Re: [BUG REPORT] ktime_get_ts64 causes Hard Lockup

From: Jeff Merkey
Date: Thu Jan 21 2016 - 11:57:16 EST


One other item worth mentioning is the effect of this bug on the user
space daemons. Since this bug is in all kernels from 3.6-v4.4, folks
will see the following behavior of systemd when this bug fires off and
the value wraps:

id 0a0c809bcdc0812145d9c35456e46335650601b7
reason: systemd-journald killed by SIGABRT
time: Sun 10 Jan 2016 11:56:16 AM MST
cmdline: /usr/lib/systemd/systemd-journald
package: systemd-219-19.el7
uid: 0 (root)
count: 2
Directory: /var/tmp/abrt/ccpp-2016-01-10-11:56:16-6281

id 38f7bb372d0c14cb912e9d225fb474b931889e77
reason: __epoll_wait_nocancel(): systemd-logind killed by SIGABRT
time: Fri 08 Jan 2016 01:33:42 PM MST
cmdline: /usr/lib/systemd/systemd-logind
package: systemd-219-19.el7
uid: 0 (root)
count: 39
Directory: /var/tmp/abrt/ccpp-2016-01-08-13:33:42-630
Reported:
https://retrace.fedoraproject.org/faf/reports/bthash/307b26a77cc6d5005ce2fdf18ff010fe3dc94401

So if folks see aborts in systemd that look like the above, it's this
bug. I traced the systemd crash as well and its caused by the
system returning garbage via system calls to systemd when that rax
value wraps.

Jeff.



