Juri reported this from the -rt kernel, but I can easily trigger it in
mainline. By simply doing:
# cd /sys/kernel/tracing
# echo 1 > options/userstacktrace
# echo 1 > events/irq/enable
With CONFIG_DEBUG_ATOMIC_SLEEP enabled, we get a WARNING splat of:
WARNING: CPU: 1 PID: 1688 at arch/x86/kernel/stacktrace.c:180 save_stack_trace_user+0x12b/0x140
Modules linked in: iptable_mangle xt_CHECKSUM tun
CPU: 1 PID: 1688 Comm: sshd Not tainted 5.1.0-rc3-test+ #14
Hardware name: Hewlett-Packard HP Compaq Pro 6300 SFF/339A, BIOS K01 v03.03 07/14/2016
RIP: 0010:save_stack_trace_user+0x12b/0x140
Code: 44 8d 58 01 44 89 1f 48 89 34 c2 8b 07 89 c2 8b 77 04 49 39 ca 74 ad 4c 89 d1 39 f0 0f 82 37 ff ff ff c3 89 c2 8b 77 04 eb 9a <0f> 0b e9 3b ff ff ff 31 f6 e9 70 ff ff ff 8b 17 8b 77 04 eb 85 0f
RSP: 0000:ffff8880d0687e40 EFLAGS: 00010006
RAX: 0000000000000001 RBX: 1ffff1101a0d0fc9 RCX: 00007fff28a44f12
RDX: 0000000000010000 RSI: 00007faa5f81f098 RDI: ffff8880d0687e68
RBP: ffff8880cc058368 R08: ffff888085ae7f58 R09: 0000000000000000
R10: ffffed100f207c97 R11: ffff88807903e4bb R12: ffff88807903e468
R13: ffff88807903e46c R14: 0000000000000698 R15: ffff88807903e46c
FS: 00007faa5f43cdc0(0000) GS:ffff8880d0680000(0000) knlGS:0000000000000000
CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 0000000000000081 CR3: 0000000085296004 CR4: 00000000001606e0
Call Trace:
<IRQ>
ftrace_trace_userstack+0x17e/0x1f0
? __trace_stack+0x80/0x80
trace_event_buffer_commit+0xd0/0x300
? trace_event_buffer_reserve+0x107/0x130
trace_event_raw_event_x86_irq_vector+0xb8/0x120
? __bpf_trace_vector_setup+0x10/0x10
? irqtime_account_irq+0x46/0xe0
? __bpf_trace_vector_setup+0x10/0x10
smp_apic_timer_interrupt+0x208/0x270
apic_timer_interrupt+0xf/0x20
</IRQ>
RIP: 0033:0x7faa5f81f098
Code: 89 d1 40 0f b6 c6 48 89 fa f3 aa 48 89 d0 c3 0f 1f 00 f3 0f 1e fa 48 39 d1 0f 82 43 80 06 00 0f 1f 00 f3 0f 1e fa 66 0f 6e c6 <48> 89 f8 66 0f 60 c0 66 0f 61 c0 66 0f 70 c0 00 48 83 fa 10 0f 82
RSP: 002b:00007fff28a44ef8 EFLAGS: 00000246 ORIG_RAX: ffffffffffffff13
RAX: 0000000000000000 RBX: 00005649f9d12830 RCX: 0000000000000001
RDX: 0000000000000100 RSI: 0000000000000000 RDI: 00005649f9d36240
RBP: 00007fff28a44f12 R08: 0000000000000000 R09: 0000000000000100
R10: 0000000000000008 R11: 00007fff28a40f90 R12: 0000000000000000
R13: 00007fff28a44f58 R14: 00005649f9d12260 R15: 00005649f9d0ab40
---[ end trace cd684bbd8c344b2a ]---
hrtimer: interrupt took 3268750 ns
------------[ cut here ]------------
This is simply caused by the irq trace events doing a user stack trace:
ftrace_trace_userstack {
save_stack_trace_user {
__save_stack_trace_user {
copy_stack_frame {
access_ok {
WARN_ON_IN_IRQ()
BOOM! Warn on.
Can we make that access_ok() call in the copy_stack_frame not trigger
the warning just if we are in an interrupt?
