x86: Retry to remove vmalloc/ioremap synchronzation
cover x86: Retry to remove vmalloc/ioremap synchronzation
From: Joerg Roedel <[email protected]>
Hi,
as discussed here are the updates to the recent patches and fixes to pre-allocate the vmalloc/ioremap second-level page-table pages on x86-64.
Patch one is a re-send of
commit 8bb9bf242d1f ("x86/mm/64: Do not sync vmalloc/ioremap mappings")with more explanations about what broke, what fixed it and why its now safe to apply it again.
Patch two updates the comment in preallocate_vmalloc_pages(), it is mostly the wording from Dave Hansen, so he really deserved the authorship of it. I just didn't want to commit/send it in his name. Feel free to change authorship of this patch to him.
Regards,
Joerg
Joerg Roedel (2):
x86/mm/64: Do not sync vmalloc/ioremap mappings x86/mm/64: Update comment in preallocate_vmalloc_pages()
arch/x86/include/asm/pgtable_64_types.h | 2 -- arch/x86/mm/init_64.c | 20 ++++++++++---------- 2 files changed, 10 insertions(+), 12 deletions(-)
Signature · 1 line
2.28.0
PATCH 1/2 x86/mm/64: Do not sync vmalloc/ioremap mappings
From: Joerg Roedel <[email protected]>
Remove the code to sync the vmalloc and ioremap ranges for x86-64. The page-table pages are all pre-allocated so that synchronization is no longer necessary.
This is a patch that already went into the kernel as:
commit 8bb9bf242d1f ("x86/mm/64: Do not sync vmalloc/ioremap mappings")But it had to be reverted later because it unveiled a bug from:
commit 6eb82f994026 ("x86/mm: Pre-allocate P4D/PUD pages for vmalloc area")The bug in that commit causes the P4D/PUD pages not to be correctly allocated, making the synchronization still necessary. That issue got fixed meanwhile upstream:
commit 995909a4e22b ("x86/mm/64: Do not dereference non-present PGD entries")With that fix it is safe again to remove the page-table synchronization for vmalloc/ioremap ranges on x86-64.
--- arch/x86/include/asm/pgtable_64_types.h | 2 -- arch/x86/mm/init_64.c | 5 ----- 2 files changed, 7 deletions(-)
arch/x86/include/asm/pgtable_64_types.h +0 -2
@@ -159,6 +159,4 @@ extern unsigned int ptrs_per_p4d;#define PGD_KERNEL_START ((PAGE_SIZE / 2) / sizeof(pgd_t)) -#define ARCH_PAGE_TABLE_SYNC_MASK (pgtable_l5_enabled() ? PGTBL_PGD_MODIFIED : PGTBL_P4D_MODIFIED) -#endif /* _ASM_X86_PGTABLE_64_DEFS_H */
arch/x86/mm/init_64.c +0 -5
@@ -217,11 +217,6 @@ static void sync_global_pgds(unsigned long start, unsigned long end)sync_global_pgds_l4 (start, end); } -void arch_sync_kernel_mappings (unsigned long start,unsigned long end) -{ -sync_global_pgds (start, end); -} -/* * NOTE: This function is marked __ref because it calls __init function * (alloc_bootmem_pages). It's safe to do it ONLY when after_bootmem == 0.
Signature · 1 line
2.28.0
[tip: x86/mm] x86/mm/64: Do not sync vmalloc/ioremap mappings
The following commit has been merged into the x86/mm branch of tip:
The following commit has been merged into the x86/mm branch of tip:
Commit-ID: 58a18fe95e83b8396605154db04d73b08063f31b Gitweb: https://git.kernel.org/tip/58a18fe95e83b8396605154db04d73b08063f31b Author: Joerg Roedel <[email protected]> AuthorDate: Fri, 14 Aug 2020 17:19:46 +02:00 Committer: Ingo Molnar <[email protected]> CommitterDate: Sat, 15 Aug 2020 13:56:16 +02:00
x86/mm/64: Do not sync vmalloc/ioremap mappings
Remove the code to sync the vmalloc and ioremap ranges for x86-64. The page-table pages are all pre-allocated so that synchronization is no longer necessary.
This is a patch that already went into the kernel as:
commit 8bb9bf242d1f ("x86/mm/64: Do not sync vmalloc/ioremap mappings")But it had to be reverted later because it unveiled a bug from:
commit 6eb82f994026 ("x86/mm: Pre-allocate P4D/PUD pages for vmalloc area")The bug in that commit causes the P4D/PUD pages not to be correctly allocated, making the synchronization still necessary. That issue got fixed meanwhile upstream:
commit 995909a4e22b ("x86/mm/64: Do not dereference non-present PGD entries")With that fix it is safe again to remove the page-table synchronization for vmalloc/ioremap ranges on x86-64.
