mm: shmem: Rename mTHP shmem counters
Subsystems (5)
Ryan Roberts
reviewerPATCH 1/1 mm: shmem: Rename mTHP shmem counters
The legacy PMD-sized THP counters at /proc/vmstat include thp_file_alloc, thp_file_fallback and thp_file_fallback_charge, which rather confusingly refer to shmem THP and do not include any other types of file pages. This is inconsistent since in most other places in the kernel, THP counters are explicitly separated for anon, shmem and file flavours. However, we are stuck with it since it constitutes a user ABI.
Recently, commit 66f44583f9b6 ("mm: shmem: add mTHP counters for anonymous shmem") added equivalent mTHP stats for shmem, keeping the same "file_" prefix in the names. But in future, we may want to add extra stats to cover actual file pages, at which point, it would all become very confusing.
So let's take the opportunity to rename these new counters "shmem_" before the change makes it upstream and the ABI becomes immutable.
---
Hi All,
Applies on top of today's mm-unstable (2073cda629a4) and tested with mm selftests; no regressions observed.
The backstory here is that I'd like to introduce some counters for regular file folio allocations to observe how often large folio allocation succeeds, but these shmem counters are named "file" which is going to make things confusing. So hoping to solve that before commit 66f44583f9b6 ("mm: shmem: add mTHP counters for anonymous shmem") goes upstream (it is currently in mm-stable).
Admittedly, this change means the mTHP stat names are not the same as the legacy PMD-size THP names, but I think that's a smaller issue than having "file_" mTHP stats that only count shmem, then having to introduce "file2_" or "pgcache_" stats for the regular file memory, which is even more inconsistent IMHO. I guess the alternative is to count both shmem and file in these mTHP stats (that's how they were documented anyway) but I think it's better to be able to consider them separately like we do for all the other counters.
Thanks,
Ryan
Documentation/admin-guide/mm/transhuge.rst | 12 ++++++------ include/linux/huge_mm.h | 6 +++--- mm/huge_memory.c | 12 ++++++------ mm/shmem.c | 8 ++++---- 4 files changed, 19 insertions(+), 19 deletions(-)
Documentation/admin-guide/mm/transhuge.rst +6 -6
@@ -496,16 +496,16 @@ 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 +shmem_alloc + is incremented every time a shmem huge page is successfully allocated. -file_fallback - is incremented if a file huge page is attempted to be allocated +shmem_fallback + is incremented if a shmem 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 +shmem_fallback_charge + is incremented if a shmem huge page cannot be charged and instead falls back to using small pages even though the allocation was successful.
include/linux/huge_mm.h +3 -3
@@ -269,9 +269,9 @@ 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_SHMEM_ALLOC , +MTHP_STAT_SHMEM_FALLBACK , +MTHP_STAT_SHMEM_FALLBACK_CHARGE ,MTHP_STAT_SPLIT ,MTHP_STAT_SPLIT_FAILED , MTHP_STAT_SPLIT_DEFERRED,
mm/huge_memory.c +6 -6
@@ -568,9 +568,9 @@ 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 ); +DEFINE_MTHP_STAT_ATTR (shmem_alloc,MTHP_STAT_SHMEM_ALLOC ); +DEFINE_MTHP_STAT_ATTR (shmem_fallback,MTHP_STAT_SHMEM_FALLBACK ); +DEFINE_MTHP_STAT_ATTR (shmem_fallback_charge,MTHP_STAT_SHMEM_FALLBACK_CHARGE );DEFINE_MTHP_STAT_ATTR (split,MTHP_STAT_SPLIT );DEFINE_MTHP_STAT_ATTR (split_failed,MTHP_STAT_SPLIT_FAILED );DEFINE_MTHP_STAT_ATTR (split_deferred,MTHP_STAT_SPLIT_DEFERRED );
@@ -581,9 +581,9 @@ 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, + &shmem_alloc_attr.attr, + &shmem_fallback_attr.attr, + &shmem_fallback_charge_attr.attr, &split_attr.attr, &split_failed_attr.attr, &split_deferred_attr.attr,
mm/shmem.c +4 -4
@@ -1777,7 +1777,7 @@ 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 ); +count_mthp_stat (order,MTHP_STAT_SHMEM_FALLBACK );#endif order =next_order (&suitable_orders, order); }
@@ -1804,8 +1804,8 @@ static struct folio *shmem_alloc_and_add_folio(struct vm_fault *vmf,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 ); +count_mthp_stat (folio_order (folio),MTHP_STAT_SHMEM_FALLBACK ); +count_mthp_stat (folio_order (folio),MTHP_STAT_SHMEM_FALLBACK_CHARGE );#endif } goto unlock;
@@ -2181,7 +2181,7 @@ static int shmem_get_folio_gfp(struct inode *inode, pgoff_t index,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 ); +count_mthp_stat (folio_order (folio),MTHP_STAT_SHMEM_ALLOC );#endif goto alloced; }
--
2.43.0
Barry Song
reviewerQuoted 8 lines · Ryan Roberts
The legacy PMD-sized THP counters at
/proc/vmstatincludethp_file_alloc,thp_file_fallbackandthp_file_fallback_charge, which rather confusingly refer to shmem THP and do not include any other types of file pages. This is inconsistent since in most other places in the kernel, THP counters are explicitly separated for anon, shmem and file flavours. However, we are stuck with it since it constitutes a user ABI.
Quoted 8 lines
Recently, commit
66f44583f9b6("mm: shmem: add mTHP counters for anonymous shmem") added equivalent mTHP stats for shmem, keeping the same "file_" prefix in the names. But in future, we may want to add extra stats to cover actual file pages, at which point, it would all become very confusing.So let's take the opportunity to rename these new counters "
shmem_" before the change makes it upstream and the ABI becomes immutable.
Personally, I think this approach is much clearer. However, I recall we discussed this
before [1], and it seems that inconsistency is a concern?
Quoted 140 lines
Signed-off-by Ryan Roberts <[email protected]>---
Hi All,
Applies on top of today's mm-unstable (
2073cda629a4) and tested with mm selftests; no regressions observed.The backstory here is that I'd like to introduce some counters for regular file folio allocations to observe how often large folio allocation succeeds, but these shmem counters are named "file" which is going to make things confusing. So hoping to solve that before commit
66f44583f9b6("mm: shmem: add mTHP counters for anonymous shmem") goes upstream (it is currently in mm-stable).Admittedly, this change means the mTHP stat names are not the same as the legacy PMD-size THP names, but I think that's a smaller issue than having "
file_" mTHP stats that only count shmem, then having to introduce "file2_" or "pgcache_" stats for the regular file memory, which is even more inconsistent IMHO. I guess the alternative is to count both shmem and file in these mTHP stats (that's how they were documented anyway) but I think it's better to be able to consider them separately like we do for all the other counters.Thanks,
RyanDocumentation/admin-guide/mm/transhuge.rst | 12 ++++++------ include/linux/huge_mm.h | 6 +++--- mm/huge_memory.c | 12 ++++++------ mm/shmem.c | 8 ++++---- 4 files changed, 19 insertions(+), 19 deletions(-)Documentation/admin-guide/mm/transhuge.rst +6
-6@@ -496,16 +496,16 @@ 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 +shmem_alloc + is incremented every time a shmem huge page is successfully allocated. -file_fallback - is incremented if a file huge page is attempted to be allocated +shmem_fallback + is incremented if a shmem 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 +shmem_fallback_charge + is incremented if a shmem huge page cannot be charged and instead falls back to using small pages even though the allocation was successful.include/linux/huge_mm.h +3
-3@@ -269,9 +269,9 @@ 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_SHMEM_ALLOC , +MTHP_STAT_SHMEM_FALLBACK , +MTHP_STAT_SHMEM_FALLBACK_CHARGE ,MTHP_STAT_SPLIT ,MTHP_STAT_SPLIT_FAILED , MTHP_STAT_SPLIT_DEFERRED,mm/huge_memory.c +6
-6@@ -568,9 +568,9 @@ 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 ); +DEFINE_MTHP_STAT_ATTR (shmem_alloc,MTHP_STAT_SHMEM_ALLOC ); +DEFINE_MTHP_STAT_ATTR (shmem_fallback,MTHP_STAT_SHMEM_FALLBACK ); +DEFINE_MTHP_STAT_ATTR (shmem_fallback_charge,MTHP_STAT_SHMEM_FALLBACK_CHARGE );DEFINE_MTHP_STAT_ATTR (split,MTHP_STAT_SPLIT );DEFINE_MTHP_STAT_ATTR (split_failed,MTHP_STAT_SPLIT_FAILED );DEFINE_MTHP_STAT_ATTR (split_deferred,MTHP_STAT_SPLIT_DEFERRED );@@ -581,9 +581,9 @@ 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, + &shmem_alloc_attr.attr, + &shmem_fallback_attr.attr, + &shmem_fallback_charge_attr.attr, &split_attr.attr, &split_failed_attr.attr, &split_deferred_attr.attr,mm/shmem.c +4
-5@@ -1777,7 +1777,7 @@ 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 ); +count_mthp_stat (order,MTHP_STAT_SHMEM_FALLBACK );#endif order =next_order (&suitable_orders, order); }@@ -1804,8 +1804,8 @@ static struct folio *shmem_alloc_and_add_folio(struct vm_fault *vmf,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 ); +count_mthp_stat (folio_order (folio),MTHP_STAT_SHMEM_FALLBACK ); +count_mthp_stat (folio_order (folio),MTHP_STAT_SHMEM_FALLBACK_CHARGE );#endif } goto unlock;@@ -2181,7 +2181,7 @@ static int shmem_get_folio_gfp(struct inode *inode, pgoff_t index,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 ); +count_mthp_stat (folio_order (folio),MTHP_STAT_SHMEM_ALLOC );#endif goto alloced; } --
Quoted 2 lines
2.43.0
Thanks
Barry
Ryan Roberts
reviewerQuoted 15 lines · Barry Song
Quoted 14 lines · Ryan Roberts
The legacy PMD-sized THP counters at
/proc/vmstatincludethp_file_alloc,thp_file_fallbackandthp_file_fallback_charge, which rather confusingly refer to shmem THP and do not include any other types of file pages. This is inconsistent since in most other places in the kernel, THP counters are explicitly separated for anon, shmem and file flavours. However, we are stuck with it since it constitutes a user ABI.Recently, commit
66f44583f9b6("mm: shmem: add mTHP counters for anonymous shmem") added equivalent mTHP stats for shmem, keeping the same "file_" prefix in the names. But in future, we may want to add extra stats to cover actual file pages, at which point, it would all become very confusing.
