| Message ID | 20240508125808.28882-1-hailong.liu@oppo.com (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | [RFC] mm/vmalloc: fix vmalloc which may return null if called with __GFP_NOFAIL | expand |
+Cc Michal, On 2024/5/8 20:58, hailong.liu@oppo.com wrote: > From: "Hailong.Liu" <hailong.liu@oppo.com> > > Commit a421ef303008 ("mm: allow !GFP_KERNEL allocations for kvmalloc") > includes support for __GFP_NOFAIL, but it presents a conflict with > commit dd544141b9eb ("vmalloc: back off when the current task is > OOM-killed"). A possible scenario is as belows: > > process-a > kvcalloc(n, m, GFP_KERNEL | __GFP_NOFAIL) > __vmalloc_node_range() > __vmalloc_area_node() > vm_area_alloc_pages() > --> oom-killer send SIGKILL to process-a > if (fatal_signal_pending(current)) break; > --> return NULL; > > to fix this, do not check fatal_signal_pending() in vm_area_alloc_pages() > if __GFP_NOFAIL set. > > Reported-by: Oven <liyangouwen1@oppo.com> > Signed-off-by: Hailong.Liu <hailong.liu@oppo.com> Why taging this as RFC here? It seems a corner-case fix of commit a421ef303008 Thanks, Gao Xiang > --- > mm/vmalloc.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/mm/vmalloc.c b/mm/vmalloc.c > index 6641be0ca80b..2f359d08bf8d 100644 > --- a/mm/vmalloc.c > +++ b/mm/vmalloc.c > @@ -3560,7 +3560,7 @@ vm_area_alloc_pages(gfp_t gfp, int nid, > > /* High-order pages or fallback path if "bulk" fails. */ > while (nr_allocated < nr_pages) { > - if (fatal_signal_pending(current)) > + if (!(gfp & __GFP_NOFAIL) && fatal_signal_pending(current)) > break; > > if (nid == NUMA_NO_NODE) > --- > This issue occurred during OPLUS KASAN test. Below is part of the log > > -> send signal > [65731.222840] [ T1308] oom-kill:constraint=CONSTRAINT_NONE,nodemask=(null),cpuset=/,mems_allowed=0,global_oom,task_memcg=/apps/uid_10198,task=gs.intelligence,pid=32454,uid=10198 > > [65731.259685] [T32454] Call trace: > [65731.259698] [T32454] dump_backtrace+0xf4/0x118 > [65731.259734] [T32454] show_stack+0x18/0x24 > [65731.259756] [T32454] dump_stack_lvl+0x60/0x7c > [65731.259781] [T32454] dump_stack+0x18/0x38 > [65731.259800] [T32454] mrdump_common_die+0x250/0x39c [mrdump] > [65731.259936] [T32454] ipanic_die+0x20/0x34 [mrdump] > [65731.260019] [T32454] atomic_notifier_call_chain+0xb4/0xfc > [65731.260047] [T32454] notify_die+0x114/0x198 > [65731.260073] [T32454] die+0xf4/0x5b4 > [65731.260098] [T32454] die_kernel_fault+0x80/0x98 > [65731.260124] [T32454] __do_kernel_fault+0x160/0x2a8 > [65731.260146] [T32454] do_bad_area+0x68/0x148 > [65731.260174] [T32454] do_mem_abort+0x151c/0x1b34 > [65731.260204] [T32454] el1_abort+0x3c/0x5c > [65731.260227] [T32454] el1h_64_sync_handler+0x54/0x90 > [65731.260248] [T32454] el1h_64_sync+0x68/0x6c > [65731.260269] [T32454] z_erofs_decompress_queue+0x7f0/0x2258 > --> be->decompressed_pages = kvcalloc(be->nr_pages, sizeof(struct page *), GFP_KERNEL | __GFP_NOFAIL); > kernel panic by NULL pointer dereference. > erofs assume kvmalloc with __GFP_NOFAIL never return NULL. > > [65731.260293] [T32454] z_erofs_runqueue+0xf30/0x104c > [65731.260314] [T32454] z_erofs_readahead+0x4f0/0x968 > [65731.260339] [T32454] read_pages+0x170/0xadc > [65731.260364] [T32454] page_cache_ra_unbounded+0x874/0xf30 > [65731.260388] [T32454] page_cache_ra_order+0x24c/0x714 > [65731.260411] [T32454] filemap_fault+0xbf0/0x1a74 > [65731.260437] [T32454] __do_fault+0xd0/0x33c > [65731.260462] [T32454] handle_mm_fault+0xf74/0x3fe0 > [65731.260486] [T32454] do_mem_abort+0x54c/0x1b34 > [65731.260509] [T32454] el0_da+0x44/0x94 > [65731.260531] [T32454] el0t_64_sync_handler+0x98/0xb4 > [65731.260553] [T32454] el0t_64_sync+0x198/0x19c > > -- > 2.34.1
On 2024/5/8 21:41, Gao Xiang wrote: > > +Cc Michal, > > On 2024/5/8 20:58, hailong.liu@oppo.com wrote: >> From: "Hailong.Liu" <hailong.liu@oppo.com> >> >> Commit a421ef303008 ("mm: allow !GFP_KERNEL allocations for kvmalloc") >> includes support for __GFP_NOFAIL, but it presents a conflict with >> commit dd544141b9eb ("vmalloc: back off when the current task is >> OOM-killed"). A possible scenario is as belows: >> >> process-a >> kvcalloc(n, m, GFP_KERNEL | __GFP_NOFAIL) >> __vmalloc_node_range() >> __vmalloc_area_node() >> vm_area_alloc_pages() >> --> oom-killer send SIGKILL to process-a >> if (fatal_signal_pending(current)) break; >> --> return NULL; >> >> to fix this, do not check fatal_signal_pending() in vm_area_alloc_pages() >> if __GFP_NOFAIL set. >> >> Reported-by: Oven <liyangouwen1@oppo.com> >> Signed-off-by: Hailong.Liu <hailong.liu@oppo.com> > > Why taging this as RFC here? It seems a corner-case fix of > commit a421ef303008 BTW, I guess maybe commit e9c3cda4d86e ("mm, vmalloc: fix high order __GFP_NOFAIL allocations") misses fatal_signal_pending() handling anyway.. > > Thanks, > Gao Xiang >
On Wed, 08. May 21:41, Gao Xiang wrote: > > +Cc Michal, > > On 2024/5/8 20:58, hailong.liu@oppo.com wrote: > > From: "Hailong.Liu" <hailong.liu@oppo.com> > > > > Commit a421ef303008 ("mm: allow !GFP_KERNEL allocations for kvmalloc") > > includes support for __GFP_NOFAIL, but it presents a conflict with > > commit dd544141b9eb ("vmalloc: back off when the current task is > > OOM-killed"). A possible scenario is as belows: > > > > process-a > > kvcalloc(n, m, GFP_KERNEL | __GFP_NOFAIL) > > __vmalloc_node_range() > > __vmalloc_area_node() > > vm_area_alloc_pages() > > --> oom-killer send SIGKILL to process-a > > if (fatal_signal_pending(current)) break; > > --> return NULL; > > > > to fix this, do not check fatal_signal_pending() in vm_area_alloc_pages() > > if __GFP_NOFAIL set. > > > > Reported-by: Oven <liyangouwen1@oppo.com> > > Signed-off-by: Hailong.Liu <hailong.liu@oppo.com> > > Why taging this as RFC here? It seems a corner-case fix of > commit a421ef303008 > > Thanks, > Gao Xiang > Hi Gao Xiang: RFC here to wait for a better way to handle this case :). IMO, if vmalloc support __GFP_NOFAIL it should not return null even system is deadlock on memory. -- Best Regards, Hailong.
Hi, On 2024/5/8 22:43, Hailong Liu wrote: > On Wed, 08. May 21:41, Gao Xiang wrote: >> >> +Cc Michal, >> >> On 2024/5/8 20:58, hailong.liu@oppo.com wrote: >>> From: "Hailong.Liu" <hailong.liu@oppo.com> >>> >>> Commit a421ef303008 ("mm: allow !GFP_KERNEL allocations for kvmalloc") >>> includes support for __GFP_NOFAIL, but it presents a conflict with >>> commit dd544141b9eb ("vmalloc: back off when the current task is >>> OOM-killed"). A possible scenario is as belows: >>> >>> process-a >>> kvcalloc(n, m, GFP_KERNEL | __GFP_NOFAIL) >>> __vmalloc_node_range() >>> __vmalloc_area_node() >>> vm_area_alloc_pages() >>> --> oom-killer send SIGKILL to process-a >>> if (fatal_signal_pending(current)) break; >>> --> return NULL; >>> >>> to fix this, do not check fatal_signal_pending() in vm_area_alloc_pages() >>> if __GFP_NOFAIL set. >>> >>> Reported-by: Oven <liyangouwen1@oppo.com> >>> Signed-off-by: Hailong.Liu <hailong.liu@oppo.com> >> >> Why taging this as RFC here? It seems a corner-case fix of >> commit a421ef303008 >> >> Thanks, >> Gao Xiang >> > > Hi Gao Xiang: > > RFC here to wait for a better way to handle this case :). > IMO, if vmalloc support __GFP_NOFAIL it should not return > null even system is deadlock on memory. The starting point is that kmalloc doesn't support __GFP_NOFAIL if order > 1 (even for very short temporary uses), see: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/mm/page_alloc.c?h=v6.8#n2896 but it is possible if we have such page pointer array (since two (order-1) pages can only keep 1024 8-byte entries, it can happen if compression ratios are high), and kvmalloc(__GFP_NOFAIL) has already been supported for almost two years, it will fallback to order-0 allocation as described in commit e9c3cda4d86e ("mm, vmalloc: fix high order __GFP_NOFAIL allocations"). With my limited understanding, I'm not sure why it can cause deadlock here since it will fallback to order-0 allocation then, and such allocation is just for short temporary uses again because kmalloc doesn't support order > 1 short memory allocation strictly. Thanks, Gao Xiang > > -- > > Best Regards, > Hailong.
