diff mbox

xen/arm: Assertion 'timer->status >= TIMER_STATUS_inactive' failed at timer.c:279

Message ID 1461692950.3525.71.camel@citrix.com (mailing list archive)
State New, archived
Headers show

Commit Message

Dario Faggioli April 26, 2016, 5:49 p.m. UTC
On Tue, 2016-04-26 at 15:25 +0100, Julien Grall wrote:
> Hi Dario,
> 
Hi,

> A couple of people have been reported Xen crash on the ARM64
> Foundation Model [1] with recent unstable.
> 
Ok, thanks for reporting.

> The crash seems to happen when Xen fails to bring up secondary CPUs
> (see stack trace below).
> 
Ah... I see.

> From my understanding, csched_free_pdata is trying to kill the
> timer spc->ticker. However the status of this timer is
> TIMER_STATUS_invalid.
> 
> This is because csched_init_pdata has set a deadline for the
> timer (set_timer) and the softirq to schedule the timer has
> not yet happen (indeed Xen is still in early boot).
> 
Yes, this is sort of what happens (only slightly different, see the
changelog of the atached patch patch).


> I am not sure how to fix this issue. How will you recommend
> to fix it?
> 
Yeah, well, doing it cleanly includes a slight change in the scheduler
hooks API, IMO... and we indeed should do it cleanly :-))

George, what do you think?

Honestly, this is similar to what I was thinking to do already (I mean,
having an deinit_pdata hook, "symmetric" with the init_pdata one), when
working on that series, because I do think it's cleaner... then, I
abandoned the idea, as it looked to not be necessary... But apparently
it may actually be! :-)

Let me know, and I'll resubmit the patch properly (together with
another bugfix I have in my queue).

Dario
---
commit eca4d65fb67a71c0f6563aafbfdd68e566c53c32
Author: Dario Faggioli <dario.faggioli@citrix.com>
Date:   Tue Apr 26 17:42:36 2016 +0200

    xen: sched: fix killing an uninitialized timer in free_pdata.
    
    commit 64269d9365 "sched: implement .init_pdata in Credit,
    Credit2 and RTDS" helped fixing Credit2 runqueues, and
    the races we had in switching scheduler for pCPUs, but
    introduced another issue. In fact, if CPU bringup fails
    during __cpu_up() (and, more precisely, after CPU_UP_PREPARE,
    but before CPU_STARTING) the CPU_UP_CANCELED notifier
    would be executed, which calls the free_pdata hook.
    
    Such hook does two things: (1) undo the initialization
    done inside the init_pdata hook; (2) free the memory
    allocated by the alloc_pdata hook.
    
    However, in the failure path just described, it is possible
    that only alloc_pdata has really been called, which is
    potentially and issue (depending on what actually happens
    inside the implementation of free_pdata).
    
    In fact, for Credit1 (the only scheduler that actually
    allocates per-pCPU data), this result in calling kill_timer()
    on a timer that had not yet been initialized, which causes
    the following:
    
    (XEN) Xen call trace:
    (XEN)    [<000000000022e304>] timer.c#active_timer+0x8/0x24 (PC)
    (XEN)    [<000000000022f624>] kill_timer+0x108/0x2e0 (LR)
    (XEN)    [<00000000002208c0>] sched_credit.c#csched_free_pdata+0xd8/0x114
    (XEN)    [<0000000000227a18>] schedule.c#cpu_schedule_callback+0xc0/0x12c
    (XEN)    [<0000000000219944>] notifier_call_chain+0x78/0x9c
    (XEN)    [<00000000002015fc>] cpu_up+0x104/0x130
    (XEN)    [<000000000028f7c0>] start_xen+0xaf8/0xce0
    (XEN)    [<00000000810021d8>] 00000000810021d8
    (XEN)
    (XEN)
    (XEN) ****************************************
    (XEN) Panic on CPU 0:
    (XEN) Assertion 'timer->status >= TIMER_STATUS_inactive' failed at timer.c:279
    (XEN) ****************************************
    
    Solve this by making the scheduler hooks API symmetric again,
    i.e., by adding an deinit_pdata hook and making it responsible
    of undoing what init_pdata did, rather than asking to free_pdata
    to do everything.
    
    This is cleaner and, in the case at hand, makes it possible to
    only call free_pdata, which is the right thing to do, as only
    allocation and no initialization was performed.
    
    Reported-by: Julien Grall <julien.grall@arm.com>
    Signed-off-by: Dario Faggioli <dario.faggioli@citrix.com>
    ---
    Cc: George Dunlap <george.dunlap@citrix.com>
    Cc: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
    Cc: Varun.Swara@arm.com
    Cc: Steve Capper <Steve.Capper@arm.com>

Comments

Julien Grall April 26, 2016, 6:05 p.m. UTC | #1
On 26/04/16 18:49, Dario Faggioli wrote:
> On Tue, 2016-04-26 at 15:25 +0100, Julien Grall wrote:
> Hi,

Hi Dario,

[...]

> commit eca4d65fb67a71c0f6563aafbfdd68e566c53c32
> Author: Dario Faggioli <dario.faggioli@citrix.com>
> Date:   Tue Apr 26 17:42:36 2016 +0200
>
>      xen: sched: fix killing an uninitialized timer in free_pdata.
>
>      commit 64269d9365 "sched: implement .init_pdata in Credit,
>      Credit2 and RTDS" helped fixing Credit2 runqueues, and
>      the races we had in switching scheduler for pCPUs, but
>      introduced another issue. In fact, if CPU bringup fails
>      during __cpu_up() (and, more precisely, after CPU_UP_PREPARE,
>      but before CPU_STARTING) the CPU_UP_CANCELED notifier
>      would be executed, which calls the free_pdata hook.
>
>      Such hook does two things: (1) undo the initialization
>      done inside the init_pdata hook; (2) free the memory
>      allocated by the alloc_pdata hook.
>
>      However, in the failure path just described, it is possible
>      that only alloc_pdata has really been called, which is
>      potentially and issue (depending on what actually happens
>      inside the implementation of free_pdata).
>
>      In fact, for Credit1 (the only scheduler that actually
>      allocates per-pCPU data), this result in calling kill_timer()
>      on a timer that had not yet been initialized, which causes
>      the following:
>
>      (XEN) Xen call trace:
>      (XEN)    [<000000000022e304>] timer.c#active_timer+0x8/0x24 (PC)
>      (XEN)    [<000000000022f624>] kill_timer+0x108/0x2e0 (LR)
>      (XEN)    [<00000000002208c0>] sched_credit.c#csched_free_pdata+0xd8/0x114
>      (XEN)    [<0000000000227a18>] schedule.c#cpu_schedule_callback+0xc0/0x12c
>      (XEN)    [<0000000000219944>] notifier_call_chain+0x78/0x9c
>      (XEN)    [<00000000002015fc>] cpu_up+0x104/0x130
>      (XEN)    [<000000000028f7c0>] start_xen+0xaf8/0xce0
>      (XEN)    [<00000000810021d8>] 00000000810021d8
>      (XEN)
>      (XEN)
>      (XEN) ****************************************
>      (XEN) Panic on CPU 0:
>      (XEN) Assertion 'timer->status >= TIMER_STATUS_inactive' failed at timer.c:279
>      (XEN) ****************************************
>
>      Solve this by making the scheduler hooks API symmetric again,
>      i.e., by adding an deinit_pdata hook and making it responsible
>      of undoing what init_pdata did, rather than asking to free_pdata
>      to do everything.
>
>      This is cleaner and, in the case at hand, makes it possible to
>      only call free_pdata, which is the right thing to do, as only
>      allocation and no initialization was performed.
>
>      Reported-by: Julien Grall <julien.grall@arm.com>
>      Signed-off-by: Dario Faggioli <dario.faggioli@citrix.com>
>      ---
>      Cc: George Dunlap <george.dunlap@citrix.com>
>      Cc: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
>      Cc: Varun.Swara@arm.com
>      Cc: Steve Capper <Steve.Capper@arm.com>