Quoted 6 lines
Quoted 2 lines
So let's take the opportunity to rename these new counters "
shmem_" before the change makes it upstream and the ABI becomes immutable.Personally, I think this approach is much clearer. However, I recall we discussed this
before [1], and it seems that inconsistency is a concern?
Embarrassingly, I don't recall that converstation at all :-| but at least what I said then is consistent with what I've done in this patch.
I think David's conclusion from that thread was to call them FILE_, and add both shmem and pagecache counts to those counters, meaning we can keep the same name as legacy THP counters. But those legacy THP counters only count shmem, and I don't think we would get away with adding pagecache counts to those at this point? (argument: they have been around for long time and there is a risk that user space relies on them and if they were to dramatically increase due to pagecache addition now that could break things). In that case, there is still inconsistency, but its worse; the names are consistent but the semantics are inconsistent.
So my vote is to change to SHMEM_ as per this patch :)
Quoted 149 lines
Quoted 140 lines
Signed-off-by Ryan Roberts <[email protected]>---
Hi All,
Applies on top of today's mm-unstable (
2073cda629a4) and tested with mm selftests; no regressions observed.The backstory here is that I'd like to introduce some counters for regular file folio allocations to observe how often large folio allocation succeeds, but these shmem counters are named "file" which is going to make things confusing. So hoping to solve that before commit
66f44583f9b6("mm: shmem: add mTHP counters for anonymous shmem") goes upstream (it is currently in mm-stable).Admittedly, this change means the mTHP stat names are not the same as the legacy PMD-size THP names, but I think that's a smaller issue than having "
file_" mTHP stats that only count shmem, then having to introduce "file2_" or "pgcache_" stats for the regular file memory, which is even more inconsistent IMHO. I guess the alternative is to count both shmem and file in these mTHP stats (that's how they were documented anyway) but I think it's better to be able to consider them separately like we do for all the other counters.Thanks,
RyanDocumentation/admin-guide/mm/transhuge.rst | 12 ++++++------ include/linux/huge_mm.h | 6 +++--- mm/huge_memory.c | 12 ++++++------ mm/shmem.c | 8 ++++---- 4 files changed, 19 insertions(+), 19 deletions(-)Documentation/admin-guide/mm/transhuge.rst +6
-6@@ -496,16 +496,16 @@ 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 +shmem_alloc + is incremented every time a shmem huge page is successfully allocated. -file_fallback - is incremented if a file huge page is attempted to be allocated +shmem_fallback + is incremented if a shmem 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 +shmem_fallback_charge + is incremented if a shmem huge page cannot be charged and instead falls back to using small pages even though the allocation was successful.include/linux/huge_mm.h +3
-3@@ -269,9 +269,9 @@ 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_SHMEM_ALLOC , +MTHP_STAT_SHMEM_FALLBACK , +MTHP_STAT_SHMEM_FALLBACK_CHARGE ,MTHP_STAT_SPLIT ,MTHP_STAT_SPLIT_FAILED , MTHP_STAT_SPLIT_DEFERRED,mm/huge_memory.c +6
-6@@ -568,9 +568,9 @@ 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 ); +DEFINE_MTHP_STAT_ATTR (shmem_alloc,MTHP_STAT_SHMEM_ALLOC ); +DEFINE_MTHP_STAT_ATTR (shmem_fallback,MTHP_STAT_SHMEM_FALLBACK ); +DEFINE_MTHP_STAT_ATTR (shmem_fallback_charge,MTHP_STAT_SHMEM_FALLBACK_CHARGE );DEFINE_MTHP_STAT_ATTR (split,MTHP_STAT_SPLIT );DEFINE_MTHP_STAT_ATTR (split_failed,MTHP_STAT_SPLIT_FAILED );DEFINE_MTHP_STAT_ATTR (split_deferred,MTHP_STAT_SPLIT_DEFERRED );@@ -581,9 +581,9 @@ 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, + &shmem_alloc_attr.attr, + &shmem_fallback_attr.attr, + &shmem_fallback_charge_attr.attr, &split_attr.attr, &split_failed_attr.attr, &split_deferred_attr.attr,mm/shmem.c +4
-5@@ -1777,7 +1777,7 @@ 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 ); +count_mthp_stat (order,MTHP_STAT_SHMEM_FALLBACK );#endif order =next_order (&suitable_orders, order); }@@ -1804,8 +1804,8 @@ static struct folio *shmem_alloc_and_add_folio(struct vm_fault *vmf,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 ); +count_mthp_stat (folio_order (folio),MTHP_STAT_SHMEM_FALLBACK ); +count_mthp_stat (folio_order (folio),MTHP_STAT_SHMEM_FALLBACK_CHARGE );#endif } goto unlock;@@ -2181,7 +2181,7 @@ static int shmem_get_folio_gfp(struct inode *inode, pgoff_t index,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 ); +count_mthp_stat (folio_order (folio),MTHP_STAT_SHMEM_ALLOC );#endif goto alloced; } --Quoted 2 lines
2.43.0
Thanks
Barry
Quoted 36 lines · Ryan Roberts
Quoted 15 lines · Barry Song
Quoted 14 lines · Ryan Roberts
The legacy PMD-sized THP counters at
/proc/vmstatincludethp_file_alloc,thp_file_fallbackandthp_file_fallback_charge, which rather confusingly refer to shmem THP and do not include any other types of file pages. This is inconsistent since in most other places in the kernel, THP counters are explicitly separated for anon, shmem and file flavours. However, we are stuck with it since it constitutes a user ABI.Recently, commit
66f44583f9b6("mm: shmem: add mTHP counters for anonymous shmem") added equivalent mTHP stats for shmem, keeping the same "file_" prefix in the names. But in future, we may want to add extra stats to cover actual file pages, at which point, it would all become very confusing.Quoted 6 lines
Quoted 2 lines
So let's take the opportunity to rename these new counters "
shmem_" before the change makes it upstream and the ABI becomes immutable.Personally, I think this approach is much clearer. However, I recall we discussed this
before [1], and it seems that inconsistency is a concern?Embarrassingly, I don't recall that converstation at all :-| but at least what I said then is consistent with what I've done in this patch.
I think David's conclusion from that thread was to call them
FILE_, and add both shmem and pagecache counts to those counters, meaning we can keep the same name as legacy THP counters. But those legacy THP counters only count shmem, and I don't think we would get away with adding pagecache counts to those at this point? (argument: they have been around for long time and there is a risk that user space relies on them and if they were to dramatically increase due to pagecache addition now that could break things). In that case, there is still inconsistency, but its worse; the names are consistent but the semantics are inconsistent.
Quoted 1 line
So my vote is to change to
SHMEM_as per this patch :)
I also forgot most of the discussion, but these 3 legacy counters are really only (currently) incremented for shmem. I think my idea was to keep everything as FILE_ for now, maybe at some point make the pagecache also use them, and then maybe have separate FILE_ + SHMEM_.
But yeah, likely it's best to only have "shmem" here for now, because who knows what we can actually change about the legacy counters. But it's always though messing with legacy stuff that is clearly suboptimal ...
Signature · 3 lines
Cheers, David / dhildenb
Quoted 40 lines · David Hildenbrand
Quoted 36 lines · Ryan Roberts
Quoted 15 lines · Barry Song
Quoted 14 lines · Ryan Roberts
The legacy PMD-sized THP counters at
/proc/vmstatincludethp_file_alloc,thp_file_fallbackandthp_file_fallback_charge, which rather confusingly refer to shmem THP and do not include any other types of file pages. This is inconsistent since in most other places in the kernel, THP counters are explicitly separated for anon, shmem and file flavours. However, we are stuck with it since it constitutes a user ABI.Recently, commit
66f44583f9b6("mm: shmem: add mTHP counters for anonymous shmem") added equivalent mTHP stats for shmem, keeping the same "file_" prefix in the names. But in future, we may want to add extra stats to cover actual file pages, at which point, it would all become very confusing.Quoted 6 lines
Quoted 2 lines
So let's take the opportunity to rename these new counters "
shmem_" before the change makes it upstream and the ABI becomes immutable.Personally, I think this approach is much clearer. However, I recall we discussed this
before [1], and it seems that inconsistency is a concern?Embarrassingly, I don't recall that converstation at all :-| but at least what I said then is consistent with what I've done in this patch.