On Wed, 08. May 23:10, Gao Xiang wrote: > Hi, > > On 2024/5/8 22:43, Hailong Liu wrote: > > On Wed, 08. May 21:41, Gao Xiang wrote: > > > > > > +Cc Michal, > > > > > > On 2024/5/8 20:58, hailong.liu@oppo.com wrote: > > > > From: "Hailong.Liu" <hailong.liu@oppo.com> > > > > > > > > Commit a421ef303008 ("mm: allow !GFP_KERNEL allocations for kvmalloc") > > > > includes support for __GFP_NOFAIL, but it presents a conflict with > > > > commit dd544141b9eb ("vmalloc: back off when the current task is > > > > OOM-killed"). A possible scenario is as belows: > > > > > > > > process-a > > > > kvcalloc(n, m, GFP_KERNEL | __GFP_NOFAIL) > > > > __vmalloc_node_range() > > > > __vmalloc_area_node() > > > > vm_area_alloc_pages() > > > > --> oom-killer send SIGKILL to process-a > > > > if (fatal_signal_pending(current)) break; > > > > --> return NULL; > > > > > > > > to fix this, do not check fatal_signal_pending() in vm_area_alloc_pages() > > > > if __GFP_NOFAIL set. > > > > > > > > Reported-by: Oven <liyangouwen1@oppo.com> > > > > Signed-off-by: Hailong.Liu <hailong.liu@oppo.com> > > > > > > Why taging this as RFC here? It seems a corner-case fix of > > > commit a421ef303008 > > > > > > Thanks, > > > Gao Xiang > > > > > > > Hi Gao Xiang: > > > > RFC here to wait for a better way to handle this case :). > > IMO, if vmalloc support __GFP_NOFAIL it should not return > > null even system is deadlock on memory. > > The starting point is that kmalloc doesn't support __GFP_NOFAIL > if order > 1 (even for very short temporary uses), see: > https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/mm/page_alloc.c?h=v6.8#n2896 > > but it is possible if we have such page pointer array (since two > (order-1) pages can only keep 1024 8-byte entries, it can happen > if compression ratios are high), and kvmalloc(__GFP_NOFAIL) has > already been supported for almost two years, it will fallback to > order-0 allocation as described in commit e9c3cda4d86e > ("mm, vmalloc: fix high order __GFP_NOFAIL allocations"). > > With my limited understanding, I'm not sure why it can cause > deadlock here since it will fallback to order-0 allocation then, > and such allocation is just for short temporary uses again > because kmalloc doesn't support order > 1 short memory > allocation strictly. > deadlock on memory meands there is a memory leak causing system to be unable to allocate memory not actual *deadlock*. > Thanks, > Gao Xiang > -- Best Regards, Hailong.
On 2024/5/8 23:31, Hailong Liu wrote: > On Wed, 08. May 23:10, Gao Xiang wrote: >> Hi, >> >> On 2024/5/8 22:43, Hailong Liu wrote: >>> On Wed, 08. May 21:41, Gao Xiang wrote: >>>> >>>> +Cc Michal, >>>> >>>> On 2024/5/8 20:58, hailong.liu@oppo.com wrote: >>>>> From: "Hailong.Liu" <hailong.liu@oppo.com> >>>>> >>>>> Commit a421ef303008 ("mm: allow !GFP_KERNEL allocations for kvmalloc") >>>>> includes support for __GFP_NOFAIL, but it presents a conflict with >>>>> commit dd544141b9eb ("vmalloc: back off when the current task is >>>>> OOM-killed"). A possible scenario is as belows: >>>>> >>>>> process-a >>>>> kvcalloc(n, m, GFP_KERNEL | __GFP_NOFAIL) >>>>> __vmalloc_node_range() >>>>> __vmalloc_area_node() >>>>> vm_area_alloc_pages() >>>>> --> oom-killer send SIGKILL to process-a >>>>> if (fatal_signal_pending(current)) break; >>>>> --> return NULL; >>>>> >>>>> to fix this, do not check fatal_signal_pending() in vm_area_alloc_pages() >>>>> if __GFP_NOFAIL set. >>>>> >>>>> Reported-by: Oven <liyangouwen1@oppo.com> >>>>> Signed-off-by: Hailong.Liu <hailong.liu@oppo.com> >>>> >>>> Why taging this as RFC here? It seems a corner-case fix of >>>> commit a421ef303008 >>>> >>>> Thanks, >>>> Gao Xiang >>>> >>> >>> Hi Gao Xiang: >>> >>> RFC here to wait for a better way to handle this case :). >>> IMO, if vmalloc support __GFP_NOFAIL it should not return >>> null even system is deadlock on memory. >> >> The starting point is that kmalloc doesn't support __GFP_NOFAIL >> if order > 1 (even for very short temporary uses), see: >> https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/mm/page_alloc.c?h=v6.8#n2896 >> >> but it is possible if we have such page pointer array (since two >> (order-1) pages can only keep 1024 8-byte entries, it can happen >> if compression ratios are high), and kvmalloc(__GFP_NOFAIL) has >> already been supported for almost two years, it will fallback to >> order-0 allocation as described in commit e9c3cda4d86e >> ("mm, vmalloc: fix high order __GFP_NOFAIL allocations"). >> >> With my limited understanding, I'm not sure why it can cause >> deadlock here since it will fallback to order-0 allocation then, >> and such allocation is just for short temporary uses again >> because kmalloc doesn't support order > 1 short memory >> allocation strictly. >> > > deadlock on memory meands there is a memory leak causing > system to be unable to allocate memory not actual > *deadlock*. Where is memory leak? If it's caused by kvmalloc(__GFP_NOFAIL) callers, then it's bugs of callers and we should fix the callers. Also why kmalloc(__GFP_NOFAIL) (for example, also order-0 allocation) differs? Thanks, Gao Xiang > >> Thanks, >> Gao Xiang >> > > -- > > Best Regards, > Hailong.
On Wed, 08. May 23:40, Gao Xiang wrote: > > > On 2024/5/8 23:31, Hailong Liu wrote: > > On Wed, 08. May 23:10, Gao Xiang wrote: > > > Hi, > > > > > > On 2024/5/8 22:43, Hailong Liu wrote: > > > > On Wed, 08. May 21:41, Gao Xiang wrote: > > > > > > > > > > +Cc Michal, > > > > > > > > > > On 2024/5/8 20:58, hailong.liu@oppo.com wrote: > > > > > > From: "Hailong.Liu" <hailong.liu@oppo.com> > > > > > > > > > > > > Commit a421ef303008 ("mm: allow !GFP_KERNEL allocations for kvmalloc") > > > > > > includes support for __GFP_NOFAIL, but it presents a conflict with > > > > > > commit dd544141b9eb ("vmalloc: back off when the current task is > > > > > > OOM-killed"). A possible scenario is as belows: > > > > > > > > > > > > process-a > > > > > > kvcalloc(n, m, GFP_KERNEL | __GFP_NOFAIL) > > > > > > __vmalloc_node_range() > > > > > > __vmalloc_area_node() > > > > > > vm_area_alloc_pages() > > > > > > --> oom-killer send SIGKILL to process-a > > > > > > if (fatal_signal_pending(current)) break; > > > > > > --> return NULL; > > > > > > > > > > > > to fix this, do not check fatal_signal_pending() in vm_area_alloc_pages() > > > > > > if __GFP_NOFAIL set. > > > > > > > > > > > > Reported-by: Oven <liyangouwen1@oppo.com> > > > > > > Signed-off-by: Hailong.Liu <hailong.liu@oppo.com> > > > > > > > > > > Why taging this as RFC here? It seems a corner-case fix of > > > > > commit a421ef303008 > > > > > > > > > > Thanks, > > > > > Gao Xiang > > > > > > > > > > > > > Hi Gao Xiang: > > > > > > > > RFC here to wait for a better way to handle this case :). > > > > IMO, if vmalloc support __GFP_NOFAIL it should not return > > > > null even system is deadlock on memory. > > > > > > The starting point is that kmalloc doesn't support __GFP_NOFAIL > > > if order > 1 (even for very short temporary uses), see: > > > https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/mm/page_alloc.c?h=v6.8#n2896 > > > > > > but it is possible if we have such page pointer array (since two > > > (order-1) pages can only keep 1024 8-byte entries, it can happen > > > if compression ratios are high), and kvmalloc(__GFP_NOFAIL) has > > > already been supported for almost two years, it will fallback to > > > order-0 allocation as described in commit e9c3cda4d86e > > > ("mm, vmalloc: fix high order __GFP_NOFAIL allocations"). > > > > > > With my limited understanding, I'm not sure why it can cause > > > deadlock here since it will fallback to order-0 allocation then, > > > and such allocation is just for short temporary uses again > > > because kmalloc doesn't support order > 1 short memory > > > allocation strictly. > > > > > > > deadlock on memory meands there is a memory leak causing > > system to be unable to allocate memory not actual > > *deadlock*. > > Where is memory leak? If it's caused by kvmalloc(__GFP_NOFAIL) > callers, then it's bugs of callers and we should fix the callers. > > Also why kmalloc(__GFP_NOFAIL) (for example, also order-0 > allocation) differs? > > Thanks, > Gao Xiang > I’m not suggesting that erofs would cause a memleak. What I mean is that if kvmalloc is invoked with __GFP_NOFAIL, it must ensure a non-NULL return, even in scenarios where memory leaks caused by other processes result in the inability to allocate a page. In such a situation, it should result in “Kernel panic - not syncing: System is deadlocked on memory”. > > > > > > Thanks, > > > Gao Xiang > > > > > > > -- > > > > Best Regards, > > Hailong. -- Best Regards, Hailong.
