add mTHP support for anonymous shmem
Subsystems (6)
Baolin Wang
reviewerv5 cover add mTHP support for anonymous shmem
Anonymous pages have already been supported for multi-size (mTHP) allocation through commit 19eaf44954df, that can allow THP to be configured through the sysfs interface located at '/sys/kernel/mm/transparent_hugepage/hugepage-XXkb/enabled'.
However, the anonymous shmem will ignore the anonymous mTHP rule configured through the sysfs interface, and can only use the PMD-mapped THP, that is not reasonable. Many implement anonymous page sharing through mmap(MAP_SHARED | MAP_ANONYMOUS), especially in database usage scenarios, therefore, users expect to apply an unified mTHP strategy for anonymous pages, also including the anonymous shared pages, in order to enjoy the benefits of mTHP. For example, lower latency than PMD-mapped THP, smaller memory bloat than PMD-mapped THP, contiguous PTEs on ARM architecture to reduce TLB miss etc.
As discussed in the bi-weekly MM meeting[1], the mTHP controls should control all of shmem, not only anonymous shmem, but support will be added iteratively. Therefore, this patch set starts with support for anonymous shmem.
The primary strategy is similar to supporting anonymous mTHP. Introduce a new interface '/mm/transparent_hugepage/hugepage-XXkb/shmem_enabled', which can have almost the same values as the top-level '/sys/kernel/mm/transparent_hugepage/shmem_enabled', with adding a new additional "inherit" option and dropping the testing options 'force' and 'deny'. By default all sizes will be set to "never" except PMD size, which is set to "inherit". This ensures backward compatibility with the anonymous shmem enabled of the top level, meanwhile also allows independent control of anonymous shmem enabled for each mTHP.
Use the page fault latency tool to measure the performance of 1G anonymous shmem with 32 threads on my machine environment with: ARM64 Architecture, 32 cores, 125G memory:
base: mm-unstable
user-time sys_time faults_per_sec_per_cpu faults_per_sec
0.04s 3.10s 83516.416 2669684.890
mm-unstable + patchset, anon shmem mTHP disabled
user-time sys_time faults_per_sec_per_cpu faults_per_sec
0.02s 3.14s 82936.359 2630746.027
mm-unstable + patchset, anon shmem 64K mTHP enabled
user-time sys_time faults_per_sec_per_cpu faults_per_sec
0.08s 0.31s 678630.231 17082522.495
From the data above, it is observed that the patchset has a minimal impact when mTHP is not enabled (some fluctuations observed during testing). When enabling 64K mTHP, there is a significant improvement of the page fault latency.
[1] https://lore.kernel.org/all/[email protected]/
Changes from v4:
- Fix the unused variable warning reported by kernel test robot.
- Drop the 'anon' prefix for variables and functions, per Daniel.
Changes from v3:
- Drop 'force' and 'deny' testing options for each mTHP.
- Use new helper update_mmu_tlb_range(), per Lance.
- Update documentation to drop "anonymous thp" terminology, per David.
- Initialize the 'suitable_orders' in shmem_alloc_and_add_folio(), reported by kernel test robot.
- Fix the highest mTHP order in shmem_get_unmapped_area().
- Update some commit message.
Changes from v2:
- Rebased to mm/mm-unstable.
- Remove 'huge' parameter for shmem_alloc_and_add_folio(), per Lance.
Changes from v1:
- Drop the patch that re-arranges the position of highest_order() and next_order(), per Ryan.
- Modify the finish_fault() to fix VA alignment issue, per Ryan and David.
- Fix some building issues, reported by Lance and kernel test robot.
- Update some commit message.
Changes from RFC:
- Rebase the patch set against the new mm-unstable branch, per Lance.
- Add a new patch to export highest_order() and next_order().
- Add a new patch to align mTHP size in shmem_get_unmapped_area().
- Handle the uffd case and the VMA limits case when building mapping for large folio in the finish_fault() function, per Ryan.
- Remove unnecessary 'order' variable in patch 3, per Kefeng.
- Keep the anon shmem counters' name consistency.
- Modify the strategy to support mTHP for anonymous shmem, discussed with Ryan and David.
- Add reviewed tag from Barry.
- Update the commit message.
Baolin Wang (6):
mm: memory: extend finish_fault() to support large folio
mm: shmem: add THP validation for PMD-mapped THP related statistics
mm: shmem: add multi-size THP sysfs interface for anonymous shmem
mm: shmem: add mTHP support for anonymous shmem
mm: shmem: add mTHP size alignment in shmem_get_unmapped_area
mm: shmem: add mTHP counters for anonymous shmem
Documentation/admin-guide/mm/transhuge.rst | 23 ++ include/linux/huge_mm.h | 23 ++ mm/huge_memory.c | 17 +- mm/memory.c | 57 +++- mm/shmem.c | 344 ++++++++++++++++++--- 5 files changed, 403 insertions(+), 61 deletions(-)
Signature · 1 line
2.39.3
Baolin Wang
reviewerPATCH v5 1/6 mm: memory: extend finish_fault() to support large folio
Add large folio mapping establishment support for finish_fault() as a preparation, to support multi-size THP allocation of anonymous shmem pages in the following patches.
Keep the same behavior (per-page fault) for non-anon shmem to avoid inflating the RSS unintentionally, and we can discuss what size of mapping to build when extending mTHP to control non-anon shmem in the future.
--- mm/memory.c | 57 +++++++++++++++++++++++++++++++++++++++++++---------- 1 file changed, 47 insertions(+), 10 deletions(-)
mm/memory.c +47 -10
@@ -4831,9 +4831,12 @@ vm_fault_t finish_fault(struct vm_fault *vmf) {struct vm_area_struct *vma = vmf->vma;struct page *page; +struct folio *folio;vm_fault_t ret;bool is_cow = (vmf->flags &FAULT_FLAG_WRITE ) && !(vma->vm_flags &VM_SHARED ); +int type, nr_pages; +unsigned long addr = vmf->address;/* Did we COW the page? */ if (is_cow)
@@ -4864,24 +4867,58 @@ vm_fault_t finish_fault(struct vm_fault *vmf)return VM_FAULT_OOM ; } + folio =page_folio (page); + nr_pages =folio_nr_pages (folio); + +/* +* Using per-page fault to maintain the uffd semantics, and same +* approach also applies to non-anonymous-shmem faults to avoid +* inflating the RSS of the process. +*/ +if (!vma_is_anon_shmem (vma) ||unlikely (userfaultfd_armed (vma))) { + nr_pages =1 ; + }else if (nr_pages >1 ) { +pgoff_t idx =folio_page_idx (folio, page); +/* The page offset of vmf->address within the VMA. */ +pgoff_t vma_off = vmf->pgoff - vmf->vma->vm_pgoff; + +/* +* Fallback to per-page fault in case the folio size in page +* cache beyond the VMA limits. +*/ +if (unlikely (vma_off < idx || + vma_off + (nr_pages - idx) >vma_pages (vma))) { + nr_pages =1 ; + }else { +/* Now we can set mappings for the whole large folio. */ + addr = vmf->address - idx *PAGE_SIZE ; + page = &folio->page; + } + } + vmf->pte =pte_offset_map_lock (vma->vm_mm, vmf->pmd, - vmf->address, &vmf->ptl); + addr, &vmf->ptl);if (!vmf->pte)return VM_FAULT_NOPAGE ;/* Re-check under ptl */ -if (likely (!vmf_pte_changed (vmf))) { -struct folio *folio =page_folio (page); -int type = is_cow ?MM_ANONPAGES :mm_counter_file (folio); - -set_pte_range (vmf, folio, page,1 , vmf->address); -add_mm_counter (vma->vm_mm, type,1 ); - ret =0 ; - }else { -update_mmu_tlb (vma, vmf->address, vmf->pte); +if (nr_pages ==1 &&unlikely (vmf_pte_changed (vmf))) { +update_mmu_tlb (vma, addr, vmf->pte); + ret =VM_FAULT_NOPAGE ; + goto unlock; + }else if (nr_pages >1 && !pte_range_none (vmf->pte, nr_pages)) { +update_mmu_tlb_range (vma, addr, vmf->pte, nr_pages); ret =VM_FAULT_NOPAGE ; + goto unlock; } +folio_ref_add (folio, nr_pages -1 ); +set_pte_range (vmf, folio, page, nr_pages, addr); + type = is_cow ?MM_ANONPAGES :mm_counter_file (folio); +add_mm_counter (vma->vm_mm, type, nr_pages); + ret =0 ; + +unlock:pte_unmap_unlock (vmf->pte, vmf->ptl);return ret; }
Signature · 1 line
2.39.3
Zi Yan
reviewerRe: [PATCH v5 1/6] mm: memory: extend finish_fault() to support large folio
Quoted 4 lines · Baolin Wang
Add large folio mapping establishment support for
finish_fault()as a preparation, to support multi-size THP allocation of anonymous shmem pages in the following patches.
Quoted 9 lines
Keep the same behavior (per-page fault) for non-anon shmem to avoid inflating the RSS unintentionally, and we can discuss what size of mapping to build when extending mTHP to control non-anon shmem in the future.
Signed-off-by Baolin Wang <[email protected]>--- mm/memory.c | 57 +++++++++++++++++++++++++++++++++++++++++++---------- 1 file changed, 47 insertions(+), 10 deletions(-)
LGTM. Reviewed-by: Zi Yan <[email protected]>
--
Best Regards,
Yan, Zi
Baolin Wang
reviewerRe: [PATCH v5 1/6] mm: memory: extend finish_fault() to support large folio
Quoted 10 lines · Zi Yan
Quoted 8 lines · Baolin Wang
Add large folio mapping establishment support for
finish_fault()as a preparation, to support multi-size THP allocation of anonymous shmem pages in the following patches.Keep the same behavior (per-page fault) for non-anon shmem to avoid inflating the RSS unintentionally, and we can discuss what size of mapping to build when extending mTHP to control non-anon shmem in the future.
Quoted 6 lines
Quoted 5 lines
Signed-off-by Baolin Wang <[email protected]>--- mm/memory.c | 57 +++++++++++++++++++++++++++++++++++++++++++---------- 1 file changed, 47 insertions(+), 10 deletions(-)LGTM. Reviewed-by: Zi Yan <[email protected]>
Thanks Zi for reviewing.
Re: [PATCH v5 1/6] mm: memory: extend finish_fault() to support large folio
In reply to Baolin WangQuoted 39 lines · Baolin Wang
Add large folio mapping establishment support for
finish_fault()as a preparation, to support multi-size THP allocation of anonymous shmem pages in the following patches.Keep the same behavior (per-page fault) for non-anon shmem to avoid inflating the RSS unintentionally, and we can discuss what size of mapping to build when extending mTHP to control non-anon shmem in the future.
Signed-off-by Baolin Wang <[email protected]>--- mm/memory.c | 57 +++++++++++++++++++++++++++++++++++++++++++---------- 1 file changed, 47 insertions(+), 10 deletions(-)mm/memory.c +8
-0@@ -4831,9 +4831,12 @@ vm_fault_t finish_fault(struct vm_fault *vmf) {struct vm_area_struct *vma = vmf->vma;struct page *page; +struct folio *folio;vm_fault_t ret;bool is_cow = (vmf->flags &FAULT_FLAG_WRITE ) && !(vma->vm_flags &VM_SHARED ); +int type, nr_pages; +unsigned long addr = vmf->address;/* Did we COW the page? */ if (is_cow)@@ -4864,24 +4867,58 @@ vm_fault_t finish_fault(struct vm_fault *vmf)return VM_FAULT_OOM ; } + folio =page_folio (page); + nr_pages =folio_nr_pages (folio); + +/* +* Using per-page fault to maintain the uffd semantics, and same
mm/memory.c vm_fault_t finish_fault(struct vm_fault *vmf) · line 4884
Quoted 10 lines
mm/memory.c +10
-0+* approach also applies to non-anonymous-shmem faults to avoid +* inflating the RSS of the process. +*/ +if (!vma_is_anon_shmem (vma) ||unlikely (userfaultfd_armed (vma))) { + nr_pages =1 ; + }else if (nr_pages >1 ) { +pgoff_t idx =folio_page_idx (folio, page); +/* The page offset of vmf->address within the VMA. */ +pgoff_t vma_off = vmf->pgoff - vmf->vma->vm_pgoff; +
vma->vm_pgoff
mm/memory.c vm_fault_t finish_fault(struct vm_fault *vmf) · line 4894
Quoted 10 lines
mm/memory.c +10
-0+/* +* Fallback to per-page fault in case the folio size in page +* cache beyond the VMA limits. +*/ +if (unlikely (vma_off < idx || + vma_off + (nr_pages - idx) >vma_pages (vma))) { + nr_pages =1 ; + }else { +/* Now we can set mappings for the whole large folio. */ + addr = vmf->address - idx *PAGE_SIZE ;
addr -= idx * PAGE_SIZE;
mm/memory.c vm_fault_t finish_fault(struct vm_fault *vmf) · line 4900
Quoted 7 lines
mm/memory.c +5
-1+ page = &folio->page; + } + } + vmf->pte =pte_offset_map_lock (vma->vm_mm, vmf->pmd, - vmf->address, &vmf->ptl); + addr, &vmf->ptl);
no newline now,
Quoted 12 lines
mm/memory.c +0
-8if (!vmf->pte)return VM_FAULT_NOPAGE ;/* Re-check under ptl */ -if (likely (!vmf_pte_changed (vmf))) { -struct folio *folio =page_folio (page); -int type = is_cow ?MM_ANONPAGES :mm_counter_file (folio); - -set_pte_range (vmf, folio, page,1 , vmf->address); -add_mm_counter (vma->vm_mm, type,1 ); - ret =0 ; - }else {
mm/memory.c vm_fault_t finish_fault(struct vm_fault *vmf) · line 4913
Quoted 10 lines
mm/memory.c +7
-1-update_mmu_tlb (vma, vmf->address, vmf->pte); +if (nr_pages ==1 &&unlikely (vmf_pte_changed (vmf))) { +update_mmu_tlb (vma, addr, vmf->pte); + ret =VM_FAULT_NOPAGE ; + goto unlock; + }else if (nr_pages >1 && !pte_range_none (vmf->pte, nr_pages)) { +update_mmu_tlb_range (vma, addr, vmf->pte, nr_pages); ret =VM_FAULT_NOPAGE ; + goto unlock; }
We may add a vmf_pte_range_changed(), but separate it.
Some very small nits, up to you,
Quoted 11 lines
mm/memory.c +7
-0+folio_ref_add (folio, nr_pages -1 ); +set_pte_range (vmf, folio, page, nr_pages, addr); + type = is_cow ?MM_ANONPAGES :mm_counter_file (folio); +add_mm_counter (vma->vm_mm, type, nr_pages); + ret =0 ; + +unlock:pte_unmap_unlock (vmf->pte, vmf->ptl);return ret; }
Baolin Wang
reviewerRe: [PATCH v5 1/6] mm: memory: extend finish_fault() to support large folio
Quoted 97 lines · Kefeng Wang
Quoted 37 lines · Baolin Wang
Add large folio mapping establishment support for
finish_fault()as a preparation, to support multi-size THP allocation of anonymous shmem pages in the following patches.Keep the same behavior (per-page fault) for non-anon shmem to avoid inflating the RSS unintentionally, and we can discuss what size of mapping to build when extending mTHP to control non-anon shmem in the future.
Signed-off-by Baolin Wang <[email protected]>---
mm/memory.c| 57 +++++++++++++++++++++++++++++++++++++++++++----------1 file changed, 47 insertions(+), 10 deletions(-)mm/memory.c +1
-0@@ -4831,9 +4831,12 @@ vm_fault_t finish_fault(struct vm_fault *vmf) { struct vm_area_struct *vma = vmf->vma; struct page *page; + struct folio *folio; vm_fault_t ret;bool
is_cow= (vmf->flags&FAULT_FLAG_WRITE) && !(vma->vm_flags &VM_SHARED); + int type,nr_pages;
+ unsigned long addr =vmf->address;
/* Did we COW the page? */
if (is_cow)mm/memory.c +5
-0@@ -4864,24 +4867,58 @@ vm_fault_t finish_fault(struct vm_fault *vmf) return VM_FAULT_OOM; } + folio = page_folio(page); + nr_pages = folio_nr_pages(folio); + + /* +* Using per-page fault to maintain the uffd semantics, and same mm/memory.c vm_fault_t finish_fault(struct vm_fault *vmf) · line 4883
Quoted 10 lines
mm/memory.c +10
-0+* approach also applies to non-anonymous-shmem faults to avoid +* inflating the RSS of the process. +*/ + if (!vma_is_anon_shmem(vma) || unlikely(userfaultfd_armed(vma))) { + nr_pages = 1; + } else if (nr_pages > 1) { + pgoff_t idx = folio_page_idx(folio, page); + /* The page offset of vmf->address within the VMA. */ + pgoff_t vma_off = vmf->pgoff - vmf->vma->vm_pgoff; +
vma->vm_pgoffmm/memory.c vm_fault_t finish_fault(struct vm_fault *vmf) · line 4894
Quoted 10 lines
mm/memory.c +10
-0+ /* + * Fallback to per-page fault in case the folio size in page + * cache beyond the VMA limits. + */ + if (unlikely(vma_off < idx || + vma_off + (nr_pages - idx) > vma_pages(vma))) { + nr_pages = 1; + } else { + /* Now we can set mappings for the whole large folio. */ + addr = vmf->address - idx * PAGE_SIZE;addr -= idx *
PAGE_SIZE;mm/memory.c vm_fault_t finish_fault(struct vm_fault *vmf) · line 4900
Quoted 7 lines
mm/memory.c +5
-1+ page = &folio->page; + } + } + vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, - vmf->address, &vmf->ptl); + addr, &vmf->ptl);no newline now,
Quoted 11 lines
mm/memory.c +0
-8if (!vmf->pte) return VM_FAULT_NOPAGE; /* Re-check under ptl */ - if (likely(!vmf_pte_changed(vmf))) { - struct folio *folio = page_folio(page); - int type = is_cow ? MM_ANONPAGES : mm_counter_file(folio); - - set_pte_range(vmf, folio, page, 1, vmf->address); - add_mm_counter(vma->vm_mm, type, 1); - ret = 0; - } else {mm/memory.c vm_fault_t finish_fault(struct vm_fault *vmf) · line 4913
Quoted 10 lines
mm/memory.c +7
-1- update_mmu_tlb(vma, vmf->address, vmf->pte); + if (nr_pages == 1 && unlikely(vmf_pte_changed(vmf))) { + update_mmu_tlb(vma, addr, vmf->pte); + ret = VM_FAULT_NOPAGE; + goto unlock; + } else if (nr_pages > 1 && !pte_range_none(vmf->pte, nr_pages)) { + update_mmu_tlb_range(vma, addr, vmf->pte, nr_pages); ret = VM_FAULT_NOPAGE; + goto unlock; }
Quoted 5 lines
We may add a
vmf_pte_range_changed(), but separate it.Some very small nits, up to you,
Reviewed-by Kefeng Wang <[email protected]>
Thanks for reviewing. If a new version is needed, then I will clean up these coding style issues.
Baolin Wang
reviewerPATCH v5 2/6 mm: shmem: add THP validation for PMD-mapped THP related statistics
In reply to Baolin WangIn order to extend support for mTHP, add THP validation for PMD-mapped THP related statistics to avoid statistical confusion.
--- mm/shmem.c | 7 ++++--- 1 file changed, 4 insertions(+), 3 deletions(-)
mm/shmem.c +4 -3
@@ -1647,7 +1647,7 @@ static struct folio *shmem_alloc_and_add_folio(gfp_t gfp,return ERR_PTR (-E2BIG ); folio =shmem_alloc_folio (gfp,HPAGE_PMD_ORDER , info, index); -if (!folio) +if (!folio && pages ==HPAGE_PMD_NR )count_vm_event (THP_FILE_FALLBACK ); }else { pages =1 ;
@@ -1665,7 +1665,7 @@ static struct folio *shmem_alloc_and_add_folio(gfp_t gfp,if (xa_find (&mapping->i_pages, &index, index + pages -1 ,XA_PRESENT )) { error = -EEXIST ; - }else if (huge) { + }else if (pages ==HPAGE_PMD_NR ) {count_vm_event (THP_FILE_FALLBACK );count_vm_event (THP_FILE_FALLBACK_CHARGE ); }
@@ -2031,7 +2031,8 @@ static int shmem_get_folio_gfp(struct inode *inode, pgoff_t index, folio =shmem_alloc_and_add_folio (huge_gfp, inode, index, fault_mm, true);if (!IS_ERR (folio)) { -count_vm_event (THP_FILE_ALLOC ); +if (folio_test_pmd_mappable (folio)) +count_vm_event (THP_FILE_ALLOC ); goto alloced; }if (PTR_ERR (folio) == -EEXIST )
Signature · 1 line
2.39.3
Baolin Wang
reviewerPATCH v5 3/6 mm: shmem: add multi-size THP sysfs interface for anonymous shmem
In reply to Baolin WangTo support the use of mTHP with anonymous shmem, add a new sysfs interface 'shmem_enabled' in the '/sys/kernel/mm/transparent_hugepage/hugepages-kB/' directory for each mTHP to control whether shmem is enabled for that mTHP, with a value similar to the top level 'shmem_enabled', which can be set to:
"always", "inherit (to inherit the top level setting)", "within_size", "advise", "never". An 'inherit' option is added to ensure compatibility with these global settings, and the options 'force' and 'deny' are dropped, which are rather testing artifacts from the old ages.
