lkml.org 
[lkml]   [2019]   [Jun]   [3]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH] perf: Fix oops when kthread execs user process
On Fri, May 31, 2019 at 04:37:03PM +0100, Will Deacon wrote:
> On Thu, May 30, 2019 at 03:57:36PM +0530, Ravi Bangoria wrote:
> > On 5/30/19 2:08 PM, Ravi Bangoria wrote:

> > >> --- a/kernel/events/core.c
> > >> +++ b/kernel/events/core.c
> > >> @@ -5923,7 +5923,7 @@ static void perf_sample_regs_user(struct
> > >> if (user_mode(regs)) {
> > >> regs_user->abi = perf_reg_abi(current);
> > >> regs_user->regs = regs;
> > >> - } else if (current->mm) {
> > >> + } else if (!(current->flags & PF_KTHREAD)) {
> >
> > With this patch applied and my patch reverted, I still see it crashing
> > because current->flags does not have PF_KTHREAD set. Sample trace with
> > v5.0 kernel:
> >
> >
> > BUG: Kernel NULL pointer dereference at 0x00000000
> > Faulting instruction address: 0xc0000000000f1a6c
> > Oops: Kernel access of bad area, sig: 11 [#1]
> > LE SMP NR_CPUS=2048 NUMA pSeries
> > CPU: 17 PID: 3241 Comm: systemd-cgroups Kdump: loaded Not tainted 5.0.0+ #7
> > NIP: c0000000000f1a6c LR: c0000000002acc7c CTR: c0000000002a8f90
> > REGS: c0000001e80469a0 TRAP: 0300 Not tainted (5.0.0+)
> > MSR: 8000000000001033 <SF,ME,IR,DR,RI,LE> CR: 48022448 XER: 20000000
> > CFAR: c00000000000deb4 DAR: 0000000000000000 DSISR: 40000000 IRQMASK: 1
> > GPR00: c0000000002acc7c c0000001e8046c30 c000000001607500 0000000000000000
> > GPR04: 0000000000000000 0000000000000000 0000000000000000 c000000000128618
> > GPR08: 000007ffffffffff 0000000000000000 ffffffffffffffff c00000000001cd40
> > GPR12: c000000000446fd8 c00000003ffdf080 00000000ffff0000 0000000000000007
> > GPR16: c0000001edd74988 c0000001edd60400 00007fff89801130 000000000005e1b0
> > GPR20: c0000001edb77a08 c0000001e8047208 c0000001f03d9800 c0000001e8046e00
> > GPR24: 000000000000b1af c000000001126938 c0000001f03d9b28 0000000000010000
> > GPR28: c0000001e8046d30 0000000000000000 0000000000000000 0000000000000000
> > NIP [c0000000000f1a6c] perf_reg_value+0x5c/0xc0
> > LR [c0000000002acc7c] perf_output_sample_regs+0x6c/0xd0
> > Call Trace:
> > [c0000001e8046c30] [c0000000002acc7c] perf_output_sample_regs+0x6c/0xd0 (unreliable)
> > [c0000001e8046c80] [c0000000002b9cd0] perf_output_sample+0x620/0x8c0
> > [c0000001e8046d10] [c0000000002ba6b4] perf_event_output_forward+0x64/0x90
> > [c0000001e8046d80] [c0000000002b2908] __perf_event_overflow+0x88/0x1e0
> > [c0000001e8046dd0] [c0000000000f3d18] record_and_restart+0x288/0x670
> > [c0000001e8047180] [c0000000000f4c18] perf_event_interrupt+0x2b8/0x4b0
> > [c0000001e8047280] [c00000000002b380] performance_monitor_exception+0x50/0x70
> > [c0000001e80472a0] [c000000000009ca0] performance_monitor_common+0x110/0x120
> > --- interrupt: f01 at slice_scan_available+0x20/0xc0
> > LR = slice_find_area+0x174/0x210
> > [c0000001e8047630] [c000000000083ea0] slice_get_unmapped_area+0x3d0/0x7f0
> > [c0000001e8047ae0] [c00000000032d5b0] get_unmapped_area+0xa0/0x170
> > [c0000001e8047b10] [c00000000001cd40] arch_setup_additional_pages+0xc0/0x1c0
> > [c0000001e8047b60] [c000000000446fd8] load_elf_binary+0xb48/0x1580
> > [c0000001e8047c80] [c0000000003c3938] search_binary_handler+0xe8/0x2a0
> > [c0000001e8047d10] [c0000000003c42f4] __do_execve_file.isra.13+0x694/0x980
> > [c0000001e8047de0] [c000000000128618] call_usermodehelper_exec_async+0x248/0x290
> > [c0000001e8047e20] [c00000000000b65c] ret_from_kernel_thread+0x5c/0x80
> > Instruction dump:
> > 7c9e2378 7c7f1b78 f8010010 f821ffd1 419e0044 3d22ff6b 7bc41764 3929ae10
> > 7d29202e 2b890150 419d003c 38210030 <7c7f482a> e8010010 ebc1fff0 ebe1fff8
> > ---[ end trace 54f3492ad1d403d8 ]---
>
> Oh, nice! I think this happens because Power doesn't actually initialise
> the regs after a kthread execs() until late in start_thread(). But the plot
> thickens somewhat, since current_pt_regs() is different to
> task_pt_regs(current) on Power (the former cannot return NULL).

So one possibility would be to have activate_mm() initialize the user
regs set. Doing it there ensure that the moment PF_KTHREAD gets cleared,
task_pt_regs() is valid.

\
 
 \ /
  Last update: 2019-06-03 13:49    [W:0.132 / U:0.304 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site