On Thu, May 9, 2024 at 12:58 AM <hailong.liu@oppo.com> wrote: > > From: "Hailong.Liu" <hailong.liu@oppo.com> > > Commit a421ef303008 ("mm: allow !GFP_KERNEL allocations for kvmalloc") > includes support for __GFP_NOFAIL, but it presents a conflict with > commit dd544141b9eb ("vmalloc: back off when the current task is > OOM-killed"). A possible scenario is as belows: > > process-a > kvcalloc(n, m, GFP_KERNEL | __GFP_NOFAIL) > __vmalloc_node_range() > __vmalloc_area_node() > vm_area_alloc_pages() > --> oom-killer send SIGKILL to process-a > if (fatal_signal_pending(current)) break; > --> return NULL; > > to fix this, do not check fatal_signal_pending() in vm_area_alloc_pages() > if __GFP_NOFAIL set. > > Reported-by: Oven <liyangouwen1@oppo.com> > Signed-off-by: Hailong.Liu <hailong.liu@oppo.com> > --- > mm/vmalloc.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/mm/vmalloc.c b/mm/vmalloc.c > index 6641be0ca80b..2f359d08bf8d 100644 > --- a/mm/vmalloc.c > +++ b/mm/vmalloc.c > @@ -3560,7 +3560,7 @@ vm_area_alloc_pages(gfp_t gfp, int nid, > > /* High-order pages or fallback path if "bulk" fails. */ > while (nr_allocated < nr_pages) { > - if (fatal_signal_pending(current)) > + if (!(gfp & __GFP_NOFAIL) && fatal_signal_pending(current)) > break; why not !nofail ? This seems a correct fix, but it undermines the assumption made in commit dd544141b9eb ("vmalloc: back off when the current task is OOM-killed") " This may trigger some hidden problems, when caller does not handle vmalloc failures, or when rollaback after failed vmalloc calls own vmallocs inside. However all of these scenarios are incorrect: vmalloc does not guarantee successful allocation, it has never been called with __GFP_NOFAIL and threfore either should not be used for any rollbacks or should handle such errors correctly and not lead to critical failures. " If a significant kvmalloc operation is performed with the NOFAIL flag, it risks reverting the fix intended to address the OOM-killer issue in commit dd544141b9eb. Should we indeed permit the NOFAIL flag for large kvmalloc allocations? > > if (nid == NUMA_NO_NODE) > --- > This issue occurred during OPLUS KASAN test. Below is part of the log > > -> send signal > [65731.222840] [ T1308] oom-kill:constraint=CONSTRAINT_NONE,nodemask=(null),cpuset=/,mems_allowed=0,global_oom,task_memcg=/apps/uid_10198,task=gs.intelligence,pid=32454,uid=10198 > > [65731.259685] [T32454] Call trace: > [65731.259698] [T32454] dump_backtrace+0xf4/0x118 > [65731.259734] [T32454] show_stack+0x18/0x24 > [65731.259756] [T32454] dump_stack_lvl+0x60/0x7c > [65731.259781] [T32454] dump_stack+0x18/0x38 > [65731.259800] [T32454] mrdump_common_die+0x250/0x39c [mrdump] > [65731.259936] [T32454] ipanic_die+0x20/0x34 [mrdump] > [65731.260019] [T32454] atomic_notifier_call_chain+0xb4/0xfc > [65731.260047] [T32454] notify_die+0x114/0x198 > [65731.260073] [T32454] die+0xf4/0x5b4 > [65731.260098] [T32454] die_kernel_fault+0x80/0x98 > [65731.260124] [T32454] __do_kernel_fault+0x160/0x2a8 > [65731.260146] [T32454] do_bad_area+0x68/0x148 > [65731.260174] [T32454] do_mem_abort+0x151c/0x1b34 > [65731.260204] [T32454] el1_abort+0x3c/0x5c > [65731.260227] [T32454] el1h_64_sync_handler+0x54/0x90 > [65731.260248] [T32454] el1h_64_sync+0x68/0x6c > [65731.260269] [T32454] z_erofs_decompress_queue+0x7f0/0x2258 > --> be->decompressed_pages = kvcalloc(be->nr_pages, sizeof(struct page *), GFP_KERNEL | __GFP_NOFAIL); > kernel panic by NULL pointer dereference. > erofs assume kvmalloc with __GFP_NOFAIL never return NULL. > > [65731.260293] [T32454] z_erofs_runqueue+0xf30/0x104c > [65731.260314] [T32454] z_erofs_readahead+0x4f0/0x968 > [65731.260339] [T32454] read_pages+0x170/0xadc > [65731.260364] [T32454] page_cache_ra_unbounded+0x874/0xf30 > [65731.260388] [T32454] page_cache_ra_order+0x24c/0x714 > [65731.260411] [T32454] filemap_fault+0xbf0/0x1a74 > [65731.260437] [T32454] __do_fault+0xd0/0x33c > [65731.260462] [T32454] handle_mm_fault+0xf74/0x3fe0 > [65731.260486] [T32454] do_mem_abort+0x54c/0x1b34 > [65731.260509] [T32454] el0_da+0x44/0x94 > [65731.260531] [T32454] el0t_64_sync_handler+0x98/0xb4 > [65731.260553] [T32454] el0t_64_sync+0x198/0x19c > > -- > 2.34.1 Thanks Barry
On Thu, May 9, 2024 at 2:20 PM Barry Song <21cnbao@gmail.com> wrote: > > On Thu, May 9, 2024 at 12:58 AM <hailong.liu@oppo.com> wrote: > > > > From: "Hailong.Liu" <hailong.liu@oppo.com> > > > > Commit a421ef303008 ("mm: allow !GFP_KERNEL allocations for kvmalloc") > > includes support for __GFP_NOFAIL, but it presents a conflict with > > commit dd544141b9eb ("vmalloc: back off when the current task is > > OOM-killed"). A possible scenario is as belows: > > > > process-a > > kvcalloc(n, m, GFP_KERNEL | __GFP_NOFAIL) > > __vmalloc_node_range() > > __vmalloc_area_node() > > vm_area_alloc_pages() > > --> oom-killer send SIGKILL to process-a > > if (fatal_signal_pending(current)) break; > > --> return NULL; > > > > to fix this, do not check fatal_signal_pending() in vm_area_alloc_pages() > > if __GFP_NOFAIL set. > > > > Reported-by: Oven <liyangouwen1@oppo.com> > > Signed-off-by: Hailong.Liu <hailong.liu@oppo.com> > > --- > > mm/vmalloc.c | 2 +- > > 1 file changed, 1 insertion(+), 1 deletion(-) > > > > diff --git a/mm/vmalloc.c b/mm/vmalloc.c > > index 6641be0ca80b..2f359d08bf8d 100644 > > --- a/mm/vmalloc.c > > +++ b/mm/vmalloc.c > > @@ -3560,7 +3560,7 @@ vm_area_alloc_pages(gfp_t gfp, int nid, > > > > /* High-order pages or fallback path if "bulk" fails. */ > > while (nr_allocated < nr_pages) { > > - if (fatal_signal_pending(current)) > > + if (!(gfp & __GFP_NOFAIL) && fatal_signal_pending(current)) > > break; > > why not !nofail ? > > This seems a correct fix, but it undermines the assumption made in > commit dd544141b9eb > ("vmalloc: back off when the current task is OOM-killed") > > " > This may trigger some hidden problems, when caller does not handle > vmalloc failures, or when rollaback after failed vmalloc calls own > vmallocs inside. However all of these scenarios are incorrect: vmalloc > does not guarantee successful allocation, it has never been called with > __GFP_NOFAIL and threfore either should not be used for any rollbacks or > should handle such errors correctly and not lead to critical failures. > " > > If a significant kvmalloc operation is performed with the NOFAIL flag, it risks > reverting the fix intended to address the OOM-killer issue in commit > dd544141b9eb. > Should we indeed permit the NOFAIL flag for large kvmalloc allocations? + Vasily, Michal. > > > > > if (nid == NUMA_NO_NODE) > > --- > > This issue occurred during OPLUS KASAN test. Below is part of the log > > > > -> send signal > > [65731.222840] [ T1308] oom-kill:constraint=CONSTRAINT_NONE,nodemask=(null),cpuset=/,mems_allowed=0,global_oom,task_memcg=/apps/uid_10198,task=gs.intelligence,pid=32454,uid=10198 > > > > [65731.259685] [T32454] Call trace: > > [65731.259698] [T32454] dump_backtrace+0xf4/0x118 > > [65731.259734] [T32454] show_stack+0x18/0x24 > > [65731.259756] [T32454] dump_stack_lvl+0x60/0x7c > > [65731.259781] [T32454] dump_stack+0x18/0x38 > > [65731.259800] [T32454] mrdump_common_die+0x250/0x39c [mrdump] > > [65731.259936] [T32454] ipanic_die+0x20/0x34 [mrdump] > > [65731.260019] [T32454] atomic_notifier_call_chain+0xb4/0xfc > > [65731.260047] [T32454] notify_die+0x114/0x198 > > [65731.260073] [T32454] die+0xf4/0x5b4 > > [65731.260098] [T32454] die_kernel_fault+0x80/0x98 > > [65731.260124] [T32454] __do_kernel_fault+0x160/0x2a8 > > [65731.260146] [T32454] do_bad_area+0x68/0x148 > > [65731.260174] [T32454] do_mem_abort+0x151c/0x1b34 > > [65731.260204] [T32454] el1_abort+0x3c/0x5c > > [65731.260227] [T32454] el1h_64_sync_handler+0x54/0x90 > > [65731.260248] [T32454] el1h_64_sync+0x68/0x6c > > [65731.260269] [T32454] z_erofs_decompress_queue+0x7f0/0x2258 > > --> be->decompressed_pages = kvcalloc(be->nr_pages, sizeof(struct page *), GFP_KERNEL | __GFP_NOFAIL); > > kernel panic by NULL pointer dereference. > > erofs assume kvmalloc with __GFP_NOFAIL never return NULL. > > > > [65731.260293] [T32454] z_erofs_runqueue+0xf30/0x104c > > [65731.260314] [T32454] z_erofs_readahead+0x4f0/0x968 > > [65731.260339] [T32454] read_pages+0x170/0xadc > > [65731.260364] [T32454] page_cache_ra_unbounded+0x874/0xf30 > > [65731.260388] [T32454] page_cache_ra_order+0x24c/0x714 > > [65731.260411] [T32454] filemap_fault+0xbf0/0x1a74 > > [65731.260437] [T32454] __do_fault+0xd0/0x33c > > [65731.260462] [T32454] handle_mm_fault+0xf74/0x3fe0 > > [65731.260486] [T32454] do_mem_abort+0x54c/0x1b34 > > [65731.260509] [T32454] el0_da+0x44/0x94 > > [65731.260531] [T32454] el0t_64_sync_handler+0x98/0xb4 > > [65731.260553] [T32454] el0t_64_sync+0x198/0x19c > > > > -- > > 2.34.1 > > Thanks > Barry