Thank you for the patch. It fixes the bug when secondary CPUs fail to 
come online on the Foundation Model:

Tested-by: Julien Grall <julien.grall@arm.com>

Regards,
George Dunlap April 27, 2016, 1:43 p.m. UTC | #2
On 26/04/16 18:49, Dario Faggioli wrote:
> On Tue, 2016-04-26 at 15:25 +0100, Julien Grall wrote:
>> Hi Dario,
>>
> Hi,
> 
>> A couple of people have been reported Xen crash on the ARM64
>> Foundation Model [1] with recent unstable.
>>
> Ok, thanks for reporting.
> 
>> The crash seems to happen when Xen fails to bring up secondary CPUs
>> (see stack trace below).
>>
> Ah... I see.
> 
>> From my understanding, csched_free_pdata is trying to kill the
>> timer spc->ticker. However the status of this timer is
>> TIMER_STATUS_invalid.
>>
>> This is because csched_init_pdata has set a deadline for the
>> timer (set_timer) and the softirq to schedule the timer has
>> not yet happen (indeed Xen is still in early boot).
>>
> Yes, this is sort of what happens (only slightly different, see the
> changelog of the atached patch patch).
> 
> 
>> I am not sure how to fix this issue. How will you recommend
>> to fix it?
>>
> Yeah, well, doing it cleanly includes a slight change in the scheduler
> hooks API, IMO... and we indeed should do it cleanly :-))
> 
> George, what do you think?
> 
> Honestly, this is similar to what I was thinking to do already (I mean,
> having an deinit_pdata hook, "symmetric" with the init_pdata one), when
> working on that series, because I do think it's cleaner... then, I
> abandoned the idea, as it looked to not be necessary... But apparently
> it may actually be! :-)
> 
> Let me know, and I'll resubmit the patch properly (together with
> another bugfix I have in my queue).

Yeah, assuming the description in your changeset is accurate, this seems
like the right approach.

The main thing to add here I think is that we need to document what
different circumstances under which the various functions may be called
-- for instance, in credit1 free_pdata(), it seems to expect that spc
may == null at some point.  Future schedulers need to know the
circumstances under which this might happen so they can DTRT.

It might be nice at some point to have the alloc / free / init / deinit
functions in credit1 ordered in a rational way so that they could be
understood by glancing at them, rather than having to jump around, but
that's probably a nice-to-have clean-up for another time. :-)

 -George
Dario Faggioli April 27, 2016, 2:05 p.m. UTC | #3
On Wed, 2016-04-27 at 14:43 +0100, George Dunlap wrote:
> On 26/04/16 18:49, Dario Faggioli wrote:
> > 
> > Let me know, and I'll resubmit the patch properly (together with
> > another bugfix I have in my queue).
> Yeah, assuming the description in your changeset is accurate, this
> seems
> like the right approach.
> 
Ok, thanks for having a look, I'll submit a proper series.

> The main thing to add here I think is that we need to document what
> different circumstances under which the various functions may be
> called
> -- for instance, in credit1 free_pdata(), it seems to expect that spc
> may == null at some point.  Future schedulers need to know the
> circumstances under which this might happen so they can DTRT.
> 
I saw that too (many times). And in fact, I'm not sure whether that can
actually happen or not, but I certainly can look at this.

And if by "document what different circumstances under which the
various functions may be called" you mean adding comments to that
effect somewhere, I'm up for that (I just need to figure out where it
would be best to put such comments).

> It might be nice at some point to have the alloc / free / init /
> deinit
> functions in credit1 ordered in a rational way so that they could be
> understood by glancing at them, rather than having to jump around,
> but
> that's probably a nice-to-have clean-up for another time. :-)
> 
When you say "ordered" you mean the order in which they appear in the
source file? If yes, I agree, but no, I'm not doing that right now (but
I can queue this for when 4.8 opens).

Regards,
Dario
George Dunlap April 27, 2016, 2:29 p.m. UTC | #4
On 27/04/16 15:05, Dario Faggioli wrote:
> On Wed, 2016-04-27 at 14:43 +0100, George Dunlap wrote:
>> On 26/04/16 18:49, Dario Faggioli wrote:
>>>  
>>> Let me know, and I'll resubmit the patch properly (together with
>>> another bugfix I have in my queue).
>> Yeah, assuming the description in your changeset is accurate, this
>> seems
>> like the right approach.
>>
> Ok, thanks for having a look, I'll submit a proper series.
> 
>> The main thing to add here I think is that we need to document what
>> different circumstances under which the various functions may be
>> called
>> -- for instance, in credit1 free_pdata(), it seems to expect that spc
>> may == null at some point.  Future schedulers need to know the
>> circumstances under which this might happen so they can DTRT.
>>
> I saw that too (many times). And in fact, I'm not sure whether that can
> actually happen or not, but I certainly can look at this.
> 
> And if by "document what different circumstances under which the
> various functions may be called" you mean adding comments to that
> effect somewhere, I'm up for that (I just need to figure out where it
> would be best to put such comments).

Yes, comments is what I meant.  At very least it should be documented
that alloc/free might be called without init/deinit netsed inside.  If
you're also not sure how free_pdata() could be called with spc == null,
then I think we can probably punt an investigation into that until later
as well.


>> It might be nice at some point to have the alloc / free / init /
>> deinit
>> functions in credit1 ordered in a rational way so that they could be
>> understood by glancing at them, rather than having to jump around,
>> but
>> that's probably a nice-to-have clean-up for another time. :-)
>>
> When you say "ordered" you mean the order in which they appear in the
> source file? If yes, I agree, but no, I'm not doing that right now (but
> I can queue this for when 4.8 opens).

Sorry, yes I meant ordered in the source file.

 -George
Julien Grall May 3, 2016, 1:03 p.m. UTC | #5
Hi Dario and George,

What is the status of this patch? It would be nice to have it for Xen 
4.7 to avoid unwanted crash when secondary CPUs fails to come online.