I think David's conclusion from that thread was to call them
FILE_, and add both shmem and pagecache counts to those counters, meaning we can keep the same name as legacy THP counters. But those legacy THP counters only count shmem, and I don't think we would get away with adding pagecache counts to those at this point? (argument: they have been around for long time and there is a risk that user space relies on them and if they were to dramatically increase due to pagecache addition now that could break things). In that case, there is still inconsistency, but its worse; the names are consistent but the semantics are inconsistent.Quoted 1 line
So my vote is to change to
SHMEM_as per this patch :)
Quoted 8 lines
I also forgot most of the discussion, but these 3 legacy counters are really only (currently) incremented for shmem. I think my idea was to keep everything as
FILE_for now, maybe at some point make the pagecache also use them, and then maybe have separateFILE_+SHMEM_.But yeah, likely it's best to only have "shmem" here for now, because who knows what we can actually change about the legacy counters. But it's always though messing with legacy stuff that is clearly suboptimal ...
Couldn't agree more! It's never an easy task to handle such matters :) Perhaps, the time has come for us to separate FILE_ and SHMEM_.
Thanks,
Lance
Quoted 6 lines
--
Cheers,David / dhildenb
Ryan Roberts
reviewerQuoted 39 lines · David Hildenbrand
Quoted 36 lines · Ryan Roberts
Quoted 15 lines · Barry Song
Quoted 14 lines · Ryan Roberts
The legacy PMD-sized THP counters at
/proc/vmstatincludethp_file_alloc,thp_file_fallbackandthp_file_fallback_charge, which rather confusingly refer to shmem THP and do not include any other types of file pages. This is inconsistent since in most other places in the kernel, THP counters are explicitly separated for anon, shmem and file flavours. However, we are stuck with it since it constitutes a user ABI.Recently, commit
66f44583f9b6("mm: shmem: add mTHP counters for anonymous shmem") added equivalent mTHP stats for shmem, keeping the same "file_" prefix in the names. But in future, we may want to add extra stats to cover actual file pages, at which point, it would all become very confusing.Quoted 6 lines
Quoted 2 lines
So let's take the opportunity to rename these new counters "
shmem_" before the change makes it upstream and the ABI becomes immutable.Personally, I think this approach is much clearer. However, I recall we discussed this
before [1], and it seems that inconsistency is a concern?Embarrassingly, I don't recall that converstation at all :-| but at least what I said then is consistent with what I've done in this patch.
I think David's conclusion from that thread was to call them
FILE_, and add both shmem and pagecache counts to those counters, meaning we can keep the same name as legacy THP counters. But those legacy THP counters only count shmem, and I don't think we would get away with adding pagecache counts to those at this point? (argument: they have been around for long time and there is a risk that user space relies on them and if they were to dramatically increase due to pagecache addition now that could break things). In that case, there is still inconsistency, but its worse; the names are consistent but the semantics are inconsistent.Quoted 1 line
So my vote is to change to
SHMEM_as per this patch :)
Quoted 8 lines
I also forgot most of the discussion, but these 3 legacy counters are really only (currently) incremented for shmem. I think my idea was to keep everything as
FILE_for now, maybe at some point make the pagecache also use them, and then maybe have separateFILE_+SHMEM_.But yeah, likely it's best to only have "shmem" here for now, because who knows what we can actually change about the legacy counters. But it's always though messing with legacy stuff that is clearly suboptimal ...
Sorry David, I've read your response a bunch of times and am still not completely sure what you are advocating for.
To make my proposal crystal clear, I think we should leave the legacy counters alone - neither change their name nor what they count (which is _only_ shmem). I think we should rename the new mTHP counters to "shmem" and have them continue to only count shmem. Then additionally, I think we should introduce new "file" mTHP counters that count the page cache allocations (that's a future set of patches, which I'm working on together with user controls to determine which mTHP sizes can be used by page cache).
As suggested by Barry, I propose to also improve the documentation for the legacy counters to make it clear that dispite their name being "file" they are actually counting "shmem". I'll do this for v2.
David, would you support this approach? If so, I'd like to push this forwards asap so that it gets into v6.11 to avoid ever exposing the mthp counters with the "file" name.
Thanks,
Ryan
Quoted 60 lines · Ryan Roberts
Quoted 39 lines · David Hildenbrand
Quoted 36 lines · Ryan Roberts
Quoted 15 lines · Barry Song
Quoted 14 lines · Ryan Roberts
The legacy PMD-sized THP counters at
/proc/vmstatincludethp_file_alloc,thp_file_fallbackandthp_file_fallback_charge, which rather confusingly refer to shmem THP and do not include any other types of file pages. This is inconsistent since in most other places in the kernel, THP counters are explicitly separated for anon, shmem and file flavours. However, we are stuck with it since it constitutes a user ABI.Recently, commit
66f44583f9b6("mm: shmem: add mTHP counters for anonymous shmem") added equivalent mTHP stats for shmem, keeping the same "file_" prefix in the names. But in future, we may want to add extra stats to cover actual file pages, at which point, it would all become very confusing.Quoted 6 lines
Quoted 2 lines
So let's take the opportunity to rename these new counters "
shmem_" before the change makes it upstream and the ABI becomes immutable.Personally, I think this approach is much clearer. However, I recall we discussed this
before [1], and it seems that inconsistency is a concern?Embarrassingly, I don't recall that converstation at all :-| but at least what I said then is consistent with what I've done in this patch.
I think David's conclusion from that thread was to call them
FILE_, and add both shmem and pagecache counts to those counters, meaning we can keep the same name as legacy THP counters. But those legacy THP counters only count shmem, and I don't think we would get away with adding pagecache counts to those at this point? (argument: they have been around for long time and there is a risk that user space relies on them and if they were to dramatically increase due to pagecache addition now that could break things). In that case, there is still inconsistency, but its worse; the names are consistent but the semantics are inconsistent.Quoted 1 line
So my vote is to change to
SHMEM_as per this patch :)Quoted 8 lines
I also forgot most of the discussion, but these 3 legacy counters are really only (currently) incremented for shmem. I think my idea was to keep everything as
FILE_for now, maybe at some point make the pagecache also use them, and then maybe have separateFILE_+SHMEM_.But yeah, likely it's best to only have "shmem" here for now, because who knows what we can actually change about the legacy counters. But it's always though messing with legacy stuff that is clearly suboptimal ...
Sorry David, I've read your response a bunch of times and am still not completely sure what you are advocating for.
To make my proposal crystal clear, I think we should leave the legacy counters alone - neither change their name nor what they count (which is _only_ shmem). I think we should rename the new mTHP counters to "shmem" and have them continue to only count shmem. Then additionally, I think we should introduce new "file" mTHP counters that count the page cache allocations (that's a future set of patches, which I'm working on together with user controls to determine which mTHP sizes can be used by page cache).
Quoted 3 lines
As suggested by Barry, I propose to also improve the documentation for the legacy counters to make it clear that dispite their name being "file" they are actually counting "shmem". I'll do this for v2.
Yes, and please. Likely we should document for the legacy ones (if not already done) that they only track PMD-sized THPs.
Quoted 4 lines
David, would you support this approach? If so, I'd like to push this forwards asap so that it gets into v6.11 to avoid ever exposing the mthp counters with the "file" name.
Yes, sorry for not being clear.
Signature · 3 lines
Cheers, David / dhildenb
Ryan Roberts
reviewerQuoted 63 lines · David Hildenbrand
Quoted 62 lines · Ryan Roberts
Quoted 41 lines · David Hildenbrand
Quoted 38 lines · Ryan Roberts
Quoted 15 lines · Barry Song
Quoted 14 lines · Ryan Roberts
The legacy PMD-sized THP counters at
/proc/vmstatincludethp_file_alloc,thp_file_fallbackandthp_file_fallback_charge, which rather confusingly refer to shmem THP and do not include any other types of file pages. This is inconsistent since in most other places in the kernel, THP counters are explicitly separated for anon, shmem and file flavours. However, we are stuck with it since it constitutes a user ABI.Recently, commit
66f44583f9b6("mm: shmem: add mTHP counters for anonymous shmem") added equivalent mTHP stats for shmem, keeping the same "file_" prefix in the names. But in future, we may want to add extra stats to cover actual file pages, at which point, it would all become very confusing.Quoted 6 lines
Quoted 2 lines
So let's take the opportunity to rename these new counters "
shmem_" before the change makes it upstream and the ABI becomes immutable.Personally, I think this approach is much clearer. However, I recall we discussed this
before [1], and it seems that inconsistency is a concern?Embarrassingly, I don't recall that converstation at all :-| but at least what I
said then is consistent with what I've done in this patch.I think David's conclusion from that thread was to call them
FILE_, and add both
shmem and pagecache counts to those counters, meaning we can keep the same name as legacy THP counters. But those legacy THP counters only count shmem, and I don't think we would get away with adding pagecache counts to those at this point? (argument: they have been around for long time and there is a risk that user space relies on them and if they were to dramatically increase due to pagecache addition now that could break things). In that case, there is still inconsistency, but its worse; the names are consistent but the semantics are inconsistent.Quoted 1 line
So my vote is to change to
SHMEM_as per this patch :)Quoted 8 lines
I also forgot most of the discussion, but these 3 legacy counters are really only (currently) incremented for shmem. I think my idea was to keep everything as
FILE_for now, maybe at some point make the pagecache also use them, and then maybe have separateFILE_+SHMEM_.But yeah, likely it's best to only have "shmem" here for now, because who knows what we can actually change about the legacy counters. But it's always though messing with legacy stuff that is clearly suboptimal ...