On Thu, May 9, 2024 at 2:26 PM Barry Song <21cnbao@gmail.com> wrote: > > On Thu, May 9, 2024 at 2:20 PM Barry Song <21cnbao@gmail.com> wrote: > > > > On Thu, May 9, 2024 at 12:58 AM <hailong.liu@oppo.com> wrote: > > > > > > From: "Hailong.Liu" <hailong.liu@oppo.com> > > > > > > Commit a421ef303008 ("mm: allow !GFP_KERNEL allocations for kvmalloc") > > > includes support for __GFP_NOFAIL, but it presents a conflict with > > > commit dd544141b9eb ("vmalloc: back off when the current task is > > > OOM-killed"). A possible scenario is as belows: > > > > > > process-a > > > kvcalloc(n, m, GFP_KERNEL | __GFP_NOFAIL) > > > __vmalloc_node_range() > > > __vmalloc_area_node() > > > vm_area_alloc_pages() > > > --> oom-killer send SIGKILL to process-a > > > if (fatal_signal_pending(current)) break; > > > --> return NULL; > > > > > > to fix this, do not check fatal_signal_pending() in vm_area_alloc_pages() > > > if __GFP_NOFAIL set. > > > > > > Reported-by: Oven <liyangouwen1@oppo.com> > > > Signed-off-by: Hailong.Liu <hailong.liu@oppo.com> > > > --- > > > mm/vmalloc.c | 2 +- > > > 1 file changed, 1 insertion(+), 1 deletion(-) > > > > > > diff --git a/mm/vmalloc.c b/mm/vmalloc.c > > > index 6641be0ca80b..2f359d08bf8d 100644 > > > --- a/mm/vmalloc.c > > > +++ b/mm/vmalloc.c > > > @@ -3560,7 +3560,7 @@ vm_area_alloc_pages(gfp_t gfp, int nid, > > > > > > /* High-order pages or fallback path if "bulk" fails. */ > > > while (nr_allocated < nr_pages) { > > > - if (fatal_signal_pending(current)) > > > + if (!(gfp & __GFP_NOFAIL) && fatal_signal_pending(current)) > > > break; > > > > why not !nofail ? > > > > This seems a correct fix, but it undermines the assumption made in > > commit dd544141b9eb > > ("vmalloc: back off when the current task is OOM-killed") > > > > " > > This may trigger some hidden problems, when caller does not handle > > vmalloc failures, or when rollaback after failed vmalloc calls own > > vmallocs inside. However all of these scenarios are incorrect: vmalloc > > does not guarantee successful allocation, it has never been called with > > __GFP_NOFAIL and threfore either should not be used for any rollbacks or > > should handle such errors correctly and not lead to critical failures. > > " > > > > If a significant kvmalloc operation is performed with the NOFAIL flag, it risks > > reverting the fix intended to address the OOM-killer issue in commit > > dd544141b9eb. > > Should we indeed permit the NOFAIL flag for large kvmalloc allocations? > > + Vasily, Michal. Sorry for my oversight. Fixed the email of Vasily. > > > > > > > > > if (nid == NUMA_NO_NODE) > > > --- > > > This issue occurred during OPLUS KASAN test. Below is part of the log > > > > > > -> send signal > > > [65731.222840] [ T1308] oom-kill:constraint=CONSTRAINT_NONE,nodemask=(null),cpuset=/,mems_allowed=0,global_oom,task_memcg=/apps/uid_10198,task=gs.intelligence,pid=32454,uid=10198 > > > > > > [65731.259685] [T32454] Call trace: > > > [65731.259698] [T32454] dump_backtrace+0xf4/0x118 > > > [65731.259734] [T32454] show_stack+0x18/0x24 > > > [65731.259756] [T32454] dump_stack_lvl+0x60/0x7c > > > [65731.259781] [T32454] dump_stack+0x18/0x38 > > > [65731.259800] [T32454] mrdump_common_die+0x250/0x39c [mrdump] > > > [65731.259936] [T32454] ipanic_die+0x20/0x34 [mrdump] > > > [65731.260019] [T32454] atomic_notifier_call_chain+0xb4/0xfc > > > [65731.260047] [T32454] notify_die+0x114/0x198 > > > [65731.260073] [T32454] die+0xf4/0x5b4 > > > [65731.260098] [T32454] die_kernel_fault+0x80/0x98 > > > [65731.260124] [T32454] __do_kernel_fault+0x160/0x2a8 > > > [65731.260146] [T32454] do_bad_area+0x68/0x148 > > > [65731.260174] [T32454] do_mem_abort+0x151c/0x1b34 > > > [65731.260204] [T32454] el1_abort+0x3c/0x5c > > > [65731.260227] [T32454] el1h_64_sync_handler+0x54/0x90 > > > [65731.260248] [T32454] el1h_64_sync+0x68/0x6c > > > [65731.260269] [T32454] z_erofs_decompress_queue+0x7f0/0x2258 > > > --> be->decompressed_pages = kvcalloc(be->nr_pages, sizeof(struct page *), GFP_KERNEL | __GFP_NOFAIL); > > > kernel panic by NULL pointer dereference. > > > erofs assume kvmalloc with __GFP_NOFAIL never return NULL. > > > > > > [65731.260293] [T32454] z_erofs_runqueue+0xf30/0x104c > > > [65731.260314] [T32454] z_erofs_readahead+0x4f0/0x968 > > > [65731.260339] [T32454] read_pages+0x170/0xadc > > > [65731.260364] [T32454] page_cache_ra_unbounded+0x874/0xf30 > > > [65731.260388] [T32454] page_cache_ra_order+0x24c/0x714 > > > [65731.260411] [T32454] filemap_fault+0xbf0/0x1a74 > > > [65731.260437] [T32454] __do_fault+0xd0/0x33c > > > [65731.260462] [T32454] handle_mm_fault+0xf74/0x3fe0 > > > [65731.260486] [T32454] do_mem_abort+0x54c/0x1b34 > > > [65731.260509] [T32454] el0_da+0x44/0x94 > > > [65731.260531] [T32454] el0t_64_sync_handler+0x98/0xb4 > > > [65731.260553] [T32454] el0t_64_sync+0x198/0x19c > > > > > > -- > > > 2.34.1 > > > > Thanks > > Barry
Hi, On 2024/5/9 10:20, Barry Song wrote: > On Thu, May 9, 2024 at 12:58 AM <hailong.liu@oppo.com> wrote: >> >> From: "Hailong.Liu" <hailong.liu@oppo.com> >> >> Commit a421ef303008 ("mm: allow !GFP_KERNEL allocations for kvmalloc") >> includes support for __GFP_NOFAIL, but it presents a conflict with >> commit dd544141b9eb ("vmalloc: back off when the current task is >> OOM-killed"). A possible scenario is as belows: >> >> process-a >> kvcalloc(n, m, GFP_KERNEL | __GFP_NOFAIL) >> __vmalloc_node_range() >> __vmalloc_area_node() >> vm_area_alloc_pages() >> --> oom-killer send SIGKILL to process-a >> if (fatal_signal_pending(current)) break; >> --> return NULL; >> >> to fix this, do not check fatal_signal_pending() in vm_area_alloc_pages() >> if __GFP_NOFAIL set. >> >> Reported-by: Oven <liyangouwen1@oppo.com> >> Signed-off-by: Hailong.Liu <hailong.liu@oppo.com> >> --- >> mm/vmalloc.c | 2 +- >> 1 file changed, 1 insertion(+), 1 deletion(-) >> >> diff --git a/mm/vmalloc.c b/mm/vmalloc.c >> index 6641be0ca80b..2f359d08bf8d 100644 >> --- a/mm/vmalloc.c >> +++ b/mm/vmalloc.c >> @@ -3560,7 +3560,7 @@ vm_area_alloc_pages(gfp_t gfp, int nid, >> >> /* High-order pages or fallback path if "bulk" fails. */ >> while (nr_allocated < nr_pages) { >> - if (fatal_signal_pending(current)) >> + if (!(gfp & __GFP_NOFAIL) && fatal_signal_pending(current)) >> break; > > why not !nofail ? > > This seems a correct fix, but it undermines the assumption made in > commit dd544141b9eb > ("vmalloc: back off when the current task is OOM-killed") > > " > This may trigger some hidden problems, when caller does not handle > vmalloc failures, or when rollaback after failed vmalloc calls own > vmallocs inside. However all of these scenarios are incorrect: vmalloc > does not guarantee successful allocation, it has never been called with > __GFP_NOFAIL and threfore either should not be used for any rollbacks or > should handle such errors correctly and not lead to critical failures. > " > > If a significant kvmalloc operation is performed with the NOFAIL flag, it risks > reverting the fix intended to address the OOM-killer issue in commit > dd544141b9eb. > Should we indeed permit the NOFAIL flag for large kvmalloc allocations? Just from my perspective, I don't really care about kmalloc, vmalloc or kvmalloc (__GFP_NOFAIL). I even don't care if it returns three order-0 pages or a high-order page. I just would like to need a virtual consecutive buffer (even it works slowly.) with __GFP_NOFAIL. Because in some cases, writing fallback code may be tough and hard to test if such fallback path is correct since it only triggers in extreme workloads, and even such buffers are just used in a very short lifetime. Also see other FS discussion of __GFP_NOFAIL, e.g. https://lore.kernel.org/all/ZcUQfzfQ9R8X0s47@tiehlicka/ In the worst cases, it usually just needs < 5 order-0 pages (for many cases it only needs one page), but with kmalloc it will trigger WARN if it occurs to > order-1 allocation. as I mentioned before. With my limited understanding I don't see why it could any problem with kvmalloc(__GFP_NOFAIL) since it has no difference of kmalloc(GFP_NOFAIL) with order-0 allocation. Thanks, Gao XIang
On Thu, May 9, 2024 at 2:39 PM Gao Xiang <hsiangkao@linux.alibaba.com> wrote: > > Hi, > > On 2024/5/9 10:20, Barry Song wrote: > > On Thu, May 9, 2024 at 12:58 AM <hailong.liu@oppo.com> wrote: > >> > >> From: "Hailong.Liu" <hailong.liu@oppo.com> > >> > >> Commit a421ef303008 ("mm: allow !GFP_KERNEL allocations for kvmalloc") > >> includes support for __GFP_NOFAIL, but it presents a conflict with > >> commit dd544141b9eb ("vmalloc: back off when the current task is > >> OOM-killed"). A possible scenario is as belows: > >> > >> process-a > >> kvcalloc(n, m, GFP_KERNEL | __GFP_NOFAIL) > >> __vmalloc_node_range() > >> __vmalloc_area_node() > >> vm_area_alloc_pages() > >> --> oom-killer send SIGKILL to process-a > >> if (fatal_signal_pending(current)) break; > >> --> return NULL; > >> > >> to fix this, do not check fatal_signal_pending() in vm_area_alloc_pages() > >> if __GFP_NOFAIL set. > >> > >> Reported-by: Oven <liyangouwen1@oppo.com> > >> Signed-off-by: Hailong.Liu <hailong.liu@oppo.com> > >> --- > >> mm/vmalloc.c | 2 +- > >> 1 file changed, 1 insertion(+), 1 deletion(-) > >> > >> diff --git a/mm/vmalloc.c b/mm/vmalloc.c > >> index 6641be0ca80b..2f359d08bf8d 100644 > >> --- a/mm/vmalloc.c > >> +++ b/mm/vmalloc.c > >> @@ -3560,7 +3560,7 @@ vm_area_alloc_pages(gfp_t gfp, int nid, > >> > >> /* High-order pages or fallback path if "bulk" fails. */ > >> while (nr_allocated < nr_pages) { > >> - if (fatal_signal_pending(current)) > >> + if (!