From mboxrd@z Thu Jan 1 00:00:00 1970 From: zhichang.yuan@linaro.org (zhichang.yuan) Date: Tue, 28 Oct 2014 14:11:22 +0800 Subject: [PATCH v0] ARMv8:mm:Support the DEBUG_PAGEALLOC In-Reply-To: <544EC65E.4060009@codeaurora.org> References: <1414382488-916-1-git-send-email-zhichang.yuan@linaro.org> <544EC65E.4060009@codeaurora.org> Message-ID: <544F338A.2060600@linaro.org> To: linux-arm-kernel@lists.infradead.org List-Id: linux-arm-kernel.lists.infradead.org Hi, Laura, Thanks for your comments! On 2014?10?28? 06:25, Laura Abbott wrote: > Hi, > > On 10/26/2014 9:01 PM, zhichang.yuan at linaro.org wrote: >> From: "zhichang.yuan" >> >> This patch targets to support DEBUG_PAGEALLOC on ARMv8. >> >> Meanwhile, it adds the processing to free some unused >> page tables created in direct kernel mapping. >> > > It might be good to split out the freeing of the page > tables into a separate patch for review since it looks to be > separate of DEBUG_PAGEALLOC. > Yes. It is not relative to DEBUG_PAGEALLOC. I just think it is small change, so do not break down it. But it is not right. I will separate it. >> The patch was tested based on the following code @: >> https://git.linaro.org/people/zhichang.yuan/pgalloc.git/shortlog/refs/heads/test_pgalloc_v1 >> >> Signed-off-by: Zhichang Yuan >> --- >> arch/arm64/Kconfig | 3 + >> arch/arm64/include/asm/pgtable-hwdef.h | 6 + >> arch/arm64/include/asm/pgtable.h | 21 +++ >> arch/arm64/mm/mmu.c | 263 +++++++++++++++++++++++++++++++- >> 4 files changed, 288 insertions(+), 5 deletions(-) >> >> diff --git a/arch/arm64/Kconfig b/arch/arm64/Kconfig >> index fd4e81a..da072d4 100644 >> --- a/arch/arm64/Kconfig >> +++ b/arch/arm64/Kconfig >> @@ -128,6 +128,9 @@ config KERNEL_MODE_NEON >> config FIX_EARLYCON_MEM >> def_bool y >> >> +config ARCH_SUPPORTS_DEBUG_PAGEALLOC >> + def_bool y >> + >> source "init/Kconfig" >> >> source "kernel/Kconfig.freezer" >> diff --git a/arch/arm64/include/asm/pgtable-hwdef.h b/arch/arm64/include/asm/pgtable-hwdef.h >> index 88174e0..0a62e9a 100644 >> --- a/arch/arm64/include/asm/pgtable-hwdef.h >> +++ b/arch/arm64/include/asm/pgtable-hwdef.h >> @@ -65,6 +65,12 @@ >> #define PUD_TYPE_SECT (_AT(pgdval_t, 1) << 0) >> >> /* >> + * Section >> + */ >> +#define PUD_SECT_VALID (_AT(pmdval_t, 1) << 0) >> +#define PUD_SECT_PROT_NONE (_AT(pmdval_t, 1) << 58) >> + >> +/* >> * Level 2 descriptor (PMD). >> */ >> #define PMD_TYPE_MASK (_AT(pmdval_t, 3) << 0) >> diff --git a/arch/arm64/include/asm/pgtable.h b/arch/arm64/include/asm/pgtable.h >> index ffe1ba0..4246f3b 100644 >> --- a/arch/arm64/include/asm/pgtable.h >> +++ b/arch/arm64/include/asm/pgtable.h >> @@ -313,9 +313,12 @@ extern pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn, >> >> #ifdef CONFIG_ARM64_64K_PAGES >> #define pud_sect(pud) (0) >> +#define pud_table(pud) (1) >> #else >> #define pud_sect(pud) ((pud_val(pud) & PUD_TYPE_MASK) == \ >> PUD_TYPE_SECT) >> +#define pud_table(pud) ((pud_val(pud) & PUD_TYPE_MASK) == \ >> + PUD_TYPE_TABLE) >> #endif >> >> static inline void set_pmd(pmd_t *pmdp, pmd_t pmd) >> @@ -422,6 +425,14 @@ static inline pud_t *pud_offset(pgd_t *pgd, unsigned long addr) >> /* to find an entry in a kernel page-table-directory */ >> #define pgd_offset_k(addr) pgd_offset(&init_mm, addr) >> >> + >> +#define PHYSICAL_PAGE_MASK (((signed long)PAGE_MASK) & PHYS_MASK) >> +/* PTE_PFN_MASK extracts the PFN from a (pte|pmd|pud|pgd)val_t */ >> +#define PTE_PFN_MASK ((pteval_t)PHYSICAL_PAGE_MASK) >> + >> +/* PTE_FLAGS_MASK extracts the flags from a (pte|pmd|pud|pgd)val_t */ >> +#define PTE_FLAGS_MASK (~PTE_PFN_MASK) >> + >> static inline pte_t pte_modify(pte_t pte, pgprot_t newprot) >> { >> const pteval_t mask = PTE_USER | PTE_PXN | PTE_UXN | PTE_RDONLY | >> @@ -477,6 +488,16 @@ extern pgd_t idmap_pg_dir[PTRS_PER_PGD]; >> >> #define PTE_FILE_MAX_BITS 55 >> >> +extern pte_t *lookup_kaddress(unsigned long address, unsigned int *level); >> + >> +enum pg_level { >> + PG_LEVEL_NONE, >> + PG_LEVEL_PAGE, >> + PG_LEVEL_PMD, >> + PG_LEVEL_PUD, >> + PG_LEVEL_NUM >> +}; >> + >> extern int kern_addr_valid(unsigned long addr); >> >> #include >> diff --git a/arch/arm64/mm/mmu.c b/arch/arm64/mm/mmu.c >> index c555672..6bc5f70 100644 >> --- a/arch/arm64/mm/mmu.c >> +++ b/arch/arm64/mm/mmu.c >> @@ -132,10 +132,21 @@ EXPORT_SYMBOL(phys_mem_access_prot); >> static void __init *early_alloc(unsigned long sz) >> { >> void *ptr = __va(memblock_alloc(sz, sz)); >> + >> memset(ptr, 0, sz); >> return ptr; >> } >> >> +/*Free the page used as pmd table entry in direct mapping*/ >> +static inline void __init pmd_table_free(pmd_t *pmd, unsigned long addr) >> +{ >> + if (pmd_table(*pmd)) { >> + phys_addr_t table = __pa(pte_offset_kernel(pmd, 0)); >> + > > pte_offset_kernel -> pmd_page_vaddr which calls __va already. Is > there another way to get the physical address without having to go > pa -> va -> pa again? > Yes. It seems to be verbose when offset is 0. I want to use existed macro. I think i had other two options. First, directly get the physical address just like that: pmd_val(*pmd) & PHYS_MASK & (s32)PAGE_MASK or pte_pfn(pmd_pte(*pmd)) << PAGE_SHIFT Maybe the front one is better. >> + memblock_free(table, PAGE_SIZE); >> + } >> +} >> + >> static void __init alloc_init_pte(pmd_t *pmd, unsigned long addr, >> unsigned long end, unsigned long pfn, >> pgprot_t prot) >> @@ -185,14 +196,16 @@ static void __init alloc_init_pmd(pud_t *pud, unsigned long addr, >> next = pmd_addr_end(addr, end); >> /* try section mapping first */ >> if (((addr | next | phys) & ~SECTION_MASK) == 0) { >> - pmd_t old_pmd =*pmd; >> + pmd_t old_pmd = *pmd; >> + >> set_pmd(pmd, __pmd(phys | prot_sect)); >> /* >> * Check for previous table entries created during >> * boot (__create_page_tables) and flush them. >> */ >> if (!pmd_none(old_pmd)) >> - flush_tlb_all(); >> + flush_tlb_kernel_range(addr, next); > Was the switch from flush_tlb_all -> flush_tlb_kernel_range found from > code inspection or as part of something else? Do you have any performance > data about from this switch? > As for this issue, i have no performance data. But you can check the implementation of flush_tlb_kernel_range. when the pages size to be flushed is over MAX_TLB_RANGE, it will call __flush_tlb_kernel_range; otherwise, same as flush_tlb_all. I think someone had evaluated the performance difference. >> + pmd_table_free(&old_pmd, addr); >> } else { >> alloc_init_pte(pmd, addr, next, __phys_to_pfn(phys), >> prot_pte); >> @@ -224,6 +237,7 @@ static void __init alloc_init_pud(pgd_t *pgd, unsigned long addr, >> if (!map_io && (PAGE_SHIFT == 12) && >> ((addr | next | phys) & ~PUD_MASK) == 0) { >> pud_t old_pud = *pud; >> + >> set_pud(pud, __pud(phys | PROT_SECT_NORMAL_EXEC)); >> >> /* >> @@ -234,9 +248,15 @@ static void __init alloc_init_pud(pgd_t *pgd, unsigned long addr, >> * Look up the old pmd table and free it. >> */ >> if (!pud_none(old_pud)) { >> - phys_addr_t table = __pa(pmd_offset(&old_pud, 0)); >> - memblock_free(table, PAGE_SIZE); >> - flush_tlb_all(); >> + flush_tlb_kernel_range(addr, next); >> + if (pud_table(old_pud)) { >> + phys_addr_t table = >> + __pa(pmd_offset(&old_pud, 0)); >> + pmd_t *pmd = pmd_offset(pud, addr); >> + >> + pmd_table_free(pmd, addr); >> + memblock_free(table, PAGE_SIZE); >> + } >> } >> } else { >> alloc_init_pmd(pud, addr, next, phys, map_io); >> @@ -287,6 +307,219 @@ void __init create_id_mapping(phys_addr_t addr, phys_addr_t size, int map_io) >> addr, addr, size, map_io); >> } >> >> +static inline pte_t *lookup_kaddress_in_pgd(pgd_t *pgd, >> + unsigned long address, >> + unsigned int *level) >> +{ >> + pud_t *pud; >> + pmd_t *pmd; >> + >> + *level = PG_LEVEL_NONE; >> + >> + if (pgd_none(*pgd)) >> + return NULL; >> + >> + pud = pud_offset(pgd, address); >> + if (pud_none(*pud)) >> + return NULL; >> + >> + *level = PG_LEVEL_PUD; >> + if (pud_sect(*pud) || !(pud_val(*pud) & PUD_SECT_VALID)) >> + return (pte_t *)pud; >> + >> + pmd = pmd_offset(pud, address); >> + if (pmd_none(*pmd)) >> + return NULL; >> + >> + /* >> + * !(pmd_val(*pmd) & PMD_SECT_VALID) means PMD_TYPE_MASK &(*pmd) >> + * is 0, but *pmd is non-zero. >> + * For Huge page split. >> + */ >> + *level = PG_LEVEL_PMD; >> + if (pmd_sect(*pmd) || !(pmd_val(*pmd) & PMD_SECT_VALID)) >> + return (pte_t *)pmd; >> + >> + *level = PG_LEVEL_PAGE; >> + >> + return pte_offset_kernel(pmd, address); >> +} >> + >> + >> +pte_t *lookup_kaddress(unsigned long address, unsigned int *level) >> +{ >> + return lookup_kaddress_in_pgd(pgd_offset_k(address), address, level); >> +} >> +EXPORT_SYMBOL_GPL(lookup_kaddress); > > Do you need the EXPORT_SYMBOL right now? Because my original test code is module, and call this function. > >> + >> + >> +#ifdef CONFIG_DEBUG_PAGEALLOC >> + >> +static void __split_pmd_page_mapping(pmd_t *pmd, >> + unsigned long addr, >> + void *p_base) >> +{ >> + pte_t *pte_base; >> + int i; >> + >> + pgprot_t prot_val; >> + >> + unsigned long pfn; >> + >> + pte_base = (pte_t *)p_base; >> + >> + /*get the original pgprot value.