-- Steve
On Wed, 3 Apr 2019, Steven Rostedt wrote:
> Juri reported this from the -rt kernel, but I can easily trigger it in
> mainline. By simply doing:
>
> # cd /sys/kernel/tracing
> # echo 1 > options/userstacktrace
> # echo 1 > events/irq/enable
> This is simply caused by the irq trace events doing a user stack trace:
>
> ftrace_trace_userstack {
> save_stack_trace_user {
> __save_stack_trace_user {
> copy_stack_frame {
> access_ok {
> WARN_ON_IN_IRQ()
>
> BOOM! Warn on.
>
> Can we make that access_ok() call in the copy_stack_frame not trigger
> the warning just if we are in an interrupt?
You really want to have access_ok_atomic() or such which does not have the
WARN and use that in copy_stack_frame(). That's fine here because the
actual copy is inside a pagefault disabled region.
Thanks,
tglx
On Fri, 5 Apr 2019 10:12:27 +0200 (CEST)
Thomas Gleixner <[email protected]> wrote:
> > BOOM! Warn on.
> >
> > Can we make that access_ok() call in the copy_stack_frame not trigger
> > the warning just if we are in an interrupt?
>
> You really want to have access_ok_atomic() or such which does not have the
> WARN and use that in copy_stack_frame(). That's fine here because the
> actual copy is inside a pagefault disabled region.
I was thinking the same.
Masami, did you post patches to do something like this?
"access_ok_inatomic()" or something?
-- Steve
On Fri, 5 Apr 2019 09:32:09 -0400
Steven Rostedt <[email protected]> wrote:
> On Fri, 5 Apr 2019 10:12:27 +0200 (CEST)
> Thomas Gleixner <[email protected]> wrote:
>
> > > BOOM! Warn on.
> > >
> > > Can we make that access_ok() call in the copy_stack_frame not trigger
> > > the warning just if we are in an interrupt?
> >
> > You really want to have access_ok_atomic() or such which does not have the
> > WARN and use that in copy_stack_frame(). That's fine here because the
> > actual copy is inside a pagefault disabled region.
>
> I was thinking the same.
>
> Masami, did you post patches to do something like this?
> "access_ok_inatomic()" or something?
Yeah, last month I sent
"x86/uaccess: Allow access_ok() in irq context if pagefault_disabled"
If you correctly disables the pagefault, access_ok() shouldn't warn it.
Ah, I see.
copy_stack_frame(const void __user *fp, struct stack_frame_user *frame)
{
int ret;
if (!access_ok(fp, sizeof(*frame))) <== this is out of pagefault_disable()!
return 0;
ret = 1;
pagefault_disable();
if (__copy_from_user_inatomic(frame, fp, sizeof(*frame)))
ret = 0;
pagefault_enable();
return ret;
}
How is below patch?
---
diff --git a/arch/x86/kernel/stacktrace.c b/arch/x86/kernel/stacktrace.c
index 2abf27d7df6b..36ff77c801f7 100644
--- a/arch/x86/kernel/stacktrace.c
+++ b/arch/x86/kernel/stacktrace.c
@@ -98,14 +98,11 @@ struct stack_frame_user {
static int
copy_stack_frame(const void __user *fp, struct stack_frame_user *frame)
{
- int ret;
+ int ret = 1;
- if (!access_ok(fp, sizeof(*frame)))
- return 0;
-
- ret = 1;
pagefault_disable();
- if (__copy_from_user_inatomic(frame, fp, sizeof(*frame)))
+ if (!access_ok(fp, sizeof(*frame)) ||
+ __copy_from_user_inatomic(frame, fp, sizeof(*frame)))
ret = 0;
pagefault_enable();
--
Masami Hiramatsu <[email protected]>
Hello
I also hit the same WARNING previously repored by Juri.
Hiramatsu san's patch looks good to me but I found that perf and
oprofile code do the similar thing by just directly calling
__range_not_ok().
perf: perf_callchain_user()@arch/x86/events/core.c
oprofile: dump_user_backtrace()@arch/x86/oprofile/backtrace.c
So for simplicity, I wrote a patch to fix the warning as other
codes do.
Ideally, we should merge these similar stacktrace codes(perf, ftrace,
oprofile) into one, but this time I made the minimum fix.
Eiichi Tsukata (1):
x86/stacktrace: Fix userstacktrace access_ok() WARNING in irq events
arch/x86/kernel/stacktrace.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
--
2.21.0