diff options
Diffstat (limited to 'arch/arc/mm')
| -rw-r--r-- | arch/arc/mm/Makefile | 1 | ||||
| -rw-r--r-- | arch/arc/mm/cache.c | 91 | ||||
| -rw-r--r-- | arch/arc/mm/fault.c | 13 | ||||
| -rw-r--r-- | arch/arc/mm/highmem.c | 140 | ||||
| -rw-r--r-- | arch/arc/mm/init.c | 106 | ||||
| -rw-r--r-- | arch/arc/mm/tlb.c | 238 | ||||
| -rw-r--r-- | arch/arc/mm/tlbex.S | 55 |
7 files changed, 527 insertions, 117 deletions
diff --git a/arch/arc/mm/Makefile b/arch/arc/mm/Makefile index 7beb94155..3703a4969 100644 --- a/arch/arc/mm/Makefile +++ b/arch/arc/mm/Makefile @@ -8,3 +8,4 @@ obj-y := extable.o ioremap.o dma.o fault.o init.o obj-y += tlb.o tlbex.o cache.o mmap.o +obj-$(CONFIG_HIGHMEM) += highmem.o diff --git a/arch/arc/mm/cache.c b/arch/arc/mm/cache.c index 0d1a6e968..ff7ff6cbb 100644 --- a/arch/arc/mm/cache.c +++ b/arch/arc/mm/cache.c @@ -25,7 +25,7 @@ static int l2_line_sz; int ioc_exists; volatile int slc_enable = 1, ioc_enable = 1; -void (*_cache_line_loop_ic_fn)(unsigned long paddr, unsigned long vaddr, +void (*_cache_line_loop_ic_fn)(phys_addr_t paddr, unsigned long vaddr, unsigned long sz, const int cacheop); void (*__dma_cache_wback_inv)(unsigned long start, unsigned long sz); @@ -37,7 +37,6 @@ char *arc_cache_mumbojumbo(int c, char *buf, int len) int n = 0; struct cpuinfo_arc_cache *p; -#define IS_USED_RUN(v) ((v) ? "" : "(disabled) ") #define PR_CACHE(p, cfg, str) \ if (!(p)->ver) \ n += scnprintf(buf + n, len - n, str"\t\t: N/A\n"); \ @@ -47,7 +46,7 @@ char *arc_cache_mumbojumbo(int c, char *buf, int len) (p)->sz_k, (p)->assoc, (p)->line_len, \ (p)->vipt ? "VIPT" : "PIPT", \ (p)->alias ? " aliasing" : "", \ - IS_ENABLED(cfg) ? "" : " (not used)"); + IS_USED_CFG(cfg)); PR_CACHE(&cpuinfo_arc700[c].icache, CONFIG_ARC_HAS_ICACHE, "I-Cache"); PR_CACHE(&cpuinfo_arc700[c].dcache, CONFIG_ARC_HAS_DCACHE, "D-Cache"); @@ -63,7 +62,7 @@ char *arc_cache_mumbojumbo(int c, char *buf, int len) if (ioc_exists) n += scnprintf(buf + n, len - n, "IOC\t\t:%s\n", - IS_USED_RUN(ioc_enable)); + IS_DISABLED_RUN(ioc_enable)); return buf; } @@ -217,7 +216,7 @@ slc_chk: */ static inline -void __cache_line_loop_v2(unsigned long paddr, unsigned long vaddr, +void __cache_line_loop_v2(phys_addr_t paddr, unsigned long vaddr, unsigned long sz, const int op) { unsigned int aux_cmd; @@ -254,8 +253,12 @@ void __cache_line_loop_v2(unsigned long paddr, unsigned long vaddr, } } +/* + * For ARC700 MMUv3 I-cache and D-cache flushes + * Also reused for HS38 aliasing I-cache configuration + */ static inline -void __cache_line_loop_v3(unsigned long paddr, unsigned long vaddr, +void __cache_line_loop_v3(phys_addr_t paddr, unsigned long vaddr, unsigned long sz, const int op) { unsigned int aux_cmd, aux_tag; @@ -290,6 +293,16 @@ void __cache_line_loop_v3(unsigned long paddr, unsigned long vaddr, if (full_page) write_aux_reg(aux_tag, paddr); + /* + * This is technically for MMU v4, using the MMU v3 programming model + * Special work for HS38 aliasing I-cache configuratino with PAE40 + * - upper 8 bits of paddr need to be written into PTAG_HI + * - (and needs to be written before the lower 32 bits) + * Note that PTAG_HI is hoisted outside the line loop + */ + if (is_pae40_enabled() && op == OP_INV_IC) + write_aux_reg(ARC_REG_IC_PTAG_HI, (u64)paddr >> 32); + while (num_lines-- > 0) { if (!full_page) { write_aux_reg(aux_tag, paddr); @@ -302,14 +315,20 @@ void __cache_line_loop_v3(unsigned long paddr, unsigned long vaddr, } /* - * In HS38x (MMU v4), although icache is VIPT, only paddr is needed for cache - * maintenance ops (in IVIL reg), as long as icache doesn't alias. + * In HS38x (MMU v4), I-cache is VIPT (can alias), D-cache is PIPT + * Here's how cache ops are implemented + * + * - D-cache: only paddr needed (in DC_IVDL/DC_FLDL) + * - I-cache Non Aliasing: Despite VIPT, only paddr needed (in IC_IVIL) + * - I-cache Aliasing: Both vaddr and paddr needed (in IC_IVIL, IC_PTAG + * respectively, similar to MMU v3 programming model, hence + * __cache_line_loop_v3() is used) * - * For Aliasing icache, vaddr is also needed (in IVIL), while paddr is - * specified in PTAG (similar to MMU v3) + * If PAE40 is enabled, independent of aliasing considerations, the higher bits + * needs to be written into PTAG_HI */ static inline -void __cache_line_loop_v4(unsigned long paddr, unsigned long vaddr, +void __cache_line_loop_v4(phys_addr_t paddr, unsigned long vaddr, unsigned long sz, const int cacheop) { unsigned int aux_cmd; @@ -336,6 +355,22 @@ void __cache_line_loop_v4(unsigned long paddr, unsigned long