By default, PMD-sized hugepages have enabled="inherit" and all other hugepage sizes have enabled="never" for '/sys/kernel/mm/transparent_hugepage/hugepages-xxkB/shmem_enabled'.
In addition, if top level value is 'force', then only PMD-sized hugepages have enabled="inherit", otherwise configuration will be failed and vice versa. That means now we will avoid using non-PMD sized THP to override the global huge allocation.
--- Documentation/admin-guide/mm/transhuge.rst | 23 ++++++ include/linux/huge_mm.h | 10 +++ mm/huge_memory.c | 11 +-- mm/shmem.c | 96 ++++++++++++++++++++++ 4 files changed, 132 insertions(+), 8 deletions(-)
Documentation/admin-guide/mm/transhuge.rst +23 -0
@@ -332,6 +332,29 @@ deny force Force the huge option on for all - very useful for testing; +Shmem can also use "multi-size THP" (mTHP) by adding a new sysfs knob to control +mTHP allocation: '/sys/kernel/mm/transparent_hugepage/hugepages-<size>kB/shmem_enabled', +and its value for each mTHP is essentially consistent with the global setting. +An 'inherit' option is added to ensure compatibility with these global settings. +Conversely, the options 'force' and 'deny' are dropped, which are rather testing +artifacts from the old ages. +always + Attempt to allocate <size> huge pages every time we need a new page; + +inherit + Inherit the top-level "shmem_enabled" value. By default, PMD-sized hugepages + have enabled="inherit" and all other hugepage sizes have enabled="never"; + +never + Do not allocate <size> huge pages; + +within_size + Only allocate <size> huge page if it will be fully within i_size. + Also respect fadvise()/madvise() hints; + +advise + Only allocate <size> huge pages if requested with fadvise()/madvise(); + Need of application restart ===========================
include/linux/huge_mm.h +10 -0
@@ -6,6 +6,7 @@#include <linux/mm_types.h> #include <linux/fs.h> /* only for vma_is_dax() */ +#include <linux/kobject.h> vm_fault_t do_huge_pmd_anonymous_page (struct vm_fault *vmf); int copy_huge_pmd(struct mm_struct *dst_mm, struct mm_struct *src_mm,
@@ -63,6 +64,7 @@ ssize_t single_hugepage_flag_show(struct kobject *kobj, struct kobj_attribute *attr, char *buf, enum transparent_hugepage_flag flag);extern struct kobj_attribute shmem_enabled_attr; +extern struct kobj_attribute thpsize_shmem_enabled_attr;/* * Mask of all large folio orders supported for anonymous THP; all orders up to
@@ -265,6 +267,14 @@ unsigned long thp_vma_allowable_orders(struct vm_area_struct *vma,return __thp_vma_allowable_orders (vma, vm_flags, tva_flags, orders); } +struct thpsize { +struct kobject kobj; +struct list_head node; +int order; +}; + +#define to_thpsize (kobj) container_of(kobj, struct thpsize, kobj) +enum mthp_stat_item {MTHP_STAT_ANON_FAULT_ALLOC ,MTHP_STAT_ANON_FAULT_FALLBACK ,
mm/huge_memory.c +3 -8
@@ -449,14 +449,6 @@ static void thpsize_release(struct kobject *kobj);static DEFINE_SPINLOCK (huge_anon_orders_lock);static LIST_HEAD (thpsize_list); -struct thpsize { -struct kobject kobj; -struct list_head node; -int order; -}; - -#define to_thpsize (kobj) container_of(kobj, struct thpsize, kobj) -static ssize_t thpsize_enabled_show (struct kobject *kobj,struct kobj_attribute *attr,char *buf) {
@@ -517,6 +509,9 @@ static struct kobj_attribute thpsize_enabled_attr =static struct attribute *thpsize_attrs[] = { &thpsize_enabled_attr.attr, +#ifdef CONFIG_SHMEM + &thpsize_shmem_enabled_attr.attr, +#endifNULL , };
mm/shmem.c +96 -0
120 lines not on this page · raw message
Signature · 1 line
2.39.3
Baolin Wang
reviewerPATCH v5 4/6 mm: shmem: add mTHP support for anonymous shmem
In reply to Baolin WangCommit 19eaf44954df adds multi-size THP (mTHP) for anonymous pages, that can allow THP to be configured through the sysfs interface located at '/sys/kernel/mm/transparent_hugepage/hugepage-XXkb/enabled'.
However, the anonymous shmem will ignore the anonymous mTHP rule configured through the sysfs interface, and can only use the PMD-mapped THP, that is not reasonable. Users expect to apply the mTHP rule for all anonymous pages, including the anonymous shmem, in order to enjoy the benefits of mTHP. For example, lower latency than PMD-mapped THP, smaller memory bloat than PMD-mapped THP, contiguous PTEs on ARM architecture to reduce TLB miss etc. In addition, the mTHP interfaces can be extended to support all shmem/tmpfs scenarios in the future, especially for the shmem mmap() case.
The primary strategy is similar to supporting anonymous mTHP. Introduce a new interface '/mm/transparent_hugepage/hugepage-XXkb/shmem_enabled', which can have almost the same values as the top-level '/sys/kernel/mm/transparent_hugepage/shmem_enabled', with adding a new additional "inherit" option and dropping the testing options 'force' and 'deny'. By default all sizes will be set to "never" except PMD size, which is set to "inherit". This ensures backward compatibility with the anonymous shmem enabled of the top level, meanwhile also allows independent control of anonymous shmem enabled for each mTHP.
--- include/linux/huge_mm.h | 10 +++ mm/shmem.c | 187 +++++++++++++++++++++++++++++++++------- 2 files changed, 167 insertions(+), 30 deletions(-)
include/linux/huge_mm.h +10 -0
@@ -575,6 +575,16 @@ static inline bool thp_migration_supported(void) {return false; } + +static inline int highest_order (unsigned long orders) +{ +return 0 ; +} + +static inline int next_order (unsigned long *orders,int prev) +{ +return 0 ; +}#endif /* CONFIG_TRANSPARENT_HUGEPAGE */ static inline int split_folio_to_list_to_order(struct folio *folio,
mm/shmem.c +157 -30
242 lines not on this page · raw message
Signature · 1 line
2.39.3
Ryan Roberts
reviewerRe: [PATCH v5 4/6] mm: shmem: add mTHP support for anonymous shmem
Quoted 24 lines · Baolin Wang
Commit
19eaf44954dfadds multi-size THP (mTHP) for anonymous pages, that can allow THP to be configured through the sysfs interface located at '/sys/kernel/mm/transparent_hugepage/hugepage-XXkb/enabled'.However, the anonymous shmem will ignore the anonymous mTHP rule configured through the sysfs interface, and can only use the PMD-mapped THP, that is not reasonable. Users expect to apply the mTHP rule for all anonymous pages, including the anonymous shmem, in order to enjoy the benefits of mTHP. For example, lower latency than PMD-mapped THP, smaller memory bloat than PMD-mapped THP, contiguous PTEs on ARM architecture to reduce TLB miss etc. In addition, the mTHP interfaces can be extended to support all shmem/tmpfs scenarios in the future, especially for the shmem
mmap()case.The primary strategy is similar to supporting anonymous mTHP. Introduce a new interface '/mm/transparent_hugepage/hugepage-XXkb/shmem_enabled', which can have almost the same values as the top-level '
/sys/kernel/mm/transparent_hugepage/shmem_enabled', with adding a new additional "inherit" option and dropping the testing options 'force' and 'deny'. By default all sizes will be set to "never" except PMD size, which is set to "inherit". This ensures backward compatibility with the anonymous shmem enabled of the top level, meanwhile also allows independent control of anonymous shmem enabled for each mTHP.
Quoted 1 line
Signed-off-by Baolin Wang <[email protected]>
[...]
Hi Baolin,
I'm currently trying to fix a bug where khugepaged is not started if only shmem is enabled for THP. See discussion at [1]. It's been broken like this forever.
Assuming anon and shmem THP are initially both disabled:
# echo never > /sys/kernel/mm/transparent_hugepage/shmem_enabled # echo never > /sys/kernel/mm/transparent_hugepage/enabled <khugepaged is stopped here>
Then shemem THP is enabled:
# echo always > /sys/kernel/mm/transparent_hugepage/shmem_enabled <khugepaged is not started, this is a bug>
As part of investigating the fix, I stumbled upon this patch, which I remember reviewing an early version of but I've been out for a while and missed the latter versions. See below for comments and questions; the answers to which will help me figure out how to fix the above...
[1] https://lore.kernel.org/linux-mm/[email protected]/
Quoted 21 lines
mm/shmem.c +20
-0+#ifdef CONFIG_TRANSPARENT_HUGEPAGE +static unsigned long shmem_allowable_huge_orders (struct inode *inode, +struct vm_area_struct *vma,pgoff_t index, +bool global_huge) +{ +unsigned long mask =READ_ONCE (huge_shmem_orders_always); +unsigned long within_size_orders =READ_ONCE (huge_shmem_orders_within_size); +unsigned long vm_flags = vma->vm_flags; +/* +* Check all the (large) orders below HPAGE_PMD_ORDER + 1 that +* are enabled for this vma. +*/ +unsigned long orders =BIT (PMD_ORDER +1 ) -1 ; +loff_t i_size; +int order; + +if ((vm_flags &VM_NOHUGEPAGE ) || +test_bit (MMF_DISABLE_THP , &vma->vm_mm->flags)) +return 0 ; +
mm/shmem.c static gfp_t limit_gfp_mask(gfp_t huge_gfp, gfp_t limit_gfp) · line 1643
Quoted 10 lines
mm/shmem.c +10
-0+/* If the hardware/firmware marked hugepage support disabled. */ +if (transparent_hugepage_flags & (1 <<TRANSPARENT_HUGEPAGE_UNSUPPORTED )) +return 0 ; + +/* +* Following the 'deny' semantics of the top level, force the huge +* option off from all mounts. +*/ +if (shmem_huge ==SHMEM_HUGE_DENY ) +return 0 ;
I don't quite get this, I don't think its the desirable behaviour. This is saying that if the top-level shemem_enabled control is set to 'deny', then all mTHP sizes, regardless of their control's setting are disabled?
The anon controls don't work like that; you can set the top-level control to anything you like, but its only consumed if the per-size controls are inheriting it.
So for the deny case, wouldn't it be better to allow that as an option on all the per-size controls (and implicitly let it be inherrited from the top-level)?
mm/shmem.c static gfp_t limit_gfp_mask(gfp_t huge_gfp, gfp_t limit_gfp) · line 1650
Quoted 7 lines
mm/shmem.c +7
-0+ +/* +* Only allow inherit orders if the top-level value is 'force', which +* means non-PMD sized THP can not override 'huge' mount option now. +*/ +if (shmem_huge ==SHMEM_HUGE_FORCE ) +return READ_ONCE (huge_shmem_orders_inherit);
I vaguely recall that we originally discussed that trying to set 'force' on the top level control while any per-size controls were set to 'inherit' would be an error, and trying to set 'force' on any per-size control except the PMD-size would be an error?
I don't really understand this logic. Shouldn't we just be looking at the per-size control settings (or the top-level control as a proxy for every per-size control that has 'inherit' set)?
Then for tmpfs, which doesn't support non-PMD-sizes yet, we just always use the PMD-size control for decisions.
I'm also really struggling with the concept of shmem_is_huge() existing along side shmem_allowable_huge_orders(). Surely this needs to all be refactored into shmem_allowable_huge_orders()?
Quoted 8 lines
mm/shmem.c +8
-0+ +/* Allow mTHP that will be fully within i_size. */ + order =highest_order (within_size_orders); +while (within_size_orders) { + index =round_up (index +1 , order); + i_size =round_up (i_size_read (inode),PAGE_SIZE ); +if (i_size >>PAGE_SHIFT >= index) { + mask |= within_size_orders;
mm/shmem.c static gfp_t limit_gfp_mask(gfp_t huge_gfp, gfp_t limit_gfp) · line 1668
Quoted 10 lines
mm/shmem.c +10
-0+break ; + } + + order =next_order (&within_size_orders, order); + } + +if (vm_flags &VM_HUGEPAGE ) + mask |=READ_ONCE (huge_shmem_orders_madvise); + +if (global_huge)
Perhaps I've misunderstood global_huge, but I think its just the return value from shmem_is_huge()? But you're also using shmem_huge directly in this function. I find it all rather confusing.
Sorry if all this was discussed and agreed in my absence! But I think this needs to be sorted in order to do the bug fix for khugepaged properly.
Thanks,
Ryan
Quoted 190 lines
mm/shmem.c +102
-30+ mask |=READ_ONCE (huge_shmem_orders_inherit); + +return orders & mask; +} + +static unsigned long shmem_suitable_orders (struct inode *inode,struct vm_fault *vmf, +struct address_space *mapping,pgoff_t index, +unsigned long orders) +{ +struct vm_area_struct *vma = vmf->vma; +unsigned long pages; +int order; + + orders =thp_vma_suitable_orders (vma, vmf->address, orders); +if (!orders) +return 0 ; + +/* Find the highest order that can add into the page cache */ + order =highest_order (orders); +while (orders) { + pages =1UL << order; + index =round_down (index, pages); +if (!xa_find (&mapping->i_pages, &index, + index + pages -1 ,XA_PRESENT )) +break ; + order =next_order (&orders, order); + } + +return orders; +} +#else +static unsigned long shmem_allowable_huge_orders (struct inode *inode, +struct vm_area_struct *vma,pgoff_t index, +bool global_huge) +{ +return 0 ; +} + +static unsigned long shmem_suitable_orders (struct inode *inode,struct vm_fault *vmf, +struct address_space *mapping,pgoff_t index, +unsigned long orders) +{ +return 0 ; +} +#endif /* CONFIG_TRANSPARENT_HUGEPAGE */ +static struct folio *shmem_alloc_folio (gfp_t gfp,int order,struct shmem_inode_info *info,pgoff_t index) {@@ -1625,38 +1726,55 @@ static struct folio *shmem_alloc_folio(gfp_t gfp, int order,return folio; } -static struct folio *shmem_alloc_and_add_folio (gfp_t gfp, -struct inode *inode,pgoff_t index, -struct mm_struct *fault_mm,bool huge) +static struct folio *shmem_alloc_and_add_folio (struct vm_fault *vmf, +gfp_t gfp,struct inode *inode,pgoff_t index, +struct mm_struct *fault_mm,unsigned long orders) {struct address_space *mapping = inode->i_mapping;struct shmem_inode_info *info =SHMEM_I (inode); -struct folio *folio; +struct vm_area_struct *vma = vmf ? vmf->vma :NULL ; +unsigned long suitable_orders =0 ; +struct folio *folio =NULL ;long pages; -int error; +int error, order;if (!IS_ENABLED (CONFIG_TRANSPARENT_HUGEPAGE )) - huge = false; + orders =0 ; -if (huge) { - pages =HPAGE_PMD_NR ; - index =round_down (index,HPAGE_PMD_NR ); +if (orders >0 ) { +if (vma &&vma_is_anon_shmem (vma)) { + suitable_orders =shmem_suitable_orders (inode, vmf, + mapping, index, orders); + }else if (orders &BIT (HPAGE_PMD_ORDER )) { + pages =HPAGE_PMD_NR ; + suitable_orders =BIT (HPAGE_PMD_ORDER ); + index =round_down (index,HPAGE_PMD_NR ); -/* -* Check for conflict before waiting on a huge allocation. -* Conflict might be that a huge page has just been allocated -* and added to page cache by a racing thread, or that there -* is already at least one small page in the huge extent. -* Be careful to retry when appropriate, but not forever! -* Elsewhere -EEXIST would be the right code, but not here. -*/ -if (xa_find (&mapping->i_pages, &index, - index +HPAGE_PMD_NR -1 ,XA_PRESENT )) -return ERR_PTR (-E2BIG ); +/* +* Check for conflict before waiting on a huge allocation. +* Conflict might be that a huge page has just been allocated +* and added to page cache by a racing thread, or that there +* is already at least one small page in the huge extent. +* Be careful to retry when appropriate, but not forever! +* Elsewhere -EEXIST would be the right code, but not here. +*/ +if (xa_find (&mapping->i_pages, &index, + index +HPAGE_PMD_NR -1 ,XA_PRESENT )) +return ERR_PTR (-E2BIG ); + } - folio =shmem_alloc_folio (gfp,HPAGE_PMD_ORDER , info, index); -if (!folio && pages ==HPAGE_PMD_NR ) -count_vm_event (THP_FILE_FALLBACK ); + order =highest_order (suitable_orders); +while (suitable_orders) { + pages =1UL << order; + index =round_down (index, pages); + folio =shmem_alloc_folio (gfp, order, info, index); +if (folio) + goto allocated; + +if (pages ==HPAGE_PMD_NR ) +count_vm_event (THP_FILE_FALLBACK ); + order =next_order (&suitable_orders, order); + } }else { pages =1 ; folio =shmem_alloc_folio (gfp,0 , info, index);@@ -1664,6 +1782,7 @@ static struct folio *shmem_alloc_and_add_folio(gfp_t gfp,if (!folio)return ERR_PTR (-ENOMEM ); +allocated:__folio_set_locked (folio);__folio_set_swapbacked (folio);@@ -1958,7 +2077,8 @@ static int shmem_get_folio_gfp(struct inode *inode, pgoff_t index,struct mm_struct *fault_mm;struct folio *folio;int error; -bool alloced; +bool alloced, huge; +unsigned long orders =0 ;if (WARN_ON_ONCE (!shmem_mapping (inode->i_mapping)))return -EINVAL ;@@ -2030,14 +2150,21 @@ static int shmem_get_folio_gfp(struct inode *inode, pgoff_t index,return 0 ; } -if (shmem_is_huge (inode, index, false, fault_mm, - vma ? vma->vm_flags :0 )) { + huge =shmem_is_huge (inode, index, false, fault_mm, + vma ? vma->vm_flags :0 ); +/* Find hugepage orders that are allowed for anonymous shmem. */ +if (vma &&vma_is_anon_shmem (vma)) + orders =shmem_allowable_huge_orders (inode, vma, index, huge); +else if (huge) + orders =BIT (HPAGE_PMD_ORDER ); + +if (orders >0 ) {gfp_t huge_gfp; huge_gfp =vma_thp_gfp_mask (vma); huge_gfp =limit_gfp_mask (huge_gfp, gfp); - folio =shmem_alloc_and_add_folio (huge_gfp, - inode, index, fault_mm, true); + folio =shmem_alloc_and_add_folio (vmf, huge_gfp, + inode, index, fault_mm, orders);if (!IS_ERR (folio)) {if (folio_test_pmd_mappable (folio))count_vm_event (THP_FILE_ALLOC );@@ -2047,7 +2174,7 @@ static int shmem_get_folio_gfp(struct inode *inode, pgoff_t index, goto repeat; } - folio =shmem_alloc_and_add_folio (gfp, inode, index, fault_mm, false); + folio =shmem_alloc_and_add_folio (vmf, gfp, inode, index, fault_mm,0 );if (IS_ERR (folio)) { error =PTR_ERR (folio);if (error == -EEXIST )@@ -2058,7 +2185,7 @@ static int shmem_get_folio_gfp(struct inode *inode, pgoff_t index, alloced: alloced = true; -if (folio_test_pmd_mappable (folio) && +if (folio_test_large (folio) &&DIV_ROUND_UP (i_size_read (inode),PAGE_SIZE ) <folio_next_index (folio) -1 ) {struct shmem_sb_info *sbinfo =SHMEM_SB (inode->i_sb);
Baolin Wang
reviewerRe: [PATCH v5 4/6] mm: shmem: add mTHP support for anonymous shmem
Quoted 36 lines · Ryan Roberts
Quoted 24 lines · Baolin Wang
Commit
19eaf44954dfadds multi-size THP (mTHP) for anonymous pages, that can allow THP to be configured through the sysfs interface located at '/sys/kernel/mm/transparent_hugepage/hugepage-XXkb/enabled'.However, the anonymous shmem will ignore the anonymous mTHP rule configured through the sysfs interface, and can only use the PMD-mapped THP, that is not reasonable. Users expect to apply the mTHP rule for all anonymous pages, including the anonymous shmem, in order to enjoy the benefits of mTHP. For example, lower latency than PMD-mapped THP, smaller memory bloat than PMD-mapped THP, contiguous PTEs on ARM architecture to reduce TLB miss etc. In addition, the mTHP interfaces can be extended to support all shmem/tmpfs scenarios in the future, especially for the shmem
mmap()case.The primary strategy is similar to supporting anonymous mTHP. Introduce a new interface '/mm/transparent_hugepage/hugepage-XXkb/shmem_enabled', which can have almost the same values as the top-level '
/sys/kernel/mm/transparent_hugepage/shmem_enabled', with adding a new additional "inherit" option and dropping the testing options 'force' and 'deny'. By default all sizes will be set to "never" except PMD size, which is set to "inherit". This ensures backward compatibility with the anonymous shmem enabled of the top level, meanwhile also allows independent control of anonymous shmem enabled for each mTHP.Quoted 1 line
Signed-off-by Baolin Wang <[email protected]>[...]
