mm: hugetlbfs: add hwcrp_hugepages to record memory failure on hugetlbfs
Subsystems (2)
PATCH 1/1 mm: hugetlbfs: add hwcrp_hugepages to record memory failure on hugetlbfs
From: Bin Wang <[email protected]>
In the current hugetlbfs memory failure handler, reserved huge page counts are used to record the number of huge pages with hwposion. There are two problems:
1. We call hugetlb_fix_reserve_counts() to change reserved counts in hugetlbfs_error_remove_page(). But this function is only called if hugetlb_unreserve_pages() fails, and hugetlb_unreserve_pages() fails only if kmalloc in region_del() fails, which is almost impossible. As a result, the reserved count is not corrected as expected when a memory failure occurs.
2. Reserved counts is designed to display the number of hugepages reserved at mmap() time. This means that even if we fix the first issue, reserved counts will be confusing because we can't tell if it's hwposion or reserved hugepage.
This patch adds hardware corrput huge pages counts to record memory failure on hugetlbfs instead of reserved counts.
--- fs/hugetlbfs/inode.c | 3 +-- include/linux/hugetlb.h | 3 +++ mm/hugetlb.c | 30 ++++++++++++++++++++++++++++++ 3 files changed, 34 insertions(+), 2 deletions(-)
fs/hugetlbfs/inode.c +1 -2
@@ -985,8 +985,7 @@ static int hugetlbfs_error_remove_page(struct address_space *mapping,pgoff_t index = page->index;remove_huge_page (page); -if (unlikely (hugetlb_unreserve_pages (inode, index, index +1 ,1 ))) -hugetlb_fix_reserve_counts (inode); +hugetlb_fix_hwcrp_counts (page);return 0 ; }
include/linux/hugetlb.h +3 -0
@@ -153,6 +153,7 @@ void putback_active_hugepage(struct page *page);void move_hugetlb_state (struct page *oldpage,struct page *newpage,int reason);void free_huge_page (struct page *page);void hugetlb_fix_reserve_counts (struct inode *inode); +void hugetlb_fix_hwcrp_counts (struct page *page);extern struct mutex *hugetlb_fault_mutex_table;u32 hugetlb_fault_mutex_hash (struct address_space *mapping,pgoff_t idx);
@@ -576,12 +577,14 @@ struct hstate {unsigned long free_huge_pages;unsigned long resv_huge_pages;unsigned long surplus_huge_pages; +unsigned long hwcrp_huge_pages;unsigned long nr_overcommit_huge_pages;struct list_head hugepage_activelist;struct list_head hugepage_freelists[MAX_NUMNODES ];unsigned int nr_huge_pages_node[MAX_NUMNODES ];unsigned int free_huge_pages_node[MAX_NUMNODES ];unsigned int surplus_huge_pages_node[MAX_NUMNODES ]; +unsigned int hwcrp_huge_pages_node[MAX_NUMNODES ];#ifdef CONFIG_CGROUP_HUGETLB /* cgroup control files */ struct cftype cgroup_files_dfl[7 ];
mm/hugetlb.c +30 -0
@@ -761,6 +761,15 @@ void hugetlb_fix_reserve_counts(struct inode *inode)pr_warn ("hugetlb: Huge Page Reserved count may go negative.\n" ); } +void hugetlb_fix_hwcrp_counts (struct page *page) +{ +struct hstate *h = &default_hstate; +int nid =page_to_nid (page); + + h->hwcrp_huge_pages++; + h->hwcrp_huge_pages_node[nid]++; +} +/* * Count and return the number of huge pages in the reserve map * that intersect with the range [f, t).
@@ -3089,12 +3098,30 @@ static ssize_t surplus_hugepages_show(struct kobject *kobj, }HSTATE_ATTR_RO (surplus_hugepages); +static ssize_t hwcrp_hugepages_show (struct kobject *kobj, +struct kobj_attribute *attr,char *buf) +{ +struct hstate *h; +unsigned long hwcrp_huge_pages; +int nid; + + h =kobj_to_hstate (kobj, &nid); +if (nid ==NUMA_NO_NODE ) + hwcrp_huge_pages = h->hwcrp_huge_pages; +else + hwcrp_huge_pages = h->hwcrp_huge_pages_node[nid]; + +return sprintf (buf,"%lu\n" , hwcrp_huge_pages); +} +HSTATE_ATTR_RO (hwcrp_hugepages); +static struct attribute *hstate_attrs[] = { &nr_hugepages_attr.attr, &nr_overcommit_hugepages_attr.attr, &free_hugepages_attr.attr, &resv_hugepages_attr.attr, &surplus_hugepages_attr.attr, + &hwcrp_hugepages_attr.attr, #ifdef CONFIG_NUMA &nr_hugepages_mempolicy_attr.attr, #endif
@@ -3164,6 +3191,7 @@ static struct attribute *per_node_hstate_attrs[] = { &nr_hugepages_attr.attr, &free_hugepages_attr.attr, &surplus_hugepages_attr.attr, + &hwcrp_hugepages_attr.attr, NULL, };
@@ -3657,11 +3685,13 @@ void hugetlb_report_meminfo(struct seq_file *m)"HugePages_Free: %5lu\n" "HugePages_Rsvd: %5lu\n" "HugePages_Surp: %5lu\n" +"HugePages_Hwcrp: %5lu\n" "Hugepagesize: %8lu kB\n" , count, h->free_huge_pages, h->resv_huge_pages, h->surplus_huge_pages, + h->hwcrp_huge_pages, huge_page_size(h) / SZ_1K); }
Signature · 1 line
2.23.0
Quoted 4 lines · wangbin
From: Bin Wang <[email protected]>
In the current hugetlbfs memory failure handler, reserved huge page counts are used to record the number of huge pages with hwposion.