*/ >> + prot_val = pmd_val(*pmd) & PTE_FLAGS_MASK; >> + prot_val &= ~PTE_TYPE_MASK; >> + prot_val |= PTE_TYPE_PAGE; >> + >> + pfn = pmd_pfn(*pmd); >> + for (i = 0; i < PTRS_PER_PTE; i++, pfn += 1) >> + set_pte(pte_base + i, pfn_pte(pfn, prot_val)); >> + >> + __pmd_populate(pmd, __pa(pte_base), PMD_TYPE_TABLE); >> + flush_tlb_kernel_range((addr & PMD_MASK), >> + ((addr + PMD_SIZE) & PMD_MASK)); >> +} >> + >> + >> +static void __split_pud_page_mapping(pud_t *pud, >> + unsigned long addr, >> + void *p_base) >> +{ >> + int i; >> + >> + pgprot_t old_prot; >> + >> + unsigned long pfn, pfn_inc; >> + >> + pmd_t *pmd_base = (pmd_t *)p_base; >> + >> + /*get the original pgprot value.*/ >> + old_prot = pud_val(*pud) & PTE_FLAGS_MASK; >> + >> + pfn = pud_pfn(*pud); >> + pfn_inc = PMD_SIZE >> PAGE_SHIFT; >> + for (i = 0; i < PTRS_PER_PMD; i++, pfn += pfn_inc) >> + set_pmd(pmd_base + i, pfn_pmd(pfn, old_prot)); >> + >> + pud_populate(&init_mm, pud, pmd_base); >> + flush_tlb_all(); > > Everywhere else you've switched to using flush_tlb_kernel_range, > why the switch here to flush_tlb_all? > Since here process PUD, a larger size. >> +} >> + >> +void kernel_map_pages(struct page *page, int numpages, int enable) >> +{ >> + unsigned long start_addr, end_addr, addr; >> + unsigned int level; >> + >> + pte_t *kpte; >> + pteval_t old_pval, new_pval; >> + >> + int i, counter = 0; >> + >> + /*no highmem in ARMv8. */ >> + addr = start_addr = (unsigned long)page_address(page); >> + end_addr = start_addr + (numpages << PAGE_SHIFT); >> + >> + for (i = 0; i < numpages; addr += PAGE_SIZE, i++) { >> + kpte = lookup_kaddress(addr, &level); >> + /* >> + * skip the memory holes. it is impossible if the input >> + * parameter is valid. >> + */ >> + if (unlikely(!kpte || pte_none(*kpte))) { >> + pr_err("Have no kernel linear mapping for 0x%0lx\n", addr); >> + break; >> + } >> + >> + if (level != PG_LEVEL_PAGE) { >> + pr_err("Page entry for 0x%0lx is not PAGE LEVEL(%d)\n", >> + addr, level); >> + break; >> + } >> + >> + old_pval = pte_val(*kpte); >> + new_pval = (enable) ? (old_pval | PTE_VALID) : >> + (old_pval & (~PTE_VALID)); >> + if (unlikely(new_pval == old_pval)) { >> + pr_warn("Page %s: same pte value at 0x%llx", >> + (enable) ? "alloc" : "free", old_pval); >> + continue; >> + } >> + >> + set_pte(kpte, __pte(new_pval)); >> + counter++; >> + } >> + >> + if (counter) >> + flush_tlb_kernel_range(start_addr, end_addr); >> +} > > We already have some of this infrastructure to set page attributes > in arch/arm64/mm/pageattr.c . We should be leveraging that for > kernel_map_pages. > There is no pageattr.c for ARMv8. In X86, it exists. Do you mean pmd_modify? >> + >> + >> +static int __init early_split_large_page_mapping(unsigned long virt, >> + phys_addr_t phys, >> + phys_addr_t