Hi Baolin,
I'm currently trying to fix a bug where khugepaged is not started if only shmem is enabled for THP. See discussion at [1]. It's been broken like this forever.
Assuming anon and shmem THP are initially both disabled:
Quoted 8 lines
# echo never > /sys/kernel/mm/transparent_hugepage/shmem_enabled # echo never > /sys/kernel/mm/transparent_hugepage/enabled <khugepaged is stopped here>Then shemem THP is enabled:
# echo always > /sys/kernel/mm/transparent_hugepage/shmem_enabled <khugepaged is not started, this is a bug>
Thanks Ryan. Yes, this is a real problem.
Quoted 44 lines
As part of investigating the fix, I stumbled upon this patch, which I remember reviewing an early version of but I've been out for a while and missed the latter versions. See below for comments and questions; the answers to which will help me figure out how to fix the above...
[1] https://lore.kernel.org/linux-mm/[email protected]/
Quoted 21 lines
mm/shmem.c +20
-0+#ifdef CONFIG_TRANSPARENT_HUGEPAGE +static unsigned long shmem_allowable_huge_orders (struct inode *inode, +struct vm_area_struct *vma,pgoff_t index, +bool global_huge) +{ +unsigned long mask =READ_ONCE (huge_shmem_orders_always); +unsigned long within_size_orders =READ_ONCE (huge_shmem_orders_within_size); +unsigned long vm_flags = vma->vm_flags; +/* +* Check all the (large) orders below HPAGE_PMD_ORDER + 1 that +* are enabled for this vma. +*/ +unsigned long orders =BIT (PMD_ORDER +1 ) -1 ; +loff_t i_size; +int order; + +if ((vm_flags &VM_NOHUGEPAGE ) || +test_bit (MMF_DISABLE_THP , &vma->vm_mm->flags)) +return 0 ; +mm/shmem.c static gfp_t limit_gfp_mask(gfp_t huge_gfp, gfp_t limit_gfp) · line 1643
Quoted 10 lines
mm/shmem.c +10
-0+/* If the hardware/firmware marked hugepage support disabled. */ +if (transparent_hugepage_flags & (1 <<TRANSPARENT_HUGEPAGE_UNSUPPORTED )) +return 0 ; + +/* +* Following the 'deny' semantics of the top level, force the huge +* option off from all mounts. +*/ +if (shmem_huge ==SHMEM_HUGE_DENY ) +return 0 ;I don't quite get this, I don't think its the desirable behaviour. This is saying that if the top-level
shemem_enabledcontrol is set to 'deny', then all mTHP sizes, regardless of their control's setting are disabled?
Quoted 2 lines
The anon controls don't work like that; you can set the top-level control to anything you like, but its only consumed if the per-size controls are inheriting it.
IMO, 'deny' option is not similar like 'never' option.
Quoted 3 lines
So for the deny case, wouldn't it be better to allow that as an option on all the per-size controls (and implicitly let it be inherrited from the top-level)?
From 'deny' option's semantics:
"disables huge on shm_mnt and all mounts, for emergency use;"
It is usually used for testing to shut down all huge pages from the old ages. Moreover, mTHP interfaces will be used as a huge order filter to support tmpfs, so I think it will make life easier to disable all huge orders for testing or emergency use, as well as to maintain the original semantics.
Quoted 4 lines
Quoted 4 lines
mm/shmem.c +4
-0+ +/* +* Only allow inherit orders if the top-level value is 'force', which +* means non-PMD sized THP can not override 'huge' mount option now.
Quoted 8 lines
mm/shmem.c static gfp_t limit_gfp_mask(gfp_t huge_gfp, gfp_t limit_gfp) · line 1650
Quoted 3 lines
mm/shmem.c +3
-0+*/ +if (shmem_huge ==SHMEM_HUGE_FORCE ) +return READ_ONCE (huge_shmem_orders_inherit);I vaguely recall that we originally discussed that trying to set 'force' on the top level control while any per-size controls were set to 'inherit' would be an error, and trying to set 'force' on any per-size control except the PMD-size would be an error?
Right.
Quoted 3 lines
I don't really understand this logic. Shouldn't we just be looking at the per-size control settings (or the top-level control as a proxy for every per-size control that has 'inherit' set)?
‘force’ will apply the huge orders for anon shmem and tmpfs, so now we only allow pmd-mapped THP to be forced. We should not look at per-size control settings for tmpfs now (mTHP for tmpfs will be discussed in future).
Quoted 7 lines
Then for tmpfs, which doesn't support non-PMD-sizes yet, we just always use the PMD-size control for decisions.
I'm also really struggling with the concept of
shmem_is_huge()existing along sideshmem_allowable_huge_orders(). Surely this needs to all be refactored intoshmem_allowable_huge_orders()?
I understood. But now they serve different purposes: shmem_is_huge() will be used to check the huge orders for the top level, for *tmpfs* and anon shmem; whereas shmem_allowable_huge_orders() will only be used to check the per-size huge orders for anon shmem (excluding tmpfs now). However, as I plan to add mTHP support for tmpfs, I think we can perform some cleanups.
Quoted 12 lines
Quoted 12 lines
mm/shmem.c +12
-0+/* Allow mTHP that will be fully within i_size. */ + order =highest_order (within_size_orders); +while (within_size_orders) { + index =round_up (index +1 , order); + i_size =round_up (i_size_read (inode),PAGE_SIZE ); +if (i_size >>PAGE_SHIFT >= index) { + mask |= within_size_orders; +break ; + } + + order =next_order (&within_size_orders, order); + }
Quoted 8 lines
mm/shmem.c static gfp_t limit_gfp_mask(gfp_t huge_gfp, gfp_t limit_gfp) · line 1668
Quoted 5 lines
mm/shmem.c +5
-0+ +if (vm_flags &VM_HUGEPAGE ) + mask |=READ_ONCE (huge_shmem_orders_madvise); + +if (global_huge)Perhaps I've misunderstood
global_huge, but I think its just the return value fromshmem_is_huge()? But you're also usingshmem_hugedirectly in this
Yes.
Quoted 1 line
function. I find it all rather confusing.
I think I have explained why need these logics as above. Since mTHP support for shmem has just started (tmpfs is still in progress). I will make it more clear in the following patches.
Quoted 2 lines
Sorry if all this was discussed and agreed in my absence! But I think this needs to be sorted in order to do the bug fix for khugepaged properly.
No worries. Thanks for your input.
Ryan Roberts
reviewerRe: [PATCH v5 4/6] mm: shmem: add mTHP support for anonymous shmem
Quoted 99 lines · Baolin Wang
Quoted 36 lines · Ryan Roberts
Quoted 24 lines · Baolin Wang
Commit
19eaf44954dfadds multi-size THP (mTHP) for anonymous pages, that can allow THP to be configured through the sysfs interface located at '/sys/kernel/mm/transparent_hugepage/hugepage-XXkb/enabled'.However, the anonymous shmem will ignore the anonymous mTHP rule configured through the sysfs interface, and can only use the PMD-mapped THP, that is not reasonable. Users expect to apply the mTHP rule for all anonymous pages, including the anonymous shmem, in order to enjoy the benefits of mTHP. For example, lower latency than PMD-mapped THP, smaller memory bloat than PMD-mapped THP, contiguous PTEs on ARM architecture to reduce TLB miss etc. In addition, the mTHP interfaces can be extended to support all shmem/tmpfs scenarios in the future, especially for the shmem
mmap()case.The primary strategy is similar to supporting anonymous mTHP. Introduce a new interface '/mm/transparent_hugepage/hugepage-XXkb/shmem_enabled', which can have almost the same values as the top-level '
/sys/kernel/mm/transparent_hugepage/shmem_enabled', with adding a new additional "inherit" option and dropping the testing options 'force' and 'deny'. By default all sizes will be set to "never" except PMD size, which is set to "inherit". This ensures backward compatibility with the anonymous shmem enabled of the top level, meanwhile also allows independent control of anonymous shmem enabled for each mTHP.Quoted 1 line
Signed-off-by Baolin Wang <[email protected]>[...]
Hi Baolin,
I'm currently trying to fix a bug where khugepaged is not started if only shmem is enabled for THP. See discussion at [1]. It's been broken like this forever.
Assuming anon and shmem THP are initially both disabled:
Quoted 8 lines
# echo never > /sys/kernel/mm/transparent_hugepage/shmem_enabled # echo never > /sys/kernel/mm/transparent_hugepage/enabled <khugepaged is stopped here>Then shemem THP is enabled:
# echo always > /sys/kernel/mm/transparent_hugepage/shmem_enabled <khugepaged is not started, this is a bug>Thanks Ryan. Yes, this is a real problem.
Quoted 45 lines
As part of investigating the fix, I stumbled upon this patch, which I remember reviewing an early version of but I've been out for a while and missed the latter versions. See below for comments and questions; the answers to which will help me figure out how to fix the above...
[1]
https://lore.kernel.org/linux-mm/[email protected]/Quoted 21 lines
+#ifdef
CONFIG_TRANSPARENT_HUGEPAGEmm/shmem.c +6
-0+static unsigned long shmem_allowable_huge_orders (struct inode *inode, + struct vm_area_struct *vma, pgoff_t index, + bool global_huge) +{ + unsigned long mask = READ_ONCE(huge_shmem_orders_always); + unsigned long within_size_orders =
READ_ONCE(huge_shmem_orders_within_size);
+ unsigned longvm_flags=vma->vm_flags;
+ /*
+ * Check all the (large) orders belowHPAGE_PMD_ORDER+ 1 that + * are enabled for this vma.
+ */
+ unsigned long orders =BIT(PMD_ORDER + 1)- 1;
+loff_ti_size;
+ int order;mm/shmem.c +5
-0+ + if ((vm_flags & VM_NOHUGEPAGE) || + test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags)) + return 0; +mm/shmem.c static gfp_t limit_gfp_mask(gfp_t huge_gfp, gfp_t limit_gfp) · line 1643
Quoted 10 lines
mm/shmem.c +10
-0+ /* If the hardware/firmware marked hugepage support disabled. */ + if (transparent_hugepage_flags & (1 << TRANSPARENT_HUGEPAGE_UNSUPPORTED)) + return 0; + + /* + * Following the 'deny' semantics of the top level, force the huge + * option off from all mounts. + */ + if (shmem_huge == SHMEM_HUGE_DENY) + return 0;I don't quite get this, I don't think its the desirable behaviour. This is saying that if the top-level
shemem_enabledcontrol is set to 'deny', then all mTHP sizes, regardless of their control's setting are disabled?Quoted 3 lines
The anon controls don't work like that; you can set the top-level control to anything you like, but its only consumed if the per-size controls are inheriting it.
Quoted 8 lines
IMO, 'deny' option is not similar like 'never' option.
Quoted 3 lines
So for the deny case, wouldn't it be better to allow that as an option on all the per-size controls (and implicitly let it be inherrited from the top-level)?
From 'deny' option's semantics:
"disables huge onshm_mntand all mounts, for emergency use;"
Right. Today, tmpfs only supports PMD-sized THP. So for all per-size controls except the PMD-size control, 'deny' and 'never' would be the same practically speaking. For the PMD-size control, 'deny' would disable THP for both anon shmem and all tmpfs mounts, whereas 'never' would only disable THP for anon shmem. When tmpfs gains mTHP support, 'deny' in the other per-size controls would also disable that size for the tmpfs mounts.
I disagree with the current implementation where setting it up so that a top-level 'deny' overrides whatever is in the per-size controls simply because it's different to the model implemented for anon THP. That's my 2 cents. If nobody else agrees then that ok - I'll fix the above bug according to the current model.
Quoted 20 lines
It is usually used for testing to shut down all huge pages from the old ages. Moreover, mTHP interfaces will be used as a huge order filter to support tmpfs, so I think it will make life easier to disable all huge orders for testing or emergency use, as well as to maintain the original semantics.>
Quoted 8 lines
Quoted 7 lines
+
+ /*
+ * Only allow inherit orders if the top-level value is 'force', which + * means non-PMD sized THP can not override 'huge' mount option now. + */
+ if (shmem_huge==SHMEM_HUGE_FORCE)
+ returnREAD_ONCE(huge_shmem_orders_inherit);Quoted 4 lines
I vaguely recall that we originally discussed that trying to set 'force' on the top level control while any per-size controls were set to 'inherit' would be an error, and trying to set 'force' on any per-size control except the PMD-size would be an error?
Right.
Quoted 7 lines
Quoted 3 lines
I don't really understand this logic. Shouldn't we just be looking at the per-size control settings (or the top-level control as a proxy for every per-size control that has 'inherit' set)?
‘force’ will apply the huge orders for anon shmem and tmpfs, so now we only allow pmd-mapped THP to be forced. We should not look at per-size control settings for tmpfs now (mTHP for tmpfs will be discussed in future).
But why not just maintain per-size controls and refactor tmpfs to just look at the PMD-size THP control for now. It can ignore the other sizes. That's much simpler and cleaner IMHO.
Quoted 36 lines
Quoted 8 lines
Then for tmpfs, which doesn't support non-PMD-sizes yet, we just always use the PMD-size control for decisions.
I'm also really struggling with the concept of
shmem_is_huge()existing along sideshmem_allowable_huge_orders(). Surely this needs to all be refactored intoshmem_allowable_huge_orders()?I understood. But now they serve different purposes:
shmem_is_huge()will be used to check the huge orders for the top level, for *tmpfs* and anon shmem; whereasshmem_allowable_huge_orders()will only be used to check the per-size huge orders for anon shmem (excluding tmpfs now). However, as I plan to add mTHP support for tmpfs, I think we can perform some cleanups.Quoted 12 lines
Quoted 12 lines
+ /* Allow mTHP that will be fully within
i_size. */ + order =highest_order(within_size_orders);
+ while (within_size_orders) {
+ index =round_up(index + 1, order);
+i_size=round_up(i_size_read(inode), PAGE_SIZE); + if (i_size>>PAGE_SHIFT>= index) {
+ mask |=within_size_orders;
+ break;
+ }
+
+ order =next_order(&within_size_orders, order); + }Quoted 8 lines
mm/shmem.c static gfp_t limit_gfp_mask(gfp_t huge_gfp, gfp_t limit_gfp) · line 1668
Quoted 5 lines
mm/shmem.c +5
-0+ + if (vm_flags & VM_HUGEPAGE) + mask |= READ_ONCE(huge_shmem_orders_madvise); + + if (global_huge)Perhaps I've misunderstood
global_huge, but I think its just the return value fromshmem_is_huge()? But you're also usingshmem_hugedirectly in this
Quoted 7 lines
Yes.
Quoted 1 line
function. I find it all rather confusing.
I think I have explained why need these logics as above. Since mTHP support for shmem has just started (tmpfs is still in progress). I will make it more clear in the following patches.
OK as long as you have a plan for the clean up, that's good enough for me.
Quoted 5 lines
Quoted 3 lines
Sorry if all this was discussed and agreed in my absence! But I think this needs to be sorted in order to do the bug fix for khugepaged properly.
No worries. Thanks for your input.
Re: [PATCH v5 4/6] mm: shmem: add mTHP support for anonymous shmem
Quoted 116 lines · Ryan Roberts
Quoted 99 lines · Baolin Wang
Quoted 36 lines · Ryan Roberts
Quoted 24 lines · Baolin Wang
Commit
19eaf44954dfadds multi-size THP (mTHP) for anonymous pages, that can allow THP to be configured through the sysfs interface located at '/sys/kernel/mm/transparent_hugepage/hugepage-XXkb/enabled'.However, the anonymous shmem will ignore the anonymous mTHP rule configured through the sysfs interface, and can only use the PMD-mapped THP, that is not reasonable. Users expect to apply the mTHP rule for all anonymous pages, including the anonymous shmem, in order to enjoy the benefits of mTHP. For example, lower latency than PMD-mapped THP, smaller memory bloat than PMD-mapped THP, contiguous PTEs on ARM architecture to reduce TLB miss etc. In addition, the mTHP interfaces can be extended to support all shmem/tmpfs scenarios in the future, especially for the shmem
mmap()case.The primary strategy is similar to supporting anonymous mTHP. Introduce a new interface '/mm/transparent_hugepage/hugepage-XXkb/shmem_enabled', which can have almost the same values as the top-level '
/sys/kernel/mm/transparent_hugepage/shmem_enabled', with adding a new additional "inherit" option and dropping the testing options 'force' and 'deny'. By default all sizes will be set to "never" except PMD size, which is set to "inherit". This ensures backward compatibility with the anonymous shmem enabled of the top level, meanwhile also allows independent control of anonymous shmem enabled for each mTHP.Quoted 1 line
Signed-off-by Baolin Wang <[email protected]>[...]
Hi Baolin,
I'm currently trying to fix a bug where khugepaged is not started if only shmem is enabled for THP. See discussion at [1]. It's been broken like this forever.
Assuming anon and shmem THP are initially both disabled:
Quoted 8 lines
# echo never > /sys/kernel/mm/transparent_hugepage/shmem_enabled # echo never > /sys/kernel/mm/transparent_hugepage/enabled <khugepaged is stopped here>Then shemem THP is enabled:
# echo always > /sys/kernel/mm/transparent_hugepage/shmem_enabled <khugepaged is not started, this is a bug>Thanks Ryan. Yes, this is a real problem.
Quoted 45 lines
As part of investigating the fix, I stumbled upon this patch, which I remember reviewing an early version of but I've been out for a while and missed the latter versions. See below for comments and questions; the answers to which will help me figure out how to fix the above...
[1]
https://lore.kernel.org/linux-mm/[email protected]/Quoted 21 lines
+#ifdef
CONFIG_TRANSPARENT_HUGEPAGEmm/shmem.c +6
-0+static unsigned long shmem_allowable_huge_orders (struct inode *inode, + struct vm_area_struct *vma, pgoff_t index, + bool global_huge) +{ + unsigned long mask = READ_ONCE(huge_shmem_orders_always); + unsigned long within_size_orders =
READ_ONCE(huge_shmem_orders_within_size);
+ unsigned longvm_flags=vma->vm_flags;
+ /*
+ * Check all the (large) orders belowHPAGE_PMD_ORDER+ 1 that + * are enabled for this vma.
+ */
+ unsigned long orders =BIT(PMD_ORDER + 1)- 1;
+loff_ti_size;
+ int order;mm/shmem.c +5
-0+ + if ((vm_flags & VM_NOHUGEPAGE) || + test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags)) + return 0; +mm/shmem.c static gfp_t limit_gfp_mask(gfp_t huge_gfp, gfp_t limit_gfp) · line 1643
Quoted 10 lines
mm/shmem.c +10
-0+ /* If the hardware/firmware marked hugepage support disabled. */ + if (transparent_hugepage_flags & (1 << TRANSPARENT_HUGEPAGE_UNSUPPORTED)) + return 0; + + /* + * Following the 'deny' semantics of the top level, force the huge + * option off from all mounts. + */ + if (shmem_huge == SHMEM_HUGE_DENY) + return 0;I don't quite get this, I don't think its the desirable behaviour. This is saying that if the top-level
shemem_enabledcontrol is set to 'deny', then all mTHP sizes, regardless of their control's setting are disabled?Quoted 3 lines
The anon controls don't work like that; you can set the top-level control to anything you like, but its only consumed if the per-size controls are inheriting it.