Regards,

On 26/04/16 18:49, Dario Faggioli wrote:
> On Tue, 2016-04-26 at 15:25 +0100, Julien Grall wrote:
>> Hi Dario,
>>
> Hi,
>
>> A couple of people have been reported Xen crash on the ARM64
>> Foundation Model [1] with recent unstable.
>>
> Ok, thanks for reporting.
>
>> The crash seems to happen when Xen fails to bring up secondary CPUs
>> (see stack trace below).
>>
> Ah... I see.
>
>>  From my understanding, csched_free_pdata is trying to kill the
>> timer spc->ticker. However the status of this timer is
>> TIMER_STATUS_invalid.
>>
>> This is because csched_init_pdata has set a deadline for the
>> timer (set_timer) and the softirq to schedule the timer has
>> not yet happen (indeed Xen is still in early boot).
>>
> Yes, this is sort of what happens (only slightly different, see the
> changelog of the atached patch patch).
>
>
>> I am not sure how to fix this issue. How will you recommend
>> to fix it?
>>
> Yeah, well, doing it cleanly includes a slight change in the scheduler
> hooks API, IMO... and we indeed should do it cleanly :-))
>
> George, what do you think?
>
> Honestly, this is similar to what I was thinking to do already (I mean,
> having an deinit_pdata hook, "symmetric" with the init_pdata one), when
> working on that series, because I do think it's cleaner... then, I
> abandoned the idea, as it looked to not be necessary... But apparently
> it may actually be! :-)
>
> Let me know, and I'll resubmit the patch properly (together with
> another bugfix I have in my queue).
>
> Dario
> ---
> commit eca4d65fb67a71c0f6563aafbfdd68e566c53c32
> Author: Dario Faggioli <dario.faggioli@citrix.com>
> Date:   Tue Apr 26 17:42:36 2016 +0200
>
>      xen: sched: fix killing an uninitialized timer in free_pdata.
>
>      commit 64269d9365 "sched: implement .init_pdata in Credit,
>      Credit2 and RTDS" helped fixing Credit2 runqueues, and
>      the races we had in switching scheduler for pCPUs, but
>      introduced another issue. In fact, if CPU bringup fails
>      during __cpu_up() (and, more precisely, after CPU_UP_PREPARE,
>      but before CPU_STARTING) the CPU_UP_CANCELED notifier
>      would be executed, which calls the free_pdata hook.
>
>      Such hook does two things: (1) undo the initialization
>      done inside the init_pdata hook; (2) free the memory
>      allocated by the alloc_pdata hook.
>
>      However, in the failure path just described, it is possible
>      that only alloc_pdata has really been called, which is
>      potentially and issue (depending on what actually happens
>      inside the implementation of free_pdata).
>
>      In fact, for Credit1 (the only scheduler that actually
>      allocates per-pCPU data), this result in calling kill_timer()
>      on a timer that had not yet been initialized, which causes
>      the following:
>
>      (XEN) Xen call trace:
>      (XEN)    [<000000000022e304>] timer.c#active_timer+0x8/0x24 (PC)
>      (XEN)    [<000000000022f624>] kill_timer+0x108/0x2e0 (LR)
>      (XEN)    [<00000000002208c0>] sched_credit.c#csched_free_pdata+0xd8/0x114
>      (XEN)    [<0000000000227a18>] schedule.c#cpu_schedule_callback+0xc0/0x12c
>      (XEN)    [<0000000000219944>] notifier_call_chain+0x78/0x9c
>      (XEN)    [<00000000002015fc>] cpu_up+0x104/0x130
>      (XEN)    [<000000000028f7c0>] start_xen+0xaf8/0xce0
>      (XEN)    [<00000000810021d8>] 00000000810021d8
>      (XEN)
>      (XEN)
>      (XEN) ****************************************
>      (XEN) Panic on CPU 0:
>      (XEN) Assertion 'timer->status >= TIMER_STATUS_inactive' failed at timer.c:279
>      (XEN) ****************************************
>
>      Solve this by making the scheduler hooks API symmetric again,
>      i.e., by adding an deinit_pdata hook and making it responsible
>      of undoing what init_pdata did, rather than asking to free_pdata
>      to do everything.
>
>      This is cleaner and, in the case at hand, makes it possible to
>      only call free_pdata, which is the right thing to do, as only
>      allocation and no initialization was performed.
>
>      Reported-by: Julien Grall <julien.grall@arm.com>
>      Signed-off-by: Dario Faggioli <dario.faggioli@citrix.com>
>      ---
>      Cc: George Dunlap <george.dunlap@citrix.com>
>      Cc: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
>      Cc: Varun.Swara@arm.com
>      Cc: Steve Capper <Steve.Capper@arm.com>
>
> diff --git a/xen/common/sched_credit.c b/xen/common/sched_credit.c
> index bc36837..0a6a1b4 100644
> --- a/xen/common/sched_credit.c
> +++ b/xen/common/sched_credit.c
> @@ -482,15 +482,25 @@ static inline void __runq_tickle(struct csched_vcpu *new)
>   }
>
>   static void
> -csched_free_pdata(const struct scheduler *ops, void *pcpu, int cpu)
> +csched_free_pdata(const struct scheduler *ops, void *pcpu)
>   {
> -    struct csched_private *prv = CSCHED_PRIV(ops);
>       struct csched_pcpu *spc = pcpu;
> -    unsigned long flags;
>
>       if ( spc == NULL )
>           return;
>
> +    xfree(spc);
> +}
> +
> +static void
> +csched_deinit_pdata(const struct scheduler *ops, void *pcpu, int cpu)
> +{
> +    struct csched_private *prv = CSCHED_PRIV(ops);
> +    struct csched_pcpu *spc = pcpu;
> +    unsigned long flags;
> +
> +    ASSERT(spc != NULL);
> +
>       spin_lock_irqsave(&prv->lock, flags);
>
>       prv->credit -= prv->credits_per_tslice;
> @@ -507,8 +517,6 @@ csched_free_pdata(const struct scheduler *ops, void *pcpu, int cpu)
>           kill_timer(&prv->master_ticker);
>
>       spin_unlock_irqrestore(&prv->lock, flags);
> -
> -    xfree(spc);
>   }
>
>   static void *
> @@ -2091,6 +2099,7 @@ static const struct scheduler sched_credit_def = {
>       .free_vdata     = csched_free_vdata,
>       .alloc_pdata    = csched_alloc_pdata,
>       .init_pdata     = csched_init_pdata,
> +    .deinit_pdata   = csched_deinit_pdata,
>       .free_pdata     = csched_free_pdata,
>       .switch_sched   = csched_switch_sched,
>       .alloc_domdata  = csched_alloc_domdata,
> diff --git a/xen/common/sched_credit2.c b/xen/common/sched_credit2.c
> index 46b9279..f4c37b4 100644
> --- a/xen/common/sched_credit2.c
> +++ b/xen/common/sched_credit2.c
> @@ -2261,13 +2261,15 @@ csched2_switch_sched(struct scheduler *new_ops, unsigned int cpu,
>   }
>
>   static void
> -csched2_free_pdata(const struct scheduler *ops, void *pcpu, int cpu)
> +csched2_deinit_pdata(const struct scheduler *ops, void *pcpu, int cpu)
>   {
>       unsigned long flags;
>       struct csched2_private *prv = CSCHED2_PRIV(ops);
>       struct csched2_runqueue_data *rqd;
>       int rqi;
>
> +    ASSERT(pcpu == NULL);
> +
>       spin_lock_irqsave(&prv->lock, flags);
>
>       ASSERT(cpumask_test_cpu(cpu, &prv->initialized));
> @@ -2387,7 +2389,7 @@ static const struct scheduler sched_credit2_def = {
>       .alloc_vdata    = csched2_alloc_vdata,
>       .free_vdata     = csched2_free_vdata,
>       .init_pdata     = csched2_init_pdata,
> -    .free_pdata     = csched2_free_pdata,
> +    .deinit_pdata   = csched2_deinit_pdata,
>       .switch_sched   = csched2_switch_sched,
>       .alloc_domdata  = csched2_alloc_domdata,
>       .free_domdata   = csched2_free_domdata,
> diff --git a/xen/common/schedule.c b/xen/common/schedule.c
> index 5546999..1a64521 100644
> --- a/xen/common/schedule.c
> +++ b/xen/common/schedule.c
> @@ -1529,7 +1529,7 @@ static void cpu_schedule_down(unsigned int cpu)
>       struct schedule_data *sd = &per_cpu(schedule_data, cpu);
>       struct scheduler *sched = per_cpu(scheduler, cpu);
>
> -    SCHED_OP(sched, free_pdata, sd->sched_priv, cpu);
> +    SCHED_OP(sched, free_pdata, sd->sched_priv);
>       SCHED_OP(sched, free_vdata, idle_vcpu[cpu]->sched_priv);
>
>       idle_vcpu[cpu]->sched_priv = NULL;
> @@ -1554,8 +1554,10 @@ static int cpu_schedule_callback(
>       case CPU_UP_PREPARE:
>           rc = cpu_schedule_up(cpu);
>           break;
> -    case CPU_UP_CANCELED:
>       case CPU_DEAD:
> +        SCHED_OP(sched, deinit_pdata, sd->sched_priv, cpu);
> +        /* Fallthrough */
> +    case CPU_UP_CANCELED:
>           cpu_schedule_down(cpu);
>           break;
>       default:
> @@ -1684,7 +1686,7 @@ int schedule_cpu_switch(unsigned int cpu, struct cpupool *c)
>       vpriv = SCHED_OP(new_ops, alloc_vdata, idle, idle->domain->sched_priv);
>       if ( vpriv == NULL )
>       {
> -        SCHED_OP(new_ops, free_pdata, ppriv, cpu);
> +        SCHED_OP(new_ops, free_pdata, ppriv);
>           return -ENOMEM;
>       }
>
> @@ -1714,7 +1716,8 @@ int schedule_cpu_switch(unsigned int cpu, struct cpupool *c)
>       SCHED_OP(new_ops, tick_resume, cpu);
>
>       SCHED_OP(old_ops, free_vdata, vpriv_old);
> -    SCHED_OP(old_ops, free_pdata, ppriv_old, cpu);
> +    SCHED_OP(old_ops, deinit_pdata, ppriv_old, cpu);
> +    SCHED_OP(old_ops, free_pdata, ppriv_old);
>
>    out:
>       per_cpu(cpupool, cpu) = c;
> diff --git a/xen/include/xen/sched-if.h b/xen/include/xen/sched-if.h
> index 1db7c8d..240f66c 100644
> --- a/xen/include/xen/sched-if.h
> +++ b/xen/include/xen/sched-if.h
> @@ -135,9 +135,10 @@ struct scheduler {
>       void         (*free_vdata)     (const struct scheduler *, void *);
>       void *       (*alloc_vdata)    (const struct scheduler *, struct vcpu *,
>                                       void *);
> -    void         (*free_pdata)     (const struct scheduler *, void *, int);
> +    void         (*free_pdata)     (const struct scheduler *, void *);
>       void *       (*alloc_pdata)    (const struct scheduler *, int);
>       void         (*init_pdata)     (const struct scheduler *, void *, int);
> +    void         (*deinit_pdata)   (const struct scheduler *, void *, int);
>       void         (*free_domdata)   (const struct scheduler *, void *);
>       void *       (*alloc_domdata)  (const struct scheduler *, struct domain *);
>
>
George Dunlap May 3, 2016, 1:20 p.m. UTC | #6
On Tue, May 3, 2016 at 2:03 PM, Julien Grall <julien.grall@arm.com> wrote:
> Hi Dario and George,
>
> What is the status of this patch? It would be nice to have it for Xen 4.7 to
> avoid unwanted crash when secondary CPUs fails to come online.

