[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20200812080650.GA3894595@elver.google.com>
Date: Wed, 12 Aug 2020 10:06:50 +0200
From: Marco Elver <elver@...gle.com>
To: peterz@...radead.org
Cc: Jürgen Groß <jgross@...e.com>,
Borislav Petkov <bp@...en8.de>,
Dave Hansen <dave.hansen@...ux.intel.com>,
fenghua.yu@...el.com, "H. Peter Anvin" <hpa@...or.com>,
LKML <linux-kernel@...r.kernel.org>,
Ingo Molnar <mingo@...hat.com>,
syzkaller-bugs <syzkaller-bugs@...glegroups.com>,
Thomas Gleixner <tglx@...utronix.de>,
"Luck, Tony" <tony.luck@...el.com>,
the arch/x86 maintainers <x86@...nel.org>,
yu-cheng.yu@...el.com, sdeep@...are.com,
virtualization@...ts.linux-foundation.org,
kasan-dev <kasan-dev@...glegroups.com>,
syzbot <syzbot+8db9e1ecde74e590a657@...kaller.appspotmail.com>,
"Paul E. McKenney" <paulmck@...nel.org>,
Wei Liu <wei.liu@...nel.org>,
Steven Rostedt <rostedt@...dmis.org>
Subject: Re: [PATCH] x86/paravirt: Add missing noinstr to arch_local*()
helpers
On Tue, Aug 11, 2020 at 10:17PM +0200, peterz@...radead.org wrote:
> On Tue, Aug 11, 2020 at 11:46:51AM +0200, peterz@...radead.org wrote:
>
> > So let me once again see if I can't find a better solution for this all.
> > Clearly it needs one :/
>
> So the below boots without triggering the debug code from Marco -- it
> should allow nesting local_irq_save/restore under raw_local_irq_*().
>
> I tried unconditional counting, but there's some _reallly_ wonky /
> asymmetric code that wrecks that and I've not been able to come up with
> anything useful.
>
> This one starts counting when local_irq_save() finds it didn't disable
> IRQs while lockdep though it did. At that point, local_irq_restore()
> will decrement and enable things again when it reaches 0.
>
> This assumes local_irq_save()/local_irq_restore() are nested sane, which
> is mostly true.
>
> This leaves #PF, which I fixed in these other patches, but I realized it
> needs fixing for all architectures :-( No bright ideas there yet.
>
> ---
> arch/x86/entry/thunk_32.S | 5 ----
> include/linux/irqflags.h | 45 +++++++++++++++++++-------------
> init/main.c | 16 ++++++++++++
> kernel/locking/lockdep.c | 58 +++++++++++++++++++++++++++++++++++++++++
> kernel/trace/trace_preemptirq.c | 33 +++++++++++++++++++++++
> 5 files changed, 134 insertions(+), 23 deletions(-)
Testing this again with syzkaller produced some new reports:
BUG: stack guard page was hit in error_entry
BUG: stack guard page was hit in exc_int3
PANIC: double fault in error_entry
PANIC: double fault in exc_int3
Most of them have corrupted reports, but this one might be useful:
BUG: stack guard page was hit at 000000001fab0982 (stack is 00000000063f33dc..00000000bf04b0d8)
BUG: stack guard page was hit at 00000000ca97ac69 (stack is 00000000af3e6c84..000000001597e1bf)
kernel stack overflow (double-fault): 0000 [#1] PREEMPT SMP
CPU: 1 PID: 4709 Comm: kworker/1:1H Not tainted 5.8.0+ #5
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.13.0-1 04/01/2014
Workqueue: events_highpri snd_vmidi_output_work
RIP: 0010:exc_int3+0x5/0xf0 arch/x86/kernel/traps.c:636