Quoted 8 lines
IMO, 'deny' option is not similar like 'never' option.
Quoted 3 lines
So for the deny case, wouldn't it be better to allow that as an option on all the per-size controls (and implicitly let it be inherrited from the top-level)?
From 'deny' option's semantics:
"disables huge onshm_mntand all mounts, for emergency use;"Right. Today, tmpfs only supports PMD-sized THP. So for all per-size controls except the PMD-size control, 'deny' and 'never' would be the same practically speaking. For the PMD-size control, 'deny' would disable THP for both anon shmem and all tmpfs mounts, whereas 'never' would only disable THP for anon shmem. When tmpfs gains mTHP support, 'deny' in the other per-size controls would also disable that size for the tmpfs mounts.
Quoted 5 lines
I disagree with the current implementation where setting it up so that a top-level 'deny' overrides whatever is in the per-size controls simply because it's different to the model implemented for anon THP. That's my 2 cents. If nobody else agrees then that ok - I'll fix the above bug according to the current model.
IIRC, Hugh said that deny+force are rather legacy artifacts that we don't want on sub-controls (and likely we cannot remove them). I agree they shouldn't overwrite other toggles. The models we used should better match.
Signature · 3 lines
Cheers, David / dhildenb
Re: [PATCH v5 4/6] mm: shmem: add mTHP support for anonymous shmem
In reply to Ryan RobertsHey Ryan,
Quoted 31 lines · Ryan Roberts
Quoted 31 lines
Quoted 6 lines
Then for tmpfs, which doesn't support non-PMD-sizes yet, we just always use the PMD-size control for decisions.
I'm also really struggling with the concept of
shmem_is_huge()existing along sideshmem_allowable_huge_orders(). Surely this needs to all be refactored intoshmem_allowable_huge_orders()?I understood. But now they serve different purposes:
shmem_is_huge()will be used to check the huge orders for the top level, for*tmpfs* and anon shmem; whereasshmem_allowable_huge_orders()will only be used to check the per-size huge orders for anon shmem (excluding tmpfs now). However, as I plan to add mTHP support for tmpfs, I think we can perform some cleanups.Quoted 19 lines
Quoted 9 lines
+ /* Allow mTHP that will be fully within
i_size. */ + order =highest_order(within_size_orders);
+ while (within_size_orders) {
+ index =round_up(index + 1, order);
+i_size=round_up(i_size_read(inode), PAGE_SIZE); + if (i_size>>PAGE_SHIFT>= index) {
+ mask |=within_size_orders;
+ break;
+ }mm/shmem.c static gfp_t limit_gfp_mask(gfp_t huge_gfp, gfp_t limit_gfp) · line 1668
Quoted 8 lines
mm/shmem.c +8
-0+ + order = next_order(&within_size_orders, order); + } + + if (vm_flags & VM_HUGEPAGE) + mask |= READ_ONCE(huge_shmem_orders_madvise); + + if (global_huge)Perhaps I've misunderstood
global_huge, but I think its just the return value fromshmem_is_huge()? But you're also usingshmem_hugedirectly in this
Quoted 7 lines
Quoted 6 lines
Yes.
Quoted 1 line
function. I find it all rather confusing.
I think I have explained why need these logics as above. Since mTHP support for shmem has just started (tmpfs is still in progress). I will make it more clear in the following patches.
OK as long as you have a plan for the clean up, that's good enough for me.
Can I continue to push the following patch [1]? When other types of shmem mTHP are supported, we will perform cleanups uniformly.
[1] https://lore.kernel.org/linux-mm/[email protected]/
Thanks,
Bang
Ryan Roberts
reviewerRe: [PATCH v5 4/6] mm: shmem: add mTHP support for anonymous shmem
Quoted 36 lines · Bang Li
Hey Ryan,
Quoted 33 lines · Ryan Roberts
Quoted 33 lines
Quoted 6 lines
Then for tmpfs, which doesn't support non-PMD-sizes yet, we just always use the PMD-size control for decisions.
I'm also really struggling with the concept of
shmem_is_huge()existing along sideshmem_allowable_huge_orders(). Surely this needs to all be refactored intoshmem_allowable_huge_orders()?I understood. But now they serve different purposes:
shmem_is_huge()will be used to check the huge orders for the top level, for*tmpfs* and anon shmem; whereasshmem_allowable_huge_orders()will only be used to check the per-size huge orders for anon shmem (excluding tmpfs now). However, as I plan to add mTHP support for tmpfs, I think we can perform some cleanups.Quoted 11 lines
Quoted 11 lines
+ /* Allow mTHP that will be fully within
i_size. */ + order =highest_order(within_size_orders);
+ while (within_size_orders) {
+ index =round_up(index + 1, order);
+i_size=round_up(i_size_read(inode), PAGE_SIZE); + if (i_size>>PAGE_SHIFT>= index) {
+ mask |=within_size_orders;
+ break;
+ }
+
+ order =next_order(&within_size_orders, order);Quoted 8 lines
mm/shmem.c static gfp_t limit_gfp_mask(gfp_t huge_gfp, gfp_t limit_gfp) · line 1668
Quoted 6 lines
+ }
mm/shmem.c +5
-0+ + if (vm_flags & VM_HUGEPAGE) + mask |= READ_ONCE(huge_shmem_orders_madvise); + + if (global_huge)Perhaps I've misunderstood
global_huge, but I think its just the return value fromshmem_is_huge()? But you're also usingshmem_hugedirectly in thisYes.
Quoted 8 lines
Quoted 5 lines
Quoted 4 lines
Quoted 1 line
function. I find it all rather confusing.
I think I have explained why need these logics as above. Since mTHP support for shmem has just started (tmpfs is still in progress). I will make it more clear in the following patches.
OK as long as you have a plan for the clean up, that's good enough for me.
Can I continue to push the following patch [1]? When other types of shmem mTHP are supported, we will perform cleanups uniformly.
I guess that makes sense.
Quoted 5 lines
[1] https://lore.kernel.org/linux-mm/[email protected]/
Thanks,
Bang
Baolin Wang
reviewerRe: [PATCH v5 4/6] mm: shmem: add mTHP support for anonymous shmem
In reply to Ryan RobertsQuoted 109 lines · Ryan Roberts
Quoted 99 lines · Baolin Wang
Quoted 36 lines · Ryan Roberts
Quoted 24 lines · Baolin Wang
Commit
19eaf44954dfadds multi-size THP (mTHP) for anonymous pages, that can allow THP to be configured through the sysfs interface located at '/sys/kernel/mm/transparent_hugepage/hugepage-XXkb/enabled'.However, the anonymous shmem will ignore the anonymous mTHP rule configured through the sysfs interface, and can only use the PMD-mapped THP, that is not reasonable. Users expect to apply the mTHP rule for all anonymous pages, including the anonymous shmem, in order to enjoy the benefits of mTHP. For example, lower latency than PMD-mapped THP, smaller memory bloat than PMD-mapped THP, contiguous PTEs on ARM architecture to reduce TLB miss etc. In addition, the mTHP interfaces can be extended to support all shmem/tmpfs scenarios in the future, especially for the shmem
mmap()case.The primary strategy is similar to supporting anonymous mTHP. Introduce a new interface '/mm/transparent_hugepage/hugepage-XXkb/shmem_enabled', which can have almost the same values as the top-level '
/sys/kernel/mm/transparent_hugepage/shmem_enabled', with adding a new additional "inherit" option and dropping the testing options 'force' and 'deny'. By default all sizes will be set to "never" except PMD size, which is set to "inherit". This ensures backward compatibility with the anonymous shmem enabled of the top level, meanwhile also allows independent control of anonymous shmem enabled for each mTHP.Quoted 1 line
Signed-off-by Baolin Wang <[email protected]>[...]
Hi Baolin,
I'm currently trying to fix a bug where khugepaged is not started if only shmem is enabled for THP. See discussion at [1]. It's been broken like this forever.
Assuming anon and shmem THP are initially both disabled:
Quoted 8 lines
# echo never > /sys/kernel/mm/transparent_hugepage/shmem_enabled # echo never > /sys/kernel/mm/transparent_hugepage/enabled <khugepaged is stopped here>Then shemem THP is enabled:
# echo always > /sys/kernel/mm/transparent_hugepage/shmem_enabled <khugepaged is not started, this is a bug>Thanks Ryan. Yes, this is a real problem.
Quoted 45 lines
As part of investigating the fix, I stumbled upon this patch, which I remember reviewing an early version of but I've been out for a while and missed the latter versions. See below for comments and questions; the answers to which will help me figure out how to fix the above...
[1]
https://lore.kernel.org/linux-mm/[email protected]/Quoted 21 lines
+#ifdef
CONFIG_TRANSPARENT_HUGEPAGEmm/shmem.c +6
-0+static unsigned long shmem_allowable_huge_orders (struct inode *inode, + struct vm_area_struct *vma, pgoff_t index, + bool global_huge) +{ + unsigned long mask = READ_ONCE(huge_shmem_orders_always); + unsigned long within_size_orders =
READ_ONCE(huge_shmem_orders_within_size);
+ unsigned longvm_flags=vma->vm_flags;
+ /*
+ * Check all the (large) orders belowHPAGE_PMD_ORDER+ 1 that + * are enabled for this vma.
+ */
+ unsigned long orders =BIT(PMD_ORDER + 1)- 1;
+loff_ti_size;
+ int order;mm/shmem.c +5
-0+ + if ((vm_flags & VM_NOHUGEPAGE) || + test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags)) + return 0; +mm/shmem.c static gfp_t limit_gfp_mask(gfp_t huge_gfp, gfp_t limit_gfp) · line 1643
Quoted 10 lines
mm/shmem.c +10
-0+ /* If the hardware/firmware marked hugepage support disabled. */ + if (transparent_hugepage_flags & (1 << TRANSPARENT_HUGEPAGE_UNSUPPORTED)) + return 0; + + /* + * Following the 'deny' semantics of the top level, force the huge + * option off from all mounts. + */ + if (shmem_huge == SHMEM_HUGE_DENY) + return 0;I don't quite get this, I don't think its the desirable behaviour. This is saying that if the top-level
shemem_enabledcontrol is set to 'deny', then all mTHP sizes, regardless of their control's setting are disabled?Quoted 3 lines
The anon controls don't work like that; you can set the top-level control to anything you like, but its only consumed if the per-size controls are inheriting it.
Quoted 8 lines
IMO, 'deny' option is not similar like 'never' option.
Quoted 3 lines
So for the deny case, wouldn't it be better to allow that as an option on all the per-size controls (and implicitly let it be inherrited from the top-level)?
From 'deny' option's semantics:
"disables huge onshm_mntand all mounts, for emergency use;"
Quoted 6 lines
Right. Today, tmpfs only supports PMD-sized THP. So for all per-size controls except the PMD-size control, 'deny' and 'never' would be the same practically speaking. For the PMD-size control, 'deny' would disable THP for both anon shmem and all tmpfs mounts, whereas 'never' would only disable THP for anon shmem. When tmpfs gains mTHP support, 'deny' in the other per-size controls would also disable that size for the tmpfs mounts.
Not really. We will not add 'deny' and 'force' testing option for each per-size mTHP control as suggested by Hugh in the previous MM meeting[1].
[1] https://lore.kernel.org/all/[email protected]/T/#u
Quoted 5 lines
I disagree with the current implementation where setting it up so that a top-level 'deny' overrides whatever is in the per-size controls simply because it's different to the model implemented for anon THP. That's my 2 cents. If nobody else agrees then that ok - I'll fix the above bug according to the current model.
I remember we have customers who use the 'deny' option to forcibly turn off the huge page option for all mounts (including shm_mnt). But if shmem/tmpfs uses mTHP and users set 'deny', they cannot force all huge orders off and they must also do 'echo never > /sys/kernel/mm/transparent_hugepage/hugepages-XXXkB/shmem_enabled' to disable all huge orders option, which breaks the previous user habits, IMHO.
In additon, I do not think this creates a difference with the anon mTHP model, as anon mTHP does not have these shmem special 'deny' and 'force' options for testing purposes. In my view, the current 'deny' option fulfills the semantic definition of 'For use in emergencies, to force the *huge* option off from all mounts'.
Quoted 23 lines
Quoted 23 lines
It is usually used for testing to shut down all huge pages from the old ages. Moreover, mTHP interfaces will be used as a huge order filter to support tmpfs, so I think it will make life easier to disable all huge orders for testing or emergency use, as well as to maintain the original semantics.>
Quoted 8 lines
Quoted 7 lines
+
+ /*
+ * Only allow inherit orders if the top-level value is 'force', which + * means non-PMD sized THP can not override 'huge' mount option now. + */
+ if (shmem_huge==SHMEM_HUGE_FORCE)
+ returnREAD_ONCE(huge_shmem_orders_inherit);Quoted 4 lines
I vaguely recall that we originally discussed that trying to set 'force' on the top level control while any per-size controls were set to 'inherit' would be an error, and trying to set 'force' on any per-size control except the PMD-size would be an error?
Right.
Quoted 3 lines
I don't really understand this logic. Shouldn't we just be looking at the per-size control settings (or the top-level control as a proxy for every per-size control that has 'inherit' set)?
Quoted 7 lines
Quoted 3 lines
‘force’ will apply the huge orders for anon shmem and tmpfs, so now we only allow pmd-mapped THP to be forced. We should not look at per-size control settings for tmpfs now (mTHP for tmpfs will be discussed in future).
But why not just maintain per-size controls and refactor tmpfs to just look at the PMD-size THP control for now. It can ignore the other sizes. That's much simpler and cleaner IMHO.
As I said above, 'force' and 'deny' option will not be added for per-size controls.
Ryan Roberts
reviewerRe: [PATCH v5 4/6] mm: shmem: add mTHP support for anonymous shmem
Quoted 132 lines · Baolin Wang
Quoted 110 lines · Ryan Roberts
Quoted 100 lines · Baolin Wang
Quoted 36 lines · Ryan Roberts
Quoted 24 lines · Baolin Wang
Commit
19eaf44954dfadds multi-size THP (mTHP) for anonymous pages, that can allow THP to be configured through the sysfs interface located at '/sys/kernel/mm/transparent_hugepage/hugepage-XXkb/enabled'.However, the anonymous shmem will ignore the anonymous mTHP rule configured through the sysfs interface, and can only use the PMD-mapped THP, that is not reasonable. Users expect to apply the mTHP rule for all anonymous pages, including the anonymous shmem, in order to enjoy the benefits of mTHP. For example, lower latency than PMD-mapped THP, smaller memory bloat than PMD-mapped THP, contiguous PTEs on ARM architecture to reduce TLB miss etc. In addition, the mTHP interfaces can be extended to support all shmem/tmpfs scenarios in the future, especially for the shmem
mmap()case.The primary strategy is similar to supporting anonymous mTHP. Introduce a new interface '/mm/transparent_hugepage/hugepage-XXkb/shmem_enabled', which can have almost the same values as the top-level '
/sys/kernel/mm/transparent_hugepage/shmem_enabled', with adding a new additional "inherit" option and dropping the testing options 'force' and 'deny'. By default all sizes will be set to "never" except PMD size, which is set to "inherit". This ensures backward compatibility with the anonymous shmem enabled of the top level, meanwhile also allows independent control of anonymous shmem enabled for each mTHP.Quoted 1 line
Signed-off-by Baolin Wang <[email protected]>[...]
Hi Baolin,
I'm currently trying to fix a bug where khugepaged is not started if only shmem is enabled for THP. See discussion at [1]. It's been broken like this forever.
Assuming anon and shmem THP are initially both disabled:
Quoted 8 lines
# echo never > /sys/kernel/mm/transparent_hugepage/shmem_enabled # echo never > /sys/kernel/mm/transparent_hugepage/enabled <khugepaged is stopped here>Then shemem THP is enabled:
# echo always > /sys/kernel/mm/transparent_hugepage/shmem_enabled <khugepaged is not started, this is a bug>Thanks Ryan. Yes, this is a real problem.
Quoted 46 lines
As part of investigating the fix, I stumbled upon this patch, which I remember reviewing an early version of but I've been out for a while and missed the latter versions. See below for comments and questions; the answers to which will
help me figure out how to fix the above...[1]
https://lore.kernel.org/linux-mm/[email protected]/Quoted 21 lines
+#ifdef
CONFIG_TRANSPARENT_HUGEPAGEmm/shmem.c +6
-0+static unsigned long shmem_allowable_huge_orders (struct inode *inode, + struct vm_area_struct *vma, pgoff_t index, + bool global_huge) +{ + unsigned long mask = READ_ONCE(huge_shmem_orders_always); + unsigned long within_size_orders =
READ_ONCE(huge_shmem_orders_within_size);
+ unsigned longvm_flags=vma->vm_flags;
+ /*
+ * Check all the (large) orders belowHPAGE_PMD_ORDER+ 1 that + * are enabled for this vma.
+ */
+ unsigned long orders =BIT(PMD_ORDER + 1)- 1;
+loff_ti_size;
+ int order;mm/shmem.c +5
-0+ + if ((vm_flags & VM_NOHUGEPAGE) || + test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags)) + return 0; +mm/shmem.c static gfp_t limit_gfp_mask(gfp_t huge_gfp, gfp_t limit_gfp) · line 1643
Quoted 10 lines
mm/shmem.c +10
-0+ /* If the hardware/firmware marked hugepage support disabled. */ + if (transparent_hugepage_flags & (1 << TRANSPARENT_HUGEPAGE_UNSUPPORTED)) + return 0; + + /* + * Following the 'deny' semantics of the top level, force the huge + * option off from all mounts. + */ + if (shmem_huge == SHMEM_HUGE_DENY) + return 0;I don't quite get this, I don't think its the desirable behaviour. This is saying that if the top-level
shemem_enabledcontrol is set to 'deny', then all mTHP sizes, regardless of their control's setting are disabled?Quoted 3 lines
The anon controls don't work like that; you can set the top-level control to anything you like, but its only consumed if the per-size controls are inheriting it.
Quoted 8 lines
IMO, 'deny' option is not similar like 'never' option.
Quoted 3 lines
So for the deny case, wouldn't it be better to allow that as an option on all the per-size controls (and implicitly let it be inherrited from the top-level)?
From 'deny' option's semantics:
"disables huge onshm_mntand all mounts, for emergency use;"Quoted 6 lines
Right. Today, tmpfs only supports PMD-sized THP. So for all per-size controls except the PMD-size control, 'deny' and 'never' would be the same practically speaking. For the PMD-size control, 'deny' would disable THP for both anon shmem and all tmpfs mounts, whereas 'never' would only disable THP for anon shmem. When tmpfs gains mTHP support, 'deny' in the other per-size controls would also disable that size for the tmpfs mounts.
Not really. We will not add 'deny' and 'force' testing option for each per-size mTHP control as suggested by Hugh in the previous MM meeting[1].
[1]
https://lore.kernel.org/all/[email protected]/T/#uQuoted 5 lines
I disagree with the current implementation where setting it up so that a top-level 'deny' overrides whatever is in the per-size controls simply because it's different to the model implemented for anon THP. That's my 2 cents. If nobody else agrees then that ok - I'll fix the above bug according to the current model.
Quoted 6 lines
I remember we have customers who use the 'deny' option to forcibly turn off the huge page option for all mounts (including
shm_mnt). But if shmem/tmpfs uses mTHP and users set 'deny', they cannot force all huge orders off and they must also do 'echo never >/sys/kernel/mm/transparent_hugepage/hugepages-XXXkB/shmem_enabled' to disable all huge orders option, which breaks the previous user habits, IMHO.
But if they have enabled mTHP for shmem in the first place, they must be mTHP aware. So its not unreasonable to expect them to call:
# echo deny > /sys/kernel/mm/transparent_hugepage/hugepages-XXXkB/shmem_enabled
That's the expectation for anon mTHP anyway. If the user enables mTHP then does:
# echo never > /sys/kernel/mm/transparent_hugepage/enabled
That *does not* disable all mTHP sizes. It only disables the sizes that have been set to 'inherit'. The model is that a size is only influenced by the top-level control if it has explicitly requested 'inherit'. But here you have this special (and weird in my opinion) case where some values of the top-level control apply unconditionally to the per-size control and some don't.