Sorry David, I've read your response a bunch of times and am still not completely sure what you are advocating for.
To make my proposal crystal clear, I think we should leave the legacy counters alone - neither change their name nor what they count (which is _only_ shmem). I think we should rename the new mTHP counters to "shmem" and have them continue to only count shmem. Then additionally, I think we should introduce new "file" mTHP counters that count the page cache allocations (that's a future set of patches, which I'm working on together with user controls to determine which mTHP sizes can be used by page cache).
Quoted 6 lines
Quoted 3 lines
As suggested by Barry, I propose to also improve the documentation for the legacy counters to make it clear that dispite their name being "file" they are actually counting "shmem". I'll do this for v2.
Yes, and please. Likely we should document for the legacy ones (if not already done) that they only track PMD-sized THPs.
The PMD-ness is already documented for the legacy counters IIRC, but I'll double check.
Quoted 7 lines
Quoted 5 lines
David, would you support this approach? If so, I'd like to push this forwards asap so that it gets into v6.11 to avoid ever exposing the mthp counters with the "file" name.
Yes, sorry for not being clear.
No worries, it sounds like at least Lance understood exectly what you were saying, so likely I'm just being obtuse.
Anyway, I'll respin with the docs improvements and get v2 out shortly.
Thanks,
Ryan
Quoted 36 lines · Ryan Roberts
Quoted 15 lines · Barry Song
Quoted 14 lines · Ryan Roberts
The legacy PMD-sized THP counters at
/proc/vmstatincludethp_file_alloc,thp_file_fallbackandthp_file_fallback_charge, which rather confusingly refer to shmem THP and do not include any other types of file pages. This is inconsistent since in most other places in the kernel, THP counters are explicitly separated for anon, shmem and file flavours. However, we are stuck with it since it constitutes a user ABI.Recently, commit
66f44583f9b6("mm: shmem: add mTHP counters for anonymous shmem") added equivalent mTHP stats for shmem, keeping the same "file_" prefix in the names. But in future, we may want to add extra stats to cover actual file pages, at which point, it would all become very confusing.Quoted 6 lines
Quoted 2 lines
So let's take the opportunity to rename these new counters "
shmem_" before the change makes it upstream and the ABI becomes immutable.Personally, I think this approach is much clearer. However, I recall we discussed this
before [1], and it seems that inconsistency is a concern?Embarrassingly, I don't recall that converstation at all :-| but at least what I said then is consistent with what I've done in this patch.
I think David's conclusion from that thread was to call them
FILE_, and add both shmem and pagecache counts to those counters, meaning we can keep the same name as legacy THP counters. But those legacy THP counters only count shmem, and I don't think we would get away with adding pagecache counts to those at this point? (argument: they have been around for long time and there is a risk that user space relies on them and if they were to dramatically increase due to pagecache addition now that could break things). In that case, there is still inconsistency, but its worse; the names are consistent but the semantics are inconsistent.
Quoted 4 lines
So my vote is to change to
SHMEM_as per this patch :)
My original preference was also for SHMEM-specific counters :) So feel free to add:
PATCH 1/1 mm: shmem: Rename mTHP shmem counters
In reply to Ryan RobertsQuoted 37 lines · Ryan Roberts
Quoted 15 lines · Barry Song
Quoted 14 lines · Ryan Roberts
The legacy PMD-sized THP counters at
/proc/vmstatincludethp_file_alloc,thp_file_fallbackandthp_file_fallback_charge, which rather confusingly refer to shmem THP and do not include any other types of file pages. This is inconsistent since in most other places in the kernel, THP counters are explicitly separated for anon, shmem and file flavours. However, we are stuck with it since it constitutes a user ABI.Recently, commit
66f44583f9b6("mm: shmem: add mTHP counters for anonymous shmem") added equivalent mTHP stats for shmem, keeping the same "file_" prefix in the names. But in future, we may want to add extra stats to cover actual file pages, at which point, it would all become very confusing.Quoted 6 lines
Quoted 2 lines
So let's take the opportunity to rename these new counters "
shmem_" before the change makes it upstream and the ABI becomes immutable.Personally, I think this approach is much clearer. However, I recall we discussed this
before [1], and it seems that inconsistency is a concern?Embarrassingly, I don't recall that converstation at all :-| but at least what I said then is consistent with what I've done in this patch.
I think David's conclusion from that thread was to call them
FILE_, and add both shmem and pagecache counts to those counters, meaning we can keep the same name as legacy THP counters. But those legacy THP counters only count shmem, and I don't think we would get away with adding pagecache counts to those at this point? (argument: they have been around for long time and there is a risk that user space relies on them and if they were to dramatically increase due to pagecache addition now that could break things). In that case, there is still inconsistency, but its worse; the names are consistent but the semantics are inconsistent.
Quoted 1 line
So my vote is to change to
SHMEM_as per this patch :)
I have no objections. However, I dislike the documentation for thp_file_*. Perhaps we can clean it all up together ?
Documentation/admin-guide/mm/transhuge.rst +10 -9
@@ -417,21 +417,22 @@ thp_collapse_alloc_failed the allocation. thp_file_alloc - is incremented every time a file huge page is successfully - allocated. + is incremented every time a file (including shmem) huge page is + successfully allocated. thp_file_fallback - is incremented if a file huge page is attempted to be allocated - but fails and instead falls back to using small pages. + is incremented if a file (including shmem) huge page is attempted + to be allocated but fails and instead falls back to using small + pages. thp_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. + is incremented if a file (including shmem) huge page cannot be + charged and instead falls back to using small pages even though + the allocation was successful. thp_file_mapped - is incremented every time a file huge page is mapped into - user address space. + is incremented every time a file (including shmem) huge page is + mapped into user address space. thp_split_page is incremented every time a huge page is split into base
Quoted 140 lines
Quoted 140 lines
Quoted 135 lines
Signed-off-by Ryan Roberts <[email protected]>---
Hi All,
Applies on top of today's mm-unstable (
2073cda629a4) and tested with mm selftests; no regressions observed.The backstory here is that I'd like to introduce some counters for regular file folio allocations to observe how often large folio allocation succeeds, but these shmem counters are named "file" which is going to make things confusing. So hoping to solve that before commit
66f44583f9b6("mm: shmem: add mTHP counters for anonymous shmem") goes upstream (it is currently in mm-stable).Admittedly, this change means the mTHP stat names are not the same as the legacy PMD-size THP names, but I think that's a smaller issue than having "
file_" mTHP stats that only count shmem, then having to introduce "file2_" or "pgcache_" stats for the regular file memory, which is even more inconsistent IMHO. I guess the alternative is to count both shmem and file in these mTHP stats (that's how they were documented anyway) but I think it's better to be able to consider them separately like we do for all the other counters.Thanks,
RyanDocumentation/admin-guide/mm/transhuge.rst | 12 ++++++------
include/linux/huge_mm.h| 6 +++---mm/huge_memory.c| 12 ++++++------mm/shmem.c| 8 ++++----4 files changed, 19 insertions(+), 19 deletions(-)Documentation/admin-guide/mm/transhuge.rst +0
-0@@ -496,16 +496,16 @@ swpout_fallbackUsually because failed to allocate some continuous swap space for the huge page.
Documentation/admin-guide/mm/transhuge.rst +2
-2-file_alloc - is incremented every time a file huge page is successfully +shmem_alloc + is incremented every time a shmem huge page is successfullyallocated.
Documentation/admin-guide/mm/transhuge.rst +2
-2-file_fallback - is incremented if a file huge page is attempted to be allocated +shmem_fallback + is incremented if a shmem huge page is attempted to be allocatedbut fails and instead falls back to using small pages.
Documentation/admin-guide/mm/transhuge.rst +2
-2-file_fallback_charge - is incremented if a file huge page cannot be charged and instead +shmem_fallback_charge + is incremented if a shmem huge page cannot be charged and insteadfalls back to using small pages even though the allocation was successful.