I do not believe this is an accurate statement. Naoya is the memory error expert and may disagree, but I do not see anywhere where reserve counts are being used to track huge pages with memory errors.
IIUC, the routine hugetlbfs_error_remove_page is called after unmapping the page from all user mappings. The routine will simply, remove the page from the cache. This effectively removes the page from the file as hugetlbfs is a memory only filesystem. The subsequent call to hugetlb_unreserve_pages cleans up any reserve map entries associated with the page and adjusts the reserve count if necessary. The reserve count adjustment is based on removing the page from the file, rather than the memory error. The same adjustment would be made if the page was hole punched from the file.
What specific problem are you trying to solve? Are trying to see how many huge pages were hit by memory errors?
Signature · 1 line
Mike Kravetz
Resend with new e-mail for Naoya
Quoted 4 lines · wangbin
From: Bin Wang <[email protected]>
In the current hugetlbfs memory failure handler, reserved huge page counts are used to record the number of huge pages with hwposion.
I do not believe this is an accurate statement. Naoya is the memory error expert and may disagree, but I do not see anywhere where reserve counts are being used to track huge pages with memory errors.
IIUC, the routine hugetlbfs_error_remove_page is called after unmapping the page from all user mappings. The routine will simply, remove the page from the cache. This effectively removes the page from the file as hugetlbfs is a memory only filesystem. The subsequent call to hugetlb_unreserve_pages cleans up any reserve map entries associated with the page and adjusts the reserve count if necessary. The reserve count adjustment is based on removing the page from the file, rather than the memory error. The same adjustment would be made if the page was hole punched from the file.
What specific problem are you trying to solve? Are trying to see how many huge pages were hit by memory errors?
Signature · 1 line
Mike Kravetz
Quoted 2 lines
What specific problem are you trying to solve? Are trying to see how many huge pages were hit by memory errors?
Yes, I'd like to know how many huge pages are not available because of the memory errors. Just like HardwareCorrupted in the /proc/meminfo. But the HardwareCorrupted only adds one page size when a huge page is hit by memory errors, and mixes with normal pages. So I think we should add a new counts to track the memory errors on hugetlbfs.
--
Bin Wang
Quoted 8 lines · wangbin
Quoted 2 lines
What specific problem are you trying to solve? Are trying to see how many huge pages were hit by memory errors?
Yes, I'd like to know how many huge pages are not available because of the memory errors. Just like HardwareCorrupted in the
/proc/meminfo. But the HardwareCorrupted only adds one page size when a huge page is hit by memory errors, and mixes with normal pages. So I think we should add a new counts to track the memory errors on hugetlbfs.
If you can use root privilege in your use-case, an easy way to get the number of corrupted hugepages is to use page-types.c (which reads /proc/kpageflags) like below:
$ page-types -b huge,hwpoison=huge,hwpoison
flags page-count MB symbolic-flags long-symbolic-flags
0x00000000000a8000 1 0 _______________H_G_X_______________________ compound_head,huge,hwpoison
total 1 0But I guess that many usecases do not permit access to this interface, where some new accounting interface for corrupted hugepages could be helpful as you suggest.
Thanks,
Naoya Horiguchi
Quoted 4 lines
If you can use root privilege in your use-case, an easy way to get the number of corrupted hugepages is to use
page-types.c(which reads/proc/kpageflags) like below:
Quoted 8 lines
$ page-types -b huge,hwpoison=huge,hwpoison flags page-count MB symbolic-flags long-symbolic-flags 0x00000000000a8000 1 0 _______________H_G_X_______________________ compound_head,huge,hwpoison total 1 0But I guess that many usecases do not permit access to this interface, where some new accounting interface for corrupted hugepages could be helpful as you suggest.
Thanks for your suggestion very much. This approach is helpful to me. But as you say, root privilege is not permitted in most cases. And I also want to know the number of corrupted hugepages per node.
--
Bin Wang
Thanks for forwarding the message, Mike.
Quoted 11 lines · Mike Kravetz
Resend with new e-mail for Naoya
Quoted 4 lines · wangbin
From: Bin Wang <[email protected]>
In the current hugetlbfs memory failure handler, reserved huge page counts are used to record the number of huge pages with hwposion.
I do not believe this is an accurate statement. Naoya is the memory error expert and may disagree, but I do not see anywhere where reserve counts are being used to track huge pages with memory errors.
And Mike is right, hugetlb's reservation count is not linked to accounting of hwpoisoned pages.
Quoted 10 lines
IIUC, the routine
hugetlbfs_error_remove_pageis called after unmapping the page from all user mappings. The routine will simply, remove the page from the cache. This effectively removes the page from the file as hugetlbfs is a memory only filesystem. The subsequent call tohugetlb_unreserve_pagescleans up any reserve map entries associated with the page and adjusts the reserve count if necessary. The reserve count adjustment is based on removing the page from the file, rather than the memory error. The same adjustment would be made if the page was hole punched from the file.
This logic totally makes sense to me.
Unmapping done in memory_failure() might increment the reserve count, but that's the cancel of the consumed reservation by unmapping.
Thanks,
Naoya Horigcuhi