Quoted 6 lines
In additon, I do not think this creates a difference with the anon mTHP model, as anon mTHP does not have these shmem special 'deny' and 'force' options for testing purposes. In my view, the current 'deny' option fulfills the semantic definition of 'For use in emergencies, to force the *huge* option off from all mounts'.
OK. My opinion is logged :) But I'm not hearing anyone joining in support of that opinion. I have a better understanding of the intent of your model as a result of this discussion so thanks for that. As long as the documentation is clear on this behaviour, let's leave it as is.
Quoted 27 lines
Quoted 27 lines
Quoted 26 lines
It is usually used for testing to shut down all huge pages from the old ages. Moreover, mTHP interfaces will be used as a huge order filter to support tmpfs, so I think it will make life easier to disable all huge orders for testing or emergency use, as well as to maintain the original semantics.>
Quoted 8 lines
Quoted 7 lines
+
+ /*
+ * Only allow inherit orders if the top-level value is 'force', which + * means non-PMD sized THP can not override 'huge' mount option now. + */
+ if (shmem_huge==SHMEM_HUGE_FORCE)
+ returnREAD_ONCE(huge_shmem_orders_inherit);Quoted 4 lines
I vaguely recall that we originally discussed that trying to set 'force' on the top level control while any per-size controls were set to 'inherit' would be an error, and trying to set 'force' on any per-size control except the PMD-size would be an error?
Right.
Quoted 3 lines
I don't really understand this logic. Shouldn't we just be looking at the per-size control settings (or the top-level control as a proxy for every per-size control that has 'inherit' set)?
‘force’ will apply the huge orders for anon shmem and tmpfs, so now we only allow pmd-mapped THP to be forced. We should not look at per-size control settings for tmpfs now (mTHP for tmpfs will be discussed in future).
Quoted 5 lines
Quoted 3 lines
But why not just maintain per-size controls and refactor tmpfs to just look at the PMD-size THP control for now. It can ignore the other sizes. That's much simpler and cleaner IMHO.
As I said above, 'force' and 'deny' option will not be added for per-size controls.
Understood.
Thanks,
Ryan
Re: [PATCH v5 4/6] mm: shmem: add mTHP support for anonymous shmem
In reply to Baolin WangHi Bao lin,
Quoted 25 lines · Baolin Wang
Quoted 11 lines
Quoted 10 lines
+
+ /*
+ * Only allow inherit orders if the top-level value is 'force', which
+ * means non-PMD sized THP can not override 'huge' mount option now.
+ */
+ if (shmem_huge==SHMEM_HUGE_FORCE)
+ returnREAD_ONCE(huge_shmem_orders_inherit);Quoted 4 lines
I vaguely recall that we originally discussed that trying to set 'force' on the top level control while any per-size controls were set to 'inherit' would be an error, and trying to set 'force' on any per-size control except the PMD-size would be an error?
Right.
Quoted 3 lines
I don't really understand this logic. Shouldn't we just be looking at the per-size control settings (or the top-level control as a proxy for every per-size control that has 'inherit' set)?
‘force’ will apply the huge orders for anon shmem and tmpfs, so now we only allow pmd-mapped THP to be forced. We should not look at per-size control settings for tmpfs now (mTHP for tmpfs will be discussed in future).
Quoted 1 line
Quoted 8 lines
Quoted 6 lines
Then for tmpfs, which doesn't support non-PMD-sizes yet, we just always use the PMD-size control for decisions.
I'm also really struggling with the concept of
shmem_is_huge()existing along sideshmem_allowable_huge_orders(). Surely this needs to all be refactored intoshmem_allowable_huge_orders()?I understood. But now they serve different purposes:
shmem_is_huge()will be used to check the huge orders for the top level, for *tmpfs* and anon shmem; whereasshmem_allowable_huge_orders()will only be used to check the per-size huge orders for anon shmem (excluding tmpfs now). However, as I plan to add mTHP support for tmpfs, I think we can perform some cleanups.
Please count me in, I'd be happy to contribute to the cleanup and enhancement process if I can.
Thanks,
Bang
Baolin Wang
reviewerRe: [PATCH v5 4/6] mm: shmem: add mTHP support for anonymous shmem
Quoted 31 lines · Bang Li
Hi Bao lin,
Quoted 28 lines · Baolin Wang
Quoted 11 lines
Quoted 10 lines
+
+ /*
+ * Only allow inherit orders if the top-level value is 'force', which
+ * means non-PMD sized THP can not override 'huge' mount option now.
+ */
+ if (shmem_huge==SHMEM_HUGE_FORCE)
+ returnREAD_ONCE(huge_shmem_orders_inherit);Quoted 4 lines
I vaguely recall that we originally discussed that trying to set 'force' on the top level control while any per-size controls were set to 'inherit' would be an error, and trying to set 'force' on any per-size control except the PMD-size would be an error?
Right.
Quoted 3 lines
I don't really understand this logic. Shouldn't we just be looking at the per-size control settings (or the top-level control as a proxy for every per-size control that has 'inherit' set)?
‘force’ will apply the huge orders for anon shmem and tmpfs, so now we only allow pmd-mapped THP to be forced. We should not look at per-size control settings for tmpfs now (mTHP for tmpfs will be discussed in future).
Quoted 4 lines
Then for tmpfs, which doesn't support non-PMD-sizes yet, we just always use the PMD-size control for decisions.
Quoted 8 lines
Quoted 5 lines
Quoted 3 lines
I'm also really struggling with the concept of
shmem_is_huge()existing along sideshmem_allowable_huge_orders(). Surely this needs to all be refactored intoshmem_allowable_huge_orders()?I understood. But now they serve different purposes:
shmem_is_huge()will be used to check the huge orders for the top level, for *tmpfs* and anon shmem; whereasshmem_allowable_huge_orders()will only be used to check the per-size huge orders for anon shmem (excluding tmpfs now). However, as I plan to add mTHP support for tmpfs, I think we can perform some cleanups.Please count me in, I'd be happy to contribute to the cleanup and enhancement process if I can.
Good. If you have time, I think you can look at the shmem khugepaged issue from the previous discussion [1], which I don't have time to look at now.
"
(3) khugepaged
khugepaged needs to handle larger folios properly as well. Until fixed, using smaller THP sizes as fallback might prohibit collapsing a PMD-sized THP later. But really, khugepaged needs to be fixed to handle that.
"
Ryan Roberts
reviewerRe: [PATCH v5 4/6] mm: shmem: add mTHP support for anonymous shmem
Quoted 50 lines · Baolin Wang
Quoted 35 lines · Bang Li
Hi Bao lin,
Quoted 32 lines · Baolin Wang
Quoted 9 lines
Quoted 8 lines
+
+ /*
+ * Only allow inherit orders if the top-level value is 'force', which + * means non-PMD sized THP can not override 'huge' mount option now. + */
+ if (shmem_huge==SHMEM_HUGE_FORCE)
+ returnREAD_ONCE(huge_shmem_orders_inherit);Quoted 4 lines
I vaguely recall that we originally discussed that trying to set 'force' on the top level control while any per-size controls were set to 'inherit' would be an error, and trying to set 'force' on any per-size control except the PMD-size would be an error?
Right.
Quoted 3 lines
I don't really understand this logic. Shouldn't we just be looking at the per-size control settings (or the top-level control as a proxy for every per-size control that has 'inherit' set)?
‘force’ will apply the huge orders for anon shmem and tmpfs, so now we only allow pmd-mapped THP to be forced. We should not look at per-size control settings for tmpfs now (mTHP for tmpfs will be discussed in future).
Quoted 7 lines
Then for tmpfs, which doesn't support non-PMD-sizes yet, we just always use the PMD-size control for decisions.
I'm also really struggling with the concept of
shmem_is_huge()existing along sideshmem_allowable_huge_orders(). Surely this needs to all be refactored intoshmem_allowable_huge_orders()?Quoted 8 lines
Quoted 5 lines
I understood. But now they serve different purposes:
shmem_is_huge()will be used to check the huge orders for the top level, for *tmpfs* and anon shmem; whereasshmem_allowable_huge_orders()will only be used to check the per-size huge orders for anon shmem (excluding tmpfs now). However, as I plan to add mTHP support for tmpfs, I think we can perform some cleanups.Please count me in, I'd be happy to contribute to the cleanup and enhancement process if I can.
Good. If you have time, I think you can look at the shmem khugepaged issue from the previous discussion [1], which I don't have time to look at now.
Quoted 8 lines
"
(3) khugepagedkhugepaged needs to handle larger folios properly as well. Until fixed, using smaller THP sizes as fallback might prohibit collapsing a PMD-sized THP later. But really, khugepaged needs to be fixed to handle that.
"
khugepaged can already collapse "folios of any order less then PMD-size" to PMD-size, for anon memory. Infact I modified the selftest to be able to test that in commit 9f0704eae8a4 ("selftests/mm/khugepaged: enlighten for multi-size THP"). I'd be surprised if khugepaged can't alreay handle the same for shmem? Although the test will definitely want to be extended to test it.
Thanks,
Ryan
Quoted 3 lines
Baolin Wang
reviewerRe: [PATCH v5 4/6] mm: shmem: add mTHP support for anonymous shmem
Quoted 60 lines · Ryan Roberts
Quoted 50 lines · Baolin Wang
Quoted 35 lines · Bang Li
Hi Bao lin,
Quoted 32 lines · Baolin Wang
Quoted 9 lines
Quoted 8 lines
+
+ /*
+ * Only allow inherit orders if the top-level value is 'force', which + * means non-PMD sized THP can not override 'huge' mount option now. + */
+ if (shmem_huge==SHMEM_HUGE_FORCE)
+ returnREAD_ONCE(huge_shmem_orders_inherit);Quoted 4 lines
I vaguely recall that we originally discussed that trying to set 'force' on the top level control while any per-size controls were set to 'inherit' would be an error, and trying to set 'force' on any per-size control except the PMD-size would be an error?
Right.
Quoted 3 lines
I don't really understand this logic. Shouldn't we just be looking at the per-size control settings (or the top-level control as a proxy for every per-size control that has 'inherit' set)?
‘force’ will apply the huge orders for anon shmem and tmpfs, so now we only allow pmd-mapped THP to be forced. We should not look at per-size control settings for tmpfs now (mTHP for tmpfs will be discussed in future).
Quoted 7 lines
Then for tmpfs, which doesn't support non-PMD-sizes yet, we just always use the PMD-size control for decisions.
I'm also really struggling with the concept of
shmem_is_huge()existing along sideshmem_allowable_huge_orders(). Surely this needs to all be refactored intoshmem_allowable_huge_orders()?Quoted 8 lines
Quoted 5 lines
I understood. But now they serve different purposes:
shmem_is_huge()will be used to check the huge orders for the top level, for *tmpfs* and anon shmem; whereasshmem_allowable_huge_orders()will only be used to check the per-size huge orders for anon shmem (excluding tmpfs now). However, as I plan to add mTHP support for tmpfs, I think we can perform some cleanups.Please count me in, I'd be happy to contribute to the cleanup and enhancement process if I can.
Good. If you have time, I think you can look at the shmem khugepaged issue from the previous discussion [1], which I don't have time to look at now.
Quoted 8 lines
"
(3) khugepagedkhugepaged needs to handle larger folios properly as well. Until fixed, using smaller THP sizes as fallback might prohibit collapsing a PMD-sized THP later. But really, khugepaged needs to be fixed to handle that.
"
Quoted 4 lines
khugepaged can already collapse "folios of any order less then PMD-size" to PMD-size, for anon memory. Infact I modified the selftest to be able to test that in commit
9f0704eae8a4("selftests/mm/khugepaged: enlighten for multi-size THP"). I'd be surprised if khugepaged can't alreay handle the same for shmem?
I did not test this, but from the comment in hpage_collapse_scan_file(), seems that compacting small mTHP into a single PMD-mapped THP is not supported yet.
/*
* TODO: khugepaged should compact smaller compound pages
* into a PMD sized page
*/
if (folio_test_large(folio)) {
result = folio_order(folio) == HPAGE_PMD_ORDER &&
folio->index == start
/* Maybe PMD-mapped */
? SCAN_PTE_MAPPED_HUGEPAGE
: SCAN_PAGE_COMPOUND;
/*
* For SCAN_PTE_MAPPED_HUGEPAGE, further processing
* by the caller won't touch the page cache, and so
* it's safe to skip LRU and refcount checks before
* returning.
*/
break;
}Quoted 1 line
Although the test will definitely want to be extended to test it.
Right.
Ryan Roberts
reviewerRe: [PATCH v5 4/6] mm: shmem: add mTHP support for anonymous shmem
Quoted 85 lines · Baolin Wang
Quoted 64 lines · Ryan Roberts
Quoted 54 lines · Baolin Wang
Quoted 39 lines · Bang Li
Hi Bao lin,
Quoted 36 lines · Baolin Wang
Quoted 9 lines
Quoted 8 lines
+
+ /*
+ * Only allow inherit orders if the top-level value is 'force', which + * means non-PMD sized THP can not override 'huge' mount option now. + */
+ if (shmem_huge==SHMEM_HUGE_FORCE)
+ returnREAD_ONCE(huge_shmem_orders_inherit);Quoted 6 lines
I vaguely recall that we originally discussed that trying to set 'force' on the
top level control while any per-size controls were set to 'inherit' would be an
error, and trying to set 'force' on any per-size control except the PMD-size would be an error?Right.
Quoted 3 lines
I don't really understand this logic. Shouldn't we just be looking at the per-size control settings (or the top-level control as a proxy for every per-size control that has 'inherit' set)?
‘force’ will apply the huge orders for anon shmem and tmpfs, so now we only allow pmd-mapped THP to be forced. We should not look at per-size control settings for tmpfs now (mTHP for tmpfs will be discussed in future).
Quoted 9 lines
Then for tmpfs, which doesn't support non-PMD-sizes yet, we just always use the
PMD-size control for decisions.I'm also really struggling with the concept of
shmem_is_huge()existing along sideshmem_allowable_huge_orders(). Surely this needs to all be refactored intoshmem_allowable_huge_orders()?Quoted 8 lines
Quoted 5 lines
I understood. But now they serve different purposes:
shmem_is_huge()will be used to check the huge orders for the top level, for *tmpfs* and anon shmem; whereasshmem_allowable_huge_orders()will only be used to check the per-size huge orders for anon shmem (excluding tmpfs now). However, as I plan to add mTHP support for tmpfs, I think we can perform some cleanups.Please count me in, I'd be happy to contribute to the cleanup and enhancement process if I can.
Good. If you have time, I think you can look at the shmem khugepaged issue from the previous discussion [1], which I don't have time to look at now.
Quoted 8 lines
"
(3) khugepagedkhugepaged needs to handle larger folios properly as well. Until fixed, using smaller THP sizes as fallback might prohibit collapsing a PMD-sized THP later. But really, khugepaged needs to be fixed to handle that.
"Quoted 4 lines
khugepaged can already collapse "folios of any order less then PMD-size" to PMD-size, for anon memory. Infact I modified the selftest to be able to test that in commit
9f0704eae8a4("selftests/mm/khugepaged: enlighten for multi-size THP"). I'd be surprised if khugepaged can't alreay handle the same for shmem?I did not test this, but from the comment in
hpage_collapse_scan_file(), seems that compacting small mTHP into a single PMD-mapped THP is not supported yet./*
* TODO: khugepaged should compact smaller compound pages
* into a PMD sized page
*/
if (folio_test_large(folio)) {
result =folio_order(folio)==HPAGE_PMD_ORDER&&folio->index== start
/* Maybe PMD-mapped */
?SCAN_PTE_MAPPED_HUGEPAGE
:SCAN_PAGE_COMPOUND;
Quoted 8 lines
/*
* ForSCAN_PTE_MAPPED_HUGEPAGE, further processing
* by the caller won't touch the page cache, and so
* it's safe to skip LRU and refcount checks before
* returning.
*/
break;
}
OK, so the functionality is missing just for shmem/file-backed collapse. Got it. Sorry for the noise.
Quoted 4 lines
Quoted 2 lines
Although the test will definitely want to be extended to test it.
Right.
Baolin Wang
reviewerPATCH v5 5/6 mm: shmem: add mTHP size alignment in shmem_get_unmapped_area
In reply to Baolin WangAlthough the top-level hugepage allocation can be turned off, anonymous shmem can still use mTHP by configuring the sysfs interface located at '/sys/kernel/mm/transparent_hugepage/hugepage-XXkb/shmem_enabled'. Therefore, add alignment for mTHP size to provide a suitable alignment address in shmem_get_unmapped_area().
--- mm/shmem.c | 40 +++++++++++++++++++++++++++++++--------- 1 file changed, 31 insertions(+), 9 deletions(-)
mm/shmem.c +31 -9
@@ -2394,6 +2394,7 @@ unsigned long shmem_get_unmapped_area(struct file *file,unsigned long inflated_len;unsigned long inflated_addr;unsigned long inflated_offset; +unsigned long hpage_size;if (len >TASK_SIZE )return -ENOMEM ;
@@ -2412,8 +2413,6 @@ unsigned long shmem_get_unmapped_area(struct file *file,if (shmem_huge ==SHMEM_HUGE_DENY )return addr; -if (len <HPAGE_PMD_SIZE ) -return addr;if (flags &MAP_FIXED )return addr;/*
@@ -2425,8 +2424,11 @@ unsigned long shmem_get_unmapped_area(struct file *file,if (uaddr == addr)return addr; + hpage_size =HPAGE_PMD_SIZE ;if (shmem_huge !=SHMEM_HUGE_FORCE ) {struct super_block *sb; +unsigned long __maybe_unused hpage_orders; +int order =0 ;if (file) {VM_BUG_ON (file->f_op != &shmem_file_operations);
@@ -2439,18 +2441,38 @@ unsigned long shmem_get_unmapped_area(struct file *file,if (IS_ERR (shm_mnt))return addr; sb = shm_mnt->mnt_sb; + +/* +* Find the highest mTHP order used for anonymous shmem to +* provide a suitable alignment address. +*/ +#ifdef CONFIG_TRANSPARENT_HUGEPAGE + hpage_orders =READ_ONCE (huge_shmem_orders_always); + hpage_orders |=READ_ONCE (huge_shmem_orders_within_size); + hpage_orders |=READ_ONCE (huge_shmem_orders_madvise); +if (SHMEM_SB (sb)->huge !=SHMEM_HUGE_NEVER ) + hpage_orders |=READ_ONCE (huge_shmem_orders_inherit); + +if (hpage_orders >0 ) { + order =highest_order (hpage_orders); + hpage_size =PAGE_SIZE << order; + } +#endif } -if (SHMEM_SB (sb)->huge ==SHMEM_HUGE_NEVER ) +if (SHMEM_SB (sb)->huge ==SHMEM_HUGE_NEVER && !order)return addr; } - offset = (pgoff <<PAGE_SHIFT ) & (HPAGE_PMD_SIZE -1 ); -if (offset && offset + len <2 *HPAGE_PMD_SIZE ) +if (len < hpage_size) +return addr; + + offset = (pgoff <<PAGE_SHIFT ) & (hpage_size -1 ); +if (offset && offset + len <2 * hpage_size)return addr; -if ((addr & (HPAGE_PMD_SIZE -1 )) == offset) +if ((addr & (hpage_size -1 )) == offset)return addr; - inflated_len = len +HPAGE_PMD_SIZE -PAGE_SIZE ; + inflated_len = len + hpage_size -PAGE_SIZE ;if (inflated_len >TASK_SIZE )return addr;if (inflated_len < len)
@@ -2463,10 +2485,10 @@ unsigned long shmem_get_unmapped_area(struct file *file,if (inflated_addr & ~PAGE_MASK )return addr; - inflated_offset = inflated_addr & (HPAGE_PMD_SIZE -1 ); + inflated_offset = inflated_addr & (hpage_size -1 ); inflated_addr += offset - inflated_offset;if (inflated_offset > offset) - inflated_addr +=HPAGE_PMD_SIZE ; + inflated_addr += hpage_size;if (inflated_addr >TASK_SIZE - len)return addr;
Signature · 1 line
2.39.3
Baolin Wang
reviewerPATCH v5 6/6 mm: shmem: add mTHP counters for anonymous shmem
In reply to Baolin WangAdd mTHP counters for anonymous shmem.
