Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Wed, 5 Jul 2023 22:50:08 +0100
From:   "Richard W.M. Jones" <rjones@redhat.com>
To:     Peter Zijlstra <peterz@infradead.org>,
        Richard Henderson <richard.henderson@linaro.org>
Cc:     Arnd Bergmann <arnd@arndb.de>,
        Naresh Kamboju <naresh.kamboju@linaro.org>,
        Anders Roxell <anders.roxell@linaro.org>,
        Daniel =?iso-8859-1?Q?D=EDaz?= <daniel.diaz@linaro.org>,
        Benjamin Copeland <ben.copeland@linaro.org>,
        linux-kernel@vger.kernel.org, x86@kernel.org,
        Paolo Bonzini <pbonzini@redhat.com>
Subject: Re: qemu-x86_64 booting with 8.0.0 stil see int3: when running LTP
 tracing testing.
Message-ID: <20230705215008.GD17440@redhat.com>
References: <CA+G9fYsETJQm0Ue7hGsb+nbsiMikwycOV3V0DPr6WC2r61KRBQ@mail.gmail.com>
 <2d7595b1-b655-4425-85d3-423801bce644@app.fastmail.com>
 <20230621160655.GL2053369@hirez.programming.kicks-ass.net>
 <20230704074620.GA17440@redhat.com>
 <20230705162830.GC17440@redhat.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20230705162830.GC17440@redhat.com>
User-Agent: Mutt/1.5.21 (2010-09-15)
Precedence: bulk