Wei, can you put this on your release blockers list?  (Julien, I take
it this should be a blocker, right?)

 -George

>
> Regards,
>
> On 26/04/16 18:49, Dario Faggioli wrote:
>>
>> On Tue, 2016-04-26 at 15:25 +0100, Julien Grall wrote:
>>>
>>> Hi Dario,
>>>
>> Hi,
>>
>>> A couple of people have been reported Xen crash on the ARM64
>>> Foundation Model [1] with recent unstable.
>>>
>> Ok, thanks for reporting.
>>
>>> The crash seems to happen when Xen fails to bring up secondary CPUs
>>> (see stack trace below).
>>>
>> Ah... I see.
>>
>>>  From my understanding, csched_free_pdata is trying to kill the
>>> timer spc->ticker. However the status of this timer is
>>> TIMER_STATUS_invalid.
>>>
>>> This is because csched_init_pdata has set a deadline for the
>>> timer (set_timer) and the softirq to schedule the timer has
>>> not yet happen (indeed Xen is still in early boot).
>>>
>> Yes, this is sort of what happens (only slightly different, see the
>> changelog of the atached patch patch).
>>
>>
>>> I am not sure how to fix this issue. How will you recommend
>>> to fix it?
>>>
>> Yeah, well, doing it cleanly includes a slight change in the scheduler
>> hooks API, IMO... and we indeed should do it cleanly :-))
>>
>> George, what do you think?
>>
>> Honestly, this is similar to what I was thinking to do already (I mean,
>> having an deinit_pdata hook, "symmetric" with the init_pdata one), when
>> working on that series, because I do think it's cleaner... then, I
>> abandoned the idea, as it looked to not be necessary... But apparently
>> it may actually be! :-)
>>
>> Let me know, and I'll resubmit the patch properly (together with
>> another bugfix I have in my queue).
>>
>> Dario
>> ---
>> commit eca4d65fb67a71c0f6563aafbfdd68e566c53c32
>> Author: Dario Faggioli <dario.faggioli@citrix.com>
>> Date:   Tue Apr 26 17:42:36 2016 +0200
>>
>>      xen: sched: fix killing an uninitialized timer in free_pdata.
>>
>>      commit 64269d9365 "sched: implement .init_pdata in Credit,
>>      Credit2 and RTDS" helped fixing Credit2 runqueues, and
>>      the races we had in switching scheduler for pCPUs, but
>>      introduced another issue. In fact, if CPU bringup fails
>>      during __cpu_up() (and, more precisely, after CPU_UP_PREPARE,
>>      but before CPU_STARTING) the CPU_UP_CANCELED notifier
>>      would be executed, which calls the free_pdata hook.
>>
>>      Such hook does two things: (1) undo the initialization
>>      done inside the init_pdata hook; (2) free the memory
>>      allocated by the alloc_pdata hook.
>>
>>      However, in the failure path just described, it is possible
>>      that only alloc_pdata has really been called, which is
>>      potentially and issue (depending on what actually happens
>>      inside the implementation of free_pdata).
>>
>>      In fact, for Credit1 (the only scheduler that actually
>>      allocates per-pCPU data), this result in calling kill_timer()
>>      on a timer that had not yet been initialized, which causes
>>      the following:
>>
>>      (XEN) Xen call trace:
>>      (XEN)    [<000000000022e304>] timer.c#active_timer+0x8/0x24 (PC)
>>      (XEN)    [<000000000022f624>] kill_timer+0x108/0x2e0 (LR)
>>      (XEN)    [<00000000002208c0>]
>> sched_credit.c#csched_free_pdata+0xd8/0x114
>>      (XEN)    [<0000000000227a18>]
>> schedule.c#cpu_schedule_callback+0xc0/0x12c
>>      (XEN)    [<0000000000219944>] notifier_call_chain+0x78/0x9c
>>      (XEN)    [<00000000002015fc>] cpu_up+0x104/0x130
>>      (XEN)    [<000000000028f7c0>] start_xen+0xaf8/0xce0
>>      (XEN)    [<00000000810021d8>] 00000000810021d8
>>      (XEN)
>>      (XEN)
>>      (XEN) ****************************************
>>      (XEN) Panic on CPU 0:
>>      (XEN) Assertion 'timer->status >= TIMER_STATUS_inactive' failed at
>> timer.c:279
>>      (XEN) ****************************************
>>
>>      Solve this by making the scheduler hooks API symmetric again,
>>      i.e., by adding an deinit_pdata hook and making it responsible
>>      of undoing what init_pdata did, rather than asking to free_pdata
>>      to do everything.
>>
>>      This is cleaner and, in the case at hand, makes it possible to
>>      only call free_pdata, which is the right thing to do, as only
>>      allocation and no initialization was performed.
>>
>>      Reported-by: Julien Grall <julien.grall@arm.com>
>>      Signed-off-by: Dario Faggioli <dario.faggioli@citrix.com>
>>      ---
>>      Cc: George Dunlap <george.dunlap@citrix.com>
>>      Cc: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
>>      Cc: Varun.Swara@arm.com
>>      Cc: Steve Capper <Steve.Capper@arm.com>
>>
>> diff --git a/xen/common/sched_credit.c b/xen/common/sched_credit.c
>> index bc36837..0a6a1b4 100644
>> --- a/xen/common/sched_credit.c
>> +++ b/xen/common/sched_credit.c
>> @@ -482,15 +482,25 @@ static inline void __runq_tickle(struct csched_vcpu
>> *new)
>>   }
>>
>>   static void
>> -csched_free_pdata(const struct scheduler *ops, void *pcpu, int cpu)
>> +csched_free_pdata(const struct scheduler *ops, void *pcpu)
>>   {
>> -    struct csched_private *prv = CSCHED_PRIV(ops);
>>       struct csched_pcpu *spc = pcpu;
>> -    unsigned long flags;
>>