--- include/linux/huge_mm.h | 3 +++ mm/huge_memory.c | 6 ++++++ mm/shmem.c | 18 +++++++++++++++--- 3 files changed, 24 insertions(+), 3 deletions(-)
include/linux/huge_mm.h +3 -0
@@ -281,6 +281,9 @@ enum mthp_stat_item {MTHP_STAT_ANON_FAULT_FALLBACK_CHARGE ,MTHP_STAT_SWPOUT ,MTHP_STAT_SWPOUT_FALLBACK , +MTHP_STAT_FILE_ALLOC , +MTHP_STAT_FILE_FALLBACK , +MTHP_STAT_FILE_FALLBACK_CHARGE , __MTHP_STAT_COUNT };
mm/huge_memory.c +6 -0
@@ -555,6 +555,9 @@ DEFINE_MTHP_STAT_ATTR(anon_fault_fallback, MTHP_STAT_ANON_FAULT_FALLBACK);DEFINE_MTHP_STAT_ATTR (anon_fault_fallback_charge,MTHP_STAT_ANON_FAULT_FALLBACK_CHARGE );DEFINE_MTHP_STAT_ATTR (swpout,MTHP_STAT_SWPOUT );DEFINE_MTHP_STAT_ATTR (swpout_fallback,MTHP_STAT_SWPOUT_FALLBACK ); +DEFINE_MTHP_STAT_ATTR (file_alloc,MTHP_STAT_FILE_ALLOC ); +DEFINE_MTHP_STAT_ATTR (file_fallback,MTHP_STAT_FILE_FALLBACK ); +DEFINE_MTHP_STAT_ATTR (file_fallback_charge,MTHP_STAT_FILE_FALLBACK_CHARGE );static struct attribute *stats_attrs[] = { &anon_fault_alloc_attr.attr,
@@ -562,6 +565,9 @@ static struct attribute *stats_attrs[] = { &anon_fault_fallback_charge_attr.attr, &swpout_attr.attr, &swpout_fallback_attr.attr, + &file_alloc_attr.attr, + &file_fallback_attr.attr, + &file_fallback_charge_attr.attr, NULL, };
mm/shmem.c +15 -3
@@ -1773,6 +1773,9 @@ static struct folio *shmem_alloc_and_add_folio(struct vm_fault *vmf,if (pages ==HPAGE_PMD_NR )count_vm_event (THP_FILE_FALLBACK ); +#ifdef CONFIG_TRANSPARENT_HUGEPAGE +count_mthp_stat (order,MTHP_STAT_FILE_FALLBACK ); +#endif order =next_order (&suitable_orders, order); } }else {
@@ -1792,9 +1795,15 @@ static struct folio *shmem_alloc_and_add_folio(struct vm_fault *vmf,if (xa_find (&mapping->i_pages, &index, index + pages -1 ,XA_PRESENT )) { error = -EEXIST ; - }else if (pages ==HPAGE_PMD_NR ) { -count_vm_event (THP_FILE_FALLBACK ); -count_vm_event (THP_FILE_FALLBACK_CHARGE ); + }else if (pages >1 ) { +if (pages ==HPAGE_PMD_NR ) { +count_vm_event (THP_FILE_FALLBACK ); +count_vm_event (THP_FILE_FALLBACK_CHARGE ); + } +#ifdef CONFIG_TRANSPARENT_HUGEPAGE +count_mthp_stat (folio_order (folio),MTHP_STAT_FILE_FALLBACK ); +count_mthp_stat (folio_order (folio),MTHP_STAT_FILE_FALLBACK_CHARGE ); +#endif } goto unlock; }
@@ -2168,6 +2177,9 @@ static int shmem_get_folio_gfp(struct inode *inode, pgoff_t index,if (!IS_ERR (folio)) {if (folio_test_pmd_mappable (folio))count_vm_event (THP_FILE_ALLOC ); +#ifdef CONFIG_TRANSPARENT_HUGEPAGE +count_mthp_stat (folio_order (folio),MTHP_STAT_FILE_ALLOC ); +#endif goto alloced; }if (PTR_ERR (folio) == -EEXIST )
Signature · 1 line
2.39.3
Re: [PATCH v5 6/6] mm: shmem: add mTHP counters for anonymous shmem
Hi Baolin,
Quoted 53 lines · Baolin Wang
Add mTHP counters for anonymous shmem.
Signed-off-by Baolin Wang <[email protected]>--- include/linux/huge_mm.h | 3 +++ mm/huge_memory.c | 6 ++++++ mm/shmem.c | 18 +++++++++++++++--- 3 files changed, 24 insertions(+), 3 deletions(-)include/linux/huge_mm.h +3
-0@@ -281,6 +281,9 @@ enum mthp_stat_item {MTHP_STAT_ANON_FAULT_FALLBACK_CHARGE ,MTHP_STAT_SWPOUT ,MTHP_STAT_SWPOUT_FALLBACK , +MTHP_STAT_FILE_ALLOC , +MTHP_STAT_FILE_FALLBACK , +MTHP_STAT_FILE_FALLBACK_CHARGE , __MTHP_STAT_COUNT };mm/huge_memory.c +6
-0@@ -555,6 +555,9 @@ DEFINE_MTHP_STAT_ATTR(anon_fault_fallback, MTHP_STAT_ANON_FAULT_FALLBACK);DEFINE_MTHP_STAT_ATTR (anon_fault_fallback_charge,MTHP_STAT_ANON_FAULT_FALLBACK_CHARGE );DEFINE_MTHP_STAT_ATTR (swpout,MTHP_STAT_SWPOUT );DEFINE_MTHP_STAT_ATTR (swpout_fallback,MTHP_STAT_SWPOUT_FALLBACK ); +DEFINE_MTHP_STAT_ATTR (file_alloc,MTHP_STAT_FILE_ALLOC ); +DEFINE_MTHP_STAT_ATTR (file_fallback,MTHP_STAT_FILE_FALLBACK ); +DEFINE_MTHP_STAT_ATTR (file_fallback_charge,MTHP_STAT_FILE_FALLBACK_CHARGE );static struct attribute *stats_attrs[] = { &anon_fault_alloc_attr.attr,@@ -562,6 +565,9 @@ static struct attribute *stats_attrs[] = { &anon_fault_fallback_charge_attr.attr, &swpout_attr.attr, &swpout_fallback_attr.attr, + &file_alloc_attr.attr, + &file_fallback_attr.attr, + &file_fallback_charge_attr.attr, NULL, };mm/shmem.c +0
-0
mm/shmem.c static struct folio *shmem_alloc_and_add_folio(struct vm_fault *vmf, · line 1778
Quoted 6 lines
mm/shmem.c +3
-0@@ -1773,6 +1773,9 @@ static struct folio *shmem_alloc_and_add_folio(struct vm_fault *vmf,if (pages ==HPAGE_PMD_NR )count_vm_event (THP_FILE_FALLBACK ); +#ifdef CONFIG_TRANSPARENT_HUGEPAGE +count_mthp_stat (order,MTHP_STAT_FILE_FALLBACK ); +#endif
Using the conditional compilation directives here is a bit weird :) Would there be any issues if we were to drop them?
Since THP_FILE_FALLBACK is working as expected, MTHP_STAT_FILE_FALLBACK should work as well without the conditional compilation directives, IIUC.
Thanks,
Lance
Quoted 35 lines
mm/shmem.c +12
-4order =next_order (&suitable_orders, order); } }else {@@ -1792,9 +1795,15 @@ static struct folio *shmem_alloc_and_add_folio(struct vm_fault *vmf,if (xa_find (&mapping->i_pages, &index, index + pages -1 ,XA_PRESENT )) { error = -EEXIST ; - }else if (pages ==HPAGE_PMD_NR ) { -count_vm_event (THP_FILE_FALLBACK ); -count_vm_event (THP_FILE_FALLBACK_CHARGE ); + }else if (pages >1 ) { +if (pages ==HPAGE_PMD_NR ) { +count_vm_event (THP_FILE_FALLBACK ); +count_vm_event (THP_FILE_FALLBACK_CHARGE ); + } +#ifdef CONFIG_TRANSPARENT_HUGEPAGE +count_mthp_stat (folio_order (folio),MTHP_STAT_FILE_FALLBACK ); +count_mthp_stat (folio_order (folio),MTHP_STAT_FILE_FALLBACK_CHARGE ); +#endif } goto unlock; }@@ -2168,6 +2177,9 @@ static int shmem_get_folio_gfp(struct inode *inode, pgoff_t index,if (!IS_ERR (folio)) {if (folio_test_pmd_mappable (folio))count_vm_event (THP_FILE_ALLOC ); +#ifdef CONFIG_TRANSPARENT_HUGEPAGE +count_mthp_stat (folio_order (folio),MTHP_STAT_FILE_ALLOC ); +#endif goto alloced; }if (PTR_ERR (folio) == -EEXIST ) --2.39.3
Baolin Wang
reviewerRe: [PATCH v5 6/6] mm: shmem: add mTHP counters for anonymous shmem
Quoted 58 lines · Lance Yang
Hi Baolin,
Quoted 54 lines · Baolin Wang
Add mTHP counters for anonymous shmem.
Signed-off-by Baolin Wang <[email protected]>--- include/linux/huge_mm.h | 3 +++ mm/huge_memory.c | 6 ++++++ mm/shmem.c | 18 +++++++++++++++--- 3 files changed, 24 insertions(+), 3 deletions(-)include/linux/huge_mm.h +3
-0@@ -281,6 +281,9 @@ enum mthp_stat_item {MTHP_STAT_ANON_FAULT_FALLBACK_CHARGE ,MTHP_STAT_SWPOUT ,MTHP_STAT_SWPOUT_FALLBACK , +MTHP_STAT_FILE_ALLOC , +MTHP_STAT_FILE_FALLBACK , +MTHP_STAT_FILE_FALLBACK_CHARGE , __MTHP_STAT_COUNT };mm/huge_memory.c +6
-0@@ -555,6 +555,9 @@ DEFINE_MTHP_STAT_ATTR(anon_fault_fallback, MTHP_STAT_ANON_FAULT_FALLBACK);DEFINE_MTHP_STAT_ATTR (anon_fault_fallback_charge,MTHP_STAT_ANON_FAULT_FALLBACK_CHARGE );DEFINE_MTHP_STAT_ATTR (swpout,MTHP_STAT_SWPOUT );DEFINE_MTHP_STAT_ATTR (swpout_fallback,MTHP_STAT_SWPOUT_FALLBACK ); +DEFINE_MTHP_STAT_ATTR (file_alloc,MTHP_STAT_FILE_ALLOC ); +DEFINE_MTHP_STAT_ATTR (file_fallback,MTHP_STAT_FILE_FALLBACK ); +DEFINE_MTHP_STAT_ATTR (file_fallback_charge,MTHP_STAT_FILE_FALLBACK_CHARGE );static struct attribute *stats_attrs[] = { &anon_fault_alloc_attr.attr,@@ -562,6 +565,9 @@ static struct attribute *stats_attrs[] = { &anon_fault_fallback_charge_attr.attr, &swpout_attr.attr, &swpout_fallback_attr.attr, + &file_alloc_attr.attr, + &file_fallback_attr.attr, + &file_fallback_charge_attr.attr, NULL, };mm/shmem.c +0
-0@@ -1773,6 +1773,9 @@ static struct folio *shmem_alloc_and_add_folio(struct vm_fault *vmf,
Quoted 8 lines
mm/shmem.c static struct folio *shmem_alloc_and_add_folio(struct vm_fault *vmf, · line 1778
Quoted 5 lines
mm/shmem.c +3
-0if (pages ==HPAGE_PMD_NR )count_vm_event (THP_FILE_FALLBACK ); +#ifdef CONFIG_TRANSPARENT_HUGEPAGE +count_mthp_stat (order,MTHP_STAT_FILE_FALLBACK ); +#endif Using the conditional compilation directives here is a bit weird :) Would there be any issues if we were to drop them?
Will cause building errors if CONFIG_TRANSPARENT_HUGEPAGE is not enabled.
Quoted 3 lines
Since
THP_FILE_FALLBACKis working as expected,MTHP_STAT_FILE_FALLBACKshould work as well without the conditional compilation directives, IIUC.
No, you should take a look at how count_mthp_stat() is defined :)
Re: [PATCH v5 6/6] mm: shmem: add mTHP counters for anonymous shmem
Quoted 71 lines · Baolin Wang
Quoted 58 lines · Lance Yang
Hi Baolin,
Quoted 54 lines · Baolin Wang
Add mTHP counters for anonymous shmem.
Signed-off-by Baolin Wang <[email protected]>--- include/linux/huge_mm.h | 3 +++ mm/huge_memory.c | 6 ++++++ mm/shmem.c | 18 +++++++++++++++--- 3 files changed, 24 insertions(+), 3 deletions(-)include/linux/huge_mm.h +3
-0@@ -281,6 +281,9 @@ enum mthp_stat_item {MTHP_STAT_ANON_FAULT_FALLBACK_CHARGE ,MTHP_STAT_SWPOUT ,MTHP_STAT_SWPOUT_FALLBACK , +MTHP_STAT_FILE_ALLOC , +MTHP_STAT_FILE_FALLBACK , +MTHP_STAT_FILE_FALLBACK_CHARGE , __MTHP_STAT_COUNT };mm/huge_memory.c +6
-0@@ -555,6 +555,9 @@ DEFINE_MTHP_STAT_ATTR(anon_fault_fallback, MTHP_STAT_ANON_FAULT_FALLBACK);DEFINE_MTHP_STAT_ATTR (anon_fault_fallback_charge,MTHP_STAT_ANON_FAULT_FALLBACK_CHARGE );DEFINE_MTHP_STAT_ATTR (swpout,MTHP_STAT_SWPOUT );DEFINE_MTHP_STAT_ATTR (swpout_fallback,MTHP_STAT_SWPOUT_FALLBACK ); +DEFINE_MTHP_STAT_ATTR (file_alloc,MTHP_STAT_FILE_ALLOC ); +DEFINE_MTHP_STAT_ATTR (file_fallback,MTHP_STAT_FILE_FALLBACK ); +DEFINE_MTHP_STAT_ATTR (file_fallback_charge,MTHP_STAT_FILE_FALLBACK_CHARGE );static struct attribute *stats_attrs[] = { &anon_fault_alloc_attr.attr,@@ -562,6 +565,9 @@ static struct attribute *stats_attrs[] = { &anon_fault_fallback_charge_attr.attr, &swpout_attr.attr, &swpout_fallback_attr.attr, + &file_alloc_attr.attr, + &file_fallback_attr.attr, + &file_fallback_charge_attr.attr, NULL, };mm/shmem.c +0
-0@@ -1773,6 +1773,9 @@ static struct folio *shmem_alloc_and_add_folio(struct vm_fault *vmf,Quoted 8 lines
mm/shmem.c static struct folio *shmem_alloc_and_add_folio(struct vm_fault *vmf, · line 1778
Quoted 5 lines
mm/shmem.c +3
-0if (pages ==HPAGE_PMD_NR )count_vm_event (THP_FILE_FALLBACK ); +#ifdef CONFIG_TRANSPARENT_HUGEPAGE +count_mthp_stat (order,MTHP_STAT_FILE_FALLBACK ); +#endif Using the conditional compilation directives here is a bit weird :) Would there be any issues if we were to drop them?
Quoted 1 line
Will cause building errors if
CONFIG_TRANSPARENT_HUGEPAGEis not enabled.
Sorry, I got it wrong :p
Quoted 6 lines
Quoted 4 lines
Since
THP_FILE_FALLBACKis working as expected,MTHP_STAT_FILE_FALLBACKshould work as well without the conditional compilation directives, IIUC.No, you should take a look at how
count_mthp_stat()is defined :)
You're correct. count_mthp_stat() does cause a compilation error without them when CONFIG_TRANSPARENT_HUGEPAGE is not defined.
Thanks,
Lance
Re: [PATCH v5 6/6] mm: shmem: add mTHP counters for anonymous shmem
In reply to Baolin WangQuoted 66 lines · Baolin Wang
Add mTHP counters for anonymous shmem.
Signed-off-by Baolin Wang <[email protected]>--- include/linux/huge_mm.h | 3 +++ mm/huge_memory.c | 6 ++++++ mm/shmem.c | 18 +++++++++++++++--- 3 files changed, 24 insertions(+), 3 deletions(-)include/linux/huge_mm.h +3
-0@@ -281,6 +281,9 @@ enum mthp_stat_item {MTHP_STAT_ANON_FAULT_FALLBACK_CHARGE ,MTHP_STAT_SWPOUT ,MTHP_STAT_SWPOUT_FALLBACK , +MTHP_STAT_FILE_ALLOC , +MTHP_STAT_FILE_FALLBACK , +MTHP_STAT_FILE_FALLBACK_CHARGE , __MTHP_STAT_COUNT };mm/huge_memory.c +6
-0@@ -555,6 +555,9 @@ DEFINE_MTHP_STAT_ATTR(anon_fault_fallback, MTHP_STAT_ANON_FAULT_FALLBACK);DEFINE_MTHP_STAT_ATTR (anon_fault_fallback_charge,MTHP_STAT_ANON_FAULT_FALLBACK_CHARGE );DEFINE_MTHP_STAT_ATTR (swpout,MTHP_STAT_SWPOUT );DEFINE_MTHP_STAT_ATTR (swpout_fallback,MTHP_STAT_SWPOUT_FALLBACK ); +DEFINE_MTHP_STAT_ATTR (file_alloc,MTHP_STAT_FILE_ALLOC ); +DEFINE_MTHP_STAT_ATTR (file_fallback,MTHP_STAT_FILE_FALLBACK ); +DEFINE_MTHP_STAT_ATTR (file_fallback_charge,MTHP_STAT_FILE_FALLBACK_CHARGE );static struct attribute *stats_attrs[] = { &anon_fault_alloc_attr.attr,@@ -562,6 +565,9 @@ static struct attribute *stats_attrs[] = { &anon_fault_fallback_charge_attr.attr, &swpout_attr.attr, &swpout_fallback_attr.attr, + &file_alloc_attr.attr, + &file_fallback_attr.attr, + &file_fallback_charge_attr.attr, NULL, };mm/shmem.c +3
-1@@ -1773,6 +1773,9 @@ static struct folio *shmem_alloc_and_add_folio(struct vm_fault *vmf,if (pages ==HPAGE_PMD_NR )count_vm_event (THP_FILE_FALLBACK ); +#ifdef CONFIG_TRANSPARENT_HUGEPAGE +count_mthp_stat (order,MTHP_STAT_FILE_FALLBACK ); +#endif order =next_order (&suitable_orders, order); } }else {@@ -1792,9 +1795,15 @@ static struct folio *shmem_alloc_and_add_folio(struct vm_fault *vmf,if (xa_find (&mapping->i_pages, &index, index + pages -1 ,XA_PRESENT )) { error = -EEXIST ; - }else if (pages ==HPAGE_PMD_NR ) {
mm/shmem.c static struct folio *shmem_alloc_and_add_folio(struct vm_fault *vmf, · line 1805
Quoted 10 lines
mm/shmem.c +8
-2-count_vm_event (THP_FILE_FALLBACK ); -count_vm_event (THP_FILE_FALLBACK_CHARGE ); + }else if (pages >1 ) { +if (pages ==HPAGE_PMD_NR ) { +count_vm_event (THP_FILE_FALLBACK ); +count_vm_event (THP_FILE_FALLBACK_CHARGE ); + } +#ifdef CONFIG_TRANSPARENT_HUGEPAGE +count_mthp_stat (folio_order (folio),MTHP_STAT_FILE_FALLBACK ); +count_mthp_stat (folio_order (folio),MTHP_STAT_FILE_FALLBACK_CHARGE );
pages > 1, we have correct order, count_mthp_stat(order, MTHP_XXX) ?
Quoted 14 lines
mm/shmem.c +4
-0+#endif } goto unlock; }@@ -2168,6 +2177,9 @@ static int shmem_get_folio_gfp(struct inode *inode, pgoff_t index,if (!IS_ERR (folio)) {if (folio_test_pmd_mappable (folio))count_vm_event (THP_FILE_ALLOC ); +#ifdef CONFIG_TRANSPARENT_HUGEPAGE +count_mthp_stat (folio_order (folio),MTHP_STAT_FILE_ALLOC ); +#endif goto alloced; }if (PTR_ERR (folio) == -EEXIST )
Baolin Wang
reviewerRe: [PATCH v5 6/6] mm: shmem: add mTHP counters for anonymous shmem
Quoted 75 lines · Kefeng Wang
Quoted 72 lines · Baolin Wang
Add mTHP counters for anonymous shmem.