On Wed, Jul 05, 2023 at 05:28:30PM +0100, Richard W.M. Jones wrote:
> On Tue, Jul 04, 2023 at 08:46:20AM +0100, Richard W.M. Jones wrote:
> > We have been having the same sort of problem
> > (https://bugzilla.redhat.com/show_bug.cgi?id=2216496).  It's another
> > of those bugs that requires hundreds or thousands of boot iterations
> > before you can see it.  There is a test in comment 27 but it requires
> > guestfish and some hacking to work.  I'll try to come up with a
> > cleaner test later.
> > 
> > We see stack traces like:
> > 
> > [    3.081939] clocksource: acpi_pm: mask: 0xffffff max_cycles: 0xffffff, max_idle_ns: 2085701024 ns
> > [    3.082266] clocksource: Switched to clocksource acpi_pm
> > [    3.090201] NET: Registered PF_INET protocol family
> > [    3.093098] int3: 0000 [#1] PREEMPT SMP NOPTI
> > [    3.093098] CPU: 3 PID: 1 Comm: swapper/0 Not tainted 6.4.0-10173-ga901a3568fd2 #8
> > [    3.093098] Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS rel-1.16.2-0-gea1b7a073390-prebuilt.qemu.org 04/01/2014
> > [    3.093098] RIP: 0010:__mod_timer+0x1c3/0x370
> > [    3.093098] Code: 00 00 41 bd ff ff ff ff 31 d2 4c 89 f6 4c 89 ff e8 f2 ef ff ff 41 89 c4 85 c0 75 09 83 e3 01 0f 85 54 ff ff ff 41 8b 4f 20 66 <90> f7 c1 00 00 10 00 0f 84 23 01 00 00 48 c7 c3 40 cc 01 00 65 48
> > [    3.093098] RSP: 0018:ffffaf1600013e00 EFLAGS: 00000046
> > [    3.093098] RAX: 0000000000000000 RBX: 0000000000000000 RCX: 0000000000280003
> > [    3.093098] RDX: 0000000000000000 RSI: ffff9aa90fd9dec0 RDI: ffffffff8441e4b8
> > [    3.093098] RBP: 00000000fffc200d R08: ffffffff8441e4a0 R09: ffffffff8441e4b8
> > [    3.093098] R10: 0000000000000001 R11: 000000000002e990 R12: 0000000000000000
> > [    3.093098] R13: 00000000ffffffff R14: ffff9aa90fd9dec0 R15: ffffffff8441e4b8
> > [    3.093098] FS:  0000000000000000(0000) GS:ffff9aa90fd80000(0000) knlGS:0000000000000000
> > [    3.093098] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> > [    3.093098] CR2: 0000000000000000 CR3: 000000004e02e000 CR4: 0000000000750ee0
> > [    3.093098] PKRU: 55555554
> > [    3.093098] Call Trace:
> > [    3.093098]  <TASK>
> > [    3.093098]  ? die+0x31/0x80
> > [    3.093098]  ? exc_int3+0x10e/0x120
> > [    3.093098]  ? asm_exc_int3+0x39/0x40
> > [    3.093098]  ? __mod_timer+0x1c3/0x370
> > [    3.093098]  ? __mod_timer+0x1c3/0x370
> > [    3.093098]  queue_delayed_work_on+0x23/0x30
> > [    3.093098]  neigh_table_init+0x1bb/0x2e0
> > [    3.093098]  arp_init+0x12/0x50
> > [    3.093098]  inet_init+0x15b/0x2f0
> > [    3.093098]  ? __pfx_inet_init+0x10/0x10
> > [    3.093098]  do_one_initcall+0x58/0x230
> > [    3.093098]  kernel_init_freeable+0x199/0x2d0
> > [    3.093098]  ? __pfx_kernel_init+0x10/0x10
> > [    3.093098]  kernel_init+0x15/0x1b0
> > [    3.093098]  ret_from_fork+0x2c/0x50
> > [    3.093098]  </TASK>
> > [    3.093098] Modules linked in:
> > [    3.093098] ---[ end trace 0000000000000000 ]---
> > [    3.093098] RIP: 0010:__mod_timer+0x1c3/0x370
> > [    3.093098] Code: 00 00 41 bd ff ff ff ff 31 d2 4c 89 f6 4c 89 ff e8 f2 ef ff ff 41 89 c4 85 c0 75 09 83 e3 01 0f 85 54 ff ff ff 41 8b 4f 20 66 <90> f7 c1 00 00 10 00 0f 84 23 01 00 00 48 c7 c3 40 cc 01 00 65 48
> > [    3.093098] RSP: 0018:ffffaf1600013e00 EFLAGS: 00000046
> > [    3.093098] RAX: 0000000000000000 RBX: 0000000000000000 RCX: 0000000000280003
> > [    3.093098] RDX: 0000000000000000 RSI: ffff9aa90fd9dec0 RDI: ffffffff8441e4b8
> > [    3.093098] RBP: 00000000fffc200d R08: ffffffff8441e4a0 R09: ffffffff8441e4b8
> > [    3.093098] R10: 0000000000000001 R11: 000000000002e990 R12: 0000000000000000
> > [    3.093098] R13: 00000000ffffffff R14: ffff9aa90fd9dec0 R15: ffffffff8441e4b8
> > [    3.093098] FS:  0000000000000000(0000) GS:ffff9aa90fd80000(0000) knlGS:0000000000000000
> > [    3.093098] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> > [    3.093098] CR2: 0000000000000000 CR3: 000000004e02e000 CR4: 0000000000750ee0
> > [    3.093098] PKRU: 55555554
> > [    3.093098] Kernel panic - not syncing: Fatal exception in interrupt
> > 
> > There are many variations, but the common pattern seems to be
> > <something in the clock or timer code> -> int3 exception
> > 
> > It only happens under qemu TCG (software emulation).
> > 
> > It goes away if we recompile qemu without MTTCG support.
> > 
> > It only happens with -smp enabled, we are using qemu -smp 4
> > 
> > We are using qemu-system-x86_64 full system emulation on x86_64 host
> > (ie. forcing KVM off).
> > 
> > It happens with the latest upstream kernel and qemu, compiled from
> > source.
> 
> I got a bit further on this one.
> 
> The bug happens in the code that updates the static branch used for at
> least these two keys:
> 
>   static DEFINE_STATIC_KEY_FALSE(__sched_clock_stable);
>   DEFINE_STATIC_KEY_FALSE(timers_migration_enabled);
> 
> There are probably others (it seems a generic problem with how static
> branches are handled by TCG), but I only see the bug for those two.
> 
> When the static branch is updated we end up calling
> arch/x86/kernel/alternative.c:text_poke_bp_batch().  It's best to read
> the description of that function to see where int3 is used:
> 
>   https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/arch/x86/kernel/alternative.c#n2086
> 
> I modified the qemu TCG int3 helper so it dumps the code at %rip when
> the interrupt fires, and I can actually see the changes in the above
> function happen, first int3 being written, then the end of the nop,
> then the int3 being overwritten with the first byte of the nop.
> 
> Unfortunately the int3 still fires after the code has been completely
> rewritten to its final (ie nop) value.
> 
> This seems to indicate to me that neither the self-write to the kernel
> text segment, nor sync_core (implemented by a "iret to self" trick)
> invalidates the qemu TCG translation block containing the old int3
> helper call.  Thus we (qemu) never "see" the new nop, we keep
> emulating int3, and that causes the kernel to crash.

The following paragraph is wrong:

> I added print statements inside tb_invalidate_phys_page() and this
> function seems never to be called at all.  It's my understanding that
> at least the kernel writing to its text segment ought to cause
> tb_invalidate_phys_page() to be called, but I'm not super-familiar
> with this qemu code.

tb_invalidate_phys_range_fast() *is* called, and we end up calling
  tb_invalidate_phys_page_range__locked ->
    tb_phys_invalidate__locked ->
      do_tb_phys_invalidate

Nevertheless the old TB (containing the call to the int3 helper) is
still called after the code has been replaced with a NOP.

Of course there are 4 MTTCG threads so maybe another thread is in the
middle of executing the same TB when it gets invalidated.
tb_invalidate_phys_page_range__locked goes to some effort to check if
the current TB is being invalidated and restart the TB, but as far as
I can see the test can only work for the current core, and won't
restart the TB on other cores.

Is there any way to get the current TranslationBlock* from a helper?
It would be useful for additional debugging, I couldn't see how to do it.

The next thing I'll look at is the kernel sync_core() function and
whether TCG does the right thing, whatever that would be.

Rich.

-- 
Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones
Read my programming and virtualization blog: http://rwmj.wordpress.com
nbdkit - Flexible, fast NBD server with plugins
https://gitlab.com/nbdkit/nbdkit