>>       if ( spc == NULL )
>>           return;
>>
>> +    xfree(spc);
>> +}
>> +
>> +static void
>> +csched_deinit_pdata(const struct scheduler *ops, void *pcpu, int cpu)
>> +{
>> +    struct csched_private *prv = CSCHED_PRIV(ops);
>> +    struct csched_pcpu *spc = pcpu;
>> +    unsigned long flags;
>> +
>> +    ASSERT(spc != NULL);
>> +
>>       spin_lock_irqsave(&prv->lock, flags);
>>
>>       prv->credit -= prv->credits_per_tslice;
>> @@ -507,8 +517,6 @@ csched_free_pdata(const struct scheduler *ops, void
>> *pcpu, int cpu)
>>           kill_timer(&prv->master_ticker);
>>
>>       spin_unlock_irqrestore(&prv->lock, flags);
>> -
>> -    xfree(spc);
>>   }
>>
>>   static void *
>> @@ -2091,6 +2099,7 @@ static const struct scheduler sched_credit_def = {
>>       .free_vdata     = csched_free_vdata,
>>       .alloc_pdata    = csched_alloc_pdata,
>>       .init_pdata     = csched_init_pdata,
>> +    .deinit_pdata   = csched_deinit_pdata,
>>       .free_pdata     = csched_free_pdata,
>>       .switch_sched   = csched_switch_sched,
>>       .alloc_domdata  = csched_alloc_domdata,
>> diff --git a/xen/common/sched_credit2.c b/xen/common/sched_credit2.c
>> index 46b9279..f4c37b4 100644
>> --- a/xen/common/sched_credit2.c
>> +++ b/xen/common/sched_credit2.c
>> @@ -2261,13 +2261,15 @@ csched2_switch_sched(struct scheduler *new_ops,
>> unsigned int cpu,
>>   }
>>
>>   static void
>> -csched2_free_pdata(const struct scheduler *ops, void *pcpu, int cpu)
>> +csched2_deinit_pdata(const struct scheduler *ops, void *pcpu, int cpu)
>>   {
>>       unsigned long flags;
>>       struct csched2_private *prv = CSCHED2_PRIV(ops);
>>       struct csched2_runqueue_data *rqd;
>>       int rqi;
>>
>> +    ASSERT(pcpu == NULL);
>> +
>>       spin_lock_irqsave(&prv->lock, flags);
>>
>>       ASSERT(cpumask_test_cpu(cpu, &prv->initialized));
>> @@ -2387,7 +2389,7 @@ static const struct scheduler sched_credit2_def = {
>>       .alloc_vdata    = csched2_alloc_vdata,
>>       .free_vdata     = csched2_free_vdata,
>>       .init_pdata     = csched2_init_pdata,
>> -    .free_pdata     = csched2_free_pdata,
>> +    .deinit_pdata   = csched2_deinit_pdata,
>>       .switch_sched   = csched2_switch_sched,
>>       .alloc_domdata  = csched2_alloc_domdata,
>>       .free_domdata   = csched2_free_domdata,
>> diff --git a/xen/common/schedule.c b/xen/common/schedule.c
>> index 5546999..1a64521 100644
>> --- a/xen/common/schedule.c
>> +++ b/xen/common/schedule.c
>> @@ -1529,7 +1529,7 @@ static void cpu_schedule_down(unsigned int cpu)
>>       struct schedule_data *sd = &per_cpu(schedule_data, cpu);
>>       struct scheduler *sched = per_cpu(scheduler, cpu);
>>
>> -    SCHED_OP(sched, free_pdata, sd->sched_priv, cpu);
>> +    SCHED_OP(sched, free_pdata, sd->sched_priv);
>>       SCHED_OP(sched, free_vdata, idle_vcpu[cpu]->sched_priv);
>>
>>       idle_vcpu[cpu]->sched_priv = NULL;
>> @@ -1554,8 +1554,10 @@ static int cpu_schedule_callback(
>>       case CPU_UP_PREPARE:
>>           rc = cpu_schedule_up(cpu);
>>           break;
>> -    case CPU_UP_CANCELED:
>>       case CPU_DEAD:
>> +        SCHED_OP(sched, deinit_pdata, sd->sched_priv, cpu);
>> +        /* Fallthrough */
>> +    case CPU_UP_CANCELED:
>>           cpu_schedule_down(cpu);
>>           break;
>>       default:
>> @@ -1684,7 +1686,7 @@ int schedule_cpu_switch(unsigned int cpu, struct
>> cpupool *c)
>>       vpriv = SCHED_OP(new_ops, alloc_vdata, idle,
>> idle->domain->sched_priv);
>>       if ( vpriv == NULL )
>>       {
>> -        SCHED_OP(new_ops, free_pdata, ppriv, cpu);
>> +        SCHED_OP(new_ops, free_pdata, ppriv);
>>           return -ENOMEM;
>>       }
>>
>> @@ -1714,7 +1716,8 @@ int schedule_cpu_switch(unsigned int cpu, struct
>> cpupool *c)
>>       SCHED_OP(new_ops, tick_resume, cpu);
>>
>>       SCHED_OP(old_ops, free_vdata, vpriv_old);
>> -    SCHED_OP(old_ops, free_pdata, ppriv_old, cpu);
>> +    SCHED_OP(old_ops, deinit_pdata, ppriv_old, cpu);
>> +    SCHED_OP(old_ops, free_pdata, ppriv_old);
>>
>>    out:
>>       per_cpu(cpupool, cpu) = c;
>> diff --git a/xen/include/xen/sched-if.h b/xen/include/xen/sched-if.h
>> index 1db7c8d..240f66c 100644
>> --- a/xen/include/xen/sched-if.h
>> +++ b/xen/include/xen/sched-if.h
>> @@ -135,9 +135,10 @@ struct scheduler {
>>       void         (*free_vdata)     (const struct scheduler *, void *);
>>       void *       (*alloc_vdata)    (const struct scheduler *, struct
>> vcpu *,
>>                                       void *);
>> -    void         (*free_pdata)     (const struct scheduler *, void *,
>> int);
>> +    void         (*free_pdata)     (const struct scheduler *, void *);
>>       void *       (*alloc_pdata)    (const struct scheduler *, int);
>>       void         (*init_pdata)     (const struct scheduler *, void *,
>> int);
>> +    void         (*deinit_pdata)   (const struct scheduler *, void *,
>> int);
>>       void         (*free_domdata)   (const struct scheduler *, void *);
>>       void *       (*alloc_domdata)  (const struct scheduler *, struct
>> domain *);
>>
>>
>
> --
> Julien Grall
>
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xen.org
> http://lists.xen.org/xen-devel
Julien Grall May 3, 2016, 1:22 p.m. UTC | #7
Hi George,