vaddr, num_lines = DIV_ROUND_UP(sz, L1_CACHE_BYTES); + /* + * For HS38 PAE40 configuration + * - upper 8 bits of paddr need to be written into PTAG_HI + * - (and needs to be written before the lower 32 bits) + */ + if (is_pae40_enabled()) { + if (cacheop == OP_INV_IC) + /* + * Non aliasing I-cache in HS38, + * aliasing I-cache handled in __cache_line_loop_v3() + */ + write_aux_reg(ARC_REG_IC_PTAG_HI, (u64)paddr >> 32); + else + write_aux_reg(ARC_REG_DC_PTAG_HI, (u64)paddr >> 32); + } + while (num_lines-- > 0) { write_aux_reg(aux_cmd, paddr); paddr += L1_CACHE_BYTES; @@ -413,7 +448,7 @@ static inline void __dc_entire_op(const int op) /* * D-Cache Line ops: Per Line INV (discard or wback+discard) or FLUSH (wback) */ -static inline void __dc_line_op(unsigned long paddr, unsigned long vaddr, +static inline void __dc_line_op(phys_addr_t paddr, unsigned long vaddr, unsigned long sz, const int op) { unsigned long flags; @@ -446,7 +481,7 @@ static inline void __ic_entire_inv(void) } static inline void -__ic_line_inv_vaddr_local(unsigned long paddr, unsigned long vaddr, +__ic_line_inv_vaddr_local(phys_addr_t paddr, unsigned long vaddr, unsigned long sz) { unsigned long flags; @@ -463,7 +498,7 @@ __ic_line_inv_vaddr_local(unsigned long paddr, unsigned long vaddr, #else struct ic_inv_args { - unsigned long paddr, vaddr; + phys_addr_t paddr, vaddr; int sz; }; @@ -474,7 +509,7 @@ static void __ic_line_inv_vaddr_helper(void *info) __ic_line_inv_vaddr_local(ic_inv->paddr, ic_inv->vaddr, ic_inv->sz); } -static void __ic_line_inv_vaddr(unsigned long paddr, unsigned long vaddr, +static void __ic_line_inv_vaddr(phys_addr_t paddr, unsigned long vaddr, unsigned long sz) { struct ic_inv_args ic_inv = { @@ -495,7 +530,7 @@ static void __ic_line_inv_vaddr(unsigned long paddr, unsigned long vaddr, #endif /* CONFIG_ARC_HAS_ICACHE */ -noinline void slc_op(unsigned long paddr, unsigned long sz, const int op) +noinline void slc_op(phys_addr_t paddr, unsigned long sz, const int op) { #ifdef CONFIG_ISA_ARCV2 /* @@ -585,7 +620,7 @@ void flush_dcache_page(struct page *page) } else if (page_mapped(page)) { /* kernel reading from page with U-mapping */ - unsigned long paddr = (unsigned long)page_address(page); + phys_addr_t paddr = (unsigned long)page_address(page); unsigned long vaddr = page->index << PAGE_CACHE_SHIFT; if (addr_not_cache_congruent(paddr, vaddr)) @@ -733,14 +768,14 @@ EXPORT_SYMBOL(flush_icache_range); * builtin kernel page will not have any virtual mappings. * kprobe on loadable module will be kernel vaddr. */ -void __sync_icache_dcache(unsigned long paddr, unsigned long vaddr, int len) +void __sync_icache_dcache(phys_addr_t paddr, unsigned long vaddr, int len) { __dc_line_op(paddr, vaddr, len, OP_FLUSH_N_INV); __ic_line_inv_vaddr(paddr, vaddr, len); } /* wrapper to compile time eliminate alignment checks in flush loop */ -void __inv_icache_page(unsigned long paddr, unsigned long vaddr) +void __inv_icache_page(phys_addr_t paddr, unsigned long vaddr) { __ic_line_inv_vaddr(paddr, vaddr, PAGE_SIZE); } @@ -749,7 +784,7 @@ void __inv_icache_page(unsigned long paddr, unsigned long vaddr) * wrapper to clearout kernel or userspace mappings of a page * For kernel mappings @vaddr == @paddr */ -void __flush_dcache_page(unsigned long paddr, unsigned long vaddr) +void __flush_dcache_page(phys_addr_t paddr, unsigned long vaddr) { __dc_line_op(paddr, vaddr & PAGE_MASK, PAGE_SIZE, OP_FLUSH_N_INV); } @@ -807,8 +842,8 @@ void flush_anon_page(struct vm_area_struct *vma, struct page *page, void copy_user_highpage(struct page *to, struct page *from, unsigned long u_vaddr, struct vm_area_struct *vma) { - unsigned long kfrom = (unsigned long)page_address(from); - unsigned long kto = (unsigned long)page_address(to); + void *kfrom = kmap_atomic(from); + void *kto = kmap_atomic(to); int clean_src_k_mappings = 0; /* @@ -818,13 +853,16 @@ void copy_user_highpage(struct page *to, struct page *from, * * Note that while @u_vaddr refers to DST page's userspace vaddr, it is * equally valid for SRC page as well + * + * For !VIPT cache, all of this gets compiled out as + * addr_not_cache_congruent() is 0 */ if (page_mapped(from) && addr_not_cache_congruent(kfrom, u_vaddr)) { - __flush_dcache_page(kfrom, u_vaddr); + __flush_dcache_page((unsigned long)kfrom, u_vaddr); clean_src_k_mappings = 1; } - copy_page((void *)kto, (void *)kfrom); + copy_page(kto, kfrom); /* * Mark DST page K-mapping as dirty for a later finalization by @@ -841,11 +879,14 @@ void copy_user_highpage(struct page *to, struct page *from, * sync the kernel mapping back to physical page */ if (clean_src_k_mappings) { - __flush_dcache_page(kfrom, kfrom); + __flush_dcache_page((unsigned long)kfrom, (unsigned long)kfrom); set_bit(PG_dc_clean, &from->flags); } else { clear_bit(PG_dc_clean, &from->flags); } + + kunmap_atomic(kto); + kunmap_atomic(kfrom); } void clear_user_page(void *to, unsigned long u_vaddr, struct page *page) diff --git a/arch/arc/mm/fault.c b/arch/arc/mm/fault.c index d948e4e9d..af63f4a13 100644 --- a/arch/arc/mm/fault.c +++ b/arch/arc/mm/fault.c @@ -18,7 +18,14 @@ #include <asm/pgalloc.h> #include <asm/mmu.h> -static int handle_vmalloc_fault(unsigned long address) +/* + * kernel virtual address is required to implement vmalloc/pkmap/fixmap + * Refer to asm/processor.h for System Memory Map + * + * It simply copies the PMD entry (pointer to 2nd level page table or hugepage) + * from swapper pgdir to task pgdir. The 2nd level table/page is thus shared + */ +noinline static int handle_kernel_vaddr_fault(unsigned long address) { /* * Synchronize this task's top level page-table @@ -72,8 +79,8 @@ void do_page_fault(unsigned long address, struct pt_regs *regs) * only copy the information from the master page table, * nothing more. */ - if (address >= VMALLOC_START && address <= VMALLOC_END) { - ret = handle_vmalloc_fault(address); + if (address >= VMALLOC_START) { + ret = handle_kernel_vaddr_fault(address); if (unlikely(ret)) goto bad_area_nosemaphore; else diff --git a/arch/arc/mm/highmem.c b/arch/arc/mm/highmem.c new file mode 100644 index 000000000..92dd92cad --- /dev/null +++ b/arch/arc/mm/highmem.c @@ -0,0 +1,140 @@ +/* + * Copyright (C) 2015 Synopsys, Inc. (www.synopsys.com) + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + */ + +#include <linux/bootmem.h> +#include <linux/export.h> +#include <linux/highmem.h> +#include <asm/processor.h> +#include <asm/pgtable.h> +#include <asm/pgalloc.h> +#include <asm/tlbflush.h> + +/* + * HIGHMEM API: + * + * kmap() API provides sleep semantics hence refered to as "permanent maps" + * It allows mapping LAST_PKMAP pages, using @last_pkmap_nr as the cursor + * for book-keeping + * + * kmap_atomic() can't sleep (calls pagefault_disable()), thus it provides + * shortlived ala "temporary mappings" which historically were implemented as + * fixmaps (compile time addr etc). Their book-keeping is done per cpu. + * + * Both these facts combined (preemption disabled and per-cpu allocation) + * means the total number of concurrent fixmaps will be limited to max + * such allocations in a single control path. Thus KM_TYPE_NR (another + * historic relic) is a small'ish number which caps max percpu fixmaps + * + * ARC HIGHMEM Details + * + * - the kernel vaddr space from 0x7z to 0x8z (currently used by vmalloc/module) + * is now shared between vmalloc and kmap (non overlapping though) + * + * - Both fixmap/pkmap use a dedicated page table each, hooked up to swapper PGD + * This means each only has 1 PGDIR_SIZE worth of kvaddr mappings, which means + * 2M of kvaddr space for typical config (8K page and 11:8:13 traversal split) + * + * - fixmap anyhow needs a limited number of mappings. So 2M kvaddr == 256 PTE + * slots across NR_CPUS would be more than sufficient (generic code defines + * KM_TYPE_NR as 20). + * + * - pkmap being preemptible, in theory could do with more than 256 concurrent + * mappings. However, generic pkmap code: map_new_virtual(), doesn't traverse + * the PGD and only works with a single page table @pkmap_page_table, hence + * sets the limit + */ + +extern pte_t * pkmap_page_table; +static pte_t * fixmap_page_table; + +void *kmap(struct page *page) +{ + BUG_ON(in_interrupt()); + if (!PageHighMem(page)) + return page_address(page); + + return kmap_high(page); +} + +void *kmap_atomic(struct page *page) +{ + int idx, cpu_idx; + unsigned long vaddr; + + preempt_disable(); + pagefault_disable(); + if (!PageHighMem(page)) + return page_address(page); + + cpu_idx = kmap_atomic_idx_push(); + idx = cpu_idx + KM_TYPE_NR * smp_processor_id(); + vaddr = FIXMAP_ADDR(idx); + + set_pte_at(&init_mm, vaddr, fixmap_page_table + idx, + mk_pte(page, kmap_prot)); + + return (void *)vaddr; +} +EXPORT_SYMBOL(kmap_atomic); + +void __kunmap_atomic(void *kv) +{ + unsigned long kvaddr = (unsigned long)kv; + + if (kvaddr >= FIXMAP_BASE && kvaddr < (FIXMAP_BASE + FIXMAP_SIZE)) { + + /* + * Because preemption is disabled, this vaddr can be associated + * with the current allocated index. + * But in case of multiple live kmap_atomic(), it still relies on + * callers to unmap in right order. + */ + int cpu_idx = kmap_atomic_idx(); + int idx = cpu_idx + KM_TYPE_NR * smp_processor_id(); + + WARN_ON(kvaddr != FIXMAP_ADDR(idx)); + + pte_clear(&init_mm, kvaddr, fixmap_page_table + idx); + local_flush_tlb_kernel_range(kvaddr, kvaddr + PAGE_SIZE); + + kmap_atomic_idx_pop(); + } + + pagefault_enable(); + preempt_enable(); +} +EXPORT_SYMBOL(__kunmap_atomic); + +static noinline pte_t * __init alloc_kmap_pgtable(unsigned long kvaddr) +{ + pgd_t *pgd_k; + pud_t *pud_k; + pmd_t *pmd_k; + pte_t *pte_k; + + pgd_k = pgd_offset_k(kvaddr); + pud_k = pud_offset(pgd_k, kvaddr); + pmd_k = pmd_offset(pud_k, kvaddr); + + pte_k = (pte_t *)alloc_bootmem_low_pages(PAGE_SIZE); + pmd_populate_kernel(&init_mm, pmd_k, pte_k); + return pte_k; +} + +void __init kmap_init(void) +{ + /* Due to recursive include hell, we can't do this in processor.h */ + BUILD_BUG_ON(PAGE_OFFSET < (VMALLOC_END + FIXMAP_SIZE + PKMAP_SIZE)); + + BUILD_BUG_ON(KM_TYPE_NR > PTRS_PER_PTE); + pkmap_page_table = alloc_kmap_pgtable(PKMAP_BASE); + + BUILD_BUG_ON(LAST_PKMAP > PTRS_PER_PTE); + fixmap_page_table = alloc_kmap_pgtable(FIXMAP_BASE); +} diff --git a/arch/arc/mm/init.c b/arch/arc/mm/init.c index d44eedd8c..7d2c4fbf4 100644 --- a/arch/arc/mm/init.c +++ b/arch/arc/mm/init.c @@ -15,6 +15,7 @@ #endif #include <linux/swap.h> #include <linux/module.h> +#include <linux/highmem.h> #include <asm/page.h> #include <asm/pgalloc.h> #include <asm/sections.h> @@ -24,16 +25,22 @@ pgd_t swapper_pg_dir[PTRS_PER_PGD] __aligned(PAGE_SIZE); char empty_zero_page[PAGE_SIZE] __aligned(PAGE_SIZE); EXPORT_SYMBOL(empty_zero_page); -/* Default tot mem from .config */ -static unsigned long arc_mem_sz = 0x20000000; /* some default */ +static const unsigned long low_mem_start = CONFIG_LINUX_LINK_BASE; +static unsigned long low_mem_sz; + +#ifdef CONFIG_HIGHMEM +static unsigned long min_high_pfn; +static u64 high_mem_start; +static u64 high_mem_sz; +#endif /* User can over-ride above with "mem=nnn[KkMm]" in cmdline */ static int __init setup_mem_sz(char *str) { - arc_mem_sz = memparse(str, NULL) & PAGE_MASK; + low_mem_sz = memparse(str, NULL) & PAGE_MASK; /* early console might not be setup yet - it will show up later */ - pr_info("\"mem=%s\": mem sz set to %ldM\n", str, TO_MB(arc_mem_sz)); + pr_info("\"mem=%s\": mem sz set to %ldM\n", str, TO_MB(low_mem_sz)); return 0; } @@ -41,8 +48,24 @@ early_param("mem", setup_mem_sz); void __init early_init_dt_add_memory_arch(u64 base, u64 size) { - arc_mem_sz = size & PAGE_MASK; - pr_info("Memory size set via devicetree %ldM\n", TO_MB(arc_mem_sz)); + int in_use = 0; + + if (!low_mem_sz) { + if (base != low_mem_start) + panic("CONFIG_LINUX_LINK_BASE != DT memory { }"); + + low_mem_sz = size; + in_use = 1; + } else { +#ifdef CONFIG_HIGHMEM + high_mem_start = base; + high_mem_sz = size; + in_use = 1; +#endif + } + + pr_info("Memory @ %llx [%lldM] %s\n", + base, TO_MB(size), !in_use ? "Not used":""); } #ifdef CONFIG_BLK_DEV_INITRD @@ -72,46 +95,62 @@ early_param("initrd", early_initrd); void __init setup_arch_memory(void) { unsigned long zones_size[MAX_NR_ZONES]; - unsigned long end_mem = CONFIG_LINUX_LINK_BASE + arc_mem_sz; + unsigned long zones_holes[MAX_NR_ZONES]; init_mm.start_code = (unsigned long)_text; init_mm.end_code = (unsigned long)_etext; init_mm.end_data = (unsigned long)_edata; init_mm.brk = (unsigned long)_end; - /* - * We do it here, so that memory is correctly instantiated - * even if "mem=xxx" cmline over-ride is given and/or - * DT has memory node. Each causes an update to @arc_mem_sz - * and we finally add memory one here - */ - memblock_add(CONFIG_LINUX_LINK_BASE, arc_mem_sz); - - /*------------- externs in mm need setting up ---------------*/ - /* first page of system - kernel .vector starts here */ min_low_pfn = ARCH_PFN_OFFSET; - /* Last usable page of low mem (no HIGHMEM yet for ARC port) */ - max_low_pfn = max_pfn = PFN_DOWN(end_mem); + /* Last usable page of low mem */ + max_low_pfn = max_pfn = PFN_DOWN(low_mem_start + low_mem_sz); - max_mapnr = max_low_pfn - min_low_pfn; +#ifdef CONFIG_HIGHMEM + min_high_pfn = PFN_DOWN(high_mem_start); + max_pfn = PFN_DOWN(high_mem_start + high_mem_sz); +#endif + + max_mapnr = max_pfn - min_low_pfn; - /*------------- reserve kernel image -----------------------*/ - memblock_reserve(CONFIG_LINUX_LINK_BASE, - __pa(_end) - CONFIG_LINUX_LINK_BASE); + /*------------- bootmem allocator setup -----------------------*/ + + /* + * seed the bootmem allocator after any DT memory node parsing or + * "mem=xxx" cmdline overrides have potentially updated @arc_mem_sz + * + * Only low mem is added, otherwise we have crashes when allocating + * mem_map[] itself. NO_BOOTMEM allocates mem_map[] at the end of + * avail memory, ending in highmem with a > 32-bit address. However + * it then tries to memset it with a truncaed 32-bit handle, causing + * the crash + */ + + memblock_add(low_mem_start, low_mem_sz); + memblock_reserve(low_mem_start, __pa(_end) - low_mem_start); #ifdef CONFIG_BLK_DEV_INITRD - /*------------- reserve initrd image -----------------------*/ if (initrd_start) memblock_reserve(__pa(initrd_start), initrd_end - initrd_start); #endif memblock_dump_all(); - /*-------------- node setup --------------------------------*/ + /*----------------- node/zones setup --------------------------*/ memset(zones_size, 0, sizeof(zones_size)); - zones_size[ZONE_NORMAL] = max_mapnr; + memset(zones_holes, 0, sizeof(zones_holes)); + + zones_size[ZONE_NORMAL] = max_low_pfn - min_low_pfn; + zones_holes[ZONE_NORMAL] = 0; + +#ifdef CONFIG_HIGHMEM + zones_size[ZONE_HIGHMEM] = max_pfn - max_low_pfn; + + /* This handles the peripheral address space hole */ + zones_holes[ZONE_HIGHMEM] = min_high_pfn - max_low_pfn; +#endif /* * We can't use the helper free_area_init(zones[]) because it uses @@ -122,9 +161,12 @@ void __init setup_arch_memory(void) free_area_init_node(0, /* node-id */ zones_size, /* num pages per zone */ min_low_pfn, /* first pfn of node */ - NULL); /* NO holes */ + zones_holes); /* holes */ - high_memory = (void *)end_mem; +#ifdef CONFIG_HIGHMEM + high_memory = (void *)(min_high_pfn << PAGE_SHIFT); + kmap_init(); +#endif } /* @@ -135,6 +177,14 @@ void __init setup_arch_memory(void) */ void __init mem_init(void) { +#ifdef CONFIG_HIGHMEM + unsigned long tmp; + + reset_all_zones_managed_pages(); + for (tmp = min_high_pfn; tmp < max_pfn; tmp++) + free_highmem_page(pfn_to_page(tmp)); +#endif + free_all_bootmem(); mem_init_print_info(NULL); } diff --git a/arch/arc/mm/tlb.c b/arch/arc/mm/tlb.c index 2c7ce8bb7..daf2bf52b 100644 --- a/arch/arc/mm/tlb.c +++ b/arch/arc/mm/tlb.c @@ -109,6 +109,10 @@ DEFINE_PER_CPU(unsigned int, asid_cache) = MM_CTXT_FIRST_CYCLE; static inline void __tlb_entry_erase(void) { write_aux_reg(ARC_REG_TLBPD1, 0); + + if (is_pae40_enabled()) + write_aux_reg(ARC_REG_TLBPD1HI, 0); + write_aux_reg(ARC_REG_TLBPD0, 0); write_aux_reg(ARC_REG_TLBCOMMAND, TLBWrite); } @@ -182,7 +186,7 @@ static void utlb_invalidate(void) } -static void tlb_entry_insert(unsigned int pd0, unsigned int pd1) +static void tlb_entry_insert(unsigned int pd0, pte_t pd1) { unsigned int idx; @@ -225,10 +229,14 @@ static void tlb_entry_erase(unsigned int vaddr_n_asid) write_aux_reg(ARC_REG_TLBCOMMAND, TLBDeleteEntry); } -static void tlb_entry_insert(unsigned int pd0, unsigned int pd1) +static void tlb_entry_insert(unsigned int pd0, pte_t pd1) { write_aux_reg(ARC_REG_TLBPD0, pd0); write_aux_reg(ARC_REG_TLBPD1, pd1); + + if (is_pae40_enabled()) + write_aux_reg(ARC_REG_TLBPD1HI, (u64)pd1 >> 32); + write_aux_reg(ARC_REG_TLBCOMMAND, TLBInsertEntry); } @@ -240,22 +248,39 @@ static void tlb_entry_insert(unsigned int pd0, unsigned int pd1) noinline void local_flush_tlb_all(void) { + struct cpuinfo_arc_mmu *mmu = &cpuinfo_arc700[smp_processor_id()].mmu; unsigned long flags; unsigned int entry; - struct cpuinfo_arc_mmu *mmu = &cpuinfo_arc700[smp_processor_id()].mmu; + int num_tlb = mmu->sets * mmu->ways; local_irq_save(flags); /* Load PD0 and PD1 with template for a Blank Entry */ write_aux_reg(ARC_REG_TLBPD1, 0); + + if (is_pae40_enabled()) + write_aux_reg(ARC_REG_TLBPD1HI, 0); + write_aux_reg(ARC_REG_TLBPD0, 0); - for (entry = 0; entry < mmu->num_tlb; entry++) { + for (entry = 0; entry < num_tlb; entry++) { /* write this entry to the TLB */ write_aux_reg(ARC_REG_TLBINDEX, entry); write_aux_reg(ARC_REG_TLBCOMMAND, TLBWrite); } + if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) { + const int stlb_idx = 0x800; + + /* Blank sTLB entry */ + write_aux_reg(ARC_REG_TLBPD0, _PAGE_HW_SZ); + + for (entry = stlb_idx; entry < stlb_idx + 16; entry++) { + write_aux_reg(ARC_REG_TLBINDEX, entry); + write_aux_reg(ARC_REG_TLBCOMMAND, TLBWrite); + } + } + utlb_invalidate(); local_irq_restore(flags); @@ -409,6 +434,15 @@ static inline void ipi_flush_tlb_range(void *arg) local_flush_tlb_range(ta->ta_vma, ta->ta_start, ta->ta_end); } +#ifdef CONFIG_TRANSPARENT_HUGEPAGE +static inline void ipi_flush_pmd_tlb_range(void *arg) +{ + struct tlb_args *ta = arg; + + local_flush_pmd_tlb_range(ta->ta_vma, ta->ta_start, ta->ta_end); +} +#endif + static inline void ipi_flush_tlb_kernel_range(void *arg) { struct tlb_args *ta = (struct tlb_args *)arg; @@ -449,6 +483,20 @@ void flush_tlb_range(struct vm_area_struct *vma, unsigned long start, on_each_cpu_mask(mm_cpumask(vma->vm_mm), ipi_flush_tlb_range, &ta, 1); } +#ifdef CONFIG_TRANSPARENT_HUGEPAGE +void flush_pmd_tlb_range(struct vm_area_struct *vma, unsigned long start, + unsigned long end) +{ + struct tlb_args ta = { + .ta_vma = vma, + .ta_start = start, + .ta_end = end + }; + + on_each_cpu_mask(mm_cpumask(vma->vm_mm), ipi_flush_pmd_tlb_range, &ta, 1); +} +#endif + void flush_tlb_kernel_range(unsigned long start, unsigned long end) { struct tlb_args ta = { @@ -463,11 +511,12 @@ void flush_tlb_kernel_range(unsigned long start, unsigned long end) /* * Routine to create a TLB entry */ -void create_tlb(struct vm_area_struct *vma, unsigned long address, pte_t *ptep) +void create_tlb(struct vm_area_struct *vma, unsigned long vaddr, pte_t *ptep) { unsigned long flags; unsigned int asid_or_sasid, rwx; - unsigned long pd0, pd1; + unsigned long pd0; + pte_t pd1; /* * create_tlb() assumes that current->mm == vma->mm, since @@ -499,9 +548,9 @@ void create_tlb(struct vm_area_struct *vma, unsigned long address, pte_t *ptep) local_irq_save(flags); - tlb_paranoid_check(asid_mm(vma->vm_mm, smp_processor_id()), address); + tlb_paranoid_check(asid_mm(vma->vm_mm, smp_processor_id()), vaddr); - address &= PAGE_MASK; + vaddr &= PAGE_MASK; /* update this PTE credentials */ pte_val(*ptep) |= (_PAGE_PRESENT | _PAGE_ACCESSED); @@ -511,7 +560,7 @@ void create_tlb(struct vm_area_struct *vma, unsigned long address, pte_t *ptep) /* ASID for this task */ asid_or_sasid = read_aux_reg(ARC_REG_PID) & 0xff; - pd0 = address | asid_or_sasid | (pte_val(*ptep) & PTE_BITS_IN_PD0); + pd0 = vaddr | asid_or_sasid | (pte_val(*ptep) & PTE_BITS_IN_PD0); /* * ARC MMU provides fully orthogonal access bits for K/U mode, @@ -547,7 +596,7 @@ void update_mmu_cache(struct vm_area_struct *vma, unsigned long vaddr_unaligned, pte_t *ptep) { unsigned long vaddr = vaddr_unaligned & PAGE_MASK; - unsigned long paddr = pte_val(*ptep) & PAGE_MASK; + phys_addr_t paddr = pte_val(*ptep) & PAGE_MASK; struct page *page = pfn_to_page(pte_pfn(*ptep)); create_tlb(vma, vaddr, ptep); @@ -570,16 +619,105 @@ void update_mmu_cache(struct vm_area_struct *vma, unsigned long vaddr_unaligned, int dirty = !test_and_set_bit(PG_dc_clean, &page->flags); if (dirty) { - /* wback + inv dcache lines */ + /* wback + inv dcache lines (K-mapping) */ __flush_dcache_page(paddr, paddr); - /* invalidate any existing icache lines */ + /* invalidate any existing icache lines (U-mapping) */ if (vma->vm_flags & VM_EXEC) __inv_icache_page(paddr, vaddr); } } } +#ifdef CONFIG_TRANSPARENT_HUGEPAGE + +/* + * MMUv4 in HS38x cores supports Super Pages which are basis for Linux THP + * support. + * + * Normal and Super pages can co-exist (ofcourse not overlap) in TLB with a + * new bit "SZ" in TLB page desciptor to distinguish between them. + * Super Page size is configurable in hardware (4K to 16M), but fixed once + * RTL builds. + * + * The exact THP size a Linx configuration will support is a function of: + * - MMU page size (typical 8K, RTL fixed) + * - software page walker address split between PGD:PTE:PFN (typical + * 11:8:13, but can be changed with 1 line) + * So for above default, THP size supported is 8K * (2^8) = 2M + * + * Default Page Walker is 2 levels, PGD:PTE:PFN, which in THP regime + * reduces to 1 level (as PTE is folded into PGD and canonically referred + * to as PMD). + * Thus THP PMD accessors are implemented in terms of PTE (just like sparc) + */ + +void update_mmu_cache_pmd(struct vm_area_struct *vma, unsigned long addr, + pmd_t *pmd) +{ + pte_t pte = __pte(pmd_val(*pmd)); + update_mmu_cache(vma, addr, &pte); +} + +void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp, + pgtable_t pgtable) +{ + struct list_head *lh = (struct list_head *) pgtable; + + assert_spin_locked(&mm->page_table_lock); + + /* FIFO */ + if (!pmd_huge_pte(mm, pmdp)) + INIT_LIST_HEAD(lh); + else + list_add(lh, (struct list_head *) pmd_huge_pte(mm, pmdp)); + pmd_huge_pte(mm, pmdp) = pgtable; +} + +pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp) +{ + struct list_head *lh; + pgtable_t pgtable; + + assert_spin_locked(&mm->page_table_lock); + + pgtable = pmd_huge_pte(mm, pmdp); + lh = (struct list_head *) pgtable; + if (list_empty(lh)) + pmd_huge_pte(mm, pmdp) = NULL; + else { + pmd_huge_pte(mm, pmdp) = (pgtable_t) lh->next; + list_del(lh); + } + + pte_val(pgtable[0]) = 0; + pte_val(pgtable[1]) = 0; + + return pgtable; +} + +void local_flush_pmd_tlb_range(struct vm_area_struct *vma, unsigned long start, + unsigned long end) +{ + unsigned int cpu; + unsigned long flags; + + local_irq_save(flags); + + cpu = smp_processor_id(); + + if (likely(asid_mm(vma->vm_mm, cpu) != MM_CTXT_NO_ASID)) { + unsigned int asid = hw_pid(vma->vm_mm, cpu); + + /* No need to loop here: this will always be for 1 Huge Page */ + tlb_entry_erase(start | _PAGE_HW_SZ | asid); + } + + local_irq_restore(flags); +} + +#endif + /* Read the Cache Build Confuration Registers, Decode them and save into * the cpuinfo structure for later use. * No Validation is done here, simply read/convert the BCRs @@ -598,10 +736,10 @@ void read_decode_mmu_bcr(void) struct bcr_mmu_3 { #ifdef CONFIG_CPU_BIG_ENDIAN - unsigned int ver:8, ways:4, sets:4, osm:1, reserv:3, pg_sz:4, + unsigned int ver:8, ways:4, sets:4, res:3, sasid:1, pg_sz:4, u_itlb:4, u_dtlb:4; #else - unsigned int u_dtlb:4, u_itlb:4, pg_sz:4, reserv:3, osm:1, sets:4, + unsigned int u_dtlb:4, u_itlb:4, pg_sz:4, sasid:1, res:3, sets:4, ways:4, ver:8; #endif } *mmu3; @@ -622,7 +760,7 @@ void read_decode_mmu_bcr(void) if (mmu->ver <= 2) { mmu2 = (struct bcr_mmu_1_2 *)&tmp; - mmu->pg_sz_k = TO_KB(PAGE_SIZE); + mmu->pg_sz_k = TO_KB(0x2000); mmu->sets = 1 << mmu2->sets; mmu->ways = 1 << mmu2->ways; mmu->u_dtlb = mmu2->u_dtlb; @@ -634,6 +772,7 @@ void read_decode_mmu_bcr(void) mmu->ways = 1 << mmu3->ways; mmu->u_dtlb = mmu3->u_dtlb; mmu->u_itlb = mmu3->u_itlb; + mmu->sasid = mmu3->sasid; } else { mmu4 = (struct bcr_mmu_4 *)&tmp; mmu->pg_sz_k = 1 << (mmu4->sz0 - 1); @@ -642,9 +781,9 @@ void read_decode_mmu_bcr(void) mmu->ways = mmu4->n_ways * 2; mmu->u_dtlb = mmu4->u_dtlb * 4; mmu->u_itlb = mmu4->u_itlb * 4; + mmu->sasid = mmu4->sasid; + mmu->pae = mmu4->pae; } - - mmu->num_tlb = mmu->sets * mmu->ways; } char *arc_mmu_mumbojumbo(int cpu_id, char *buf, int len) @@ -655,14 +794,15 @@ char *arc_mmu_mumbojumbo(int cpu_id, char *buf, int len) if (p_mmu->s_pg_sz_m) scnprintf(super_pg, 64, "%dM Super Page%s, ", - p_mmu->s_pg_sz_m, " (not used)"); + p_mmu->s_pg_sz_m, + IS_USED_CFG(CONFIG_TRANSPARENT_HUGEPAGE)); n += scnprintf(buf + n, len - n, - "MMU [v%x]\t: %dk PAGE, %sJTLB %d (%dx%d), uDTLB %d, uITLB %d %s\n", + "MMU [v%x]\t: %dk PAGE, %sJTLB %d (%dx%d), uDTLB %d, uITLB %d %s%s\n", p_mmu->ver, p_mmu->pg_sz_k, super_pg, - p_mmu->num_tlb, p_mmu->sets, p_mmu->ways, + p_mmu->sets * p_mmu->ways, p_mmu->sets, p_mmu->ways, p_mmu->u_dtlb, p_mmu->u_itlb, - IS_ENABLED(CONFIG_ARC_MMU_SASID) ? ",SASID" : ""); + IS_AVAIL2(p_mmu->pae, "PAE40 ", CONFIG_ARC_HAS_PAE40)); return buf; } @@ -690,6 +830,14 @@ void arc_mmu_init(void) if (mmu->pg_sz_k != TO_KB(PAGE_SIZE)) panic("MMU pg size != PAGE_SIZE (%luk)\n", TO_KB(PAGE_SIZE)); + if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) && + mmu->s_pg_sz_m != TO_MB(HPAGE_PMD_SIZE)) + panic("MMU Super pg size != Linux HPAGE_PMD_SIZE (%luM)\n", + (unsigned long)TO_MB(HPAGE_PMD_SIZE)); + + if (IS_ENABLED(CONFIG_ARC_HAS_PAE40) && !mmu->pae) + panic("Hardware doesn't support PAE40\n"); + /* Enable the MMU */ write_aux_reg(ARC_REG_PID, MMU_ENABLE); @@ -725,15 +873,15 @@ void arc_mmu_init(void) * the duplicate one. * -Knob to be verbose abt it.(TODO: hook them up to debugfs) */ -volatile int dup_pd_verbose = 1;/* Be slient abt it or complain (default) */ +volatile int dup_pd_silent; /* Be slient abt it or complain (default) */ void do_tlb_overlap_fault(unsigned long cause, unsigned long address, struct pt_regs *regs) { - int set, way, n; - unsigned long flags, is_valid; struct cpuinfo_arc_mmu *mmu = &cpuinfo_arc700[smp_processor_id()].mmu; - unsigned int pd0[mmu->ways], pd1[mmu->ways]; + unsigned int pd0[mmu->ways]; + unsigned long flags; + int set; local_irq_save(flags); @@ -743,14 +891,16 @@ void do_tlb_overlap_fault(unsigned long cause, unsigned long address, /* loop thru all sets of TLB */ for (set = 0; set < mmu->sets; set++) { + int is_valid, way; + /* read out all the ways of current set */ for (way = 0, is_valid = 0; way < mmu->ways; way++) { write_aux_reg(ARC_REG_TLBINDEX, SET_WAY_TO_IDX(mmu, set, way)); write_aux_reg(ARC_REG_TLBCOMMAND, TLBRead); pd0[way] = read_aux_reg(ARC_REG_TLBPD0); - pd1[way] = read_aux_reg(ARC_REG_TLBPD1); is_valid |= pd0[way] & _PAGE_PRESENT; + pd0[way] &= PAGE_MASK; } /* If all the WAYS in SET are empty, skip to next SET */ @@ -759,30 +909,28 @@ void do_tlb_overlap_fault(unsigned long cause, unsigned