On Wed, Apr 27, 2016 at 1:50 AM, Hannes Reinecke wrote:
>> that, I'd be happy to apply it to wq/for-3.7.
>> ^^^
> Ah. Time warp.
> I knew it would happen eventually :-)
lol wrong tree too. Sorry about that.
--
tejun
On Wed, Apr 27, 2016 at 1:50 AM, Hannes Reinecke wrote:
>> that, I'd be happy to apply it to wq/for-3.7.
>> ^^^
> Ah. Time warp.
> I knew it would happen eventually :-)
lol wrong tree too. Sorry about that.
--
tejun
On 04/26/2016 07:45 PM, Tejun Heo wrote:
> Hello, Peter.
>
> On Tue, Apr 26, 2016 at 10:27:59AM -0700, Peter Hurley wrote:
>>> It's unlikely to make any measureable difference. Is xchg() actually
>>> cheaper than store + rmb?
>>
>> store + mfence (full barrier), yes. Roughly 2x faster.
>>
>>
On 04/26/2016 07:45 PM, Tejun Heo wrote:
> Hello, Peter.
>
> On Tue, Apr 26, 2016 at 10:27:59AM -0700, Peter Hurley wrote:
>>> It's unlikely to make any measureable difference. Is xchg() actually
>>> cheaper than store + rmb?
>>
>> store + mfence (full barrier), yes. Roughly 2x faster.
>>
>>
On 04/26/2016 10:45 AM, Tejun Heo wrote:
> As long as what's happening is clearly documented, I think either is
> fine. I'm gonna go with Roman's mb patch for -stable fix but think
> it'd be nice to have a separate patch to consolidate the paths which
> clear PENDING and make them use xchg. If
On 04/26/2016 10:45 AM, Tejun Heo wrote:
> As long as what's happening is clearly documented, I think either is
> fine. I'm gonna go with Roman's mb patch for -stable fix but think
> it'd be nice to have a separate patch to consolidate the paths which
> clear PENDING and make them use xchg. If
Hello, Peter.
On Tue, Apr 26, 2016 at 10:27:59AM -0700, Peter Hurley wrote:
> > It's unlikely to make any measureable difference. Is xchg() actually
> > cheaper than store + rmb?
>
> store + mfence (full barrier), yes. Roughly 2x faster.
>
> https://lkml.org/lkml/2015/11/2/607
Ah, didn't know
Hello, Peter.
On Tue, Apr 26, 2016 at 10:27:59AM -0700, Peter Hurley wrote:
> > It's unlikely to make any measureable difference. Is xchg() actually
> > cheaper than store + rmb?
>
> store + mfence (full barrier), yes. Roughly 2x faster.
>
> https://lkml.org/lkml/2015/11/2/607
Ah, didn't know
Hi Tejun,
On 04/26/2016 08:15 AM, Tejun Heo wrote:
> Hello, Peter.
>
> On Mon, Apr 25, 2016 at 06:22:01PM -0700, Peter Hurley wrote:
>> This is the same bug I wrote about 2 yrs ago (but with the wrong fix).
>>
>> http://lkml.iu.edu/hypermail/linux/kernel/1402.2/04697.html
>>
>> Unfortunately I
Hi Tejun,
On 04/26/2016 08:15 AM, Tejun Heo wrote:
> Hello, Peter.
>
> On Mon, Apr 25, 2016 at 06:22:01PM -0700, Peter Hurley wrote:
>> This is the same bug I wrote about 2 yrs ago (but with the wrong fix).
>>
>> http://lkml.iu.edu/hypermail/linux/kernel/1402.2/04697.html
>>
>> Unfortunately I
Hello, Peter.
On Mon, Apr 25, 2016 at 06:22:01PM -0700, Peter Hurley wrote:
> This is the same bug I wrote about 2 yrs ago (but with the wrong fix).
>
> http://lkml.iu.edu/hypermail/linux/kernel/1402.2/04697.html
>
> Unfortunately I didn't have a reproducer at all :/
Ah, bummer.
> The
Hello, Peter.
On Mon, Apr 25, 2016 at 06:22:01PM -0700, Peter Hurley wrote:
> This is the same bug I wrote about 2 yrs ago (but with the wrong fix).
>
> http://lkml.iu.edu/hypermail/linux/kernel/1402.2/04697.html
>
> Unfortunately I didn't have a reproducer at all :/
Ah, bummer.
> The
On 04/25/2016 08:48 AM, Tejun Heo wrote:
> Hello, Roman.
>
> On Mon, Apr 25, 2016 at 05:22:51PM +0200, Roman Pen wrote:
> ...
>> CPU#6CPU#2
>> reqeust 884000343600 inserted
>> hctx marked as pended
>> kblockd_schedule...() returns 1
>>
>> ***
On 04/25/2016 08:48 AM, Tejun Heo wrote:
> Hello, Roman.
>
> On Mon, Apr 25, 2016 at 05:22:51PM +0200, Roman Pen wrote:
> ...
>> CPU#6CPU#2
>> reqeust 884000343600 inserted
>> hctx marked as pended
>> kblockd_schedule...() returns 1
>>
>> ***
Hello, Roman.