include/linux/huge_mm.h +3
-3@@ -269,9 +269,9 @@ 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_SHMEM_ALLOC, + MTHP_STAT_SHMEM_FALLBACK, + MTHP_STAT_SHMEM_FALLBACK_CHARGE, MTHP_STAT_SPLIT, MTHP_STAT_SPLIT_FAILED, MTHP_STAT_SPLIT_DEFERRED,mm/huge_memory.c +6
-6@@ -568,9 +568,9 @@ 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 ); +DEFINE_MTHP_STAT_ATTR (shmem_alloc,MTHP_STAT_SHMEM_ALLOC ); +DEFINE_MTHP_STAT_ATTR (shmem_fallback,MTHP_STAT_SHMEM_FALLBACK ); +DEFINE_MTHP_STAT_ATTR (shmem_fallback_charge,MTHP_STAT_SHMEM_FALLBACK_CHARGE ); DEFINE_MTHP_STAT_ATTR(split, MTHP_STAT_SPLIT); DEFINE_MTHP_STAT_ATTR(split_failed, MTHP_STAT_SPLIT_FAILED); DEFINE_MTHP_STAT_ATTR(split_deferred, MTHP_STAT_SPLIT_DEFERRED);@@ -581,9 +581,9 @@ 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, + &shmem_alloc_attr.attr, + &shmem_fallback_attr.attr, + &shmem_fallback_charge_attr.attr, &split_attr.attr, &split_failed_attr.attr, &split_deferred_attr.attr,mm/shmem.c +3
-4@@ -1777,7 +1777,7 @@ 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); + count_mthp_stat(order, MTHP_STAT_SHMEM_FALLBACK); #endif order = next_order(&suitable_orders, order); }@@ -1804,8 +1804,8 @@ static struct folio *shmem_alloc_and_add_folio(struct vm_fault *vmf, 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); + count_mthp_stat(folio_order(folio), MTHP_STAT_SHMEM_FALLBACK); + count_mthp_stat(folio_order(folio), MTHP_STAT_SHMEM_FALLBACK_CHARGE); #endif } goto unlock;@@ -2181,7 +2181,7 @@ static int shmem_get_folio_gfp(struct inode *inode, pgoff_t index, 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);
Quoted 8 lines
Quoted 8 lines
Quoted 7 lines
mm/shmem.c +1
-1+ count_mthp_stat(folio_order(folio), MTHP_STAT_SHMEM_ALLOC); #endif goto alloced; } --2.43.0
Thanks
Barry
Ryan Roberts
reviewerQuoted 38 lines · Barry Song
Quoted 37 lines · Ryan Roberts
Quoted 15 lines · Barry Song
Quoted 14 lines · Ryan Roberts
The legacy PMD-sized THP counters at
/proc/vmstatincludethp_file_alloc,thp_file_fallbackandthp_file_fallback_charge, which rather confusingly refer to shmem THP and do not include any other types of file pages. This is inconsistent since in most other places in the kernel, THP counters are explicitly separated for anon, shmem and file flavours. However, we are stuck with it since it constitutes a user ABI.Recently, commit
66f44583f9b6("mm: shmem: add mTHP counters for anonymous shmem") added equivalent mTHP stats for shmem, keeping the same "file_" prefix in the names. But in future, we may want to add extra stats to cover actual file pages, at which point, it would all become very confusing.Quoted 6 lines
Quoted 2 lines
So let's take the opportunity to rename these new counters "
shmem_" before the change makes it upstream and the ABI becomes immutable.Personally, I think this approach is much clearer. However, I recall we discussed this
before [1], and it seems that inconsistency is a concern?Embarrassingly, I don't recall that converstation at all :-| but at least what I said then is consistent with what I've done in this patch.
I think David's conclusion from that thread was to call them
FILE_, and add both shmem and pagecache counts to those counters, meaning we can keep the same name as legacy THP counters. But those legacy THP counters only count shmem, and I don't think we would get away with adding pagecache counts to those at this point? (argument: they have been around for long time and there is a risk that user space relies on them and if they were to dramatically increase due to pagecache addition now that could break things). In that case, there is still inconsistency, but its worse; the names are consistent but the semantics are inconsistent.
Quoted 4 lines
Quoted 1 line
So my vote is to change to
SHMEM_as per this patch :)I have no objections. However, I dislike the documentation for thp_file_*. Perhaps we can clean it all up together ?
I agree that we should clean this documentation up and I'm happy to roll it into v2. However, I don't think what you have suggested is quite right.
thp_file_alloc, thp_file_fallback and thp_file_fallback_charge *only* count shmem. They don't count pagecache. So perhaps the change should be "...every time a shmem huge page (dispite being named after "file", the counter measures only shmem) is..."?
thp_file_mapped includes both file and shmem, so agree with your change there.
What do you think?
Quoted 189 lines
Documentation/admin-guide/mm/transhuge.rst +10
-9@@ -417,21 +417,22 @@ thp_collapse_alloc_failed the allocation. thp_file_alloc - is incremented every time a file huge page is successfully - allocated. + is incremented every time a file (including shmem) huge page is + successfully allocated. thp_file_fallback - is incremented if a file huge page is attempted to be allocated - but fails and instead falls back to using small pages. + is incremented if a file (including shmem) huge page is attempted + to be allocated but fails and instead falls back to using small + pages. thp_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. + is incremented if a file (including shmem) huge page cannot be + charged and instead falls back to using small pages even though + the allocation was successful. thp_file_mapped - is incremented every time a file huge page is mapped into - user address space. + is incremented every time a file (including shmem) huge page is + mapped into user address space. thp_split_page is incremented every time a huge page is split into baseQuoted 141 lines
Quoted 141 lines
Quoted 135 lines
Signed-off-by Ryan Roberts <[email protected]>---
Hi All,
Applies on top of today's mm-unstable (
2073cda629a4) and tested with mm selftests; no regressions observed.The backstory here is that I'd like to introduce some counters for regular file folio allocations to observe how often large folio allocation succeeds, but these shmem counters are named "file" which is going to make things confusing. So hoping to solve that before commit
66f44583f9b6("mm: shmem: add mTHP counters for anonymous shmem") goes upstream (it is currently in mm-stable).Admittedly, this change means the mTHP stat names are not the same as the legacy PMD-size THP names, but I think that's a smaller issue than having "
file_" mTHP stats that only count shmem, then having to introduce "file2_" or "pgcache_" stats for the regular file memory, which is even more inconsistent IMHO. I guess the alternative is to count both shmem and file in these mTHP stats (that's how they were documented anyway) but I think it's better to be able to consider them separately like we do for all the other counters.Thanks,
RyanDocumentation/admin-guide/mm/transhuge.rst | 12 ++++++------
include/linux/huge_mm.h| 6 +++---mm/huge_memory.c| 12 ++++++------mm/shmem.c| 8 ++++----4 files changed, 19 insertions(+), 19 deletions(-)Documentation/admin-guide/mm/transhuge.rst +0
-0@@ -496,16 +496,16 @@ swpout_fallbackUsually because failed to allocate some continuous swap space for the huge page.
Documentation/admin-guide/mm/transhuge.rst +2
-2-file_alloc - is incremented every time a file huge page is successfully +shmem_alloc + is incremented every time a shmem huge page is successfullyallocated.
Documentation/admin-guide/mm/transhuge.rst +2
-2-file_fallback - is incremented if a file huge page is attempted to be allocated +shmem_fallback + is incremented if a shmem huge page is attempted to be allocatedbut fails and instead falls back to using small pages.
Documentation/admin-guide/mm/transhuge.rst +2
-2-file_fallback_charge - is incremented if a file huge page cannot be charged and instead +shmem_fallback_charge + is incremented if a shmem huge page cannot be charged and insteadfalls back to using small pages even though the allocation was successful.
include/linux/huge_mm.h +3
-3@@ -269,9 +269,9 @@ 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_SHMEM_ALLOC, + MTHP_STAT_SHMEM_FALLBACK, + MTHP_STAT_SHMEM_FALLBACK_CHARGE, MTHP_STAT_SPLIT, MTHP_STAT_SPLIT_FAILED, MTHP_STAT_SPLIT_DEFERRED,mm/huge_memory.c +6
-6@@ -568,9 +568,9 @@ 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 ); +DEFINE_MTHP_STAT_ATTR (shmem_alloc,MTHP_STAT_SHMEM_ALLOC ); +DEFINE_MTHP_STAT_ATTR (shmem_fallback,MTHP_STAT_SHMEM_FALLBACK ); +DEFINE_MTHP_STAT_ATTR (shmem_fallback_charge,MTHP_STAT_SHMEM_FALLBACK_CHARGE ); DEFINE_MTHP_STAT_ATTR(split, MTHP_STAT_SPLIT); DEFINE_MTHP_STAT_ATTR(split_failed, MTHP_STAT_SPLIT_FAILED); DEFINE_MTHP_STAT_ATTR(split_deferred, MTHP_STAT_SPLIT_DEFERRED);@@ -581,9 +581,9 @@ 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, + &shmem_alloc_attr.attr, + &shmem_fallback_attr.attr, + &shmem_fallback_charge_attr.attr, &split_attr.attr, &split_failed_attr.attr, &split_deferred_attr.attr,mm/shmem.c +3
-4@@ -1777,7 +1777,7 @@ 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); + count_mthp_stat(order, MTHP_STAT_SHMEM_FALLBACK); #endif order = next_order(&suitable_orders, order); }@@ -1804,8 +1804,8 @@ static struct folio *shmem_alloc_and_add_folio(struct vm_fault *vmf, 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); + count_mthp_stat(folio_order(folio), MTHP_STAT_SHMEM_FALLBACK); + count_mthp_stat(folio_order(folio), MTHP_STAT_SHMEM_FALLBACK_CHARGE); #endif } goto unlock;@@ -2181,7 +2181,7 @@ static int shmem_get_folio_gfp(struct inode *inode, pgoff_t index, 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);Quoted 8 lines
Quoted 8 lines
Quoted 7 lines
mm/shmem.c +1
-1+ count_mthp_stat(folio_order(folio), MTHP_STAT_SHMEM_ALLOC); #endif goto alloced; } --2.43.0
Thanks
Barry
Quoted 45 lines · Ryan Roberts
Quoted 38 lines · Barry Song
Quoted 37 lines · Ryan Roberts
Quoted 15 lines · Barry Song
Quoted 14 lines · Ryan Roberts
The legacy PMD-sized THP counters at
/proc/vmstatincludethp_file_alloc,thp_file_fallbackandthp_file_fallback_charge, which rather confusingly refer to shmem THP and do not include any other types of file pages. This is inconsistent since in most other places in the kernel, THP counters are explicitly separated for anon, shmem and file flavours. However, we are stuck with it since it constitutes a user ABI.Recently, commit
66f44583f9b6("mm: shmem: add mTHP counters for anonymous shmem") added equivalent mTHP stats for shmem, keeping the same "file_" prefix in the names. But in future, we may want to add extra stats to cover actual file pages, at which point, it would all become very confusing.Quoted 6 lines
Quoted 2 lines
So let's take the opportunity to rename these new counters "
shmem_" before the change makes it upstream and the ABI becomes immutable.Personally, I think this approach is much clearer. However, I recall we discussed this
before [1], and it seems that inconsistency is a concern?Embarrassingly, I don't recall that converstation at all :-| but at least what I said then is consistent with what I've done in this patch.