(gfp & __GFP_NOFAIL) && fatal_signal_pending(current)) > >> break; > > > > why not !nofail ? > > > > This seems a correct fix, but it undermines the assumption made in > > commit dd544141b9eb > > ("vmalloc: back off when the current task is OOM-killed") > > > > " > > This may trigger some hidden problems, when caller does not handle > > vmalloc failures, or when rollaback after failed vmalloc calls own > > vmallocs inside. However all of these scenarios are incorrect: vmalloc > > does not guarantee successful allocation, it has never been called with > > __GFP_NOFAIL and threfore either should not be used for any rollbacks or > > should handle such errors correctly and not lead to critical failures. > > " > > > > If a significant kvmalloc operation is performed with the NOFAIL flag, it risks > > reverting the fix intended to address the OOM-killer issue in commit > > dd544141b9eb. > > Should we indeed permit the NOFAIL flag for large kvmalloc allocations? > > Just from my perspective, I don't really care about kmalloc, vmalloc > or kvmalloc (__GFP_NOFAIL). I even don't care if it returns three > order-0 pages or a high-order page. I just would like to need a > virtual consecutive buffer (even it works slowly.) with __GFP_NOFAIL. > > Because in some cases, writing fallback code may be tough and hard to > test if such fallback path is correct since it only triggers in extreme > workloads, and even such buffers are just used in a very short lifetime. > Also see other FS discussion of __GFP_NOFAIL, e.g. > https://lore.kernel.org/all/ZcUQfzfQ9R8X0s47@tiehlicka/ > > In the worst cases, it usually just needs < 5 order-0 pages (for many > cases it only needs one page), but with kmalloc it will trigger WARN > if it occurs to > order-1 allocation. as I mentioned before. > > With my limited understanding I don't see why it could any problem with > kvmalloc(__GFP_NOFAIL) since it has no difference of kmalloc(GFP_NOFAIL) > with order-0 allocation. I completely understand that you're not concerned about the origin of the memory, such as whether it's organized by all zero-order pages. However, in the event that someone else allocates a large memory, like several megabytes with the NOFAIL flag, commit dd544141b9eb aims to halt the allocation before success if the process being allocated is targeted for termination of OOM-killer. With the current patch, we miss the opportunity for early allocation termination. However, if the size of the kvmalloc() is small, as in your case, I believe it should be perfectly fine. but do we have any way to prevent large size allocation with NOFAIL? > > > Thanks, > Gao XIang Thanks Barry
On 2024/5/9 10:39, Gao Xiang wrote: > Hi, > > On 2024/5/9 10:20, Barry Song wrote: >> On Thu, May 9, 2024 at 12:58 AM <hailong.liu@oppo.com> wrote: >>> >>> From: "Hailong.Liu" <hailong.liu@oppo.com> >>> >>> Commit a421ef303008 ("mm: allow !GFP_KERNEL allocations for kvmalloc") >>> includes support for __GFP_NOFAIL, but it presents a conflict with >>> commit dd544141b9eb ("vmalloc: back off when the current task is >>> OOM-killed"). A possible scenario is as belows: >>> >>> process-a >>> kvcalloc(n, m, GFP_KERNEL | __GFP_NOFAIL) >>> __vmalloc_node_range() >>> __vmalloc_area_node() >>> vm_area_alloc_pages() >>> --> oom-killer send SIGKILL to process-a >>> if (fatal_signal_pending(current)) break; >>> --> return NULL; >>> >>> to fix this, do not check fatal_signal_pending() in vm_area_alloc_pages() >>> if __GFP_NOFAIL set. >>> >>> Reported-by: Oven <liyangouwen1@oppo.com> >>> Signed-off-by: Hailong.Liu <hailong.liu@oppo.com> >>> --- >>> mm/vmalloc.c | 2 +- >>> 1 file changed, 1 insertion(+), 1 deletion(-) >>> >>> diff --git a/mm/vmalloc.c b/mm/vmalloc.c >>> index 6641be0ca80b..2f359d08bf8d 100644 >>> --- a/mm/vmalloc.c >>> +++ b/mm/vmalloc.c >>> @@ -3560,7 +3560,7 @@ vm_area_alloc_pages(gfp_t gfp, int nid, >>> >>> /* High-order pages or fallback path if "bulk" fails. */ >>> while (nr_allocated < nr_pages) { >>> - if (fatal_signal_pending(current)) >>> + if (!(gfp & __GFP_NOFAIL) && fatal_signal_pending(current)) >>> break; >> >> why not !nofail ? >> >> This seems a correct fix, but it undermines the assumption made in >> commit dd544141b9eb >> ("vmalloc: back off when the current task is OOM-killed") >> >> " >> This may trigger some hidden problems, when caller does not handle >> vmalloc failures, or when rollaback after failed vmalloc calls own >> vmallocs inside. However all of these scenarios are incorrect: vmalloc >> does not guarantee successful allocation, it has never been called with >> __GFP_NOFAIL and threfore either should not be used for any rollbacks or >> should handle such errors correctly and not lead to critical failures. >> " >> >> If a significant kvmalloc operation is performed with the NOFAIL flag, it risks >> reverting the fix intended to address the OOM-killer issue in commit >> dd544141b9eb. >> Should we indeed permit the NOFAIL flag for large kvmalloc allocations? > > Just from my perspective, I don't really care about kmalloc, vmalloc > or kvmalloc (__GFP_NOFAIL). I even don't care if it returns three > order-0 pages or a high-order page. I just would like to need a > virtual consecutive buffer (even it works slowly.) with __GFP_NOFAIL. > > Because in some cases, writing fallback code may be tough and hard to > test if such fallback path is correct since it only triggers in extreme > workloads, and even such buffers are just used in a very short lifetime. add some words... ^ here extreme cases were mostly just generated by syzkaller fuzzing tests, but if real users try to use some configuration to compress more, I still think it needs to be handled (even such kvmalloc may be slow if falling back to order-0 allocations due to memory pressures) > Also see other FS discussion of __GFP_NOFAIL, e.g. > https://lore.kernel.org/all/ZcUQfzfQ9R8X0s47@tiehlicka/ > > In the worst cases, it usually just needs < 5 order-0 pages (for many > cases it only needs one page), but with kmalloc it will trigger WARN > if it occurs to > order-1 allocation. as I mentioned before. > > With my limited understanding I don't see why it could any problem with > kvmalloc(__GFP_NOFAIL) since it has no difference of kmalloc(GFP_NOFAIL) > with order-0 allocation. .. kvmalloc with order-0 pages to form a virtual consecutive buffer just like several kmalloc(__GFP_NOFAIL) allocations together in the callers, I don't see any difference of memory pressure here. Thanks, Gao Xiang > > > Thanks, > Gao XIang
On 2024/5/9 11:09, Barry Song wrote: > On Thu, May 9, 2024 at 2:39 PM Gao Xiang <hsiangkao@linux.alibaba.com> wrote: >> >> Hi, >> >> On 2024/5/9 10:20, Barry Song wrote: >>> On Thu, May 9, 2024 at 12:58 AM <hailong.liu@oppo.com> wrote: >>>> >>>> From: "Hailong.Liu" <hailong.liu@oppo.com> >>>> >>>> Commit a421ef303008 ("mm: allow !GFP_KERNEL allocations for kvmalloc") >>>> includes support for __GFP_NOFAIL, but it presents a conflict with >>>> commit dd544141b9eb ("vmalloc: back off when the current task is >>>> OOM-killed"). A possible scenario is as belows: >>>> >>>> process-a >>>> kvcalloc(n, m, GFP_KERNEL | __GFP_NOFAIL) >>>> __vmalloc_node_range() >>>> __vmalloc_area_node() >>>> vm_area_alloc_pages() >>>> --> oom-killer send SIGKILL to process-a >>>> if (fatal_signal_pending(current)) break; >>>> --> return NULL; >>>> >>>> to fix this, do not check fatal_signal_pending() in vm_area_alloc_pages() >>>> if __GFP_NOFAIL set. >>>> >>>> Reported-by: Oven <liyangouwen1@oppo.com> >>>> Signed-off-by: Hailong.Liu <hailong.liu@oppo.com> >>>> --- >>>> mm/vmalloc.c | 2 +- >>>> 1 file changed, 1 insertion(+), 1 deletion(-) >>>> >>>> diff --git a/mm/vmalloc.c b/mm/vmalloc.c >>>> index 6641be0ca80b..2f359d08bf8d 100644 >>>> --- a/mm/vmalloc.c >>>> +++ b/mm/vmalloc.c >>>> @@ -3560,7 +3560,7 @@ vm_area_alloc_pages(gfp_t gfp, int nid, >>>> >>>> /* High-order pages or fallback path if "bulk" fails. */ >>>> while (nr_allocated < nr_pages) { >>>> - if (fatal_signal_pending(current)) >>>> + if (!(gfp & __GFP_NOFAIL) && fatal_signal_pending(current)) >>>> break; >>> >>> why not !nofail ? >>> >>> This seems a correct fix, but it undermines the assumption made in >>> commit dd544141b9eb >>> ("vmalloc: back off when the current task is OOM-killed") >>> >>> " >>> This may trigger some hidden problems, when caller does not handle >>> vmalloc failures, or when rollaback after failed vmalloc calls own >>> vmallocs inside. However all of these scenarios are incorrect: vmalloc >>> does not guarantee successful allocation, it has never been called with >>> __GFP_NOFAIL and threfore either should not be used for any rollbacks or >>> should handle such errors correctly and not lead to critical failures. >>> " >>> >>> If a significant kvmalloc operation is performed with the NOFAIL flag, it risks >>> reverting the fix intended to address the OOM-killer issue in commit >>> dd544141b9eb. >>> Should we indeed permit the NOFAIL flag for large kvmalloc allocations? >> >> Just from my perspective, I don't really care about kmalloc, vmalloc >> or kvmalloc (__GFP_NOFAIL). I even don't care if it returns three >> order-0 pages or a high-order page. I just would like to need a >> virtual consecutive buffer (even it works slowly.) with __GFP_NOFAIL. >> >> Because in some cases, writing fallback code may be tough and hard to >> test if such fallback path is correct since it only triggers in extreme >> workloads, and even such buffers are just used in a very short lifetime. >> Also see other FS discussion of __GFP_NOFAIL, e.g. >> https://lore.kernel.org/all/ZcUQfzfQ9R8X0s47@tiehlicka/ >> >> In the worst cases, it usually just needs < 5 order-0 pages (for many >> cases it only needs one page), but with kmalloc it will trigger WARN >> if it occurs to > order-1 allocation. as I mentioned before. >> >> With my limited understanding I don't see why it could any problem with >> kvmalloc(__GFP_NOFAIL) since it has no difference of kmalloc(GFP_NOFAIL) >> with order-0 allocation. > > I completely understand that you're not concerned about the origin of > the memory, > such as whether it's organized by all zero-order pages. However, in the event > that someone else allocates a large memory, like several megabytes with the > NOFAIL flag, commit dd544141b9eb aims to halt the allocation before success > if the process being allocated is targeted for termination of OOM-killer. > > With the current patch, we miss the opportunity for early allocation > termination. > However, if the size of the kvmalloc() is small, as in your case, I > believe it should > be perfectly fine. but do we have any way to prevent large size allocation with > NOFAIL? I think large size order-0 virtual consecutive allocation prevention should be done by the callers (e.g. EROFS) other than memory subsystem, since it totally sounds like caller bugs. It doesn't make sense for me to block getting a virtual consecutive buffer which only needs more than two (e.g. three or five) order-0 pages in the extreme cases. Yes, I need to prevent very insane allocations but it's an on-disk hard limitation of a filesystem itself. Thanks, Gao Xiang > >> >> >> Thanks, >> Gao XIang > > Thanks > Barry