Signed-off-by Baolin Wang <[email protected]>---
include/linux/huge_mm.h| 3 +++
mm/huge_memory.c | 6 ++++++
mm/shmem.c | 18 +++++++++++++++---3 files changed, 24 insertions(+), 3 deletions(-)include/linux/huge_mm.h +3
-0@@ -281,6 +281,9 @@ enum mthp_stat_item { MTHP_STAT_ANON_FAULT_FALLBACK_CHARGE, MTHP_STAT_SWPOUT, MTHP_STAT_SWPOUT_FALLBACK, + MTHP_STAT_FILE_ALLOC, + MTHP_STAT_FILE_FALLBACK, + MTHP_STAT_FILE_FALLBACK_CHARGE,
__MTHP_STAT_COUNT
};mm/huge_memory.c +3
-0@@ -555,6 +555,9 @@ DEFINE_MTHP_STAT_ATTR(anon_fault_fallback, MTHP_STAT_ANON_FAULT_FALLBACK); DEFINE_MTHP_STAT_ATTR(anon_fault_fallback_charge, MTHP_STAT_ANON_FAULT_FALLBACK_CHARGE); DEFINE_MTHP_STAT_ATTR(swpout, MTHP_STAT_SWPOUT); DEFINE_MTHP_STAT_ATTR(swpout_fallback, MTHP_STAT_SWPOUT_FALLBACK); +DEFINE_MTHP_STAT_ATTR (file_alloc,MTHP_STAT_FILE_ALLOC ); +DEFINE_MTHP_STAT_ATTR (file_fallback,MTHP_STAT_FILE_FALLBACK ); +DEFINE_MTHP_STAT_ATTR (file_fallback_charge,MTHP_STAT_FILE_FALLBACK_CHARGE ); static struct attribute *stats_attrs[] = {&anon_fault_alloc_attr.attr,
mm/huge_memory.c +3
-0@@ -562,6 +565,9 @@ static struct attribute *stats_attrs[] = { &anon_fault_fallback_charge_attr.attr, &swpout_attr.attr, &swpout_fallback_attr.attr, + &file_alloc_attr.attr, + &file_fallback_attr.attr, + &file_fallback_charge_attr.attr, NULL, };mm/shmem.c +7
-3@@ -1773,6 +1773,9 @@ static struct folio *shmem_alloc_and_add_folio(struct vm_fault *vmf, if (pages == HPAGE_PMD_NR) count_vm_event(THP_FILE_FALLBACK); +#ifdef CONFIG_TRANSPARENT_HUGEPAGE + count_mthp_stat(order, MTHP_STAT_FILE_FALLBACK); +#endif order = next_order(&suitable_orders, order); } } else {@@ -1792,9 +1795,15 @@ static struct folio *shmem_alloc_and_add_folio(struct vm_fault *vmf, if (xa_find(&mapping->i_pages, &index, index + pages - 1, XA_PRESENT)) { error = -EEXIST; - } else if (pages == HPAGE_PMD_NR) { - count_vm_event(THP_FILE_FALLBACK); - count_vm_event(THP_FILE_FALLBACK_CHARGE); + } else if (pages > 1) { + if (pages == HPAGE_PMD_NR) { + count_vm_event(THP_FILE_FALLBACK); + count_vm_event(THP_FILE_FALLBACK_CHARGE);
Quoted 8 lines
mm/shmem.c static struct folio · line 1808
Quoted 6 lines
mm/shmem.c +4
-0+ } +#ifdef CONFIG_TRANSPARENT_HUGEPAGE + count_mthp_stat(folio_order(folio),MTHP_STAT_FILE_FALLBACK ); + count_mthp_stat(folio_order(folio),MTHP_STAT_FILE_FALLBACK_CHARGE );pages > 1, we have correct order,
count_mthp_stat(order, MTHP_XXX)?
Yes, I can use 'order' instead if a new version is needed.
Re: [PATCH v5 6/6] mm: shmem: add mTHP counters for anonymous shmem
In reply to Baolin WangQuoted 4 lines · Baolin Wang
Add mTHP counters for anonymous shmem.
Signed-off-by Baolin Wang <[email protected]>
LGTM. Feel free to add:
Just a friendly reminder: We also need to update the documentation for the counters in transhuge.rst.
Thanks,
Lance
Quoted 90 lines
--- include/linux/huge_mm.h | 3 +++ mm/huge_memory.c | 6 ++++++ mm/shmem.c | 18 +++++++++++++++--- 3 files changed, 24 insertions(+), 3 deletions(-)include/linux/huge_mm.h +3
-0@@ -281,6 +281,9 @@ enum mthp_stat_item {MTHP_STAT_ANON_FAULT_FALLBACK_CHARGE ,MTHP_STAT_SWPOUT ,MTHP_STAT_SWPOUT_FALLBACK , +MTHP_STAT_FILE_ALLOC , +MTHP_STAT_FILE_FALLBACK , +MTHP_STAT_FILE_FALLBACK_CHARGE , __MTHP_STAT_COUNT };mm/huge_memory.c +6
-0@@ -555,6 +555,9 @@ DEFINE_MTHP_STAT_ATTR(anon_fault_fallback, MTHP_STAT_ANON_FAULT_FALLBACK);DEFINE_MTHP_STAT_ATTR (anon_fault_fallback_charge,MTHP_STAT_ANON_FAULT_FALLBACK_CHARGE );DEFINE_MTHP_STAT_ATTR (swpout,MTHP_STAT_SWPOUT );DEFINE_MTHP_STAT_ATTR (swpout_fallback,MTHP_STAT_SWPOUT_FALLBACK ); +DEFINE_MTHP_STAT_ATTR (file_alloc,MTHP_STAT_FILE_ALLOC ); +DEFINE_MTHP_STAT_ATTR (file_fallback,MTHP_STAT_FILE_FALLBACK ); +DEFINE_MTHP_STAT_ATTR (file_fallback_charge,MTHP_STAT_FILE_FALLBACK_CHARGE );static struct attribute *stats_attrs[] = { &anon_fault_alloc_attr.attr,@@ -562,6 +565,9 @@ static struct attribute *stats_attrs[] = { &anon_fault_fallback_charge_attr.attr, &swpout_attr.attr, &swpout_fallback_attr.attr, + &file_alloc_attr.attr, + &file_fallback_attr.attr, + &file_fallback_charge_attr.attr, NULL, };mm/shmem.c +15
-4@@ -1773,6 +1773,9 @@ static struct folio *shmem_alloc_and_add_folio(struct vm_fault *vmf,if (pages ==HPAGE_PMD_NR )count_vm_event (THP_FILE_FALLBACK ); +#ifdef CONFIG_TRANSPARENT_HUGEPAGE +count_mthp_stat (order,MTHP_STAT_FILE_FALLBACK ); +#endif order =next_order (&suitable_orders, order); } }else {@@ -1792,9 +1795,15 @@ static struct folio *shmem_alloc_and_add_folio(struct vm_fault *vmf,if (xa_find (&mapping->i_pages, &index, index + pages -1 ,XA_PRESENT )) { error = -EEXIST ; - }else if (pages ==HPAGE_PMD_NR ) { -count_vm_event (THP_FILE_FALLBACK ); -count_vm_event (THP_FILE_FALLBACK_CHARGE ); + }else if (pages >1 ) { +if (pages ==HPAGE_PMD_NR ) { +count_vm_event (THP_FILE_FALLBACK ); +count_vm_event (THP_FILE_FALLBACK_CHARGE ); + } +#ifdef CONFIG_TRANSPARENT_HUGEPAGE +count_mthp_stat (folio_order (folio),MTHP_STAT_FILE_FALLBACK ); +count_mthp_stat (folio_order (folio),MTHP_STAT_FILE_FALLBACK_CHARGE ); +#endif } goto unlock; }@@ -2168,6 +2177,9 @@ static int shmem_get_folio_gfp(struct inode *inode, pgoff_t index,if (!IS_ERR (folio)) {if (folio_test_pmd_mappable (folio))count_vm_event (THP_FILE_ALLOC ); +#ifdef CONFIG_TRANSPARENT_HUGEPAGE +count_mthp_stat (folio_order (folio),MTHP_STAT_FILE_ALLOC ); +#endif goto alloced; }if (PTR_ERR (folio) == -EEXIST ) --2.39.3
Baolin Wang
reviewerRe: [PATCH v5 6/6] mm: shmem: add mTHP counters for anonymous shmem
Quoted 9 lines · Lance Yang
Quoted 4 lines · Baolin Wang
Add mTHP counters for anonymous shmem.
Signed-off-by Baolin Wang <[email protected]>LGTM. Feel free to add:
Reviewed-by Lance Yang <[email protected]>
Thanks.
Quoted 3 lines
Just a friendly reminder: We also need to update the documentation for the counters in
transhuge.rst.
Indeed.
Andrew, could you help to fold following changes into this patch? Thanks.
Documentation/admin-guide/mm/transhuge.rst +13 -0
@@ -501,6 +501,19 @@ swpout_fallback Usually because failed to allocate some continuous swap space for the huge page. +file_alloc + is incremented every time a file huge page is successfully + allocated. + +file_fallback + is incremented if a file huge page is attempted to be allocated + but fails and instead falls back to using small pages. + +file_fallback_charge + is incremented if a file huge page cannot be charged and instead + falls back to using small pages even though the allocation was + successful. + As the system ages, allocating huge pages may be expensive as the system uses memory compaction to copy data around memory to free a huge page for use. There are some counters in ``/proc/vmstat`` to help
Quoted 4 lines · Baolin Wang
Anonymous pages have already been supported for multi-size (mTHP) allocation through commit
19eaf44954df, that can allow THP to be configured through the sysfs interface located at '/sys/kernel/mm/transparent_hugepage/hugepage-XXkb/enabled'.
Quoted 8 lines
However, the anonymous shmem will ignore the anonymous mTHP rule configured through the sysfs interface, and can only use the PMD-mapped THP, that is not reasonable. Many implement anonymous page sharing through
mmap(MAP_SHARED | MAP_ANONYMOUS), especially in database usage scenarios, therefore, users expect to apply an unified mTHP strategy for anonymous pages, also including the anonymous shared pages, in order to enjoy the benefits of mTHP. For example, lower latency than PMD-mapped THP, smaller memory bloat than PMD-mapped THP, contiguous PTEs on ARM architecture to reduce TLB miss etc.
OK, this makes sense.
Quoted 3 lines
As discussed in the bi-weekly MM meeting[1], the mTHP controls should control all of shmem, not only anonymous shmem, but support will be added iteratively. Therefore, this patch set starts with support for anonymous shmem.
But then this doesn't. You say first that users want the same controls to control all anonymous memory, then you introduce a completely separate set of controls for shared anonymous memory.
shmem has two uses:
- MAP_ANONYMOUS | MAP_SHARED (this patch set)
- tmpfs
For the second use case we don't want controls *at all*, we want the same heiristics used for all other filesystems to apply to tmpfs.
There's no reason to have separate controls for choosing folio size in shmem.
Quoted 9 lines
The primary strategy is similar to supporting anonymous mTHP. Introduce a new interface '/mm/transparent_hugepage/hugepage-XXkb/shmem_enabled', which can have almost the same values as the top-level '
/sys/kernel/mm/transparent_hugepage/shmem_enabled', with adding a new additional "inherit" option and dropping the testing options 'force' and 'deny'. By default all sizes will be set to "never" except PMD size, which is set to "inherit". This ensures backward compatibility with the anonymous shmem enabled of the top level, meanwhile also allows independent control of anonymous shmem enabled for each mTHP.
Quoted 7 lines
shmem has two uses:
-
MAP_ANONYMOUS|MAP_SHARED(this patch set)
- tmpfsFor the second use case we don't want controls *at all*, we want the same heiristics used for all other filesystems to apply to tmpfs.
As discussed in the MM meeting, Hugh had a different opinion on that.
Signature · 3 lines
Cheers, David / dhildenb
Quoted 9 lines · David Hildenbrand
Quoted 7 lines
shmem has two uses:
-
MAP_ANONYMOUS|MAP_SHARED(this patch set)
- tmpfsFor the second use case we don't want controls *at all*, we want the same heiristics used for all other filesystems to apply to tmpfs.
As discussed in the MM meeting, Hugh had a different opinion on that.
FWIW, I just recalled that I wrote a quick summary:
https://lkml.kernel.org/r/[email protected]
I believe the meetings are recorded as well, but never looked at recordings.
Signature · 3 lines
Cheers, David / dhildenb
Quoted 9 lines · David Hildenbrand
Quoted 8 lines · David Hildenbrand
Quoted 7 lines
shmem has two uses:
-
MAP_ANONYMOUS|MAP_SHARED(this patch set)
- tmpfsFor the second use case we don't want controls *at all*, we want the same heiristics used for all other filesystems to apply to tmpfs.
Quoted 7 lines
Quoted 1 line
As discussed in the MM meeting, Hugh had a different opinion on that.
FWIW, I just recalled that I wrote a quick summary:
https://lkml.kernel.org/r/[email protected]
I believe the meetings are recorded as well, but never looked at recordings.
That's not what I understood Hugh to mean. To me, it seemed that Hugh was expressing an opinion on using shmem as shmem, not as using it as tmpfs.
If I misunderstood Hugh, well, I still disagree. We should not have separate controls for this. tmpfs is just not that special.
Baolin Wang
reviewerQuoted 16 lines · Matthew Wilcox
Quoted 15 lines · David Hildenbrand
Quoted 9 lines · David Hildenbrand
Quoted 7 lines
shmem has two uses:
-
MAP_ANONYMOUS|MAP_SHARED(this patch set)
- tmpfsFor the second use case we don't want controls *at all*, we want the same heiristics used for all other filesystems to apply to tmpfs.
As discussed in the MM meeting, Hugh had a different opinion on that.
FWIW, I just recalled that I wrote a quick summary:
Quoted 8 lines
Quoted 1 line
I believe the meetings are recorded as well, but never looked at recordings.
That's not what I understood Hugh to mean. To me, it seemed that Hugh was expressing an opinion on using shmem as shmem, not as using it as tmpfs.
If I misunderstood Hugh, well, I still disagree. We should not have separate controls for this. tmpfs is just not that special.
But now we already have a PMD-mapped THP control for tmpfs, and mTHP simply extends this control to per-size.
IIUC, as David mentioned before, for tmpfs, mTHP should act like a huge order filter which should be respected by the expected huge orders in the write() and fallocate() paths. This would also solve the issue of allocating huge orders in writable mmap() path for tmpfs, as well as unifying the interface.
Anyway, I will try to provide an RFC to discuss the mTHP for tmpfs approach.
Ryan Roberts
reviewerQuoted 19 lines · Baolin Wang
Quoted 16 lines · Matthew Wilcox
Quoted 15 lines · David Hildenbrand
Quoted 9 lines · David Hildenbrand
Quoted 7 lines
shmem has two uses:
-
MAP_ANONYMOUS|MAP_SHARED(this patch set)
- tmpfsFor the second use case we don't want controls *at all*, we want the same heiristics used for all other filesystems to apply to tmpfs.
As discussed in the MM meeting, Hugh had a different opinion on that.
FWIW, I just recalled that I wrote a quick summary:
Quoted 8 lines
Quoted 8 lines
Quoted 1 line
I believe the meetings are recorded as well, but never looked at recordings.
That's not what I understood Hugh to mean. To me, it seemed that Hugh was expressing an opinion on using shmem as shmem, not as using it as tmpfs.
If I misunderstood Hugh, well, I still disagree. We should not have separate controls for this. tmpfs is just not that special.
I wasn't at the meeting that's being referred to, but I thought we previously agreed that tmpfs *is* special because in some configurations its not backed by swap so is locked in ram?
Quoted 10 lines
But now we already have a PMD-mapped THP control for tmpfs, and mTHP simply extends this control to per-size.
IIUC, as David mentioned before, for tmpfs, mTHP should act like a huge order filter which should be respected by the expected huge orders in the
write()andfallocate()paths. This would also solve the issue of allocating huge orders in writablemmap()path for tmpfs, as well as unifying the interface.Anyway, I will try to provide an RFC to discuss the mTHP for tmpfs approach.
Quoted 26 lines · Ryan Roberts
Quoted 25 lines · Baolin Wang
Quoted 22 lines · Matthew Wilcox
Quoted 9 lines · David Hildenbrand
Quoted 8 lines · David Hildenbrand
Quoted 7 lines
shmem has two uses:
-
MAP_ANONYMOUS|MAP_SHARED(this patch set)
- tmpfsFor the second use case we don't want controls *at all*, we want the same heiristics used for all other filesystems to apply to tmpfs.
Quoted 7 lines
Quoted 1 line
As discussed in the MM meeting, Hugh had a different opinion on that.
FWIW, I just recalled that I wrote a quick summary:
https://lkml.kernel.org/r/[email protected]
I believe the meetings are recorded as well, but never looked at recordings.
That's not what I understood Hugh to mean. To me, it seemed that Hugh was expressing an opinion on using shmem as shmem, not as using it as tmpfs.
Quoted 6 lines
Quoted 2 lines
Quoted 2 lines
If I misunderstood Hugh, well, I still disagree. We should not have separate controls for this. tmpfs is just not that special.
I wasn't at the meeting that's being referred to, but I thought we previously agreed that tmpfs *is* special because in some configurations its not backed by swap so is locked in ram?
There are multiple things to that, like:
* Machines only having limited/no swap configured
* tmpfs can be configured to never go to swap
* memfd/tmpfs files getting used purely for mmap(): there is no real difference to MAP_ANON|MAP_SHARE besides the processes we share that memory with.
Especially when it comes to memory waste concerns and access behavior in some cases, tmpfs behaved much more like anonymous memory. But there are for sure other use cases where tmpfs is not that special.
My opinion is that we need to let people configure orders (if you feel like it, configure all), but *select* the order to allocate based on readahead information -- in contrast to anonymous memory where we start at the highest order and don't have readahead information available.
Maybe we need different "order allcoation" logic for read/write vs. fault, not sure.
But I don't maintain that code, so I can only give stupid suggestions and repeat what I understood from the meeting with Hugh and Kirill :)
Signature · 3 lines
Cheers, David / dhildenb
Ryan Roberts
reviewerQuoted 42 lines · David Hildenbrand
Quoted 26 lines · Ryan Roberts
Quoted 25 lines · Baolin Wang
Quoted 22 lines · Matthew Wilcox
Quoted 9 lines · David Hildenbrand
Quoted 8 lines · David Hildenbrand
Quoted 7 lines
shmem has two uses:
-
MAP_ANONYMOUS|MAP_SHARED(this patch set)
- tmpfsFor the second use case we don't want controls *at all*, we want the same heiristics used for all other filesystems to apply to tmpfs.
Quoted 7 lines
Quoted 1 line
As discussed in the MM meeting, Hugh had a different opinion on that.
FWIW, I just recalled that I wrote a quick summary:
https://lkml.kernel.org/r/[email protected]
I believe the meetings are recorded as well, but never looked at recordings.
That's not what I understood Hugh to mean. To me, it seemed that Hugh was expressing an opinion on using shmem as shmem, not as using it as tmpfs.
Quoted 6 lines
Quoted 2 lines
Quoted 2 lines
If I misunderstood Hugh, well, I still disagree. We should not have separate controls for this. tmpfs is just not that special.
I wasn't at the meeting that's being referred to, but I thought we previously agreed that tmpfs *is* special because in some configurations its not backed by swap so is locked in ram?
There are multiple things to that, like:
* Machines only having limited/no swap configured
* tmpfs can be configured to never go to swap
* memfd/tmpfs files getting used purely formmap(): there is no real difference to MAP_ANON|MAP_SHARE besides the processes we share that memory with.
Quoted 8 lines
Especially when it comes to memory waste concerns and access behavior in some cases, tmpfs behaved much more like anonymous memory. But there are for sure other use cases where tmpfs is not that special.
My opinion is that we need to let people configure orders (if you feel like it, configure all), but *select* the order to allocate based on readahead information -- in contrast to anonymous memory where we start at the highest order and don't have readahead information available.
That approach is exactly what I proposed to start playing with yesterday [1] for regular pagecache folio allocations too :)
[1] https://lore.kernel.org/linux-mm/[email protected]/
Quoted 7 lines
Maybe we need different "order allcoation" logic for read/write vs. fault, not sure.