On 1/21/16, Jeff Merkey <linux.mdb@xxxxxxxxx> wrote:
> Hi Thomas
>
> That patch works and so does the patch currently in linux-next. Yeah !
>
> The code on your machine is apparently from the linux-next tree. In
> 4.4 is busted and that was the build I was using since that's the last
> build I have MDB released on. If you compare v4.4 vs. linux-next you
> will see someone already has patched this problem -- its just not in
> Linus tree yet.
>
> v4.4
>
> static inline s64 timekeeping_get_ns(struct tk_read_base *tkr)
> {
> cycle_t delta;
> s64 nsec;
>
> delta = timekeeping_get_delta(tkr);
>
> nsec = delta * tkr->mult + tkr->xtime_nsec;
> nsec >>= tkr->shift; << causes sar to be used
>
> /* If arch requires, add in get_arch_timeoffset() */
> return nsec + arch_gettimeoffset();
> }
>
> linux-next
>
> static inline s64 timekeeping_get_ns(struct tk_read_base *tkr)
> {
> cycle_t delta;
> u64 nsec;
>
> delta = timekeeping_get_delta(tkr);
>
> nsec = (delta * tkr->mult + tkr->xtime_nsec) >> tkr->shift; << shr used
> here
>
> /* If arch requires, add in get_arch_timeoffset() */
> return (s64)(nsec + arch_gettimeoffset());
> }
>
>
> This bug goes way back in the linux revs, so I have to retro apply
> this patch back to v3.6 in the MDB patch series. This bug was
> introduced in the late linux 3.6 kernels and exists in all of them up
> to v4.4.
>
> Jeff
>
>
> On 1/21/16, Thomas Gleixner <tglx@xxxxxxxxxxxxx> wrote:
>> Jeff,
>>
>> On Thu, 21 Jan 2016, Jeff Merkey wrote:
>>> static inline s64 timekeeping_get_ns(struct tk_read_base *tkr)
>>> {
>>> cycle_t delta;
>>> s64 nsec;
>>>
>>> delta = timekeeping_get_delta(tkr);
>>>
>>> nsec = delta * tkr->mult + tkr->xtime_nsec;
>>> nsec >>= tkr->shift; << wrap caused here
>>>
>>> /* If arch requires, add in get_arch_timeoffset() */
>>> return nsec + arch_gettimeoffset();
>>> }
>>>
>>> You only have 64 bits of register and the numbers being calculated
>>> here are big. By way of example, I observed the following during
>>> normal operations:
>>>
>>> delta (RAX) | tkr->mult (RDX)
>>>
>>> 0x157876 0x65ee27
>>> 0xf1855 0x65f158
>>> 0x16cf05 0x65f408
>>> 303bc3 0x65f154
>>>
>>> When this bug occurs different story.
>>>
>>> delta (RAX) | tkr->mult (RDX)
>>>
>>> 0x243283994b8 0x65233
>>>
>>> So it goes like this:
>>>
>>> nsec = delta * tkr->mult + tkr->xtime_nsec;
>>> 0x243283994b8 * 0x65233
>>> imul rax,rdx = 0xE6A2Ce1f1ea690a8
>>>
>>> nsec >>= tkr->shift; << wrap caused here
>>> sar rax,cl = 0xFFFFFFE6BFB3B7C3
>>
>> That SAR is siomply wrong here. It must be an SHR and it is at least when
>> I'm
>> looking at the assembly of my machine.
>>
>>> the sar instruction doesn't just shift, it backfills the signedness of
>>> the value, so this instruction is not doing what the C code is asking
>>> it to do. I am guessing that somewhere in this mass of macros,
>>> something may have gotten declared wrong or incomplete (declared
>>> signed ?).
>>
>> There is no macro involved.
>>
>> timekeeping_get_ns
>> {
>> nsec = (delta * tkr->mult + tkr->xtime_nsec) >> tkr->shift;
>> }
>>
>>> The assembler output for this section that calls the macro to
>>> calculate nsecs shows the sar instruction:
>>>
>>> delta = timekeeping_get_delta(tkr);
>>>
>>> nsec = delta * tkr->mult + tkr->xtime_nsec;
>>> 29b: 48 0f af c2 imul %rdx,%rax
>>> 29f: 48 03 05 00 00 00 00 add 0x0(%rip),%rax # 2a6
>>> <ktime_get_ts64+0xc6>
>>> nsec >>= tkr->shift;
>>> 2a6: 48 d3 f8 sar %cl,%rax
>>
>> And this is fundamentally wrong. Why is the compiler emitting SAR instead
>> of
>> SHR here? Here is the assembly output from my kernel:
>>
>> nsec = (delta * tkr->mult + tkr->xtime_nsec) >> tkr->shift;
>> 27e: 48 0f af c5 imul %rbp,%rax
>> 282: 48 01 d8 add %rbx,%rax
>> 285: 48 d3 e8 shr %cl,%rax
>>
>> } while (read_seqcount_retry(&tk_core.seq, seq));
>>
>>
>> So the first thing which needs to be figured out is WHY this results in a
>> SAR
>> on your compiler.
>>
>>> There is another problem with the tkr->read returning an unchanging,
>>> unclearable number when this bug occurs for the delta value. I
>>> appears for whatever reason the clock has gone to sleep or gone away
>>> and is no longer updating its counters.
>>>
>>> static inline cycle_t timekeeping_get_delta(struct tk_read_base *tkr)
>>> {
>>> cycle_t cycle_now, delta;
>>>
>>> /* read clocksource */
>>> cycle_now = tkr->read(tkr->clock); << returns the same value after
>>> this bug happens
>>>
>>> /* calculate the delta since the last update_wall_time */
>>> delta = clocksource_delta(cycle_now, tkr->cycle_last, tkr->mask); <<
>>> cycle last is also the same value.
>>>
>>> return delta;
>>> }
>>
>> If that value does not change, then the timekeeping update is not
>> running. That might happen because the timer interrupt is not happening
>> or
>> whatever got wreckaged.
>>
>>> I would check how these structs are defined and the vars in them to
>>> see if somewhere they are declared as signed values to the compiler,
>>> because that's what it thinks it was given to compile.
>>
>> Sure. Here you go:
>>
>> nsec = (delta * tkr->mult + tkr->xtime_nsec) >> tkr->shift;
>>
>> delta, mult, xtime_nsec and shift are unsigned. The only signed value is
>> nsec.
>>
>> Does that issue go away if you apply the patch below?
>>
>> Thanks,
>>
>> tglx
>>
>> 8<-----------
>> diff --git a/kernel/time/timekeeping.c b/kernel/time/timekeeping.c
>> index 34b4cedfa80d..d405bcdf9d40 100644
>> --- a/kernel/time/timekeeping.c
>> +++ b/kernel/time/timekeeping.c
>> @@ -301,7 +301,7 @@ static inline u32 arch_gettimeoffset(void) { return
>> 0;
>> }
>> static inline s64 timekeeping_get_ns(struct tk_read_base *tkr)
>> {
>> cycle_t delta;
>> - s64 nsec;
>> + u64 nsec;
>>
>> delta = timekeeping_get_delta(tkr);
>>
>>
>