I think David's conclusion from that thread was to call them
FILE_, and add both shmem and pagecache counts to those counters, meaning we can keep the same name as legacy THP counters. But those legacy THP counters only count shmem, and I don't think we would get away with adding pagecache counts to those at this point? (argument: they have been around for long time and there is a risk that user space relies on them and if they were to dramatically increase due to pagecache addition now that could break things). In that case, there is still inconsistency, but its worse; the names are consistent but the semantics are inconsistent.Quoted 4 lines
Quoted 1 line
So my vote is to change to
SHMEM_as per this patch :)I have no objections. However, I dislike the documentation for thp_file_*. Perhaps we can clean it all up together ?
Quoted 7 lines
I agree that we should clean this documentation up and I'm happy to roll it into v2. However, I don't think what you have suggested is quite right.
thp_file_alloc,thp_file_fallbackandthp_file_fallback_charge*only* count shmem. They don't count pagecache. So perhaps the change should be "...every time a shmem huge page (dispite being named after "file", the counter measures only shmem) is..."?
I understand what you are saying, and I know that thp_file_* has only included shmem so far. My question is whether it will include regular files in the future? If not, I am perfectly fine with your approach.
READ_ONLY_THP_FOR_FS isn't applicable in this path as it is created by khugepaged collapse.
Quoted 192 lines
thp_file_mappedincludes both file and shmem, so agree with your change there.What do you think?
Quoted 186 lines
Documentation/admin-guide/mm/transhuge.rst +10
-9@@ -417,21 +417,22 @@ thp_collapse_alloc_failed the allocation. thp_file_alloc - is incremented every time a file huge page is successfully - allocated. + is incremented every time a file (including shmem) huge page is + successfully allocated. thp_file_fallback - is incremented if a file huge page is attempted to be allocated - but fails and instead falls back to using small pages. + is incremented if a file (including shmem) huge page is attempted + to be allocated but fails and instead falls back to using small + pages. thp_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. + is incremented if a file (including shmem) huge page cannot be + charged and instead falls back to using small pages even though + the allocation was successful. thp_file_mapped - is incremented every time a file huge page is mapped into - user address space. + is incremented every time a file (including shmem) huge page is + mapped into user address space. thp_split_page is incremented every time a huge page is split into baseQuoted 149 lines
Quoted 149 lines
Quoted 142 lines
Signed-off-by Ryan Roberts <[email protected]>---
Hi All,
Applies on top of today's mm-unstable (
2073cda629a4) and tested with mm selftests; no regressions observed.The backstory here is that I'd like to introduce some counters for regular file folio allocations to observe how often large folio allocation succeeds, but these shmem counters are named "file" which is going to make things confusing. So hoping to solve that before commit
66f44583f9b6("mm: shmem: add mTHP counters for anonymous shmem") goes upstream (it is currently in mm-stable).Admittedly, this change means the mTHP stat names are not the same as the legacy PMD-size THP names, but I think that's a smaller issue than having "
file_" mTHP stats that only count shmem, then having to introduce "file2_" or "pgcache_" stats for the regular file memory, which is even more inconsistent IMHO. I guess the alternative is to count both shmem and file in these mTHP stats (that's how they were documented anyway) but I think it's better to be able to consider them separately like we do for all the other counters.Thanks,
RyanDocumentation/admin-guide/mm/transhuge.rst | 12 ++++++------ include/linux/huge_mm.h | 6 +++--- mm/huge_memory.c | 12 ++++++------ mm/shmem.c | 8 ++++---- 4 files changed, 19 insertions(+), 19 deletions(-)Documentation/admin-guide/mm/transhuge.rst +6
-6@@ -496,16 +496,16 @@ 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 +shmem_alloc + is incremented every time a shmem huge page is successfully allocated. -file_fallback - is incremented if a file huge page is attempted to be allocated +shmem_fallback + is incremented if a shmem 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 +shmem_fallback_charge + is incremented if a shmem huge page cannot be charged and instead falls back to using small pages even though the allocation was successful.include/linux/huge_mm.h +3
-3@@ -269,9 +269,9 @@ 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_SHMEM_ALLOC , +MTHP_STAT_SHMEM_FALLBACK , +MTHP_STAT_SHMEM_FALLBACK_CHARGE ,MTHP_STAT_SPLIT ,MTHP_STAT_SPLIT_FAILED , MTHP_STAT_SPLIT_DEFERRED,mm/huge_memory.c +6
-6@@ -568,9 +568,9 @@ 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 ); +DEFINE_MTHP_STAT_ATTR (shmem_alloc,MTHP_STAT_SHMEM_ALLOC ); +DEFINE_MTHP_STAT_ATTR (shmem_fallback,MTHP_STAT_SHMEM_FALLBACK ); +DEFINE_MTHP_STAT_ATTR (shmem_fallback_charge,MTHP_STAT_SHMEM_FALLBACK_CHARGE );DEFINE_MTHP_STAT_ATTR (split,MTHP_STAT_SPLIT );DEFINE_MTHP_STAT_ATTR (split_failed,MTHP_STAT_SPLIT_FAILED );DEFINE_MTHP_STAT_ATTR (split_deferred,MTHP_STAT_SPLIT_DEFERRED );@@ -581,9 +581,9 @@ 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, + &shmem_alloc_attr.attr, + &shmem_fallback_attr.attr, + &shmem_fallback_charge_attr.attr, &split_attr.attr, &split_failed_attr.attr, &split_deferred_attr.attr,mm/shmem.c +4
-5@@ -1777,7 +1777,7 @@ 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 ); +count_mthp_stat (order,MTHP_STAT_SHMEM_FALLBACK );#endif order =next_order (&suitable_orders, order); }@@ -1804,8 +1804,8 @@ static struct folio *shmem_alloc_and_add_folio(struct vm_fault *vmf,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 ); +count_mthp_stat (folio_order (folio),MTHP_STAT_SHMEM_FALLBACK ); +count_mthp_stat (folio_order (folio),MTHP_STAT_SHMEM_FALLBACK_CHARGE );#endif } goto unlock;@@ -2181,7 +2181,7 @@ static int shmem_get_folio_gfp(struct inode *inode, pgoff_t index,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 ); +count_mthp_stat (folio_order (folio),MTHP_STAT_SHMEM_ALLOC );#endif goto alloced; } --2.43.0
Thanks
Barry
Ryan Roberts
reviewerQuoted 49 lines · Barry Song
Quoted 48 lines · Ryan Roberts
Quoted 38 lines · Barry Song
Quoted 37 lines · Ryan Roberts
Quoted 15 lines · Barry Song
Quoted 14 lines · Ryan Roberts
The legacy PMD-sized THP counters at
/proc/vmstatincludethp_file_alloc,thp_file_fallbackandthp_file_fallback_charge, which rather confusingly refer to shmem THP and do not include any other types of file pages. This is inconsistent since in most other places in the kernel, THP counters are explicitly separated for anon, shmem and file flavours. However, we are stuck with it since it constitutes a user ABI.Recently, commit
66f44583f9b6("mm: shmem: add mTHP counters for anonymous shmem") added equivalent mTHP stats for shmem, keeping the same "file_" prefix in the names. But in future, we may want to add extra stats to cover actual file pages, at which point, it would all become very confusing.Quoted 6 lines
Quoted 2 lines
So let's take the opportunity to rename these new counters "
shmem_" before the change makes it upstream and the ABI becomes immutable.Personally, I think this approach is much clearer. However, I recall we discussed this
before [1], and it seems that inconsistency is a concern?Embarrassingly, I don't recall that converstation at all :-| but at least what I said then is consistent with what I've done in this patch.
I think David's conclusion from that thread was to call them
FILE_, and add both shmem and pagecache counts to those counters, meaning we can keep the same name as legacy THP counters. But those legacy THP counters only count shmem, and I don't think we would get away with adding pagecache counts to those at this point? (argument: they have been around for long time and there is a risk that user space relies on them and if they were to dramatically increase due to pagecache addition now that could break things). In that case, there is still inconsistency, but its worse; the names are consistent but the semantics are inconsistent.Quoted 4 lines
Quoted 1 line
So my vote is to change to
SHMEM_as per this patch :)I have no objections. However, I dislike the documentation for thp_file_*. Perhaps we can clean it all up together ?
I agree that we should clean this documentation up and I'm happy to roll it into v2. However, I don't think what you have suggested is quite right.
Quoted 8 lines
Quoted 4 lines
thp_file_alloc,thp_file_fallbackandthp_file_fallback_charge*only* count shmem. They don't count pagecache. So perhaps the change should be "...every time a shmem huge page (dispite being named after "file", the counter measures only shmem) is..."?I understand what you are saying, and I know that thp_file_* has only included shmem so far. My question is whether it will include regular files in the future? If not, I am perfectly fine with your approach.
My whole reasoning for this patch is based on the assertion that since THP_FILE_ALLOC has been there for 8 years and in all that time has only counted shmem, then its highly likely that someone is depending on that semantic and we can't change it. I don't have any actual evidence of code that relies on it though.