On 03/05/16 14:20, George Dunlap wrote:
> On Tue, May 3, 2016 at 2:03 PM, Julien Grall <julien.grall@arm.com> wrote:
>> Hi Dario and George,
>>
>> What is the status of this patch? It would be nice to have it for Xen 4.7 to
>> avoid unwanted crash when secondary CPUs fails to come online.
>
> Wei, can you put this on your release blockers list?  (Julien, I take
> it this should be a blocker, right?)

Correct, this is a regression compare to Xen 4.6.

Regards,

>   -George
>
>>
>> Regards,
>>
>> On 26/04/16 18:49, Dario Faggioli wrote:
>>>
>>> On Tue, 2016-04-26 at 15:25 +0100, Julien Grall wrote:
>>>>
>>>> Hi Dario,
>>>>
>>> Hi,
>>>
>>>> A couple of people have been reported Xen crash on the ARM64
>>>> Foundation Model [1] with recent unstable.
>>>>
>>> Ok, thanks for reporting.
>>>
>>>> The crash seems to happen when Xen fails to bring up secondary CPUs
>>>> (see stack trace below).
>>>>
>>> Ah... I see.
>>>
>>>>   From my understanding, csched_free_pdata is trying to kill the
>>>> timer spc->ticker. However the status of this timer is
>>>> TIMER_STATUS_invalid.
>>>>
>>>> This is because csched_init_pdata has set a deadline for the
>>>> timer (set_timer) and the softirq to schedule the timer has
>>>> not yet happen (indeed Xen is still in early boot).
>>>>
>>> Yes, this is sort of what happens (only slightly different, see the
>>> changelog of the atached patch patch).
>>>
>>>
>>>> I am not sure how to fix this issue. How will you recommend
>>>> to fix it?
>>>>
>>> Yeah, well, doing it cleanly includes a slight change in the scheduler
>>> hooks API, IMO... and we indeed should do it cleanly :-))
>>>
>>> George, what do you think?
>>>
>>> Honestly, this is similar to what I was thinking to do already (I mean,
>>> having an deinit_pdata hook, "symmetric" with the init_pdata one), when
>>> working on that series, because I do think it's cleaner... then, I
>>> abandoned the idea, as it looked to not be necessary... But apparently
>>> it may actually be! :-)
>>>
>>> Let me know, and I'll resubmit the patch properly (together with
>>> another bugfix I have in my queue).
>>>
>>> Dario
>>> ---
>>> commit eca4d65fb67a71c0f6563aafbfdd68e566c53c32
>>> Author: Dario Faggioli <dario.faggioli@citrix.com>
>>> Date:   Tue Apr 26 17:42:36 2016 +0200
>>>
>>>       xen: sched: fix killing an uninitialized timer in free_pdata.
>>>
>>>       commit 64269d9365 "sched: implement .init_pdata in Credit,
>>>       Credit2 and RTDS" helped fixing Credit2 runqueues, and
>>>       the races we had in switching scheduler for pCPUs, but
>>>       introduced another issue. In fact, if CPU bringup fails
>>>       during __cpu_up() (and, more precisely, after CPU_UP_PREPARE,
>>>       but before CPU_STARTING) the CPU_UP_CANCELED notifier
>>>       would be executed, which calls the free_pdata hook.
>>>
>>>       Such hook does two things: (1) undo the initialization
>>>       done inside the init_pdata hook; (2) free the memory
>>>       allocated by the alloc_pdata hook.
>>>
>>>       However, in the failure path just described, it is possible
>>>       that only alloc_pdata has really been called, which is
>>>       potentially and issue (depending on what actually happens
>>>       inside the implementation of free_pdata).
>>>
>>>       In fact, for Credit1 (the only scheduler that actually
>>>       allocates per-pCPU data), this result in calling kill_timer()
>>>       on a timer that had not yet been initialized, which causes
>>>       the following:
>>>
>>>       (XEN) Xen call trace:
>>>       (XEN)    [<000000000022e304>] timer.c#active_timer+0x8/0x24 (PC)
>>>       (XEN)    [<000000000022f624>] kill_timer+0x108/0x2e0 (LR)
>>>       (XEN)    [<00000000002208c0>]
>>> sched_credit.c#csched_free_pdata+0xd8/0x114
>>>       (XEN)    [<0000000000227a18>]
>>> schedule.c#cpu_schedule_callback+0xc0/0x12c
>>>       (XEN)    [<0000000000219944>] notifier_call_chain+0x78/0x9c
>>>       (XEN)    [<00000000002015fc>] cpu_up+0x104/0x130
>>>       (XEN)    [<000000000028f7c0>] start_xen+0xaf8/0xce0
>>>       (XEN)    [<00000000810021d8>] 00000000810021d8
>>>       (XEN)
>>>       (XEN)
>>>       (XEN) ****************************************
>>>       (XEN) Panic on CPU 0:
>>>       (XEN) Assertion 'timer->status >= TIMER_STATUS_inactive' failed at
>>> timer.c:279
>>>       (XEN) ****************************************
>>>
>>>       Solve this by making the scheduler hooks API symmetric again,
>>>       i.e., by adding an deinit_pdata hook and making it responsible
>>>       of undoing what init_pdata did, rather than asking to free_pdata
>>>       to do everything.
>>>
>>>       This is cleaner and, in the case at hand, makes it possible to
>>>       only call free_pdata, which is the right thing to do, as only
>>>       allocation and no initialization was performed.
>>>
>>>       Reported-by: Julien Grall <julien.grall@arm.com>
>>>       Signed-off-by: Dario Faggioli <dario.faggioli@citrix.com>
>>>       ---
>>>       Cc: George Dunlap <george.dunlap@citrix.com>
>>>       Cc: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
>>>       Cc: Varun.Swara@arm.com
>>>       Cc: Steve Capper <Steve.Capper@arm.com>
>>>
>>> diff --git a/xen/common/sched_credit.c b/xen/common/sched_credit.c
>>> index bc36837..0a6a1b4 100644
>>> --- a/xen/common/sched_credit.c
>>> +++ b/xen/common/sched_credit.c
>>> @@ -482,15 +482,25 @@ static inline void __runq_tickle(struct csched_vcpu
>>> *new)
>>>    }
>>>
>>>    static void
>>> -csched_free_pdata(const struct scheduler *ops, void *pcpu, int cpu)
>>> +csched_free_pdata(const struct scheduler *ops, void *pcpu)
>>>    {
>>> -    struct csched_private *prv = CSCHED_PRIV(ops);
>>>        struct csched_pcpu *spc = pcpu;
>>> -    unsigned long flags;
>>>
>>>        if ( spc == NULL )
>>>            return;
>>>
>>> +    xfree(spc);
>>> +}
