From 5b65c4677a57a1d4414212f9995aa0e46a21ff80 Mon Sep 17 00:00:00 2001 From: "Kirill A. Shutemov" Date: Sat, 9 Sep 2017 00:56:03 +0300 Subject: [PATCH 1/6] mm, x86/mm: Fix performance regression in get_user_pages_fast() The 0-day test bot found a performance regression that was tracked down to switching x86 to the generic get_user_pages_fast() implementation: http://lkml.kernel.org/r/20170710024020.GA26389@yexl-desktop The regression was caused by the fact that we now use local_irq_save() + local_irq_restore() in get_user_pages_fast() to disable interrupts. In x86 implementation local_irq_disable() + local_irq_enable() was used. The fix is to make get_user_pages_fast() use local_irq_disable(), leaving local_irq_save() for __get_user_pages_fast() that can be called with interrupts disabled. Numbers for pinning a gigabyte of memory, one page a time, 20 repeats: Before: Average: 14.91 ms, stddev: 0.45 ms After: Average: 10.76 ms, stddev: 0.18 ms Signed-off-by: Kirill A. Shutemov Cc: Andrew Morton Cc: Huang Ying Cc: Jonathan Corbet Cc: Linus Torvalds Cc: Peter Zijlstra Cc: Thomas Gleixner Cc: Thorsten Leemhuis Cc: linux-mm@kvack.org Fixes: e585513b76f7 ("x86/mm/gup: Switch GUP to the generic get_user_page_fast() implementation") Link: http://lkml.kernel.org/r/20170908215603.9189-3-kirill.shutemov@linux.intel.com Signed-off-by: Ingo Molnar --- mm/gup.c | 97 +++++++++++++++++++++++++++++++++----------------------- 1 file changed, 58 insertions(+), 39 deletions(-) diff --git a/mm/gup.c b/mm/gup.c index 23f01c40c88f..4a789f1c6a27 100644 --- a/mm/gup.c +++ b/mm/gup.c @@ -1618,6 +1618,47 @@ static int gup_p4d_range(pgd_t pgd, unsigned long addr, unsigned long end, return 1; } +static void gup_pgd_range(unsigned long addr, unsigned long end, + int write, struct page **pages, int *nr) +{ + unsigned long next; + pgd_t *pgdp; + + pgdp = pgd_offset(current->mm, addr); + do { + pgd_t pgd = READ_ONCE(*pgdp); + + next = pgd_addr_end(addr, end); + if (pgd_none(pgd)) + return; + if (unlikely(pgd_huge(pgd))) { + if (!gup_huge_pgd(pgd, pgdp, addr, next, write, + pages, nr)) + return; + } else if (unlikely(is_hugepd(__hugepd(pgd_val(pgd))))) { + if (!gup_huge_pd(__hugepd(pgd_val(pgd)), addr, + PGDIR_SHIFT, next, write, pages, nr)) + return; + } else if (!gup_p4d_range(pgd, addr, next, write, pages, nr)) + return; + } while (pgdp++, addr = next, addr != end); +} + +#ifndef gup_fast_permitted +/* + * Check if it's allowed to use __get_user_pages_fast() for the range, or + * we need to fall back to the slow version: + */ +bool gup_fast_permitted(unsigned long start, int nr_pages, int write) +{ + unsigned long len, end; + + len = (unsigned long) nr_pages << PAGE_SHIFT; + end = start + len; + return end >= start; +} +#endif + /* * Like get_user_pages_fast() except it's IRQ-safe in that it won't fall back to * the regular GUP. It will only return non-negative values. @@ -1625,10 +1666,8 @@ static int gup_p4d_range(pgd_t pgd, unsigned long addr, unsigned long end, int __get_user_pages_fast(unsigned long start, int nr_pages, int write, struct page **pages) { - struct mm_struct *mm = current->mm; unsigned long addr, len, end; - unsigned long next, flags; - pgd_t *pgdp; + unsigned long flags; int nr = 0; start &= PAGE_MASK; @@ -1652,45 +1691,15 @@ int __get_user_pages_fast(unsigned long start, int nr_pages, int write, * block IPIs that come from THPs splitting. */ - local_irq_save(flags); - pgdp = pgd_offset(mm, addr); - do { - pgd_t pgd = READ_ONCE(*pgdp); - - next = pgd_addr_end(addr, end); - if (pgd_none(pgd)) - break; - if (unlikely(pgd_huge(pgd))) { - if (!gup_huge_pgd(pgd, pgdp, addr, next, write, - pages, &nr)) - break; - } else if (unlikely(is_hugepd(__hugepd(pgd_val(pgd))))) { - if (!gup_huge_pd(__hugepd(pgd_val(pgd)), addr, - PGDIR_SHIFT, next, write, pages, &nr)) - break; - } else