On Thu, 09. May 10:39, Gao Xiang wrote: > Hi, > > On 2024/5/9 10:20, Barry Song wrote: > > On Thu, May 9, 2024 at 12:58 AM <hailong.liu@oppo.com> wrote: > > > > > > From: "Hailong.Liu" <hailong.liu@oppo.com> > > > > > > Commit a421ef303008 ("mm: allow !GFP_KERNEL allocations for kvmalloc") > > > includes support for __GFP_NOFAIL, but it presents a conflict with > > > commit dd544141b9eb ("vmalloc: back off when the current task is > > > OOM-killed"). A possible scenario is as belows: > > > > > > process-a > > > kvcalloc(n, m, GFP_KERNEL | __GFP_NOFAIL) > > > __vmalloc_node_range() > > > __vmalloc_area_node() > > > vm_area_alloc_pages() > > > --> oom-killer send SIGKILL to process-a > > > if (fatal_signal_pending(current)) break; > > > --> return NULL; > > > > > > to fix this, do not check fatal_signal_pending() in vm_area_alloc_pages() > > > if __GFP_NOFAIL set. > > > > > > Reported-by: Oven <liyangouwen1@oppo.com> > > > Signed-off-by: Hailong.Liu <hailong.liu@oppo.com> > > > --- > > > mm/vmalloc.c | 2 +- > > > 1 file changed, 1 insertion(+), 1 deletion(-) > > > > > > diff --git a/mm/vmalloc.c b/mm/vmalloc.c > > > index 6641be0ca80b..2f359d08bf8d 100644 > > > --- a/mm/vmalloc.c > > > +++ b/mm/vmalloc.c > > > @@ -3560,7 +3560,7 @@ vm_area_alloc_pages(gfp_t gfp, int nid, > > > > > > /* High-order pages or fallback path if "bulk" fails. */ > > > while (nr_allocated < nr_pages) { > > > - if (fatal_signal_pending(current)) > > > + if (!(gfp & __GFP_NOFAIL) && fatal_signal_pending(current)) > > > break; > > > > why not !nofail ? > > > > This seems a correct fix, but it undermines the assumption made in > > commit dd544141b9eb > > ("vmalloc: back off when the current task is OOM-killed") > > > > " > > This may trigger some hidden problems, when caller does not handle > > vmalloc failures, or when rollaback after failed vmalloc calls own > > vmallocs inside. However all of these scenarios are incorrect: vmalloc > > does not guarantee successful allocation, it has never been called with > > __GFP_NOFAIL and threfore either should not be used for any rollbacks or > > should handle such errors correctly and not lead to critical failures. > > " j> > > > If a significant kvmalloc operation is performed with the NOFAIL flag, it risks > > reverting the fix intended to address the OOM-killer issue in commit > > dd544141b9eb. > > Should we indeed permit the NOFAIL flag for large kvmalloc allocations? > > Just from my perspective, I don't really care about kmalloc, vmalloc > or kvmalloc (__GFP_NOFAIL). I even don't care if it returns three > order-0 pages or a high-order page. I just would like to need a > virtual consecutive buffer (even it works slowly.) with __GFP_NOFAIL. > > Because in some cases, writing fallback code may be tough and hard to > test if such fallback path is correct since it only triggers in extreme > workloads, and even such buffers are just used in a very short lifetime. > Also see other FS discussion of __GFP_NOFAIL, e.g. > https://lore.kernel.org/all/ZcUQfzfQ9R8X0s47@tiehlicka/ > > In the worst cases, it usually just needs < 5 order-0 pages (for many > cases it only needs one page), but with kmalloc it will trigger WARN > if it occurs to > order-1 allocation. as I mentioned before. > > With my limited understanding I don't see why it could any problem with > kvmalloc(__GFP_NOFAIL) since it has no difference of kmalloc(GFP_NOFAIL) > with order-0 allocation. > I totally agree with you. so I fixed this in vmalloc or not in erofs. However, there are still some extreme scenarios that cause vmalloc to return NULL, such as vmap failure. Let's see what the mm experts think. > > Thanks, > Gao XIang > -- Best Regards, Hailong.
On Thu, 09. May 14:20, Barry Song wrote: > On Thu, May 9, 2024 at 12:58 AM <hailong.liu@oppo.com> wrote: > > > > From: "Hailong.Liu" <hailong.liu@oppo.com> > > > > Commit a421ef303008 ("mm: allow !GFP_KERNEL allocations for kvmalloc") > > includes support for __GFP_NOFAIL, but it presents a conflict with > > commit dd544141b9eb ("vmalloc: back off when the current task is > > OOM-killed"). A possible scenario is as belows: > > > > process-a > > kvcalloc(n, m, GFP_KERNEL | __GFP_NOFAIL) > > __vmalloc_node_range() > > __vmalloc_area_node() > > vm_area_alloc_pages() > > --> oom-killer send SIGKILL to process-a > > if (fatal_signal_pending(current)) break; > > --> return NULL; > > > > to fix this, do not check fatal_signal_pending() in vm_area_alloc_pages() > > if __GFP_NOFAIL set. > > > > Reported-by: Oven <liyangouwen1@oppo.com> > > Signed-off-by: Hailong.Liu <hailong.liu@oppo.com> > > --- > > mm/vmalloc.c | 2 +- > > 1 file changed, 1 insertion(+), 1 deletion(-) > > > > diff --git a/mm/vmalloc.c b/mm/vmalloc.c > > index 6641be0ca80b..2f359d08bf8d 100644 > > --- a/mm/vmalloc.c > > +++ b/mm/vmalloc.c > > @@ -3560,7 +3560,7 @@ vm_area_alloc_pages(gfp_t gfp, int nid, > > > > /* High-order pages or fallback path if "bulk" fails. */ > > while (nr_allocated < nr_pages) { > > - if (fatal_signal_pending(current)) > > + if (!(gfp & __GFP_NOFAIL) && fatal_signal_pending(current)) > > break; > > why not !nofail ? if order = 0, nofail would not be set true in bulk allocator. in such a case, it is still possible to break early > > This seems a correct fix, but it undermines the assumption made in > commit dd544141b9eb > ("vmalloc: back off when the current task is OOM-killed") > > " > This may trigger some hidden problems, when caller does not handle > vmalloc failures, or when rollaback after failed vmalloc calls own > vmallocs inside. However all of these scenarios are incorrect: vmalloc > does not guarantee successful allocation, it has never been called with > __GFP_NOFAIL and threfore either should not be used for any rollbacks or > should handle such errors correctly and not lead to critical failures. > " > > If a significant kvmalloc operation is performed with the NOFAIL flag, it risks > reverting the fix intended to address the OOM-killer issue in commit > dd544141b9eb. > Should we indeed permit the NOFAIL flag for large kvmalloc allocations? IMO, if we encounter this issue, it should be fixed by the caller, not here. > > > Thanks > Barry -- Best Regards, Hailong.
On Thu, May 9, 2024 at 3:33 PM Hailong Liu <hailong.liu@oppo.com> wrote: > > On Thu, 09. May 14:20, Barry Song wrote: > > On Thu, May 9, 2024 at 12:58 AM <hailong.liu@oppo.com> wrote: > > > > > > From: "Hailong.Liu" <hailong.liu@oppo.com> > > > > > > Commit a421ef303008 ("mm: allow !GFP_KERNEL allocations for kvmalloc") > > > includes support for __GFP_NOFAIL, but it presents a conflict with > > > commit dd544141b9eb ("vmalloc: back off when the current task is > > > OOM-killed"). A possible scenario is as belows: > > > > > > process-a > > > kvcalloc(n, m, GFP_KERNEL | __GFP_NOFAIL) > > > __vmalloc_node_range() > > > __vmalloc_area_node() > > > vm_area_alloc_pages() > > > --> oom-killer send SIGKILL to process-a > > > if (fatal_signal_pending(current)) break; > > > --> return NULL; > > > > > > to fix this, do not check fatal_signal_pending() in vm_area_alloc_pages() > > > if __GFP_NOFAIL set. > > > > > > Reported-by: Oven <liyangouwen1@oppo.com> > > > Signed-off-by: Hailong.Liu <hailong.liu@oppo.com> > > > --- > > > mm/vmalloc.c | 2 +- > > > 1 file changed, 1 insertion(+), 1 deletion(-) > > > > > > diff --git a/mm/vmalloc.c b/mm/vmalloc.c > > > index 6641be0ca80b..2f359d08bf8d 100644 > > > --- a/mm/vmalloc.c > > > +++ b/mm/vmalloc.c > > > @@ -3560,7 +3560,7 @@ vm_area_alloc_pages(gfp_t gfp, int nid, > > > > > > /* High-order pages or fallback path if "bulk" fails. */ > > > while (nr_allocated < nr_pages) { > > > - if (fatal_signal_pending(current)) > > > + if (!(gfp & __GFP_NOFAIL) && fatal_signal_pending(current)) > > > break; > > > > why not !nofail ? > > if order = 0, nofail would not be set true in bulk allocator. in such a case, > it is still possible to break early > > > > > This seems a correct fix, but it undermines the assumption made in > > commit dd544141b9eb > > ("vmalloc: back off when the current task is OOM-killed") > > > > " > > This may trigger some hidden problems, when caller does not handle > > vmalloc failures, or when rollaback after failed vmalloc calls own > > vmallocs inside. However all of these scenarios are incorrect: vmalloc > > does not guarantee successful allocation, it has never been called with > > __GFP_NOFAIL and threfore either should not be used for any rollbacks or > > should handle such errors correctly and not lead to critical failures. > > " > > > > If a significant kvmalloc operation is performed with the NOFAIL flag, it risks > > reverting the fix intended to address the OOM-killer issue in commit > > dd544141b9eb. > > Should we indeed permit the NOFAIL flag for large kvmalloc allocations? > > IMO, if we encounter this issue, it should be fixed by the > caller, not here. I agree. but could we WARN_ON a large kvmalloc(NOFAIL) allocation? > > > > > > > Thanks > > Barry > > -- > > Best Regards, > Hailong.