But I don't maintain that code, so I can only give stupid suggestions and repeat what I understood from the meeting with Hugh and Kirill :)
Quoted 47 lines · Ryan Roberts
Quoted 46 lines · David Hildenbrand
Quoted 26 lines · Ryan Roberts
Quoted 25 lines · Baolin Wang
Quoted 22 lines · Matthew Wilcox
Quoted 9 lines · David Hildenbrand
Quoted 8 lines · David Hildenbrand
Quoted 7 lines
shmem has two uses:
-
MAP_ANONYMOUS|MAP_SHARED(this patch set)
- tmpfsFor the second use case we don't want controls *at all*, we want the same heiristics used for all other filesystems to apply to tmpfs.
Quoted 7 lines
Quoted 1 line
As discussed in the MM meeting, Hugh had a different opinion on that.
FWIW, I just recalled that I wrote a quick summary:
https://lkml.kernel.org/r/[email protected]
I believe the meetings are recorded as well, but never looked at recordings.
That's not what I understood Hugh to mean. To me, it seemed that Hugh was expressing an opinion on using shmem as shmem, not as using it as tmpfs.
Quoted 6 lines
Quoted 2 lines
Quoted 2 lines
If I misunderstood Hugh, well, I still disagree. We should not have separate controls for this. tmpfs is just not that special.
I wasn't at the meeting that's being referred to, but I thought we previously agreed that tmpfs *is* special because in some configurations its not backed by swap so is locked in ram?
There are multiple things to that, like:
* Machines only having limited/no swap configured
* tmpfs can be configured to never go to swap
* memfd/tmpfs files getting used purely formmap(): there is no real difference to MAP_ANON|MAP_SHARE besides the processes we share that memory with.Especially when it comes to memory waste concerns and access behavior in some cases, tmpfs behaved much more like anonymous memory. But there are for sure other use cases where tmpfs is not that special.
Quoted 7 lines
Quoted 4 lines
My opinion is that we need to let people configure orders (if you feel like it, configure all), but *select* the order to allocate based on readahead information -- in contrast to anonymous memory where we start at the highest order and don't have readahead information available.
That approach is exactly what I proposed to start playing with yesterday [1] for regular pagecache folio allocations too :)
In German, there is this saying "zwei Dumme ein Gedanke".
The official English alternative is "great minds think alike".
... well, the direct German->English translation definitely has a "German touch" to it: "two stupid ones one thought"
Signature · 3 lines
Cheers, David / dhildenb
Ryan Roberts
reviewerQuoted 56 lines · David Hildenbrand
Quoted 47 lines · Ryan Roberts
Quoted 46 lines · David Hildenbrand
Quoted 26 lines · Ryan Roberts
Quoted 25 lines · Baolin Wang
Quoted 22 lines · Matthew Wilcox
Quoted 9 lines · David Hildenbrand
Quoted 8 lines · David Hildenbrand
Quoted 7 lines
shmem has two uses:
-
MAP_ANONYMOUS|MAP_SHARED(this patch set)
- tmpfsFor the second use case we don't want controls *at all*, we want the same heiristics used for all other filesystems to apply to tmpfs.
Quoted 7 lines
Quoted 1 line
As discussed in the MM meeting, Hugh had a different opinion on that.
FWIW, I just recalled that I wrote a quick summary:
https://lkml.kernel.org/r/[email protected]
I believe the meetings are recorded as well, but never looked at recordings.
That's not what I understood Hugh to mean. To me, it seemed that Hugh was expressing an opinion on using shmem as shmem, not as using it as tmpfs.
Quoted 6 lines
Quoted 2 lines
Quoted 2 lines
If I misunderstood Hugh, well, I still disagree. We should not have separate controls for this. tmpfs is just not that special.
I wasn't at the meeting that's being referred to, but I thought we previously agreed that tmpfs *is* special because in some configurations its not backed by swap so is locked in ram?
There are multiple things to that, like:
* Machines only having limited/no swap configured
* tmpfs can be configured to never go to swap
* memfd/tmpfs files getting used purely formmap(): there is no real difference to MAP_ANON|MAP_SHARE besides the processes we share that memory with.Especially when it comes to memory waste concerns and access behavior in some cases, tmpfs behaved much more like anonymous memory. But there are for sure other use cases where tmpfs is not that special.
Quoted 7 lines
Quoted 4 lines
My opinion is that we need to let people configure orders (if you feel like it, configure all), but *select* the order to allocate based on readahead information -- in contrast to anonymous memory where we start at the highest order and don't have readahead information available.
That approach is exactly what I proposed to start playing with yesterday [1] for regular pagecache folio allocations too :)
Quoted 6 lines
In German, there is this saying "zwei Dumme ein Gedanke".
The official English alternative is "great minds think alike".
... well, the direct
German->Englishtranslation definitely has a "German touch" to it: "two stupid ones one thought"
I definitely prefer the direct translation. :)
Quoted 42 lines · David Hildenbrand
Quoted 26 lines · Ryan Roberts
Quoted 25 lines · Baolin Wang
Quoted 22 lines · Matthew Wilcox
Quoted 9 lines · David Hildenbrand
Quoted 8 lines · David Hildenbrand
Quoted 7 lines
shmem has two uses:
-
MAP_ANONYMOUS|MAP_SHARED(this patch set)
- tmpfsFor the second use case we don't want controls *at all*, we want the same heiristics used for all other filesystems to apply to tmpfs.
Quoted 7 lines
Quoted 1 line
As discussed in the MM meeting, Hugh had a different opinion on that.
FWIW, I just recalled that I wrote a quick summary:
https://lkml.kernel.org/r/[email protected]
I believe the meetings are recorded as well, but never looked at recordings.
That's not what I understood Hugh to mean. To me, it seemed that Hugh was expressing an opinion on using shmem as shmem, not as using it as tmpfs.
Quoted 6 lines
Quoted 2 lines
Quoted 2 lines
If I misunderstood Hugh, well, I still disagree. We should not have separate controls for this. tmpfs is just not that special.
I wasn't at the meeting that's being referred to, but I thought we previously agreed that tmpfs *is* special because in some configurations its not backed by swap so is locked in ram?
There are multiple things to that, like:
* Machines only having limited/no swap configured
* tmpfs can be configured to never go to swap
* memfd/tmpfs files getting used purely formmap(): there is no real difference to MAP_ANON|MAP_SHARE besides the processes we share that memory with.
Quoted 3 lines
Especially when it comes to memory waste concerns and access behavior in some cases, tmpfs behaved much more like anonymous memory. But there are for sure other use cases where tmpfs is not that special.
Having controls to select the allowable folio order allocations for tmpfs does not address any of these issues. The suggested filesystem approach [1] involves allocating orders in larger chunks, but always the same size you would allocate when using order-0 folios. So, it's a conservative approach. Using mTHP knobs in tmpfs would cause:
* Over allocation when using mTHP and/ord THP under the 'always' flag.
* Allocate in bigger chunks in a non optimal way, when
not all mTHP and THP orders are enabled.
* Operate in a similar manner as in [1] when all mTHP and THP orders are enabled and 'within_size' flag is used (assuming we use patch 11 from [1]).
[1] Last 3 patches of these series:
https://lore.kernel.org/all/[email protected]/
My understanding of why mTHP was preferred is to raise awareness in user space and allow tmpfs mounts used at boot time to operate in 'safe' mode (no large folios). Does it make more sense to have a large folios enable flag to control order allocation as in [1], instead of every single order possible?
Quoted 8 lines
My opinion is that we need to let people configure orders (if you feel like it, configure all), but *select* the order to allocate based on readahead information -- in contrast to anonymous memory where we start at the highest order and don't have readahead information available.
Maybe we need different "order allcoation" logic for read/write vs. fault, not sure.
I would suggest [1] the file size of the write for the write and fallocate paths. But when does make sense to use readahead information? Maybe when swap is involved?
Quoted 9 lines
But I don't maintain that code, so I can only give stupid suggestions and repeat what I understood from the meeting with Hugh and Kirill :)
Signature · 3 lines
Cheers, David / dhildenb
Ryan Roberts
reviewerQuoted 43 lines · Daniel Gomez
Quoted 42 lines · David Hildenbrand
Quoted 26 lines · Ryan Roberts
Quoted 25 lines · Baolin Wang
Quoted 22 lines · Matthew Wilcox
Quoted 9 lines · David Hildenbrand
Quoted 8 lines · David Hildenbrand
Quoted 7 lines
shmem has two uses:
-
MAP_ANONYMOUS|MAP_SHARED(this patch set)
- tmpfsFor the second use case we don't want controls *at all*, we want the same heiristics used for all other filesystems to apply to tmpfs.
Quoted 7 lines
Quoted 1 line
As discussed in the MM meeting, Hugh had a different opinion on that.
FWIW, I just recalled that I wrote a quick summary:
https://lkml.kernel.org/r/[email protected]
I believe the meetings are recorded as well, but never looked at recordings.
That's not what I understood Hugh to mean. To me, it seemed that Hugh was expressing an opinion on using shmem as shmem, not as using it as tmpfs.
Quoted 6 lines
Quoted 2 lines
Quoted 2 lines
If I misunderstood Hugh, well, I still disagree. We should not have separate controls for this. tmpfs is just not that special.
I wasn't at the meeting that's being referred to, but I thought we previously agreed that tmpfs *is* special because in some configurations its not backed by swap so is locked in ram?
There are multiple things to that, like:
* Machines only having limited/no swap configured
* tmpfs can be configured to never go to swap
* memfd/tmpfs files getting used purely formmap(): there is no real difference to MAP_ANON|MAP_SHARE besides the processes we share that memory with.
Quoted 8 lines
Quoted 3 lines
Especially when it comes to memory waste concerns and access behavior in some cases, tmpfs behaved much more like anonymous memory. But there are for sure other use cases where tmpfs is not that special.
Having controls to select the allowable folio order allocations for tmpfs does not address any of these issues. The suggested filesystem approach [1] involves allocating orders in larger chunks, but always the same size you would allocate when using order-0 folios.
Well you can't know that you will never allocate more. If you allocate a 2M block, you probably have some good readahead data that tells you you are likely to keep reading sequentially, but you don't know for sure that the application won't stop after just 4K.
Quoted 8 lines
So,
it's a conservative approach. Using mTHP knobs in tmpfs would cause:
* Over allocation when using mTHP and/ord THP under the 'always' flag.
* Allocate in bigger chunks in a non optimal way, when
not all mTHP and THP orders are enabled.
* Operate in a similar manner as in [1] when all mTHP and THP orders are enabled and 'within_size' flag is used (assuming we use patch 11 from [1]).
Large folios may still be considered scarce resources even if the amount of memory allocated is still the same. And if shmem isn't backed by swap then once you have allocated a large folio for shmem, it is stuck in shmem, even if it would be better used somewhere else.
And it's possible (likely even, in my opinion) that allocating lots of different folio sizes will exacerbate memory fragmentation, leading to more order-0 fallbacks, which would hurt the overall system performance in the long run, vs restricting to a couple of folio sizes.
I'm starting some work to actually measure how limiting the folio sizes allocated for page cache memory can help reduce large folio allocation failure overall. My hypothesis is that the data will show us that in an environment like Android, where memory pressure is high, limiting everything to order-0 and order-4 will significantly improve the allocation success rate of order-4. Let's see.
Quoted 9 lines
[1] Last 3 patches of these series:
https://lore.kernel.org/all/[email protected]/My understanding of why mTHP was preferred is to raise awareness in user space and allow tmpfs mounts used at boot time to operate in 'safe' mode (no large folios). Does it make more sense to have a large folios enable flag to control order allocation as in [1], instead of every single order possible?
My intuition is towards every order possible, as per above. Let's see what the data tells us.
Quoted 22 lines
Quoted 9 lines
My opinion is that we need to let people configure orders (if you feel like it, configure all), but *select* the order to allocate based on readahead information -- in contrast to anonymous memory where we start at the highest order and don't have readahead information available.
Maybe we need different "order allcoation" logic for read/write vs. fault, not sure.
I would suggest [1] the file size of the write for the write and fallocate paths. But when does make sense to use readahead information? Maybe when swap is involved?
Quoted 8 lines
But I don't maintain that code, so I can only give stupid suggestions and repeat what I understood from the meeting with Hugh and Kirill :)
Signature · 3 lines
Cheers, David / dhildenb
Quoted 48 lines · Ryan Roberts
Quoted 47 lines · Daniel Gomez
Quoted 45 lines · David Hildenbrand
Quoted 26 lines · Ryan Roberts
Quoted 25 lines · Baolin Wang
Quoted 22 lines · Matthew Wilcox
Quoted 9 lines · David Hildenbrand
Quoted 8 lines · David Hildenbrand
Quoted 7 lines
shmem has two uses:
-
MAP_ANONYMOUS|MAP_SHARED(this patch set)
- tmpfsFor the second use case we don't want controls *at all*, we want the same heiristics used for all other filesystems to apply to tmpfs.
Quoted 7 lines
Quoted 1 line
As discussed in the MM meeting, Hugh had a different opinion on that.
FWIW, I just recalled that I wrote a quick summary:
https://lkml.kernel.org/r/[email protected]
I believe the meetings are recorded as well, but never looked at recordings.
That's not what I understood Hugh to mean. To me, it seemed that Hugh was expressing an opinion on using shmem as shmem, not as using it as tmpfs.
Quoted 6 lines
Quoted 2 lines
Quoted 2 lines
If I misunderstood Hugh, well, I still disagree. We should not have separate controls for this. tmpfs is just not that special.
I wasn't at the meeting that's being referred to, but I thought we previously agreed that tmpfs *is* special because in some configurations its not backed by swap so is locked in ram?
There are multiple things to that, like:
* Machines only having limited/no swap configured
* tmpfs can be configured to never go to swap
* memfd/tmpfs files getting used purely formmap(): there is no real difference to MAP_ANON|MAP_SHARE besides the processes we share that memory with.Especially when it comes to memory waste concerns and access behavior in some cases, tmpfs behaved much more like anonymous memory. But there are for sure other use cases where tmpfs is not that special.
Quoted 6 lines
Quoted 4 lines
Having controls to select the allowable folio order allocations for tmpfs does not address any of these issues. The suggested filesystem approach [1] involves allocating orders in larger chunks, but always the same size you would allocate when using order-0 folios.
Well you can't know that you will never allocate more. If you allocate a 2M
In the fs large folio approach implementation [1], the allocation of a 2M (or any non order-0) occurs when the size of the write/fallocate is 2M (and index is aligned).
Quoted 3 lines
block, you probably have some good readahead data that tells you you are likely to keep reading sequentially, but you don't know for sure that the application won't stop after just 4K.
Is shmem_file_read_iter() getting readahead data to perform the read? or what do you mean exactly?
In [1], read is perform in chunks of 4k, so I think this does not apply.
Quoted 10 lines
Quoted 9 lines
So,
it's a conservative approach. Using mTHP knobs in tmpfs would cause:
* Over allocation when using mTHP and/ord THP under the 'always' flag.
* Allocate in bigger chunks in a non optimal way, when
not all mTHP and THP orders are enabled.
* Operate in a similar manner as in [1] when all mTHP and THP orders are enabled and 'within_size' flag is used (assuming we use patch 11 from [1]).
Quoted 4 lines
Large folios may still be considered scarce resources even if the amount of memory allocated is still the same. And if shmem isn't backed by swap then once you have allocated a large folio for shmem, it is stuck in shmem, even if it would be better used somewhere else.
Is that true for tmpfs as well? We have shmem_unused_huge_shrink() that will reclaim unused large folios (when ENOSPC and free_cached_objects()). Can't we reuse that when the system is under memory pressure?
Quoted 5 lines
And it's possible (likely even, in my opinion) that allocating lots of different folio sizes will exacerbate memory fragmentation, leading to more order-0 fallbacks, which would hurt the overall system performance in the long run, vs restricting to a couple of folio sizes.
Since we are transitioning to large folios in other filesystems, the impact of restricting the order here will only depend on the extent of tmpfs usage relative to the rest of the system. Luis discussed the topic of mm fragmentation and measurment in a session at LSFMM this year [2].
[2] https://lore.kernel.org/all/[email protected]/
Quoted 3 lines
I'm starting some work to actually measure how limiting the folio sizes allocated for page cache memory can help reduce large folio allocation failure
It would be great to hear more about that effort.
Quoted 41 lines
overall. My hypothesis is that the data will show us that in an environment like Android, where memory pressure is high, limiting everything to order-0 and order-4 will significantly improve the allocation success rate of order-4. Let's see.
Quoted 9 lines
[1] Last 3 patches of these series:
https://lore.kernel.org/all/[email protected]/My understanding of why mTHP was preferred is to raise awareness in user space and allow tmpfs mounts used at boot time to operate in 'safe' mode (no large folios). Does it make more sense to have a large folios enable flag to control order allocation as in [1], instead of every single order possible?
My intuition is towards every order possible, as per above. Let's see what the data tells us.
Quoted 22 lines
Quoted 9 lines
My opinion is that we need to let people configure orders (if you feel like it, configure all), but *select* the order to allocate based on readahead information -- in contrast to anonymous memory where we start at the highest order and don't have readahead information available.
Maybe we need different "order allcoation" logic for read/write vs. fault, not sure.
I would suggest [1] the file size of the write for the write and fallocate paths. But when does make sense to use readahead information? Maybe when swap is involved?
Quoted 8 lines
But I don't maintain that code, so I can only give stupid suggestions and repeat what I understood from the meeting with Hugh and Kirill :)
Signature · 3 lines
Cheers, David / dhildenb
Quoted 54 lines · Daniel Gomez
Quoted 53 lines · Ryan Roberts
Quoted 51 lines · Daniel Gomez
Quoted 42 lines · David Hildenbrand
Quoted 26 lines · Ryan Roberts
Quoted 25 lines · Baolin Wang
Quoted 22 lines · Matthew Wilcox
Quoted 9 lines · David Hildenbrand
Quoted 8 lines · David Hildenbrand
Quoted 7 lines
shmem has two uses:
-
MAP_ANONYMOUS|MAP_SHARED(this patch set)
- tmpfsFor the second use case we don't want controls *at all*, we want the same heiristics used for all other filesystems to apply to tmpfs.
Quoted 7 lines
Quoted 1 line
As discussed in the MM meeting, Hugh had a different opinion on that.
FWIW, I just recalled that I wrote a quick summary:
https://lkml.kernel.org/r/[email protected]
I believe the meetings are recorded as well, but never looked at recordings.
That's not what I understood Hugh to mean. To me, it seemed that Hugh was expressing an opinion on using shmem as shmem, not as using it as tmpfs.
Quoted 6 lines
Quoted 2 lines
Quoted 2 lines
If I misunderstood Hugh, well, I still disagree. We should not have separate controls for this. tmpfs is just not that special.
I wasn't at the meeting that's being referred to, but I thought we previously agreed that tmpfs *is* special because in some configurations its not backed by swap so is locked in ram?
There are multiple things to that, like:
* Machines only having limited/no swap configured
* tmpfs can be configured to never go to swap
* memfd/tmpfs files getting used purely formmap(): there is no real difference to MAP_ANON|MAP_SHARE besides the processes we share that memory with.Quoted 3 lines
Especially when it comes to memory waste concerns and access behavior in some cases, tmpfs behaved much more like anonymous memory. But there are for sure other use cases where tmpfs is not that special.
Having controls to select the allowable folio order allocations for tmpfs does not address any of these issues. The suggested filesystem approach [1] involves allocating orders in larger chunks, but always the same size you would allocate when using order-0 folios.
Quoted 5 lines
Quoted 1 line
Well you can't know that you will never allocate more. If you allocate a 2M
In the fs large folio approach implementation [1], the allocation of a 2M (or any non order-0) occurs when the size of the write/fallocate is 2M (and index is aligned).
I don't have time right now follow the discussion in detail here (I thought we had a meeting to discuss that and received guidance from Hugh?), but I'll point out two things:
(1) We need a reasonable model for handling/allocating of large folios during page faults. shmem/tmpfs can be used just like anon-shmem if you simply only mmap that thing (hello VMs!).
(2) Hugh gave (IMHO) clear feedback during the meeting how he thinks we should approach large folios in shmem.
Maybe I got (2) all wrong and people can point out all the issues in my summary from the meeting.
Otherwise, if people don't want to accept the result from that meeting, we need further guidance from Hugh.
Signature · 3 lines
Cheers, David / dhildenb