if (!gup_p4d_range(pgd, addr, next, write, pages, &nr)) - break; - } while (pgdp++, addr = next, addr != end); - local_irq_restore(flags); + if (gup_fast_permitted(start, nr_pages, write)) { + local_irq_save(flags); + gup_pgd_range(addr, end, write, pages, &nr); + local_irq_restore(flags); + } return nr; } -#ifndef gup_fast_permitted -/* - * Check if it's allowed to use __get_user_pages_fast() for the range, or - * we need to fall back to the slow version: - */ -bool gup_fast_permitted(unsigned long start, int nr_pages, int write) -{ - unsigned long len, end; - - len = (unsigned long) nr_pages << PAGE_SHIFT; - end = start + len; - return end >= start; -} -#endif - /** * get_user_pages_fast() - pin user pages in memory * @start: starting user address @@ -1710,12 +1719,22 @@ bool gup_fast_permitted(unsigned long start, int nr_pages, int write) int get_user_pages_fast(unsigned long start, int nr_pages, int write, struct page **pages) { + unsigned long addr, len, end; int nr = 0, ret = 0; start &= PAGE_MASK; + addr = start; + len = (unsigned long) nr_pages << PAGE_SHIFT; + end = start + len; + + if (unlikely(!access_ok(write ? VERIFY_WRITE : VERIFY_READ, + (void __user *)start, len))) + return 0; if (gup_fast_permitted(start, nr_pages, write)) { - nr = __get_user_pages_fast(start, nr_pages, write, pages); + local_irq_disable(); + gup_pgd_range(addr, end, write, pages, &nr); + local_irq_enable(); ret = nr; } From 83e3c48729d9ebb7af5a31a504f3fd6aff0348c4 Mon Sep 17 00:00:00 2001 From: "Kirill A. Shutemov" Date: Fri, 29 Sep 2017 17:08:16 +0300 Subject: [PATCH 2/6] mm/sparsemem: Allocate mem_section at runtime for CONFIG_SPARSEMEM_EXTREME=y Size of the mem_section[] array depends on the size of the physical address space. In preparation for boot-time switching between paging modes on x86-64 we need to make the allocation of mem_section[] dynamic, because otherwise we waste a lot of RAM: with CONFIG_NODE_SHIFT=10, mem_section[] size is 32kB for 4-level paging and 2MB for 5-level paging mode. The patch allocates the array on the first call to sparse_memory_present_with_active_regions(). Signed-off-by: Kirill A. Shutemov Cc: Andrew Morton Cc: Andy Lutomirski Cc: Borislav Petkov Cc: Cyrill Gorcunov Cc: Linus Torvalds Cc: Peter Zijlstra Cc: Thomas Gleixner Cc: linux-mm@kvack.org Link: http://lkml.kernel.org/r/20170929140821.37654-2-kirill.shutemov@linux.intel.com Signed-off-by: Ingo Molnar --- include/linux/mmzone.h | 6 +++++- mm/page_alloc.c | 10 ++++++++++ mm/sparse.c | 17 +++++++++++------ 3 files changed, 26 insertions(+), 7 deletions(-) diff --git a/include/linux/mmzone.h b/include/linux/mmzone.h index c8f89417740b..e796edf1296f 100644 --- a/include/linux/mmzone.h +++ b/include/linux/mmzone.h @@ -1150,13 +1150,17 @@ struct mem_section { #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1) #ifdef CONFIG_SPARSEMEM_EXTREME -extern struct mem_section *mem_section[NR_SECTION_ROOTS]; +extern struct mem_section **mem_section; #else extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT]; #endif static inline struct mem_section *__nr_to_section(unsigned long nr) { +#ifdef CONFIG_SPARSEMEM_EXTREME + if (!mem_section) + return NULL; +#endif if (!mem_section[SECTION_NR_TO_ROOT(nr)]) return NULL; return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK]; diff --git a/mm/page_alloc.c b/mm/page_alloc.c index 77e4d3c5c57b..8dfd13f724d9 100644 --- a/mm/page_alloc.c +++ b/mm/page_alloc.c @@ -5646,6 +5646,16 @@ void __init sparse_memory_present_with_active_regions(int nid) unsigned long start_pfn, end_pfn; int i, this_nid; +#ifdef CONFIG_SPARSEMEM_EXTREME + if (!mem_section) { + unsigned long size, align; + + size = sizeof(struct mem_section) * NR_SECTION_ROOTS; + align = 1 << (INTERNODE_CACHE_SHIFT); + mem_section = memblock_virt_alloc(size, align); + } +#endif + for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) memory_present(this_nid, start_pfn, end_pfn); } diff --git a/mm/sparse.c b/mm/sparse.c index 83b3bf6461af..b00a97398795 100644 --- a/mm/sparse.c +++ b/mm/sparse.c @@ -22,8 +22,7 @@ * 1) mem_section - memory sections, mem_map's for valid memory */ #ifdef CONFIG_SPARSEMEM_EXTREME -struct mem_section *mem_section[NR_SECTION_ROOTS] - ____cacheline_internodealigned_in_smp; +struct mem_section **mem_section; #else struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT] ____cacheline_internodealigned_in_smp; @@ -100,7 +99,7 @@ static inline int sparse_index_init(unsigned long section_nr, int nid) int __section_nr(struct mem_section* ms) { unsigned long root_nr; - struct mem_section* root; + struct mem_section *root = NULL; for (root_nr = 0; root_nr < NR_SECTION_ROOTS; root_nr++) { root = __nr_to_section(root_nr * SECTIONS_PER_ROOT); @@ -111,7 +110,7 @@ int __section_nr(struct mem_section* ms) break; } - VM_BUG_ON(root_nr == NR_SECTION_ROOTS); + VM_BUG_ON(!root); return (root_nr * SECTIONS_PER_ROOT) + (ms - root); } @@ -329,11 +328,17 @@ again: static void __init check_usemap_section_nr(int nid, unsigned long *usemap) { unsigned long usemap_snr, pgdat_snr; - static unsigned long old_usemap_snr = NR_MEM_SECTIONS; - static unsigned long old_pgdat_snr = NR_MEM_SECTIONS; + static unsigned long old_usemap_snr; + static unsigned long old_pgdat_snr; struct pglist_data *pgdat = NODE_DATA(nid); int usemap_nid; + /* First call */ + if (!old_usemap_snr) { + old_usemap_snr = NR_MEM_SECTIONS; + old_pgdat_snr = NR_MEM_SECTIONS; + } + usemap_snr = pfn_to_section_nr(__pa(usemap) >> PAGE_SHIFT); pgdat_snr = pfn_to_section_nr(__pa(pgdat) >> PAGE_SHIFT); if (usemap_snr == pgdat_snr) From 12a8cc7fcf54a8575f094be1e99032ec38aa045c Mon Sep 17 00:00:00 2001 From: Andrey Ryabinin Date: Fri, 29 Sep 2017 17:08:18 +0300 Subject: [PATCH 3/6] x86/kasan: Use the same shadow offset for 4- and 5-level paging We are going to support boot-time switching between 4- and 5-level paging. For KASAN it means we cannot have different KASAN_SHADOW_OFFSET for different paging modes: the constant is passed to gcc to generate code and cannot be changed at runtime. This patch changes KASAN code to use 0xdffffc0000000000 as shadow offset for both 4- and 5-level paging. For 5-level paging it means that shadow memory region is not aligned to PGD boundary anymore and we have to handle unaligned parts of the region properly. In addition, we have to exclude paravirt code from KASAN instrumentation as we now use set_pgd() before KASAN is fully ready. [kirill.shutemov@linux.intel.com: clenaup, changelog message] Signed-off-by: Andrey Ryabinin Signed-off-by: Kirill A. Shutemov Cc: Andrew Morton Cc: Andy Lutomirski Cc: Borislav Petkov Cc: Cyrill Gorcunov Cc: Linus Torvalds Cc: Peter Zijlstra Cc: Thomas Gleixner Cc: linux-mm@kvack.org Link: http://lkml.kernel.org/r/20170929140821.37654-4-kirill.shutemov@linux.intel.com Signed-off-by: Ingo Molnar --- Documentation/x86/x86_64/mm.txt | 2 +- arch/x86/Kconfig | 1 - arch/x86/kernel/Makefile | 3 +- arch/x86/mm/kasan_init_64.c | 101 +++++++++++++++++++++++++------- 4 files changed, 83 insertions(+), 24 deletions(-) diff --git a/Documentation/x86/x86_64/mm.txt b/Documentation/x86/x86_64/mm.txt index b0798e281aa6..3448e675b462 100644 --- a/Documentation/x86/x86_64/mm.txt +++ b/Documentation/x86/x86_64/mm.txt @@ -34,7 +34,7 @@ ff92000000000000 - ffd1ffffffffffff (=54 bits) vmalloc/ioremap space ffd2000000000000 - ffd3ffffffffffff (=49 bits) hole ffd4000000000000 - ffd5ffffffffffff (=49 bits) virtual memory map (512TB) ... unused hole ... -ffd8000000000000 - fff7ffffffffffff (=53 bits) kasan shadow memory (8PB) +ffdf000000000000 - fffffc0000000000 (=53 bits) kasan shadow memory (8PB) ... unused hole ... ffffff0000000000 - ffffff7fffffffff (=39 bits) %esp fixup stacks ... unused hole ... diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig index 971feac13506..32779beb56e2 100644 --- a/arch/x86/Kconfig +++ b/arch/x86/Kconfig @@ -302,7 +302,6 @@ config ARCH_SUPPORTS_DEBUG_PAGEALLOC config KASAN_SHADOW_OFFSET hex depends on KASAN - default 0xdff8000000000000 if X86_5LEVEL default 0xdffffc0000000000 config HAVE_INTEL_TXT diff --git a/arch/x86/kernel/Makefile b/arch/x86/kernel/Makefile index fd0a7895b63f..a97a6b611531 100644 --- a/arch/x86/kernel/Makefile +++ b/arch/x86/kernel/Makefile @@ -24,7 +24,8 @@ endif KASAN_SANITIZE_head$(BITS).o := n KASAN_SANITIZE_dumpstack.o := n KASAN_SANITIZE_dumpstack_$(BITS).o := n -KASAN_SANITIZE_stacktrace.o := n +KASAN_SANITIZE_stacktrace.o := n +KASAN_SANITIZE_paravirt.o := n OBJECT_FILES_NON_STANDARD_head_$(BITS).o := y OBJECT_FILES_NON_STANDARD_relocate_kernel_$(BITS).o := y diff --git a/arch/x86/mm/kasan_init_64.c b/arch/x86/mm/kasan_init_64.c index bc84b73684b7..fe5760db7b19 100644 --- a/arch/x86/mm/kasan_init_64.c +++ b/arch/x86/mm/kasan_init_64.c @@ -15,6 +15,8 @@ extern struct range pfn_mapped[E820_MAX_ENTRIES]; +static p4d_t tmp_p4d_table[PTRS_PER_P4D] __initdata __aligned(PAGE_SIZE); + static int __init map_range(struct range *range) { unsigned long start; @@ -30,8 +32,10 @@ static void __init clear_pgds(unsigned long start, unsigned long end) { pgd_t *pgd; + /* See comment in kasan_init() */ + unsigned long pgd_end = end & PGDIR_MASK; - for (; start < end; start += PGDIR_SIZE) { + for (; start < pgd_end; start += PGDIR_SIZE) { pgd = pgd_offset_k(start); /* * With folded p4d, pgd_clear() is nop, use p4d_clear() @@ -42,29 +46,61 @@ static void __init clear_pgds(unsigned long start, else pgd_clear(pgd); } + + pgd = pgd_offset_k(start); + for (; start < end; start += P4D_SIZE) + p4d_clear(p4d_offset(pgd, start)); +} + +static inline p4d_t *early_p4d_offset(pgd_t *pgd, unsigned long addr) +{ + unsigned long p4d; + + if (!IS_ENABLED(CONFIG_X86_5LEVEL)) + return (p4d_t *)pgd; + + p4d = __pa_nodebug(pgd_val(*pgd)) & PTE_PFN_MASK; + p4d += __START_KERNEL_map - phys_base; + return (p4d_t *)p4d + p4d_index(addr); +} + +static void __init kasan_early_p4d_populate(pgd_t *pgd, + unsigned long addr, + unsigned long end) +{ + pgd_t pgd_entry; + p4d_t *p4d, p4d_entry; + unsigned long next; + + if (pgd_none(*pgd)) { + pgd_entry = __pgd(_KERNPG_TABLE | __pa_nodebug(kasan_zero_p4d)); + set_pgd(pgd, pgd_entry); + } + + p4d = early_p4d_offset(pgd, addr); + do { + next = p4d_addr_end(addr, end); + + if (!p4d_none(*p4d)) + continue; + + p4d_entry = __p4d(_KERNPG_TABLE | __pa_nodebug(kasan_zero_pud)); + set_p4d(p4d, p4d_entry); + } while (p4d++, addr = next, addr != end && p4d_none(*p4d)); } static void __init kasan_map_early_shadow(pgd_t *pgd) { - int i; - unsigned long start = KASAN_SHADOW_START; + /* See comment in kasan_init() */ + unsigned long addr = KASAN_SHADOW_START & PGDIR_MASK; unsigned long end = KASAN_SHADOW_END; + unsigned long next; - for (i = pgd_index(start); start < end; i++) { - switch (CONFIG_PGTABLE_LEVELS) { - case 4: - pgd[i] = __pgd(__pa_nodebug(kasan_zero_pud) | - _KERNPG_TABLE); - break; - case 5: - pgd[i] = __pgd(__pa_nodebug(kasan_zero_p4d) | - _KERNPG_TABLE); - break; - default: - BUILD_BUG(); - } - start += PGDIR_SIZE; - } + pgd += pgd_index(addr); + do { + next = pgd_addr_end(addr, end); + kasan_early_p4d_populate(pgd, addr, next); + } while (pgd++, addr = next, addr != end); } #ifdef CONFIG_KASAN_INLINE @@ -101,7 +137,7 @@ void __init kasan_early_init(void) for (i = 0; i < PTRS_PER_PUD; i++) kasan_zero_pud[i] = __pud(pud_val); - for (i = 0; CONFIG_PGTABLE_LEVELS >= 5 && i < PTRS_PER_P4D; i++) + for (i = 0; IS_ENABLED(CONFIG_X86_5LEVEL) && i < PTRS_PER_P4D; i++) kasan_zero_p4d[i] = __p4d(p4d_val); kasan_map_early_shadow(early_top_pgt); @@ -117,12 +153,35 @@ void __init kasan_init(void) #endif memcpy(early_top_pgt, init_top_pgt, sizeof(early_top_pgt)); + + /* + * We use the same shadow offset for 4- and 5-level paging to + * facilitate boot-time switching between paging modes. + * As result in 5-level paging mode KASAN_SHADOW_START and + * KASAN_SHADOW_END are not aligned to PGD boundary. + * + * KASAN_SHADOW_START doesn't share PGD with anything else. + * We claim whole PGD entry to make things easier. + * + * KASAN_SHADOW_END lands in the last PGD entry and it collides with + * bunch of things like kernel code, modules, EFI mapping, etc. + * We need to take extra steps to not overwrite them. + */ + if (IS_ENABLED(CONFIG_X86_5LEVEL)) { + void *ptr; + + ptr = (void *)pgd_page_vaddr(*pgd_offset_k(KASAN_SHADOW_END)); + memcpy(tmp_p4d_table, (void *)ptr, sizeof(tmp_p4d_table)); + set_pgd(&early_top_pgt[pgd_index(KASAN_SHADOW_END)], + __pgd(__pa(tmp_p4d_table) | _KERNPG_TABLE)); + } + load_cr3(early_top_pgt); __flush_tlb_all(); - clear_pgds(KASAN_SHADOW_START, KASAN_SHADOW_END); + clear_pgds(KASAN_SHADOW_START & PGDIR_MASK, KASAN_SHADOW_END); - kasan_populate_zero_shadow((void *)KASAN_SHADOW_START, + kasan_populate_zero_shadow((void *)(KASAN_SHADOW_START & PGDIR_MASK), kasan_mem_to_shadow((void *)PAGE_OFFSET)); for (i = 0; i < E820_MAX_ENTRIES; i++) { From 4375c29985f155d7eb2346615d84e62d1b673682 Mon Sep 17 00:00:00 2001 From: "Kirill A. Shutemov" Date: Fri, 29 Sep 2017 17:08:19 +0300 Subject: [PATCH 4/6] x86/xen: Provide pre-built page tables only for CONFIG_XEN_PV=y and CONFIG_XEN_PVH=y Looks like we only need pre-built page tables in the CONFIG_XEN_PV=y and CONFIG_XEN_PVH=y cases. Let's not provide them for other configurations. Signed-off-by: Kirill A. Shutemov Reviewed-by: Juergen Gross Cc: Andrew Morton Cc: Andy Lutomirski Cc: Borislav Petkov Cc: Cyrill Gorcunov Cc: Linus Torvalds Cc: Peter Zijlstra Cc: Thomas Gleixner Cc: linux-mm@kvack.org Link: http://lkml.kernel.org/r/20170929140821.37654-5-kirill.shutemov@linux.intel.com Signed-off-by: Ingo Molnar --- arch/x86/kernel/head_64.S | 11 ++++++----- 1 file changed, 6 insertions(+), 5 deletions(-) diff --git a/arch/x86/kernel/head_64.S b/arch/x86/kernel/head_64.S index 513cbb012ecc..2be7d1e7fcf1 100644 --- a/arch/x86/kernel/head_64.S +++ b/arch/x86/kernel/head_64.S @@ -37,11 +37,12 @@ * */ -#define p4d_index(x) (((x) >> P4D_SHIFT) & (PTRS_PER_P4D-1)) #define pud_index(x) (((x) >> PUD_SHIFT) & (PTRS_PER_PUD-1)) +#if defined(CONFIG_XEN_PV) || defined(CONFIG_XEN_PVH) PGD_PAGE_OFFSET = pgd_index(__PAGE_OFFSET_BASE) PGD_START_KERNEL = pgd_index(__START_KERNEL_map) +#endif L3_START_KERNEL = pud_index(__START_KERNEL_map) .text @@ -361,10 +362,7 @@ NEXT_PAGE(early_dynamic_pgts) .data -#ifndef CONFIG_XEN -NEXT_PAGE(init_top_pgt) - .fill 512,8,0 -#else +#if defined(CONFIG_XEN_PV) || defined(CONFIG_XEN_PVH) NEXT_PAGE(init_top_pgt) .quad level3_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE_NOENC .org init_top_pgt + PGD_PAGE_OFFSET*8, 0 @@ -381,6 +379,9 @@ NEXT_PAGE(level2_ident_pgt) * Don't set NX because code runs from these pages. */ PMDS(0, __PAGE_KERNEL_IDENT_LARGE_EXEC, PTRS_PER_PMD) +#else +NEXT_PAGE(init_top_pgt) + .fill 512,8,0 #endif #ifdef CONFIG_X86_5LEVEL From 773dd2fca581b0a80e5a33332cc8ee67e5a79cba Mon Sep 17 00:00:00 2001 From: "Kirill A. Shutemov" Date: Fri, 29 Sep 2017 17:08:20 +0300 Subject: [PATCH 5/6] x86/xen: Drop 5-level paging support code from the XEN_PV code It was decided 5-level paging is not going to be supported in XEN_PV. Let's drop the dead code from the XEN_PV code. Tested-by: Juergen Gross Signed-off-by: Kirill A. Shutemov Reviewed-by: Juergen Gross Cc: Andrew Morton Cc: Andy Lutomirski Cc: Borislav Petkov Cc: Cyrill Gorcunov Cc: Linus Torvalds Cc: Peter Zijlstra Cc: Thomas Gleixner Cc: linux-mm@kvack.org Link: http://lkml.kernel.org/r/20170929140821.37654-6-kirill.shutemov@linux.intel.com Signed-off-by: Ingo Molnar --- arch/x86/xen/mmu_pv.c | 159 ++++++++++++++++-------------------------- 1 file changed, 60 insertions(+), 99 deletions(-) diff --git a/arch/x86/xen/mmu_pv.c b/arch/x86/xen/mmu_pv.c index 71495f1a86d7..2ccdaba31a07 100644 --- a/arch/x86/xen/mmu_pv.c +++ b/arch/x86/xen/mmu_pv.c @@ -449,7 +449,7 @@ __visible pmd_t xen_make_pmd(pmdval_t pmd) } PV_CALLEE_SAVE_REGS_THUNK(xen_make_pmd); -#if CONFIG_PGTABLE_LEVELS == 4 +#ifdef CONFIG_X86_64 __visible pudval_t xen_pud_val(pud_t pud) { return pte_mfn_to_pfn(pud.pud); @@ -538,7 +538,7 @@ static void xen_set_p4d(p4d_t *ptr, p4d_t val) xen_mc_issue(PARAVIRT_LAZY_MMU); } -#endif /* CONFIG_PGTABLE_LEVELS == 4 */ +#endif /* CONFIG_X86_64 */ static int xen_pmd_walk(struct mm_struct *mm, pmd_t *pmd, int (*func)(struct mm_struct *mm, struct page *, enum pt_level), @@ -580,21 +580,17 @@ static int xen_p4d_walk(struct mm_struct *mm, p4d_t *p4d, int (*func)(struct mm_struct *mm, struct page *, enum pt_level), bool last, unsigned long limit) { - int i, nr, flush = 0; + int flush = 0; + pud_t *pud; - nr = last ? p4d_index(limit) + 1 : PTRS_PER_P4D; - for (i = 0; i < nr; i++) { - pud_t *pud; - if (p4d_none(p4d[i])) - continue; + if (p4d_none(*p4d)) + return flush; - pud = pud_offset(&p4d[i], 0); - if (PTRS_PER_PUD > 1) - flush |= (*func)(mm, virt_to_page(pud), PT_PUD); - flush |= xen_pud_walk(mm, pud, func, - last && i == nr - 1, limit); - } + pud = pud_offset(p4d, 0); + if (PTRS_PER_PUD > 1) + flush |= (*func)(mm, virt_to_page(pud), PT_PUD); + flush |= xen_pud_walk(mm, pud, func, last, limit); return flush; } @@ -644,8 +640,6 @@ static int __xen_pgd_walk(struct mm_struct *mm, pgd_t *pgd, continue; p4d = p4d_offset(&pgd[i], 0); - if (PTRS_PER_P4D > 1) - flush |= (*func)(mm, virt_to_page(p4d), PT_P4D); flush |= xen_p4d_walk(mm, p4d, func, i == nr - 1, limit); } @@ -1176,22 +1170,14 @@ static void __init xen_cleanmfnmap(unsigned long vaddr) { pgd_t *pgd; p4d_t *p4d; - unsigned int i; bool unpin; unpin = (vaddr == 2 * PGDIR_SIZE); vaddr &= PMD_MASK; pgd = pgd_offset_k(vaddr); p4d = p4d_offset(pgd, 0); - for (i = 0; i < PTRS_PER_P4D; i++) { - if (p4d_none(p4d[i])) - continue; - xen_cleanmfnmap_p4d(p4d + i, unpin); - } - if (IS_ENABLED(CONFIG_X86_5LEVEL)) { - set_pgd(pgd, __pgd(0)); - xen_cleanmfnmap_free_pgtbl(p4d, unpin); - } + if (!p4d_none(*p4d)) + xen_cleanmfnmap_p4d(p4d, unpin); } static void __init xen_pagetable_p2m_free(void) @@ -1692,7 +1678,7 @@ static void xen_release_pmd(unsigned long pfn) xen_release_ptpage(pfn, PT_PMD); } -#if CONFIG_PGTABLE_LEVELS >= 4 +#ifdef CONFIG_X86_64 static void xen_alloc_pud(struct mm_struct *mm, unsigned long pfn) { xen_alloc_ptpage(mm, pfn, PT_PUD); @@ -2029,13 +2015,12 @@ static phys_addr_t __init xen_early_virt_to_phys(unsigned long vaddr) */ void __init xen_relocate_p2m(void) { - phys_addr_t size, new_area, pt_phys, pmd_phys, pud_phys, p4d_phys; + phys_addr_t size, new_area, pt_phys, pmd_phys, pud_phys; unsigned long p2m_pfn, p2m_pfn_end, n_frames, pfn, pfn_end; - int n_pte, n_pt, n_pmd, n_pud, n_p4d, idx_pte, idx_pt, idx_pmd, idx_pud, idx_p4d; + int n_pte, n_pt, n_pmd, n_pud, idx_pte, idx_pt, idx_pmd, idx_pud; pte_t *pt; pmd_t *pmd; pud_t *pud; - p4d_t *p4d = NULL; pgd_t *pgd; unsigned long *new_p2m; int save_pud; @@ -2045,11 +2030,7 @@ void __init xen_relocate_p2m(void) n_pt = roundup(size, PMD_SIZE) >> PMD_SHIFT; n_pmd = roundup(size, PUD_SIZE) >> PUD_SHIFT; n_pud = roundup(size, P4D_SIZE) >> P4D_SHIFT; - if (PTRS_PER_P4D > 1) - n_p4d = roundup(size, PGDIR_SIZE) >> PGDIR_SHIFT; - else - n_p4d = 0; - n_frames = n_pte + n_pt + n_pmd + n_pud + n_p4d; + n_frames = n_pte + n_pt + n_pmd + n_pud; new_area = xen_find_free_area(PFN_PHYS(n_frames)); if (!new_area) { @@ -2065,76 +2046,56 @@ void __init xen_relocate_p2m(void) * To avoid any possible virtual address collision, just use * 2 * PUD_SIZE for the new area. */ - p4d_phys = new_area; - pud_phys = p4d_phys + PFN_PHYS(n_p4d); + pud_phys = new_area; pmd_phys = pud_phys + PFN_PHYS(n_pud); pt_phys = pmd_phys + PFN_PHYS(n_pmd); p2m_pfn = PFN_DOWN(pt_phys) + n_pt; pgd = __va(read_cr3_pa()); new_p2m = (unsigned long *)(2 * PGDIR_SIZE); - idx_p4d = 0; save_pud = n_pud; - do { - if (n_p4d > 0) { - p4d = early_memremap(p4d_phys, PAGE_SIZE); - clear_page(p4d); - n_pud = min(save_pud, PTRS_PER_P4D); - } - for (idx_pud = 0; idx_pud < n_pud; idx_pud++) { - pud = early_memremap(pud_phys, PAGE_SIZE); - clear_page(pud); - for (idx_pmd = 0; idx_pmd < min(n_pmd, PTRS_PER_PUD); - idx_pmd++) { - pmd = early_memremap(pmd_phys, PAGE_SIZE); - clear_page(pmd); - for (idx_pt = 0; idx_pt < min(n_pt, PTRS_PER_PMD); - idx_pt++) { - pt = early_memremap(pt_phys, PAGE_SIZE); - clear_page(pt); - for (idx_pte = 0; - idx_pte < min(n_pte, PTRS_PER_PTE); - idx_pte++) { - set_pte(pt + idx_pte, - pfn_pte(p2m_pfn, PAGE_KERNEL)); - p2m_pfn++; - } - n_pte -= PTRS_PER_PTE; - early_memunmap(pt, PAGE_SIZE); - make_lowmem_page_readonly(__va(pt_phys)); - pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, - PFN_DOWN(pt_phys)); - set_pmd(pmd + idx_pt, - __pmd(_PAGE_TABLE | pt_phys)); - pt_phys += PAGE_SIZE; + for (idx_pud = 0; idx_pud < n_pud; idx_pud++) { + pud = early_memremap(pud_phys, PAGE_SIZE); + clear_page(pud); + for (idx_pmd = 0; idx_pmd < min(n_pmd, PTRS_PER_PUD); + idx_pmd++) { + pmd = early_memremap(pmd_phys, PAGE_SIZE); + clear_page(pmd); + for (idx_pt = 0; idx_pt < min(n_pt, PTRS_PER_PMD); + idx_pt++) { + pt = early_memremap(pt_phys, PAGE_SIZE); + clear_page(pt); + for (idx_pte = 0; + idx_pte < min(n_pte, PTRS_PER_PTE); + idx_pte++) { + set_pte(pt + idx_pte, + pfn_pte(p2m_pfn, PAGE_KERNEL)); + p2m_pfn++; } - n_pt -= PTRS_PER_PMD; - early_memunmap(pmd, PAGE_SIZE); - make_lowmem_page_readonly(__va(pmd_phys)); - pin_pagetable_pfn(MMUEXT_PIN_L2_TABLE, - PFN_DOWN(pmd_phys)); - set_pud(pud + idx_pmd, __pud(_PAGE_TABLE | pmd_phys)); - pmd_phys += PAGE_SIZE; + n_pte -= PTRS_PER_PTE; + early_memunmap(pt, PAGE_SIZE); + make_lowmem_page_readonly(__va(pt_phys)); + pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, + PFN_DOWN(pt_phys)); + set_pmd(pmd + idx_pt, + __pmd(_PAGE_TABLE | pt_phys)); + pt_phys += PAGE_SIZE; } - n_pmd -= PTRS_PER_PUD; - early_memunmap(pud, PAGE_SIZE); - make_lowmem_page_readonly(__va(pud_phys)); - pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, PFN_DOWN(pud_phys)); - if (n_p4d > 0) - set_p4d(p4d + idx_pud, __p4d(_PAGE_TABLE | pud_phys)); - else - set_pgd(pgd + 2 + idx_pud, __pgd(_PAGE_TABLE | pud_phys)); - pud_phys += PAGE_SIZE; + n_pt -= PTRS_PER_PMD; + early_memunmap(pmd, PAGE_SIZE); + make_lowmem_page_readonly(__va(pmd_phys)); + pin_pagetable_pfn(MMUEXT_PIN_L2_TABLE, + PFN_DOWN(pmd_phys)); + set_pud(pud + idx_pmd, __pud(_PAGE_TABLE | pmd_phys)); + pmd_phys += PAGE_SIZE; } - if (n_p4d > 0) { - save_pud -= PTRS_PER_P4D; - early_memunmap(p4d, PAGE_SIZE); - make_lowmem_page_readonly(__va(p4d_phys)); - pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE, PFN_DOWN(p4d_phys)); - set_pgd(pgd + 2 + idx_p4d, __pgd(_PAGE_TABLE | p4d_phys)); - p4d_phys += PAGE_SIZE; - } - } while (++idx_p4d < n_p4d); + n_pmd -= PTRS_PER_PUD; + early_memunmap(pud, PAGE_SIZE); + make_lowmem_page_readonly(__va(pud_phys)); + pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, PFN_DOWN(pud_phys)); + set_pgd(pgd + 2 + idx_pud, __pgd(_PAGE_TABLE | pud_phys)); + pud_phys += PAGE_SIZE; + } /* Now copy the old p2m info to the new area. */ memcpy(new_p2m, xen_p2m_addr, size); @@ -2361,7 +2322,7 @@ static void __init xen_post_allocator_init(void) pv_mmu_ops.set_pte = xen_set_pte; pv_mmu_ops.set_pmd = xen_set_pmd; pv_mmu_ops.set_pud = xen_set_pud; -#if CONFIG_PGTABLE_LEVELS >= 4 +#ifdef CONFIG_X86_64 pv_mmu_ops.set_p4d = xen_set_p4d; #endif @@ -2371,7 +2332,7 @@ static void __init xen_post_allocator_init(void) pv_mmu_ops.alloc_pmd = xen_alloc_pmd; pv_mmu_ops.release_pte = xen_release_pte; pv_mmu_ops.release_pmd = xen_release_pmd; -#if CONFIG_PGTABLE_LEVELS >= 4 +#ifdef CONFIG_X86_64 pv_mmu_ops.alloc_pud = xen_alloc_pud; pv_mmu_ops.release_pud = xen_release_pud; #endif @@ -2435,14 +2396,14 @@ static const struct pv_mmu_ops xen_mmu_ops __initconst = { .make_pmd = PV_CALLEE_SAVE(xen_make_pmd), .pmd_val = PV_CALLEE_SAVE(xen_pmd_val), -#if CONFIG_PGTABLE_LEVELS >= 4 +#ifdef CONFIG_X86_64 .pud_val = PV_CALLEE_SAVE(xen_pud_val), .make_pud = PV_CALLEE_SAVE(xen_make_pud), .set_p4d = xen_set_p4d_hyper, .alloc_pud = xen_alloc_pmd_init, .release_pud = xen_release_pmd_init, -#endif /* CONFIG_PGTABLE_LEVELS == 4 */ +#endif /* CONFIG_X86_64 */ .activate_mm = xen_activate_mm, .dup_mmap = xen_dup_mmap, From 15670bfe19905b1dcbb63137f40d718b59d84479 Mon Sep 17 00:00:00 2001 From: Baoquan He Date: Sat, 28 Oct 2017 09:30:38 +0800 Subject: [PATCH 6/6] x86/mm/64: Rename the register_page_bootmem_memmap() 'size' parameter to 'nr_pages' register_page_bootmem_memmap()'s 3rd 'size' parameter is named in a somewhat misleading fashion - rename it to 'nr_pages' which makes the units of it much clearer. Meanwhile rename the existing local variable 'nr_pages' to 'nr_pmd_pages', a more expressive name, to avoid conflict with new function parameter 'nr_pages'. (Also clean up the unnecessary parentheses in which get_order() is called.) Signed-off-by: Baoquan He Acked-by: Thomas Gleixner Cc: Linus Torvalds Cc: Peter Zijlstra Cc: akpm@linux-foundation.org Link: http://lkml.kernel.org/r/1509154238-23250-1-git-send-email-bhe@redhat.com Signed-off-by: Ingo Molnar --- arch/x86/mm/init_64.c | 10 +++++----- include/linux/mm.h | 2 +- 2 files changed, 6 insertions(+), 6 deletions(-) diff --git a/arch/x86/mm/init_64.c b/arch/x86/mm/init_64.c index 048fbe8fc274..adcea90a2046 100644 --- a/arch/x86/mm/init_64.c +++ b/arch/x86/mm/init_64.c @@ -1426,16 +1426,16 @@ int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node) #if defined(CONFIG_MEMORY_HOTPLUG_SPARSE) && defined(CONFIG_HAVE_BOOTMEM_INFO_NODE) void register_page_bootmem_memmap(unsigned long section_nr, - struct page *start_page, unsigned long size) + struct page *start_page, unsigned long nr_pages) { unsigned long addr = (unsigned long)start_page; - unsigned long end = (unsigned long)(start_page + size); + unsigned long end = (unsigned long)(start_page + nr_pages); unsigned long next; pgd_t *pgd; p4d_t *p4d; pud_t *pud; pmd_t *pmd; - unsigned int nr_pages; + unsigned int nr_pmd_pages; struct page *page; for (; addr < end; addr = next) { @@ -1482,9 +1482,9 @@ void register_page_bootmem_memmap(unsigned long section_nr, if (pmd_none(*pmd)) continue; - nr_pages = 1 << (get_order(PMD_SIZE)); + nr_pmd_pages = 1 << get_order(PMD_SIZE); page = pmd_page(*pmd); - while (nr_pages--) + while (nr_pmd_pages--) get_page_bootmem(section_nr, page++, SECTION_INFO); } diff --git a/include/linux/mm.h b/include/linux/mm.h index 065d99deb847..b2c7045e9604 100644 --- a/include/linux/mm.h +++ b/include/linux/mm.h @@ -2495,7 +2495,7 @@ void vmemmap_populate_print_last(void); void vmemmap_free(unsigned long start, unsigned long end); #endif void register_page_bootmem_memmap(unsigned long section_nr, struct page *map, - unsigned long size); + unsigned long nr_pages); enum mf_flags { MF_COUNT_INCREASED = 1 << 0,