Re: [PATCH] printk: use kvmalloc instead of kmalloc for devkmsg_user

From: Sergey Senozhatsky
Date: Tue Aug 31 2021 - 19:02:17 EST


On (21/08/31 14:42), Petr Mladek wrote:
> On Mon 2021-08-30 16:17:01, 이용택 wrote:
> > Size of struct devkmsg_user increased to 16784 by commit 896fbe20b4e2
> > ("printk: use the lockless ringbuffer") so order3(32kb) is needed for
> > kmalloc. Under stress conditions the kernel may temporary fail to
> > allocate 32k with kmalloc. Use kvmalloc instead of kmalloc to aviod
> > this issue.
> >
> > qseecomd invoked oom-killer: gfp_mask=0x40cc0(GFP_KERNEL|__GFP_COMP), order=3, oom_score_adj=-1000
> > Call trace:
> > dump_backtrace+0x0/0x34c
> > dump_stack_lvl+0xd4/0x16c
> > dump_header+0x5c/0x338
> > out_of_memory+0x374/0x4cc
> > __alloc_pages_slowpath+0xbc8/0x1130
> > __alloc_pages_nodemask+0x170/0x1b0
> > kmalloc_order+0x5c/0x24c
> > devkmsg_open+0x1f4/0x558
> > memory_open+0x94/0xf0
> > chrdev_open+0x288/0x3dc
> > do_dentry_open+0x2b4/0x618
> > path_openat+0xce4/0xfa8
> > do_filp_open+0xb0/0x164
> > do_sys_openat2+0xa8/0x264
> > __arm64_sys_openat+0x70/0xa0
> > el0_svc_common+0xc4/0x270
> > el0_svc+0x34/0x9c
> > el0_sync_handler+0x88/0xf0
> > el0_sync+0x1bc/0x200
> >
> > DMA32: 4521*4kB (UMEC) 1377*8kB (UMECH) 73*16kB (UM) 0*32kB 0*64kB 0*128kB 0*256kB 0*512kB 0*1024kB 0*2048kB 0*4096kB = 30268kB
> > Normal: 2490*4kB (UMEH) 277*8kB (UMH) 27*16kB (UH) 1*32kB (H) 0*64kB 0*128kB 0*256kB 0*512kB 0*1024kB 0*2048kB 0*4096kB = 12640kB
> >
> > Signed-off-by: Yong-Taek Lee <ytk.lee@xxxxxxxxxxx>
>
> Reviewed-by: Petr Mladek <pmladek@xxxxxxxx>

Acked-by: Sergey Senozhatsky <senozhatsky@xxxxxxxxxxxx>