long address, /* Scan the set for duplicate ways: needs a nested loop */ for (way = 0; way < mmu->ways - 1; way++) { + + int n; + if (!pd0[way]) continue; for (n = way + 1; n < mmu->ways; n++) { - if ((pd0[way] & PAGE_MASK) == - (pd0[n] & PAGE_MASK)) { - - if (dup_pd_verbose) { - pr_info("Duplicate PD's @" - "[%d:%d]/[%d:%d]\n", - set, way, set, n); - pr_info("TLBPD0[%u]: %08x\n", - way, pd0[way]); - } - - /* - * clear entry @way and not @n. This is - * critical to our optimised loop - */ - pd0[way] = pd1[way] = 0; - write_aux_reg(ARC_REG_TLBINDEX, + if (pd0[way] != pd0[n]) + continue; + + if (!dup_pd_silent) + pr_info("Dup TLB PD0 %08x @ set %d ways %d,%d\n", + pd0[way], set, way, n); + + /* + * clear entry @way and not @n. + * This is critical to our optimised loop + */ + pd0[way] = 0; + write_aux_reg(ARC_REG_TLBINDEX, SET_WAY_TO_IDX(mmu, set, way)); - __tlb_entry_erase(); - } + __tlb_entry_erase(); } } } diff --git a/arch/arc/mm/tlbex.S b/arch/arc/mm/tlbex.S index f6f4c3cb5..f1967eeb3 100644 --- a/arch/arc/mm/tlbex.S +++ b/arch/arc/mm/tlbex.S @@ -88,7 +88,7 @@ ex_saved_reg1: #ifdef CONFIG_SMP sr r0, [ARC_REG_SCRATCH_DATA0] ; freeup r0 to code with GET_CPU_ID r0 ; get to per cpu scratch mem, - lsl r0, r0, L1_CACHE_SHIFT ; cache line wide per cpu + asl r0, r0, L1_CACHE_SHIFT ; cache line wide per cpu add r0, @ex_saved_reg1, r0 #else st r0, [@ex_saved_reg1] @@ -107,7 +107,7 @@ ex_saved_reg1: .macro TLBMISS_RESTORE_REGS #ifdef CONFIG_SMP GET_CPU_ID r0 ; get to per cpu scratch mem - lsl r0, r0, L1_CACHE_SHIFT ; each is cache line wide + asl r0, r0, L1_CACHE_SHIFT ; each is cache line wide add r0, @ex_saved_reg1, r0 ld_s r3, [r0,12] ld_s r2, [r0, 8] @@ -205,20 +205,38 @@ ex_saved_reg1: #endif lsr r0, r2, PGDIR_SHIFT ; Bits for indexing into PGD - ld.as r1, [r1, r0] ; PGD entry corresp to faulting addr - and.f r1, r1, PAGE_MASK ; Ignoring protection and other flags - ; contains Ptr to Page Table - bz.d do_slow_path_pf ; if no Page Table, do page fault + ld.as r3, [r1, r0] ; PGD entry corresp to faulting addr + tst r3, r3 + bz do_slow_path_pf ; if no Page Table, do page fault + +#ifdef CONFIG_TRANSPARENT_HUGEPAGE + and.f 0, r3, _PAGE_HW_SZ ; Is this Huge PMD (thp) + add2.nz r1, r1, r0 + bnz.d 2f ; YES: PGD == PMD has THP PTE: stop pgd walk + mov.nz r0, r3 + +#endif + and r1, r3, PAGE_MASK ; Get the PTE entry: The idea is ; (1) x = addr >> PAGE_SHIFT -> masks page-off bits from @fault-addr ; (2) y = x & (PTRS_PER_PTE - 1) -> to get index - ; (3) z = pgtbl[y] - ; To avoid the multiply by in end, we do the -2, <<2 below + ; (3) z = (pgtbl + y * 4) + +#ifdef CONFIG_ARC_HAS_PAE40 +#define PTE_SIZE_LOG 3 /* 8 == 2 ^ 3 */ +#else +#define PTE_SIZE_LOG 2 /* 4 == 2 ^ 2 */ +#endif + + ; multiply in step (3) above avoided by shifting lesser in step (1) + lsr r0, r2, ( PAGE_SHIFT - PTE_SIZE_LOG ) + and r0, r0, ( (PTRS_PER_PTE - 1) << PTE_SIZE_LOG ) + ld.aw r0, [r1, r0] ; r0: PTE (lower word only for PAE40) + ; r1: PTE ptr + +2: - lsr r0, r2, (PAGE_SHIFT - 2) - and r0, r0, ( (PTRS_PER_PTE - 1) << 2) - ld.aw r0, [r1, r0] ; get PTE and PTE ptr for fault addr #ifdef CONFIG_ARC_DBG_TLB_MISS_COUNT and.f 0, r0, _PAGE_PRESENT bz 1f @@ -233,18 +251,23 @@ ex_saved_reg1: ;----------------------------------------------------------------- ; Convert Linux PTE entry into TLB entry ; A one-word PTE entry is programmed as two-word TLB Entry [PD0:PD1] in mmu +; (for PAE40, two-words PTE, while three-word TLB Entry [PD0:PD1:PD1HI]) ; IN: r0 = PTE, r1 = ptr to PTE .macro CONV_PTE_TO_TLB - and r3, r0, PTE_BITS_RWX ; r w x - lsl r2, r3, 3 ; r w x 0 0 0 (GLOBAL, kernel only) + and r3, r0, PTE_BITS_RWX ; r w x + asl r2, r3, 3 ; Kr Kw Kx 0 0 0 (GLOBAL, kernel only) and.f 0, r0, _PAGE_GLOBAL - or.z r2, r2, r3 ; r w x r w x (!GLOBAL, user page) + or.z r2, r2, r3 ; Kr Kw Kx Ur Uw Ux (!GLOBAL, user page) and r3, r0, PTE_BITS_NON_RWX_IN_PD1 ; Extract PFN+cache bits from PTE or r3, r3, r2 - sr r3, [ARC_REG_TLBPD1] ; these go in PD1 + sr r3, [ARC_REG_TLBPD1] ; paddr[31..13] | Kr Kw Kx Ur Uw Ux | C +#ifdef CONFIG_ARC_HAS_PAE40 + ld r3, [r1, 4] ; paddr[39..32] + sr r3, [ARC_REG_TLBPD1HI] +#endif and r2, r0, PTE_BITS_IN_PD0 ; Extract other PTE flags: (V)alid, (G)lb @@ -365,7 +388,7 @@ ENTRY(EV_TLBMissD) lr r3, [ecr] or r0, r0, _PAGE_ACCESSED ; Accessed bit always btst_s r3, ECR_C_BIT_DTLB_ST_MISS ; See if it was a Write Access ? - or.nz r0, r0, _PAGE_MODIFIED ; if Write, set Dirty bit as well + or.nz r0, r0, _PAGE_DIRTY ; if Write, set Dirty bit as well st_s r0, [r1] ; Write back PTE CONV_PTE_TO_TLB |