[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <0fd27cfd-4fe3-9b33-c583-45e6e71e18b7@linux.intel.com>
Date: Sun, 5 Jul 2020 16:44:52 -0700
From: Tim Chen <tim.c.chen@...ux.intel.com>
To: Joel Fernandes <joel@...lfernandes.org>
Cc: Vineeth Remanan Pillai <vpillai@...italocean.com>,
Peter Zijlstra <peterz@...radead.org>,
Nishanth Aravamudan <naravamudan@...italocean.com>,
Julien Desfossez <jdesfossez@...italocean.com>,
tglx@...utronix.de, pjt@...gle.com, torvalds@...ux-foundation.org,
linux-kernel@...r.kernel.org, subhra.mazumdar@...cle.com,
mingo@...nel.org, fweisbec@...il.com, keescook@...omium.org,
kerrnel@...gle.com, Phil Auld <pauld@...hat.com>,
Aaron Lu <aaron.lwe@...il.com>,
Aubrey Li <aubrey.intel@...il.com>,
Valentin Schneider <valentin.schneider@....com>,
Mel Gorman <mgorman@...hsingularity.net>,
Pawan Gupta <pawan.kumar.gupta@...ux.intel.com>,
Paolo Bonzini <pbonzini@...hat.com>, vineethrp@...il.com,
Chen Yu <yu.c.chen@...el.com>,
Christian Brauner <christian.brauner@...ntu.com>,
Aaron Lu <aaron.lu@...ux.alibaba.com>, paulmck@...nel.org
Subject: Re: [RFC PATCH 06/16] sched: Add core wide task selection and
scheduling.
On 7/2/20 5:57 AM, Joel Fernandes wrote:
> On Wed, Jul 01, 2020 at 05:54:11PM -0700, Tim Chen wrote:
>>
>>
>> On 7/1/20 4:28 PM, Joel Fernandes wrote:
>>> On Tue, Jun 30, 2020 at 09:32:27PM +0000, Vineeth Remanan Pillai wrote:
>>>> From: Peter Zijlstra <peterz@...radead.org>
>>>>
>>>> Instead of only selecting a local task, select a task for all SMT
>>>> siblings for every reschedule on the core (irrespective which logical
>>>> CPU does the reschedule).
>>>>
>>>> There could be races in core scheduler where a CPU is trying to pick
>>>> a task for its sibling in core scheduler, when that CPU has just been
>>>> offlined. We should not schedule any tasks on the CPU in this case.
>>>> Return an idle task in pick_next_task for this situation.
>>>>
>>>> NOTE: there is still potential for siblings rivalry.
>>>> NOTE: this is far too complicated; but thus far I've failed to
>>>> simplify it further.
>>>>
>>>> Signed-off-by: Peter Zijlstra (Intel) <peterz@...radead.org>
>>>> Signed-off-by: Julien Desfossez <jdesfossez@...italocean.com>
>>>> Signed-off-by: Vineeth Remanan Pillai <vpillai@...italocean.com>
>>>> Signed-off-by: Aaron Lu <aaron.lu@...ux.alibaba.com>
>>>> Signed-off-by: Tim Chen <tim.c.chen@...ux.intel.com>
>>>
>>> Hi Peter, Tim, all, the below patch fixes the hotplug issue described in the
>>> below patch's Link tag. Patch description below describes the issues fixed
>>> and it applies on top of this patch.
>>>
>>> ------8<----------
>>>
>>> From: "Joel Fernandes (Google)" <joel@...lfernandes.org>
>>> Subject: [PATCH] sched: Fix CPU hotplug causing crashes in task selection logic
>>>
>>> The selection logic does not run correctly if the current CPU is not in the
>>> cpu_smt_mask (which it is not because the CPU is offlined when the stopper
>>> finishes running and needs to switch to idle). There are also other issues
>>> fixed by the patch I think such as: if some other sibling set core_pick to
>>> something, however the selection logic on current cpu resets it before
>>> selecting. In this case, we need to run the task selection logic again to
>>> make sure it picks something if there is something to run. It might end up
>>> picking the wrong task. Yet another issue was, if the stopper thread is an
>
> "It might end up picking the wrong task" needs to be: "We might end up
> picking a different task but that's Ok".
>
>>> unconstrained pick, then rq->core_pick is set. The next time task selection
>>> logic runs when stopper needs to switch to idle, the current CPU is not in
>>> the smt_mask. This causes the previous ->core_pick to be picked again which
>>> happens to be the unconstrained task! so the stopper keeps getting selected
>>> forever.
>>>
>>> That and there are a few more safe guards and checks around checking/setting
>>> rq->core_pick. To test it, I ran rcutorture and made it tag all torture
>>> threads. Then ran it in hotplug mode (hotplugging every 200ms) and it hit the
>>> issue. Now it runs for an hour or so without issue. (Torture testing debug
>>> changes: https://bit.ly/38htfqK ).
>>>
>>> Various fixes were tried causing varying degrees of crashes. Finally I found
>>> that it is easiest to just add current CPU to the smt_mask's copy always.
>>> This is so that task selection logic always runs on the current CPU which
>>> called schedule().
>>
>>
>> It looks good to me.
>
> Thank you for your review! Could I add your Reviewed-by tag to the patch?
>
Sure.
Tim
Powered by blists - more mailing lists