>>> +
>>> +static void
>>> +csched_deinit_pdata(const struct scheduler *ops, void *pcpu, int cpu)
>>> +{
>>> +    struct csched_private *prv = CSCHED_PRIV(ops);
>>> +    struct csched_pcpu *spc = pcpu;
>>> +    unsigned long flags;
>>> +
>>> +    ASSERT(spc != NULL);
>>> +
>>>        spin_lock_irqsave(&prv->lock, flags);
>>>
>>>        prv->credit -= prv->credits_per_tslice;
>>> @@ -507,8 +517,6 @@ csched_free_pdata(const struct scheduler *ops, void
>>> *pcpu, int cpu)
>>>            kill_timer(&prv->master_ticker);
>>>
>>>        spin_unlock_irqrestore(&prv->lock, flags);
>>> -
>>> -    xfree(spc);
>>>    }
>>>
>>>    static void *
>>> @@ -2091,6 +2099,7 @@ static const struct scheduler sched_credit_def = {
>>>        .free_vdata     = csched_free_vdata,
>>>        .alloc_pdata    = csched_alloc_pdata,
>>>        .init_pdata     = csched_init_pdata,
>>> +    .deinit_pdata   = csched_deinit_pdata,
>>>        .free_pdata     = csched_free_pdata,
>>>        .switch_sched   = csched_switch_sched,
>>>        .alloc_domdata  = csched_alloc_domdata,
>>> diff --git a/xen/common/sched_credit2.c b/xen/common/sched_credit2.c
>>> index 46b9279..f4c37b4 100644
>>> --- a/xen/common/sched_credit2.c
>>> +++ b/xen/common/sched_credit2.c
>>> @@ -2261,13 +2261,15 @@ csched2_switch_sched(struct scheduler *new_ops,
>>> unsigned int cpu,
>>>    }
>>>
>>>    static void
>>> -csched2_free_pdata(const struct scheduler *ops, void *pcpu, int cpu)
>>> +csched2_deinit_pdata(const struct scheduler *ops, void *pcpu, int cpu)
>>>    {
>>>        unsigned long flags;
>>>        struct csched2_private *prv = CSCHED2_PRIV(ops);
>>>        struct csched2_runqueue_data *rqd;
>>>        int rqi;
>>>
>>> +    ASSERT(pcpu == NULL);
>>> +
>>>        spin_lock_irqsave(&prv->lock, flags);
>>>
>>>        ASSERT(cpumask_test_cpu(cpu, &prv->initialized));
>>> @@ -2387,7 +2389,7 @@ static const struct scheduler sched_credit2_def = {
>>>        .alloc_vdata    = csched2_alloc_vdata,
>>>        .free_vdata     = csched2_free_vdata,
>>>        .init_pdata     = csched2_init_pdata,
>>> -    .free_pdata     = csched2_free_pdata,
>>> +    .deinit_pdata   = csched2_deinit_pdata,
>>>        .switch_sched   = csched2_switch_sched,
>>>        .alloc_domdata  = csched2_alloc_domdata,
>>>        .free_domdata   = csched2_free_domdata,
>>> diff --git a/xen/common/schedule.c b/xen/common/schedule.c
>>> index 5546999..1a64521 100644
>>> --- a/xen/common/schedule.c
>>> +++ b/xen/common/schedule.c
>>> @@ -1529,7 +1529,7 @@ static void cpu_schedule_down(unsigned int cpu)
>>>        struct schedule_data *sd = &per_cpu(schedule_data, cpu);
>>>        struct scheduler *sched = per_cpu(scheduler, cpu);
>>>
>>> -    SCHED_OP(sched, free_pdata, sd->sched_priv, cpu);
>>> +    SCHED_OP(sched, free_pdata, sd->sched_priv);
>>>        SCHED_OP(sched, free_vdata, idle_vcpu[cpu]->sched_priv);
>>>
>>>        idle_vcpu[cpu]->sched_priv = NULL;
>>> @@ -1554,8 +1554,10 @@ static int cpu_schedule_callback(
>>>        case CPU_UP_PREPARE:
>>>            rc = cpu_schedule_up(cpu);
>>>            break;
>>> -    case CPU_UP_CANCELED:
>>>        case CPU_DEAD:
>>> +        SCHED_OP(sched, deinit_pdata, sd->sched_priv, cpu);
>>> +        /* Fallthrough */
>>> +    case CPU_UP_CANCELED:
>>>            cpu_schedule_down(cpu);
>>>            break;
>>>        default:
>>> @@ -1684,7 +1686,7 @@ int schedule_cpu_switch(unsigned int cpu, struct
>>> cpupool *c)
>>>        vpriv = SCHED_OP(new_ops, alloc_vdata, idle,
>>> idle->domain->sched_priv);
>>>        if ( vpriv == NULL )
>>>        {
>>> -        SCHED_OP(new_ops, free_pdata, ppriv, cpu);
>>> +        SCHED_OP(new_ops, free_pdata, ppriv);
>>>            return -ENOMEM;
>>>        }
>>>
>>> @@ -1714,7 +1716,8 @@ int schedule_cpu_switch(unsigned int cpu, struct
>>> cpupool *c)
>>>        SCHED_OP(new_ops, tick_resume, cpu);
>>>
>>>        SCHED_OP(old_ops, free_vdata, vpriv_old);
>>> -    SCHED_OP(old_ops, free_pdata, ppriv_old, cpu);
>>> +    SCHED_OP(old_ops, deinit_pdata, ppriv_old, cpu);
>>> +    SCHED_OP(old_ops, free_pdata, ppriv_old);
>>>
>>>     out:
>>>        per_cpu(cpupool, cpu) = c;
>>> diff --git a/xen/include/xen/sched-if.h b/xen/include/xen/sched-if.h
>>> index 1db7c8d..240f66c 100644
>>> --- a/xen/include/xen/sched-if.h
>>> +++ b/xen/include/xen/sched-if.h
>>> @@ -135,9 +135,10 @@ struct scheduler {
>>>        void         (*free_vdata)     (const struct scheduler *, void *);
>>>        void *       (*alloc_vdata)    (const struct scheduler *, struct
>>> vcpu *,
>>>                                        void *);
>>> -    void         (*free_pdata)     (const struct scheduler *, void *,
>>> int);
>>> +    void         (*free_pdata)     (const struct scheduler *, void *);
>>>        void *       (*alloc_pdata)    (const struct scheduler *, int);
>>>        void         (*init_pdata)     (const struct scheduler *, void *,
>>> int);
>>> +    void         (*deinit_pdata)   (const struct scheduler *, void *,
>>> int);
>>>        void         (*free_domdata)   (const struct scheduler *, void *);
>>>        void *       (*alloc_domdata)  (const struct scheduler *, struct
>>> domain *);
>>>
>>>
>>
>> --
>> Julien Grall
>>
>> _______________________________________________
>> Xen-devel mailing list
>> Xen-devel@lists.xen.org
>> http://lists.xen.org/xen-devel
>
Wei Liu May 3, 2016, 1:23 p.m. UTC | #8
On Tue, May 03, 2016 at 02:20:05PM +0100, George Dunlap wrote:
> On Tue, May 3, 2016 at 2:03 PM, Julien Grall <julien.grall@arm.com> wrote:
> > Hi Dario and George,
> >
> > What is the status of this patch? It would be nice to have it for Xen 4.7 to
> > avoid unwanted crash when secondary CPUs fails to come online.
> 
> Wei, can you put this on your release blockers list?  (Julien, I take
> it this should be a blocker, right?)
> 

It's already on my list.