I propose I change the docs to reflect what's actually happening today (i.e. shmem *only*). If we later decide we want to also report page cache numbers through that same counter, then we can change the docs at that point. But if I get my way, we'll soon have mTHP counters for FILE, which is solely for page cache. So You'll be able to get all the fine-grained info out of those and there will be no need to mess with the legacy counters.
Quoted 199 lines
READ_ONLY_THP_FOR_FSisn't applicable in this path as it is created by khugepaged collapse.Quoted 192 lines
thp_file_mappedincludes both file and shmem, so agree with your change there.What do you think?
Quoted 186 lines
Documentation/admin-guide/mm/transhuge.rst +10
-9@@ -417,21 +417,22 @@ thp_collapse_alloc_failed the allocation. thp_file_alloc - is incremented every time a file huge page is successfully - allocated. + is incremented every time a file (including shmem) huge page is + successfully allocated. thp_file_fallback - is incremented if a file huge page is attempted to be allocated - but fails and instead falls back to using small pages. + is incremented if a file (including shmem) huge page is attempted + to be allocated but fails and instead falls back to using small + pages. thp_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. + is incremented if a file (including shmem) huge page cannot be + charged and instead falls back to using small pages even though + the allocation was successful. thp_file_mapped - is incremented every time a file huge page is mapped into - user address space. + is incremented every time a file (including shmem) huge page is + mapped into user address space. thp_split_page is incremented every time a huge page is split into baseQuoted 149 lines
Quoted 149 lines
Quoted 142 lines
Signed-off-by Ryan Roberts <[email protected]>---
Hi All,
Applies on top of today's mm-unstable (
2073cda629a4) and tested with mm selftests; no regressions observed.The backstory here is that I'd like to introduce some counters for regular file folio allocations to observe how often large folio allocation succeeds, but these shmem counters are named "file" which is going to make things confusing. So hoping to solve that before commit
66f44583f9b6("mm: shmem: add mTHP counters for anonymous shmem") goes upstream (it is currently in mm-stable).Admittedly, this change means the mTHP stat names are not the same as the legacy PMD-size THP names, but I think that's a smaller issue than having "
file_" mTHP stats that only count shmem, then having to introduce "file2_" or "pgcache_" stats for the regular file memory, which is even more inconsistent IMHO. I guess the alternative is to count both shmem and file in these mTHP stats (that's how they were documented anyway) but I think it's better to be able to consider them separately like we do for all the other counters.Thanks,
RyanDocumentation/admin-guide/mm/transhuge.rst | 12 ++++++------ include/linux/huge_mm.h | 6 +++--- mm/huge_memory.c | 12 ++++++------ mm/shmem.c | 8 ++++---- 4 files changed, 19 insertions(+), 19 deletions(-)Documentation/admin-guide/mm/transhuge.rst +6
-6@@ -496,16 +496,16 @@ 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 +shmem_alloc + is incremented every time a shmem huge page is successfully allocated. -file_fallback - is incremented if a file huge page is attempted to be allocated +shmem_fallback + is incremented if a shmem 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 +shmem_fallback_charge + is incremented if a shmem huge page cannot be charged and instead falls back to using small pages even though the allocation was successful.include/linux/huge_mm.h +3
-3@@ -269,9 +269,9 @@ 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_SHMEM_ALLOC , +MTHP_STAT_SHMEM_FALLBACK , +MTHP_STAT_SHMEM_FALLBACK_CHARGE ,MTHP_STAT_SPLIT ,MTHP_STAT_SPLIT_FAILED , MTHP_STAT_SPLIT_DEFERRED,mm/huge_memory.c +6
-6@@ -568,9 +568,9 @@ 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 ); +DEFINE_MTHP_STAT_ATTR (shmem_alloc,MTHP_STAT_SHMEM_ALLOC ); +DEFINE_MTHP_STAT_ATTR (shmem_fallback,MTHP_STAT_SHMEM_FALLBACK ); +DEFINE_MTHP_STAT_ATTR (shmem_fallback_charge,MTHP_STAT_SHMEM_FALLBACK_CHARGE );DEFINE_MTHP_STAT_ATTR (split,MTHP_STAT_SPLIT );DEFINE_MTHP_STAT_ATTR (split_failed,MTHP_STAT_SPLIT_FAILED );DEFINE_MTHP_STAT_ATTR (split_deferred,MTHP_STAT_SPLIT_DEFERRED );@@ -581,9 +581,9 @@ 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, + &shmem_alloc_attr.attr, + &shmem_fallback_attr.attr, + &shmem_fallback_charge_attr.attr, &split_attr.attr, &split_failed_attr.attr, &split_deferred_attr.attr,mm/shmem.c +4
-5@@ -1777,7 +1777,7 @@ 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 ); +count_mthp_stat (order,MTHP_STAT_SHMEM_FALLBACK );#endif order =next_order (&suitable_orders, order); }@@ -1804,8 +1804,8 @@ static struct folio *shmem_alloc_and_add_folio(struct vm_fault *vmf,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 ); +count_mthp_stat (folio_order (folio),MTHP_STAT_SHMEM_FALLBACK ); +count_mthp_stat (folio_order (folio),MTHP_STAT_SHMEM_FALLBACK_CHARGE );#endif } goto unlock;@@ -2181,7 +2181,7 @@ static int shmem_get_folio_gfp(struct inode *inode, pgoff_t index,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 ); +count_mthp_stat (folio_order (folio),MTHP_STAT_SHMEM_ALLOC );#endif goto alloced; } --2.43.0
Thanks
Barry
Quoted 65 lines · Ryan Roberts
Quoted 49 lines · Barry Song
Quoted 48 lines · Ryan Roberts
Quoted 38 lines · Barry Song
Quoted 37 lines · Ryan Roberts
Quoted 15 lines · Barry Song
Quoted 14 lines · Ryan Roberts
The legacy PMD-sized THP counters at
/proc/vmstatincludethp_file_alloc,thp_file_fallbackandthp_file_fallback_charge, which rather confusingly refer to shmem THP and do not include any other types of file pages. This is inconsistent since in most other places in the kernel, THP counters are explicitly separated for anon, shmem and file flavours. However, we are stuck with it since it constitutes a user ABI.Recently, commit
66f44583f9b6("mm: shmem: add mTHP counters for anonymous shmem") added equivalent mTHP stats for shmem, keeping the same "file_" prefix in the names. But in future, we may want to add extra stats to cover actual file pages, at which point, it would all become very confusing.Quoted 6 lines
Quoted 2 lines
So let's take the opportunity to rename these new counters "
shmem_" before the change makes it upstream and the ABI becomes immutable.Personally, I think this approach is much clearer. However, I recall we discussed this
before [1], and it seems that inconsistency is a concern?Embarrassingly, I don't recall that converstation at all :-| but at least what I said then is consistent with what I've done in this patch.
I think David's conclusion from that thread was to call them
FILE_, and add both shmem and pagecache counts to those counters, meaning we can keep the same name as legacy THP counters. But those legacy THP counters only count shmem, and I don't think we would get away with adding pagecache counts to those at this point? (argument: they have been around for long time and there is a risk that user space relies on them and if they were to dramatically increase due to pagecache addition now that could break things). In that case, there is still inconsistency, but its worse; the names are consistent but the semantics are inconsistent.Quoted 4 lines
Quoted 1 line
So my vote is to change to
SHMEM_as per this patch :)I have no objections. However, I dislike the documentation for thp_file_*. Perhaps we can clean it all up together ?
I agree that we should clean this documentation up and I'm happy to roll it into v2. However, I don't think what you have suggested is quite right.
Quoted 8 lines
Quoted 4 lines
thp_file_alloc,thp_file_fallbackandthp_file_fallback_charge*only* count shmem. They don't count pagecache. So perhaps the change should be "...every time a shmem huge page (dispite being named after "file", the counter measures only shmem) is..."?I understand what you are saying, and I know that thp_file_* has only included shmem so far. My question is whether it will include regular files in the future? If not, I am perfectly fine with your approach.
My whole reasoning for this patch is based on the assertion that since
THP_FILE_ALLOChas been there for 8 years and in all that time has only counted shmem, then its highly likely that someone is depending on that semantic and we can't change it. I don't have any actual evidence of code that relies on it though.
Quoted 6 lines
I propose I change the docs to reflect what's actually happening today (i.e. shmem *only*). If we later decide we want to also report page cache numbers through that same counter, then we can change the docs at that point. But if I get my way, we'll soon have mTHP counters for FILE, which is solely for page cache. So You'll be able to get all the fine-grained info out of those and there will be no need to mess with the legacy counters.
Make sense to me. I'd rather we go to
/sys/kernel/mm/transparent_hugepage/hugepages-2048kB/stats/file_* /sys/kernel/mm/transparent_hugepage/hugepages-2048kB/stats/shmem_*
if we later have 2MiB file counters.
Quoted 190 lines
Quoted 190 lines
READ_ONLY_THP_FOR_FSisn't applicable in this path as it is created by khugepaged collapse.Quoted 185 lines
thp_file_mappedincludes both file and shmem, so agree with your change there.What do you think?