On Mon, Apr 25, 2016 at 07:39:52PM +0200, Roman Penyaev wrote:
> Ok, that's clear now. Thanks. I was confused also by a spin lock, which
> is being released just after clear pending:
>
>set_work_pool_and_clear_pending(work, pool->id);
>spin_unlock_irq(>lock);
>...
>
Hello, Roman.
On Mon, Apr 25, 2016 at 07:39:52PM +0200, Roman Penyaev wrote:
> Ok, that's clear now. Thanks. I was confused also by a spin lock, which
> is being released just after clear pending:
>
>set_work_pool_and_clear_pending(work, pool->id);
>spin_unlock_irq(>lock);
>...
>
On Mon, Apr 25, 2016 at 7:03 PM, Tejun Heo wrote:
> Hello, Roman.
>
> On Mon, Apr 25, 2016 at 06:34:45PM +0200, Roman Penyaev wrote:
>> I can assure you that smp_mb() helps (at least running for 30 minutes
>> under IO). That was my first variant, but I did not like it because I
On Mon, Apr 25, 2016 at 7:03 PM, Tejun Heo wrote:
> Hello, Roman.
>
> On Mon, Apr 25, 2016 at 06:34:45PM +0200, Roman Penyaev wrote:
>> I can assure you that smp_mb() helps (at least running for 30 minutes
>> under IO). That was my first variant, but I did not like it because I
>> could not
Hello, Roman.
On Mon, Apr 25, 2016 at 06:34:45PM +0200, Roman Penyaev wrote:
> I can assure you that smp_mb() helps (at least running for 30 minutes
> under IO). That was my first variant, but I did not like it because I
> could not explain myself why:
>
> 1. not smp_wmb()? We need to do flush
Hello, Roman.
On Mon, Apr 25, 2016 at 06:34:45PM +0200, Roman Penyaev wrote:
> I can assure you that smp_mb() helps (at least running for 30 minutes
> under IO). That was my first variant, but I did not like it because I
> could not explain myself why:
>
> 1. not smp_wmb()? We need to do flush
On Mon, Apr 25, 2016 at 6:00 PM, Tejun Heo wrote:
> On Mon, Apr 25, 2016 at 11:48:47AM -0400, Tejun Heo wrote:
>> Heh, excellent debugging. I wonder how old this bug is. cc'ing David
>
> I just went through the history. So, we used to have clear_bit()
> instead of
On Mon, Apr 25, 2016 at 6:00 PM, Tejun Heo wrote:
> On Mon, Apr 25, 2016 at 11:48:47AM -0400, Tejun Heo wrote:
>> Heh, excellent debugging. I wonder how old this bug is. cc'ing David
>
> I just went through the history. So, we used to have clear_bit()
> instead of atomic_long_set() but
Hello, Tejun,
On Mon, Apr 25, 2016 at 5:48 PM, Tejun Heo wrote:
> Hello, Roman.
>
> On Mon, Apr 25, 2016 at 05:22:51PM +0200, Roman Pen wrote:
> ...
>> CPU#6CPU#2
>> reqeust 884000343600 inserted
>> hctx marked as pended
>>
Hello, Tejun,
On Mon, Apr 25, 2016 at 5:48 PM, Tejun Heo wrote:
> Hello, Roman.
>
> On Mon, Apr 25, 2016 at 05:22:51PM +0200, Roman Pen wrote:
> ...
>> CPU#6CPU#2
>> reqeust 884000343600 inserted
>> hctx marked as pended
>> kblockd_schedule...()
On Mon, Apr 25, 2016 at 11:48:47AM -0400, Tejun Heo wrote:
> Heh, excellent debugging. I wonder how old this bug is. cc'ing David
I just went through the history. So, we used to have clear_bit()
instead of atomic_long_set() but clear_bit() doesn't imply any barrier
either, so kudos to you.
On Mon, Apr 25, 2016 at 11:48:47AM -0400, Tejun Heo wrote:
> Heh, excellent debugging. I wonder how old this bug is. cc'ing David
I just went through the history. So, we used to have clear_bit()
instead of atomic_long_set() but clear_bit() doesn't imply any barrier
either, so kudos to you.
Hello, Roman.
On Mon, Apr 25, 2016 at 05:22:51PM +0200, Roman Pen wrote:
...
> CPU#6CPU#2
> reqeust 884000343600 inserted
> hctx marked as pended
> kblockd_schedule...() returns 1
>
> *** WORK_STRUCT_PENDING_BIT is cleared ***
> flush_busy_ctxs()
Hello, Roman.
On Mon, Apr 25, 2016 at 05:22:51PM +0200, Roman Pen wrote:
...
> CPU#6CPU#2
> reqeust 884000343600 inserted
> hctx marked as pended
> kblockd_schedule...() returns 1
>
> *** WORK_STRUCT_PENDING_BIT is cleared ***
> flush_busy_ctxs()
Hi,
This is RFC, because for last couple of days I hunt a mystery bug and
since now I do not have a strong feeling that the following story is
nothing but bug's trick and attempt to cheat me.
[Sorry, the whole story and explanation are quite long]
The bug is reproduced quite often on our server
Hi,
This is RFC, because for last couple of days I hunt a mystery bug and
since now I do not have a strong feeling that the following story is
nothing but bug's trick and attempt to cheat me.
[Sorry, the whole story and explanation are quite long]
The bug is reproduced quite often on our server
30 matches
Mail list logo