size) >> +{ >> + pte_t *pte; >> + void *pte_base; >> + >> + unsigned long addr, end, next; >> + unsigned int pg_level; >> + unsigned long size_level; >> + unsigned long mask_level; >> + >> + /*make the addr aligned to PAGE*/ >> + addr = virt & PAGE_MASK; >> + end = addr + PAGE_ALIGN(size + (virt & ~PAGE_MASK)); >> + >> + for (; addr != end; phys += (next - addr), addr = next) { >> +repeat: >> + pte = lookup_kaddress(addr, &pg_level); >> + /* >> + * support the input memory range is a wider range. If we >> + * can not find valid page entry for some addresses, we do >> + * not know the page section size. But we only care the >> + * large page, just move forward in minimal large page size >> + * (PMD size) >> + */ >> + if (!pte || pte_none(*pte)) { >> + next = pmd_addr_end(addr, end); >> + continue; >> + } >> + >> + size_level = (_AC(1, UL) << >> + ((PAGE_SHIFT - 3) * pg_level + 3)); >> + mask_level = ~(size_level - 1); >> + >> + next = (addr + size_level) & mask_level; >> + if (next > end) >> + next = end; >> + >> + /*Does it need to split it?*/ >> + if (pg_level == PG_LEVEL_PAGE) >> + continue; >> + >> + /*start the splitting...*/ >> + if (pte_pfn(*pte) != PFN_DOWN(phys & mask_level)) { >> + pr_err("Physical addr 0x%0llx mis-match with virt 0x%0lx\n", >> + pte_pfn(*pte), addr & mask_level); >> + return -1; > > Return a real error code here and not just -1 > Yes. Maybe EFAULT is more better. >> + } >> + >> + pte_base = early_alloc(PAGE_SIZE); >> + >> + if (pg_level == PG_LEVEL_PUD) { >> + __split_pud_page_mapping((pud_t *)pte, addr, pte_base); >> + goto repeat; > > This looks like a less friendly use of goto. Any change we could turn this > into a real loop? > I just want to make __split_pud_page_mapping more concise and just for the split from pud to pmd. Best, -Zhichang >> + } >> + __split_pmd_page_mapping((pmd_t *)pte, addr, pte_base); >> + } >> + >> + return 0; >> +} >> + >> + >> +#endif >> static void __init map_mem(void) >> { >> struct memblock_region *reg; >> @@ -331,6 +564,26 @@ static void __init map_mem(void) >> create_mapping(start, __phys_to_virt(start), end - start); >> } >> >> +#ifdef CONFIG_DEBUG_PAGEALLOC >> + /* >> + * the biggest direct mapping is ready, then start the PTE building. >> + * Now,there are sufficient mapped pages to store the PTE tables. >> + * And more important, doing large page splitting here can dispose >> + * the page tables in contiguous memory area. >> + */ >> + for_each_memblock(memory, reg) { >> + phys_addr_t start = reg->base; >> + phys_addr_t end = start + reg->size; >> + >> + if (start >= end || PFN_UP(start) >= PFN_DOWN(end)) >> + break; >> + >> + if (early_split_large_page_mapping(__phys_to_virt(start), >> + start, end - start)) >> + panic("map_mem:Fail to split large page[0x%0llx,0x%0llx)\n", >> + start, end); >> + } >> +#endif >> /* Limit no longer required. */ >> memblock_set_current_limit(MEMBLOCK_ALLOC_ANYWHERE); >> } >> > > Thanks, > Laura >