--- arch/x86/include/asm/pgtable_64_types.h | 2 -- arch/x86/mm/init_64.c | 5 ----- 2 files changed, 7 deletions(-)
arch/x86/include/asm/pgtable_64_types.h +0 -2
@@ -159,6 +159,4 @@ extern unsigned int ptrs_per_p4d;#define PGD_KERNEL_START ((PAGE_SIZE / 2) / sizeof(pgd_t)) -#define ARCH_PAGE_TABLE_SYNC_MASK (pgtable_l5_enabled() ? PGTBL_PGD_MODIFIED : PGTBL_P4D_MODIFIED) -#endif /* _ASM_X86_PGTABLE_64_DEFS_H */
arch/x86/mm/init_64.c +0 -5
@@ -217,11 +217,6 @@ static void sync_global_pgds(unsigned long start, unsigned long end)sync_global_pgds_l4 (start, end); } -void arch_sync_kernel_mappings (unsigned long start,unsigned long end) -{ -sync_global_pgds (start, end); -} -/* * NOTE: This function is marked __ref because it calls __init function * (alloc_bootmem_pages). It's safe to do it ONLY when after_bootmem == 0.
PATCH 2/2 x86/mm/64: Update comment in preallocate_vmalloc_pages()
In reply to Joerg RoedelFrom: Joerg Roedel <[email protected]>
The comment explaining why 4-level systems only need to allocate on the P4D level caused some confustion. Update it to better explain why on 4-level systems the allocation on PUD level is necessary.
--- arch/x86/mm/init_64.c | 15 ++++++++++----- 1 file changed, 10 insertions(+), 5 deletions(-)
arch/x86/mm/init_64.c +10 -5
@@ -1252,14 +1252,19 @@ static void __init preallocate_vmalloc_pages(void)if (!p4d) goto failed; -/* -* With 5-level paging the P4D level is not folded. So the PGDs -* are now populated and there is no need to walk down to the -* PUD level. -*/ if (pgtable_l5_enabled ())continue ; +/* +* The goal here is to allocate all possibly required +* hardware page tables pointed to by the top hardware +* level. +* +* On 4-level systems, the p4d layer is folded away and +* the above code does no preallocation. Below, go down +* to the pud _software_ level to ensure the second +* hardware level is allocated on 4-level systems too. +*/ lvl ="pud" ; pud =pud_alloc (&init_mm, p4d, addr);if (!pud)
Signature · 1 line
2.28.0
[tip: x86/mm] x86/mm/64: Update comment in preallocate_vmalloc_pages()
The following commit has been merged into the x86/mm branch of tip:
The following commit has been merged into the x86/mm branch of tip:
Commit-ID: 7a27ef5e83089090f3a4073a9157c862ef00acfc Gitweb: https://git.kernel.org/tip/7a27ef5e83089090f3a4073a9157c862ef00acfc Author: Joerg Roedel <[email protected]> AuthorDate: Fri, 14 Aug 2020 17:19:47 +02:00 Committer: Ingo Molnar <[email protected]> CommitterDate: Sat, 15 Aug 2020 13:56:16 +02:00
x86/mm/64: Update comment in preallocate_vmalloc_pages()
The comment explaining why 4-level systems only need to allocate on the P4D level caused some confustion. Update it to better explain why on 4-level systems the allocation on PUD level is necessary.
--- arch/x86/mm/init_64.c | 15 ++++++++++----- 1 file changed, 10 insertions(+), 5 deletions(-)
arch/x86/mm/init_64.c +10 -5
@@ -1252,14 +1252,19 @@ static void __init preallocate_vmalloc_pages(void)if (!p4d) goto failed; -/* -* With 5-level paging the P4D level is not folded. So the PGDs -* are now populated and there is no need to walk down to the -* PUD level. -*/ if (pgtable_l5_enabled ())continue ; +/* +* The goal here is to allocate all possibly required +* hardware page tables pointed to by the top hardware +* level. +* +* On 4-level systems, the P4D layer is folded away and +* the above code does no preallocation. Below, go down +* to the pud _software_ level to ensure the second +* hardware level is allocated on 4-level systems too. +*/ lvl ="pud" ; pud =pud_alloc (&init_mm, p4d, addr);if (!pud)