as we want to sent bsg_reply
+ * to Application */
+ rval = QLA_SUCCESS;
+ break;
+ default:
+ ql4_printk(KERN_ERR, ha, "%s: Invalid diag test: 0x%x\n",
+ __func__,
+ bsg_req->rqst_data.h_vendor.vendor_cmd[2]);
+ }
+ } else if ((diag_cmd == MBOX_CMD_SET_LED_CONFIG) ||
+ (diag_cmd == MBOX_CMD_GET_LED_CONFIG)) {
+ ql4xxx_execute_diag_cmd(bsg_job);
+ rval = QLA_SUCCESS;
+ } else {
+ ql4_printk(KERN_ERR, ha, "%s: Invalid diag cmd: 0x%x\n",
+ __func__, diag_cmd);
+ }
+
+ return rval;
+}
+
+/**
+ * qla4xxx_process_vendor_specific - handle vendor specific bsg request
+ * @job: iscsi_bsg_job to handle
+ **/
+int qla4xxx_process_vendor_specific(struct bsg_job *bsg_job)
+{
+ struct iscsi_bsg_reply *bsg_reply = bsg_job->reply;
+ struct iscsi_bsg_request *bsg_req = bsg_job->request;
+ struct Scsi_Host *host = iscsi_job_to_shost(bsg_job);
+ struct scsi_qla_host *ha = to_qla_host(host);
+
+ switch (bsg_req->rqst_data.h_vendor.vendor_cmd[0]) {
+ case QLISCSI_VND_READ_FLASH:
+ return qla4xxx_read_flash(bsg_job);
+
+ case QLISCSI_VND_UPDATE_FLASH:
+ return qla4xxx_update_flash(bsg_job);
+
+ case QLISCSI_VND_GET_ACB_STATE:
+ return qla4xxx_get_acb_state(bsg_job);
+
+ case QLISCSI_VND_READ_NVRAM:
+ return qla4xxx_read_nvram(bsg_job);
+
+ case QLISCSI_VND_UPDATE_NVRAM:
+ return qla4xxx_update_nvram(bsg_job);
+
+ case QLISCSI_VND_RESTORE_DEFAULTS:
+ return qla4xxx_restore_defaults(bsg_job);
+
+ case QLISCSI_VND_GET_ACB:
+ return qla4xxx_bsg_get_acb(bsg_job);
+
+ case QLISCSI_VND_DIAG_TEST:
+ return qla4xxx_execute_diag_test(bsg_job);
+
+ default:
+ ql4_printk(KERN_ERR, ha, "%s: invalid BSG vendor command: "
+ "0x%x\n", __func__, bsg_req->msgcode);
+ bsg_reply->result = (DID_ERROR << 16);
+ bsg_reply->reply_payload_rcv_len = 0;
+ bsg_job_done(bsg_job, bsg_reply->result,
+ bsg_reply->reply_payload_rcv_len);
+ return -ENOSYS;
+ }
+}
+
+/**
+ * qla4xxx_bsg_request - handle bsg request from ISCSI transport
+ * @job: iscsi_bsg_job to handle
+ */
+int qla4xxx_bsg_request(struct bsg_job *bsg_job)
+{
+ struct iscsi_bsg_request *bsg_req = bsg_job->request;
+ struct Scsi_Host *host = iscsi_job_to_shost(bsg_job);
+ struct scsi_qla_host *ha = to_qla_host(host);
+
+ switch (bsg_req->msgcode) {
+ case ISCSI_BSG_HST_VENDOR:
+ return qla4xxx_process_vendor_specific(bsg_job);
+
+ default:
+ ql4_printk(KERN_ERR, ha, "%s: invalid BSG command: 0x%x\n",
+ __func__, bsg_req->msgcode);
+ }
+
+ return -ENOSYS;
+}