On 2024/5/9 11:48, Barry Song wrote: > On Thu, May 9, 2024 at 3:33 PM Hailong Liu <hailong.liu@oppo.com> wrote: >> >> On Thu, 09. May 14:20, Barry Song wrote: >>> On Thu, May 9, 2024 at 12:58 AM <hailong.liu@oppo.com> wrote: >>>> >>>> From: "Hailong.Liu" <hailong.liu@oppo.com> >>>> >>>> Commit a421ef303008 ("mm: allow !GFP_KERNEL allocations for kvmalloc") >>>> includes support for __GFP_NOFAIL, but it presents a conflict with >>>> commit dd544141b9eb ("vmalloc: back off when the current task is >>>> OOM-killed"). A possible scenario is as belows: >>>> >>>> process-a >>>> kvcalloc(n, m, GFP_KERNEL | __GFP_NOFAIL) >>>> __vmalloc_node_range() >>>> __vmalloc_area_node() >>>> vm_area_alloc_pages() >>>> --> oom-killer send SIGKILL to process-a >>>> if (fatal_signal_pending(current)) break; >>>> --> return NULL; >>>> >>>> to fix this, do not check fatal_signal_pending() in vm_area_alloc_pages() >>>> if __GFP_NOFAIL set. >>>> >>>> Reported-by: Oven <liyangouwen1@oppo.com> >>>> Signed-off-by: Hailong.Liu <hailong.liu@oppo.com> >>>> --- >>>> mm/vmalloc.c | 2 +- >>>> 1 file changed, 1 insertion(+), 1 deletion(-) >>>> >>>> diff --git a/mm/vmalloc.c b/mm/vmalloc.c >>>> index 6641be0ca80b..2f359d08bf8d 100644 >>>> --- a/mm/vmalloc.c >>>> +++ b/mm/vmalloc.c >>>> @@ -3560,7 +3560,7 @@ vm_area_alloc_pages(gfp_t gfp, int nid, >>>> >>>> /* High-order pages or fallback path if "bulk" fails. */ >>>> while (nr_allocated < nr_pages) { >>>> - if (fatal_signal_pending(current)) >>>> + if (!(gfp & __GFP_NOFAIL) && fatal_signal_pending(current)) >>>> break; >>> >>> why not !nofail ? >> >> if order = 0, nofail would not be set true in bulk allocator. in such a case, >> it is still possible to break early >> >>> >>> This seems a correct fix, but it undermines the assumption made in >>> commit dd544141b9eb >>> ("vmalloc: back off when the current task is OOM-killed") >>> >>> " >>> This may trigger some hidden problems, when caller does not handle >>> vmalloc failures, or when rollaback after failed vmalloc calls own >>> vmallocs inside. However all of these scenarios are incorrect: vmalloc >>> does not guarantee successful allocation, it has never been called with >>> __GFP_NOFAIL and threfore either should not be used for any rollbacks or >>> should handle such errors correctly and not lead to critical failures. >>> " >>> >>> If a significant kvmalloc operation is performed with the NOFAIL flag, it risks >>> reverting the fix intended to address the OOM-killer issue in commit >>> dd544141b9eb. >>> Should we indeed permit the NOFAIL flag for large kvmalloc allocations? >> >> IMO, if we encounter this issue, it should be fixed by the >> caller, not here. > > I agree. but could we WARN_ON a large kvmalloc(NOFAIL) allocation? For me, kvmalloc(~24k) sounds good to handle extreme cases (for extreme compression), only specific users for file archiving (and fuzzers) are expected to use this configuration. Anyway, it's just my own ideas. Thanks, Gao Xiang > >>> >> >>> >>> Thanks >>> Barry >> >> -- >> >> Best Regards, >> Hailong.
On Thu, May 09, 2024 at 09:30:59AM +0800, Hailong Liu wrote: > I’m not suggesting that erofs would cause a memleak. What I mean is > that if kvmalloc is invoked with __GFP_NOFAIL, it must ensure a non-NULL > return, even in scenarios where memory leaks caused by other processes > result in the inability to allocate a page. In such a situation, it > should result in “Kernel panic - not syncing: System is deadlocked > on memory”. Yes. __GFP_NOFAIL is a contract that says never ever return NULL. The callers will generally not handle a NULL return and blindly dereference it, leading to all kinds of nasty security issues. Note that deadlocking would be nice, but at least it is just a denial of service and not a possible privilege escalation.
On Thu, May 09, 2024 at 02:20:03PM +1200, Barry Song wrote: > reverting the fix intended to address the OOM-killer issue in commit > dd544141b9eb. > Should we indeed permit the NOFAIL flag for large kvmalloc allocations? What is large? When you don't allow actually use cases people will just reimplement it poorly. E.g. we'd probably have to add back the XFS kmem_ wrappers.
On Thu, May 9, 2024 at 4:52 PM Christoph Hellwig <hch@infradead.org> wrote: > > On Thu, May 09, 2024 at 02:20:03PM +1200, Barry Song wrote: > > reverting the fix intended to address the OOM-killer issue in commit > > dd544141b9eb. > > Should we indeed permit the NOFAIL flag for large kvmalloc allocations? > > What is large? When you don't allow actually use cases people will > just reimplement it poorly. E.g. we'd probably have to add back the > XFS kmem_ wrappers. Xiang gave his number 24KiB in erofs, but probably we still have some "naughty" users allocating much more than 24KiB with NOFAIL in other subsystems. We should never return NULL for NOFAIL. So, in any case, we require Hailong's patch in some form. However, commit dd544141b9eb ("vmalloc: back off when the current task is OOM-killed") is also tackling an issue. If we can find a way to preserve its benefits even in the NOFAIL scenario, it would be preferable, though it seems improbable. Thus, I'm considering if we could at least include a WARN_ON_ONCE((gfp & NOFAIL) && size > LARGE) to aid in debugging potential issues we might encounter. Furthermore, compelling a large allocation with NOFAIL also appears pointless, as it could lead to unpredictable long latency. But I don't know what the proper "LARGE" value is. Thanks Barry
On Wed 08-05-24 20:58:08, hailong.liu@oppo.com wrote: > From: "Hailong.Liu" <hailong.liu@oppo.com> > > Commit a421ef303008 ("mm: allow !GFP_KERNEL allocations for kvmalloc") > includes support for __GFP_NOFAIL, but it presents a conflict with > commit dd544141b9eb ("vmalloc: back off when the current task is > OOM-killed"). A possible scenario is as belows: > > process-a > kvcalloc(n, m, GFP_KERNEL | __GFP_NOFAIL) > __vmalloc_node_range() > __vmalloc_area_node() > vm_area_alloc_pages() > --> oom-killer send SIGKILL to process-a > if (fatal_signal_pending(current)) break; > --> return NULL; > > to fix this, do not check fatal_signal_pending() in vm_area_alloc_pages() > if __GFP_NOFAIL set. > > Reported-by: Oven <liyangouwen1@oppo.com> > Signed-off-by: Hailong.Liu <hailong.liu@oppo.com> > --- > mm/vmalloc.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/mm/vmalloc.c b/mm/vmalloc.c > index 6641be0ca80b..2f359d08bf8d 100644 > --- a/mm/vmalloc.c > +++ b/mm/vmalloc.c > @@ -3560,7 +3560,7 @@ vm_area_alloc_pages(gfp_t gfp, int nid, > > /* High-order pages or fallback path if "bulk" fails. */ > while (nr_allocated < nr_pages) { > - if (fatal_signal_pending(current)) > + if (!(gfp & __GFP_NOFAIL) && fatal_signal_pending(current)) Use nofail instead of gfp & __GFP_NOFAIL. Other than that looks good to me. After that is fixed, please feel free to add Acked-by: Michal Hocko <mhocko@suse.com> I believe this should also have Fixes: 9376130c390a ("mm/vmalloc: add support for __GFP_NOFAIL")
On Thu, 09. May 09:48, Michal Hocko wrote: > On Wed 08-05-24 20:58:08, hailong.liu@oppo.com wrote: > > From: "Hailong.Liu" <hailong.liu@oppo.com> > > > > Commit a421ef303008 ("mm: allow !GFP_KERNEL allocations for kvmalloc") > > includes support for __GFP_NOFAIL, but it presents a conflict with > > commit dd544141b9eb ("vmalloc: back off when the current task is > > OOM-killed"). A possible scenario is as belows: > > > > process-a > > kvcalloc(n, m, GFP_KERNEL | __GFP_NOFAIL) > > __vmalloc_node_range() > > __vmalloc_area_node() > > vm_area_alloc_pages() > > --> oom-killer send SIGKILL to process-a > > if (fatal_signal_pending(current)) break; > > --> return NULL; > > > > to fix this, do not check fatal_signal_pending() in vm_area_alloc_pages() > > if __GFP_NOFAIL set. > > > > Reported-by: Oven <liyangouwen1@oppo.com> > > Signed-off-by: Hailong.Liu <hailong.liu@oppo.com> > > --- > > mm/vmalloc.c | 2 +- > > 1 file changed, 1 insertion(+), 1 deletion(-) > > > > diff --git a/mm/vmalloc.c b/mm/vmalloc.c > > index 6641be0ca80b..2f359d08bf8d 100644 > > --- a/mm/vmalloc.c > > +++ b/mm/vmalloc.c > > @@ -3560,7 +3560,7 @@ vm_area_alloc_pages(gfp_t gfp, int nid, > > > > /* High-order pages or fallback path if "bulk" fails. */ > > while (nr_allocated < nr_pages) { > > - if (fatal_signal_pending(current)) > > + if (!(gfp & __GFP_NOFAIL) && fatal_signal_pending(current)) > > Use nofail instead of gfp & __GFP_NOFAIL. > > Other than that looks good to me. After that is fixed, please feel free > to add Acked-by: Michal Hocko <mhocko@suse.com> > > I believe this should also have Fixes: 9376130c390a ("mm/vmalloc: add support for __GFP_NOFAIL") > -- > Michal Hocko > SUSE Labs Thanks for the review and the Ack! Add Fixes in V2 patch. IIUC, nofail could not used for this case. /* * For order-0 pages we make use of bulk allocator, if * the page array is partly or not at all populated due * to fails, fallback to a single page allocator that is * more permissive. */ if (!order) { /* bulk allocator doesn't support nofail req. officially */ xxx -> nofail = false; } else if (gfp & __GFP_NOFAIL) { /* * Higher order nofail allocations are really expensive and * potentially dangerous (pre-mature OOM, disruptive reclaim * and compaction etc. */ alloc_gfp &= ~__GFP_NOFAIL; nofail = true; } /* High-order pages or fallback path if "bulk" fails. */ while (nr_allocated < nr_pages) { -> nofail is false here if bulk allocator fails. if (fatal_signal_pending(current)) break; -- Best Regards, Hailong.