Code: c9 85 4d 89 e8 31 c0 e8 a9 7d 68 fd e9 90 fe ff ff e8 0f 35 00 00 66 66 2e 0f 1f 84 00 00 00 00 00 0f 1f 40 00 55 53 48 89 fb <e8> 76 0e 00 00 85 c0 74 03 5b 5d c3 f6 83 88 00 00 00 03 74 7e 48
RSP: 0018:ffffc90008114000 EFLAGS: 00010083
RAX: 0000000084e00e17 RBX: ffffc90008114018 RCX: ffffffff84e00e17
RDX: 0000000000000000 RSI: ffffffff84e00a39 RDI: ffffc90008114018
RBP: 0000000000000000 R08: 0000000000000000 R09: 0000000000000000
R10: 0000000000000000 R11: 0000000000000000 R12: 0000000000000000
R13: 0000000000000000 R14: 0000000000000000 R15: 0000000000000000
FS: 0000000000000000(0000) GS:ffff88807dc80000(0000) knlGS:0000000000000000
CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: ffffc90008113ff8 CR3: 000000002dae4006 CR4: 0000000000770ee0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
PKRU: 00000000
Call Trace:
asm_exc_int3+0x31/0x40 arch/x86/include/asm/idtentry.h:537
RIP: 0010:arch_static_branch include/trace/events/preemptirq.h:40 [inline]
RIP: 0010:static_key_false include/linux/jump_label.h:200 [inline]
RIP: 0010:trace_irq_enable_rcuidle+0xd/0x120 include/trace/events/preemptirq.h:40
Code: 24 08 48 89 df e8 43 8d ef ff 48 89 df 5b e9 4a 2e 99 03 66 2e 0f 1f 84 00 00 00 00 00 55 41 56 53 48 89 fb e8 84 1a fd ff cc <1f> 44 00 00 5b 41 5e 5d c3 65 8b 05 ab 74 c3 7e 89 c0 31 f6 48 0f
RSP: 0018:ffffc900081140f8 EFLAGS: 00000093
RAX: ffffffff813d9e8c RBX: ffffffff81314dd3 RCX: ffff888076ce6000
RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffffffff81314dd3
RBP: 0000000000000000 R08: ffffffff813da3d4 R09: 0000000000000001
R10: 0000000000000000 R11: 0000000000000000 R12: 0000000000000000
R13: 0000000000000082 R14: 0000000000000000 R15: ffff888076ce6000
trace_hardirqs_restore+0x59/0x80 kernel/trace/trace_preemptirq.c:106
rcu_irq_enter_irqson+0x43/0x70 kernel/rcu/tree.c:1074
trace_irq_enable_rcuidle+0x87/0x120 include/trace/events/preemptirq.h:40
trace_hardirqs_restore+0x59/0x80 kernel/trace/trace_preemptirq.c:106
rcu_irq_enter_irqson+0x43/0x70 kernel/rcu/tree.c:1074
trace_irq_enable_rcuidle+0x87/0x120 include/trace/events/preemptirq.h:40
trace_hardirqs_restore+0x59/0x80 kernel/trace/trace_preemptirq.c:106
rcu_irq_enter_irqson+0x43/0x70 kernel/rcu/tree.c:1074
trace_irq_enable_rcuidle+0x87/0x120 include/trace/events/preemptirq.h:40
trace_hardirqs_restore+0x59/0x80 kernel/trace/trace_preemptirq.c:106
rcu_irq_enter_irqson+0x43/0x70 kernel/rcu/tree.c:1074
<... repeated many many times ...>
trace_irq_enable_rcuidle+0x87/0x120 include/trace/events/preemptirq.h:40
trace_hardirqs_restore+0x59/0x80 kernel/trace/trace_preemptirq.c:106
rcu_irq_enter_irqson+0x43/0x70 kernel/rcu/tree.c:1074
Lost 500 message(s)!
BUG: stack guard page was hit at 00000000cab483ba (stack is 00000000b1442365..00000000c26f9ad3)
BUG: stack guard page was hit at 00000000318ff8d8 (stack is 00000000fd87d656..0000000058100136)
---[ end trace 4157e0bb4a65941a ]---
.config is attached. Unfortunately no C-reproducer yet.
Thanks,
-- Marco
View attachment ".config" of type "text/plain" (209317 bytes)
Powered by blists - more mailing lists