On Fri, 6 Mar 2020 15:16:22 +0900 Jaewon Kim <jaewon31.kim@xxxxxxxxxxx> wrote:
>
> Even on 64 bit kernel, the mmap failure can happen for a 32 bit task.
> Virtual memory space shortage of a task on mmap is reported to userspace
> as -ENOMEM. It can be confused as physical memory shortage of overall
> system.
>
> The vm_unmapped_area can be called to by some drivers or other kernel
> core system like filesystem. It can be hard to know which code layer
> returns the -ENOMEM.
>
> Print error log of vm_unmapped_area with rate limited. Without rate
> limited, soft lockup ocurrs on infinite mmap sytem call.
>
> i.e.)
> <4>[ 68.556470] [2: mmap_infinite:12363] mmap: vm_unmapped_area err:-12 total_vm:0xf4c08 flags:0x1 len:0x100000 low:0x8000 high:0xf4583000 mask:0x0 offset:0x0
>
> ...
>
> --- a/include/linux/mm.h
> +++ b/include/linux/mm.h
This patch was messed up by your email client (tabs expanded to spaces).
> @@ -27,6 +27,7 @@
> #include <linux/memremap.h>
> #include <linux/overflow.h>
> #include <linux/sizes.h>
> +#include <linux/ratelimit.h>
>
> struct mempolicy;
> struct anon_vma;
> @@ -2379,10 +2380,20 @@ extern unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info);
> static inline unsigned long
> vm_unmapped_area(struct vm_unmapped_area_info *info)
> {
> + unsigned long addr;
> +
> if (info->flags & VM_UNMAPPED_AREA_TOPDOWN)
> - return unmapped_area_topdown(info);
> + addr = unmapped_area_topdown(info);
> else
> - return unmapped_area(info);
> + addr = unmapped_area(info);
> +
> + if (IS_ERR_VALUE(addr)) {
> + pr_warn_ratelimited("%s err:%ld total_vm:0x%lx flags:0x%lx len:0x%lx low:0x%lx high:0x%lx mask:0x%lx offset:0x%lx\n",
> + __func__, addr, current->mm->total_vm, info->flags,
> + info->length, info->low_limit, info->high_limit,
> + info->align_mask, info->align_offset);
> + }
> + return addr;
> }
pr_warn_ratelimited() contains static state. Using it in an inlined
function means that each callsite gets its own copy of that state, so
we're ratelimiting the vm_unmapped_area() output on a per-callsite
basis, not on a kernelwide basis.
Maybe that's what we want, maybe it's not. But I think
vm_unmapped_area() has become too large to be inlined anyway, so I
suggest making it a regular out-of-line function in mmap.c. I don't
believe that function needs to be exported to modules.