On Thu, May 9, 2024 at 8:21 PM Hailong Liu <hailong.liu@oppo.com> wrote: > > On Thu, 09. May 09:48, Michal Hocko wrote: > > On Wed 08-05-24 20:58:08, hailong.liu@oppo.com wrote: > > > From: "Hailong.Liu" <hailong.liu@oppo.com> > > > > > > Commit a421ef303008 ("mm: allow !GFP_KERNEL allocations for kvmalloc") > > > includes support for __GFP_NOFAIL, but it presents a conflict with > > > commit dd544141b9eb ("vmalloc: back off when the current task is > > > OOM-killed"). A possible scenario is as belows: > > > > > > process-a > > > kvcalloc(n, m, GFP_KERNEL | __GFP_NOFAIL) > > > __vmalloc_node_range() > > > __vmalloc_area_node() > > > vm_area_alloc_pages() > > > --> oom-killer send SIGKILL to process-a > > > if (fatal_signal_pending(current)) break; > > > --> return NULL; > > > > > > to fix this, do not check fatal_signal_pending() in vm_area_alloc_pages() > > > if __GFP_NOFAIL set. > > > > > > Reported-by: Oven <liyangouwen1@oppo.com> > > > Signed-off-by: Hailong.Liu <hailong.liu@oppo.com> > > > --- > > > mm/vmalloc.c | 2 +- > > > 1 file changed, 1 insertion(+), 1 deletion(-) > > > > > > diff --git a/mm/vmalloc.c b/mm/vmalloc.c > > > index 6641be0ca80b..2f359d08bf8d 100644 > > > --- a/mm/vmalloc.c > > > +++ b/mm/vmalloc.c > > > @@ -3560,7 +3560,7 @@ vm_area_alloc_pages(gfp_t gfp, int nid, > > > > > > /* High-order pages or fallback path if "bulk" fails. */ > > > while (nr_allocated < nr_pages) { > > > - if (fatal_signal_pending(current)) > > > + if (!(gfp & __GFP_NOFAIL) && fatal_signal_pending(current)) > > > > Use nofail instead of gfp & __GFP_NOFAIL. > > > > Other than that looks good to me. After that is fixed, please feel free > > to add Acked-by: Michal Hocko <mhocko@suse.com> > > > > I believe this should also have Fixes: 9376130c390a ("mm/vmalloc: add support for __GFP_NOFAIL") > > -- > > Michal Hocko > > SUSE Labs > > Thanks for the review and the Ack! > > Add Fixes in V2 patch. > > IIUC, nofail could not used for this case. > > /* > * For order-0 pages we make use of bulk allocator, if > * the page array is partly or not at all populated due > * to fails, fallback to a single page allocator that is > * more permissive. > */ > if (!order) { > /* bulk allocator doesn't support nofail req. officially */ > xxx > -> nofail = false; isn't it another bug that needs a fix? > } else if (gfp & __GFP_NOFAIL) { > /* > * Higher order nofail allocations are really expensive and > * potentially dangerous (pre-mature OOM, disruptive reclaim > * and compaction etc. > */ > alloc_gfp &= ~__GFP_NOFAIL; > nofail = true; > } > > /* High-order pages or fallback path if "bulk" fails. */ > while (nr_allocated < nr_pages) { > > -> nofail is false here if bulk allocator fails. > if (fatal_signal_pending(current)) > break; > > -- > > Best Regards, > Hailong.
On Thu, May 9, 2024 at 8:32 PM Barry Song <21cnbao@gmail.com> wrote: > > On Thu, May 9, 2024 at 8:21 PM Hailong Liu <hailong.liu@oppo.com> wrote: > > > > On Thu, 09. May 09:48, Michal Hocko wrote: > > > On Wed 08-05-24 20:58:08, hailong.liu@oppo.com wrote: > > > > From: "Hailong.Liu" <hailong.liu@oppo.com> > > > > > > > > Commit a421ef303008 ("mm: allow !GFP_KERNEL allocations for kvmalloc") > > > > includes support for __GFP_NOFAIL, but it presents a conflict with > > > > commit dd544141b9eb ("vmalloc: back off when the current task is > > > > OOM-killed"). A possible scenario is as belows: > > > > > > > > process-a > > > > kvcalloc(n, m, GFP_KERNEL | __GFP_NOFAIL) > > > > __vmalloc_node_range() > > > > __vmalloc_area_node() > > > > vm_area_alloc_pages() > > > > --> oom-killer send SIGKILL to process-a > > > > if (fatal_signal_pending(current)) break; > > > > --> return NULL; > > > > > > > > to fix this, do not check fatal_signal_pending() in vm_area_alloc_pages() > > > > if __GFP_NOFAIL set. > > > > > > > > Reported-by: Oven <liyangouwen1@oppo.com> > > > > Signed-off-by: Hailong.Liu <hailong.liu@oppo.com> > > > > --- > > > > mm/vmalloc.c | 2 +- > > > > 1 file changed, 1 insertion(+), 1 deletion(-) > > > > > > > > diff --git a/mm/vmalloc.c b/mm/vmalloc.c > > > > index 6641be0ca80b..2f359d08bf8d 100644 > > > > --- a/mm/vmalloc.c > > > > +++ b/mm/vmalloc.c > > > > @@ -3560,7 +3560,7 @@ vm_area_alloc_pages(gfp_t gfp, int nid, > > > > > > > > /* High-order pages or fallback path if "bulk" fails. */ > > > > while (nr_allocated < nr_pages) { > > > > - if (fatal_signal_pending(current)) > > > > + if (!(gfp & __GFP_NOFAIL) && fatal_signal_pending(current)) > > > > > > Use nofail instead of gfp & __GFP_NOFAIL. > > > > > > Other than that looks good to me. After that is fixed, please feel free > > > to add Acked-by: Michal Hocko <mhocko@suse.com> > > > > > > I believe this should also have Fixes: 9376130c390a ("mm/vmalloc: add support for __GFP_NOFAIL") > > > -- > > > Michal Hocko > > > SUSE Labs > > > > Thanks for the review and the Ack! > > > > Add Fixes in V2 patch. > > > > IIUC, nofail could not used for this case. > > > > /* > > * For order-0 pages we make use of bulk allocator, if > > * the page array is partly or not at all populated due > > * to fails, fallback to a single page allocator that is > > * more permissive. > > */ > > if (!order) { > > /* bulk allocator doesn't support nofail req. officially */ > > xxx > > -> nofail = false; > > isn't it another bug that needs a fix? Upon further examination, it's not a bug, but we can still utilize 'nofail'. The current code is very hard to read about gfp and "nofail" :-) maybe: diff --git a/mm/vmalloc.c b/mm/vmalloc.c index 6641be0ca80b..7c66fe16c2ad 100644 --- a/mm/vmalloc.c +++ b/mm/vmalloc.c @@ -3498,7 +3498,7 @@ vm_area_alloc_pages(gfp_t gfp, int nid, { unsigned int nr_allocated = 0; gfp_t alloc_gfp = gfp; - bool nofail = false; + bool nofail = !!(gfp & __GFP_NOFAIL); struct page *page; int i; @@ -3555,7 +3555,6 @@ vm_area_alloc_pages(gfp_t gfp, int nid, * and compaction etc. */ alloc_gfp &= ~__GFP_NOFAIL; - nofail = true; } /* High-order pages or fallback path if "bulk" fails. */ > > > } else if (gfp & __GFP_NOFAIL) { > > /* > > * Higher order nofail allocations are really expensive and > > * potentially dangerous (pre-mature OOM, disruptive reclaim > > * and compaction etc. > > */ > > alloc_gfp &= ~__GFP_NOFAIL; > > nofail = true; > > } > > > > /* High-order pages or fallback path if "bulk" fails. */ > > while (nr_allocated < nr_pages) { > > > > -> nofail is false here if bulk allocator fails. > > if (fatal_signal_pending(current)) > > break; > > > > -- > > > > Best Regards, > > Hailong.
On Thu, 09. May 20:57, Barry Song wrote: > > Upon further examination, it's not a bug, but we can still utilize 'nofail'. > The current code is very hard to read about gfp and "nofail" :-) > > maybe: > > diff --git a/mm/vmalloc.c b/mm/vmalloc.c > index 6641be0ca80b..7c66fe16c2ad 100644 > --- a/mm/vmalloc.c > +++ b/mm/vmalloc.c > @@ -3498,7 +3498,7 @@ vm_area_alloc_pages(gfp_t gfp, int nid, > { > unsigned int nr_allocated = 0; > gfp_t alloc_gfp = gfp; > - bool nofail = false; > + bool nofail = !!(gfp & __GFP_NOFAIL); > struct page *page; > int i; > > @@ -3555,7 +3555,6 @@ vm_area_alloc_pages(gfp_t gfp, int nid, > * and compaction etc. > */ > alloc_gfp &= ~__GFP_NOFAIL; > - nofail = true; > } Thanks for suggestion. I think that makes more clearly. Will try it in next version. -- Best Regards, Hailong.
diff --git a/mm/vmalloc.c b/mm/vmalloc.c index 6641be0ca80b..2f359d08bf8d 100644 --- a/mm/vmalloc.c +++ b/mm/vmalloc.c @@ -3560,7 +3560,7 @@ vm_area_alloc_pages(gfp_t gfp, int nid, /* High-order pages or fallback path if "bulk" fails. */ while (nr_allocated < nr_pages) { - if (fatal_signal_pending(current)) + if (!(gfp & __GFP_NOFAIL) && fatal_signal_pending(current)) break; if (nid == NUMA_NO_NODE)