Quoted 179 lines
Documentation/admin-guide/mm/transhuge.rst +10
-9@@ -417,21 +417,22 @@ thp_collapse_alloc_failed the allocation. thp_file_alloc - is incremented every time a file huge page is successfully - allocated. + is incremented every time a file (including shmem) huge page is + successfully allocated. thp_file_fallback - is incremented if a file huge page is attempted to be allocated - but fails and instead falls back to using small pages. + is incremented if a file (including shmem) huge page is attempted + to be allocated but fails and instead falls back to using small + pages. thp_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. + is incremented if a file (including shmem) huge page cannot be + charged and instead falls back to using small pages even though + the allocation was successful. thp_file_mapped - is incremented every time a file huge page is mapped into - user address space. + is incremented every time a file (including shmem) huge page is + mapped into user address space. thp_split_page is incremented every time a huge page is split into baseQuoted 142 lines
Quoted 142 lines
Quoted 136 lines
Signed-off-by Ryan Roberts <[email protected]>---
Hi All,
Applies on top of today's mm-unstable (
2073cda629a4) and tested with mm selftests; no regressions observed.The backstory here is that I'd like to introduce some counters for regular file folio allocations to observe how often large folio allocation succeeds, but these shmem counters are named "file" which is going to make things confusing. So hoping to solve that before commit
66f44583f9b6("mm: shmem: add mTHP counters for anonymous shmem") goes upstream (it is currently in mm-stable).Admittedly, this change means the mTHP stat names are not the same as the legacy PMD-size THP names, but I think that's a smaller issue than having "
file_" mTHP stats that only count shmem, then having to introduce "file2_" or "pgcache_" stats for the regular file memory, which is even more inconsistent IMHO. I guess the alternative is to count both shmem and file in these mTHP stats (that's how they were documented anyway) but I think it's better to be able to consider them separately like we do for all the other counters.Thanks,
RyanDocumentation/admin-guide/mm/transhuge.rst | 12 ++++++------ include/linux/huge_mm.h | 6 +++--- mm/huge_memory.c | 12 ++++++------ mm/shmem.c | 8 ++++---- 4 files changed, 19 insertions(+), 19 deletions(-)Documentation/admin-guide/mm/transhuge.rst +6
-6@@ -496,16 +496,16 @@ 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 +shmem_alloc + is incremented every time a shmem huge page is successfully allocated. -file_fallback - is incremented if a file huge page is attempted to be allocated +shmem_fallback + is incremented if a shmem 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 +shmem_fallback_charge + is incremented if a shmem huge page cannot be charged and instead falls back to using small pages even though the allocation was successful.include/linux/huge_mm.h +3
-3@@ -269,9 +269,9 @@ 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_SHMEM_ALLOC , +MTHP_STAT_SHMEM_FALLBACK , +MTHP_STAT_SHMEM_FALLBACK_CHARGE ,MTHP_STAT_SPLIT ,MTHP_STAT_SPLIT_FAILED , MTHP_STAT_SPLIT_DEFERRED,mm/huge_memory.c +6
-6@@ -568,9 +568,9 @@ 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 ); +DEFINE_MTHP_STAT_ATTR (shmem_alloc,MTHP_STAT_SHMEM_ALLOC ); +DEFINE_MTHP_STAT_ATTR (shmem_fallback,MTHP_STAT_SHMEM_FALLBACK ); +DEFINE_MTHP_STAT_ATTR (shmem_fallback_charge,MTHP_STAT_SHMEM_FALLBACK_CHARGE );DEFINE_MTHP_STAT_ATTR (split,MTHP_STAT_SPLIT );DEFINE_MTHP_STAT_ATTR (split_failed,MTHP_STAT_SPLIT_FAILED );DEFINE_MTHP_STAT_ATTR (split_deferred,MTHP_STAT_SPLIT_DEFERRED );@@ -581,9 +581,9 @@ 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, + &shmem_alloc_attr.attr, + &shmem_fallback_attr.attr, + &shmem_fallback_charge_attr.attr, &split_attr.attr, &split_failed_attr.attr, &split_deferred_attr.attr,mm/shmem.c +4
-4@@ -1777,7 +1777,7 @@ 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 ); +count_mthp_stat (order,MTHP_STAT_SHMEM_FALLBACK );#endif order =next_order (&suitable_orders, order); }@@ -1804,8 +1804,8 @@ static struct folio *shmem_alloc_and_add_folio(struct vm_fault *vmf,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 ); +count_mthp_stat (folio_order (folio),MTHP_STAT_SHMEM_FALLBACK ); +count_mthp_stat (folio_order (folio),MTHP_STAT_SHMEM_FALLBACK_CHARGE );#endif } goto unlock;@@ -2181,7 +2181,7 @@ static int shmem_get_folio_gfp(struct inode *inode, pgoff_t index,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 ); +count_mthp_stat (folio_order (folio),MTHP_STAT_SHMEM_ALLOC );
Quoted 8 lines
Quoted 8 lines
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
Quoted 6 lines
mm/shmem.c +0
-1#endif goto alloced; } --2.43.0
Thanks
Barry
Quoted 14 lines · Ryan Roberts
The legacy PMD-sized THP counters at
/proc/vmstatincludethp_file_alloc,thp_file_fallbackandthp_file_fallback_charge, which rather confusingly refer to shmem THP and do not include any other types of file pages. This is inconsistent since in most other places in the kernel, THP counters are explicitly separated for anon, shmem and file flavours. However, we are stuck with it since it constitutes a user ABI.Recently, commit
66f44583f9b6("mm: shmem: add mTHP counters for anonymous shmem") added equivalent mTHP stats for shmem, keeping the same "file_" prefix in the names. But in future, we may want to add extra stats to cover actual file pages, at which point, it would all become very confusing.
Quoted 4 lines
So let's take the opportunity to rename these new counters "
shmem_" before the change makes it upstream and the ABI becomes immutable.Signed-off-by Ryan Roberts <[email protected]>
LGTM. Feel free to add:
Thanks,
Lance
Quoted 140 lines
---
Hi All,
Applies on top of today's mm-unstable (
2073cda629a4) and tested with mm selftests; no regressions observed.The backstory here is that I'd like to introduce some counters for regular file folio allocations to observe how often large folio allocation succeeds, but these shmem counters are named "file" which is going to make things confusing. So hoping to solve that before commit
66f44583f9b6("mm: shmem: add mTHP counters for anonymous shmem") goes upstream (it is currently in mm-stable).Admittedly, this change means the mTHP stat names are not the same as the legacy PMD-size THP names, but I think that's a smaller issue than having "
file_" mTHP stats that only count shmem, then having to introduce "file2_" or "pgcache_" stats for the regular file memory, which is even more inconsistent IMHO. I guess the alternative is to count both shmem and file in these mTHP stats (that's how they were documented anyway) but I think it's better to be able to consider them separately like we do for all the other counters.Thanks,
RyanDocumentation/admin-guide/mm/transhuge.rst | 12 ++++++------ include/linux/huge_mm.h | 6 +++--- mm/huge_memory.c | 12 ++++++------ mm/shmem.c | 8 ++++---- 4 files changed, 19 insertions(+), 19 deletions(-)Documentation/admin-guide/mm/transhuge.rst +6
-6@@ -496,16 +496,16 @@ 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 +shmem_alloc + is incremented every time a shmem huge page is successfully allocated. -file_fallback - is incremented if a file huge page is attempted to be allocated +shmem_fallback + is incremented if a shmem 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 +shmem_fallback_charge + is incremented if a shmem huge page cannot be charged and instead falls back to using small pages even though the allocation was successful.include/linux/huge_mm.h +3
-3@@ -269,9 +269,9 @@ 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_SHMEM_ALLOC , +MTHP_STAT_SHMEM_FALLBACK , +MTHP_STAT_SHMEM_FALLBACK_CHARGE ,MTHP_STAT_SPLIT ,MTHP_STAT_SPLIT_FAILED , MTHP_STAT_SPLIT_DEFERRED,mm/huge_memory.c +6
-6@@ -568,9 +568,9 @@ 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 ); +DEFINE_MTHP_STAT_ATTR (shmem_alloc,MTHP_STAT_SHMEM_ALLOC ); +DEFINE_MTHP_STAT_ATTR (shmem_fallback,MTHP_STAT_SHMEM_FALLBACK ); +DEFINE_MTHP_STAT_ATTR (shmem_fallback_charge,MTHP_STAT_SHMEM_FALLBACK_CHARGE );DEFINE_MTHP_STAT_ATTR (split,MTHP_STAT_SPLIT );DEFINE_MTHP_STAT_ATTR (split_failed,MTHP_STAT_SPLIT_FAILED );DEFINE_MTHP_STAT_ATTR (split_deferred,MTHP_STAT_SPLIT_DEFERRED );@@ -581,9 +581,9 @@ 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, + &shmem_alloc_attr.attr, + &shmem_fallback_attr.attr, + &shmem_fallback_charge_attr.attr, &split_attr.attr, &split_failed_attr.attr, &split_deferred_attr.attr,mm/shmem.c +4
-5@@ -1777,7 +1777,7 @@ 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 ); +count_mthp_stat (order,MTHP_STAT_SHMEM_FALLBACK );#endif order =next_order (&suitable_orders, order); }@@ -1804,8 +1804,8 @@ static struct folio *shmem_alloc_and_add_folio(struct vm_fault *vmf,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 ); +count_mthp_stat (folio_order (folio),MTHP_STAT_SHMEM_FALLBACK ); +count_mthp_stat (folio_order (folio),MTHP_STAT_SHMEM_FALLBACK_CHARGE );#endif } goto unlock;@@ -2181,7 +2181,7 @@ static int shmem_get_folio_gfp(struct inode *inode, pgoff_t index,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 ); +count_mthp_stat (folio_order (folio),MTHP_STAT_SHMEM_ALLOC );#endif goto alloced; } --2.43.0