Wei.
Dario Faggioli May 3, 2016, 9:52 p.m. UTC | #9
On Tue, 2016-05-03 at 14:23 +0100, Wei Liu wrote:
> On Tue, May 03, 2016 at 02:20:05PM +0100, George Dunlap wrote:
> > 
> > On Tue, May 3, 2016 at 2:03 PM, Julien Grall <julien.grall@arm.com>
> > wrote:
> > > 
> > > Hi Dario and George,
> > > 
> > > What is the status of this patch? It would be nice to have it for
> > > Xen 4.7 to
> > > avoid unwanted crash when secondary CPUs fails to come online.
> > Wei, can you put this on your release blockers list?  (Julien, I
> > take
> > it this should be a blocker, right?)
> > 
> It's already on my list.
> 
Here it comes, as a part of a small series:
 http://lists.xen.org/archives/html/xen-devel/2016-05/msg00243.html

Sorry it took a bit but:
 1. I found a few other issues, an wanted to group them in a series;
 2. I tried to do what George asked, so document and put ASSERT()-s to
    make things clear and understandable.

Regards,
Dario
diff mbox

Patch

commit eca4d65fb67a71c0f6563aafbfdd68e566c53c32
Author: Dario Faggioli <dario.faggioli@citrix.com>
Date:   Tue Apr 26 17:42:36 2016 +0200

    xen: sched: fix killing an uninitialized timer in free_pdata.
    
    commit 64269d9365 "sched: implement .init_pdata in Credit,
    Credit2 and RTDS" helped fixing Credit2 runqueues, and
    the races we had in switching scheduler for pCPUs, but
    introduced another issue. In fact, if CPU bringup fails
    during __cpu_up() (and, more precisely, after CPU_UP_PREPARE,
    but before CPU_STARTING) the CPU_UP_CANCELED notifier
    would be executed, which calls the free_pdata hook.
    
    Such hook does two things: (1) undo the initialization
    done inside the init_pdata hook; (2) free the memory
    allocated by the alloc_pdata hook.
    
    However, in the failure path just described, it is possible
    that only alloc_pdata has really been called, which is
    potentially and issue (depending on what actually happens
    inside the implementation of free_pdata).
    
    In fact, for Credit1 (the only scheduler that actually
    allocates per-pCPU data), this result in calling kill_timer()
    on a timer that had not yet been initialized, which causes
    the following:
    
    (XEN) Xen call trace:
    (XEN)    [<000000000022e304>] timer.c#active_timer+0x8/0x24 (PC)
    (XEN)    [<000000000022f624>] kill_timer+0x108/0x2e0 (LR)
    (XEN)    [<00000000002208c0>] sched_credit.c#csched_free_pdata+0xd8/0x114
    (XEN)    [<0000000000227a18>] schedule.c#cpu_schedule_callback+0xc0/0x12c
    (XEN)    [<0000000000219944>] notifier_call_chain+0x78/0x9c
    (XEN)    [<00000000002015fc>] cpu_up+0x104/0x130
    (XEN)    [<000000000028f7c0>] start_xen+0xaf8/0xce0
    (XEN)    [<00000000810021d8>] 00000000810021d8
    (XEN)
    (XEN)
    (XEN) ****************************************
    (XEN) Panic on CPU 0:
    (XEN) Assertion 'timer->status >= TIMER_STATUS_inactive' failed at timer.c:279
    (XEN) ****************************************
    
    Solve this by making the scheduler hooks API symmetric again,
    i.e., by adding an deinit_pdata hook and making it responsible
    of undoing what init_pdata did, rather than asking to free_pdata
    to do everything.
    
    This is cleaner and, in the case at hand, makes it possible to
    only call free_pdata, which is the right thing to do, as only
    allocation and no initialization was performed.
    
    Reported-by: Julien Grall <julien.grall@arm.com>
    Signed-off-by: Dario Faggioli <dario.faggioli@citrix.com>
    ---
    Cc: George Dunlap <george.dunlap@citrix.com>
    Cc: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
    Cc: Varun.Swara@arm.com
    Cc: Steve Capper <Steve.Capper@arm.com>

diff --git a/xen/common/sched_credit.c b/xen/common/sched_credit.c
index bc36837..0a6a1b4 100644
--- a/xen/common/sched_credit.c
+++ b/xen/common/sched_credit.c
@@ -482,15 +482,25 @@  static inline void __runq_tickle(struct csched_vcpu *new)
 }
 
 static void
-csched_free_pdata(const struct scheduler *ops, void *pcpu, int cpu)
+csched_free_pdata(const struct scheduler *ops, void *pcpu)
 {
-    struct csched_private *prv = CSCHED_PRIV(ops);
     struct csched_pcpu *spc = pcpu;
-    unsigned long flags;
 
     if ( spc == NULL )
         return;
 
+    xfree(spc);
+}
+
+static void
+csched_deinit_pdata(const struct scheduler *ops, void *pcpu, int cpu)
+{
+    struct csched_private *prv = CSCHED_PRIV(ops);
+    struct csched_pcpu *spc = pcpu;
+    unsigned long flags;
+
+    ASSERT(spc != NULL);
+
     spin_lock_irqsave(&prv->lock, flags);
 
     prv->credit -= prv->credits_per_tslice;
@@ -507,8 +517,6 @@  csched_free_pdata(const struct scheduler *ops, void *pcpu, int cpu)
         kill_timer(&prv->master_ticker);
 
     spin_unlock_irqrestore(&prv->lock, flags);
-
-    xfree(spc);
 }
 
 static void *
@@ -2091,6 +2099,7 @@  static const struct scheduler sched_credit_def = {
     .free_vdata     = csched_free_vdata,
     .alloc_pdata    = csched_alloc_pdata,
     .init_pdata     = csched_init_pdata,
+    .deinit_pdata   = csched_deinit_pdata,
     .free_pdata     = csched_free_pdata,
     .switch_sched   = csched_switch_sched,
     .alloc_domdata  = csched_alloc_domdata,
diff --git a/xen/common/sched_credit2.c b/xen/common/sched_credit2.c
index 46b9279..f4c37b4 100644
--- a/xen/common/sched_credit2.c
+++ b/xen/common/sched_credit2.c
@@ -2261,13 +2261,15 @@  csched2_switch_sched(struct scheduler *new_ops, unsigned int cpu,
 }
 
 static void
-csched2_free_pdata(const struct scheduler *ops, void *pcpu, int cpu)
+csched2_deinit_pdata(const struct scheduler *ops, void *pcpu, int cpu)
 {
     unsigned long flags;
     struct csched2_private *prv = CSCHED2_PRIV(ops);
     struct csched2_runqueue_data *rqd;
     int rqi;
 
+    ASSERT(pcpu == NULL);
+
     spin_lock_irqsave(&prv->lock, flags);
 
     ASSERT(cpumask_test_cpu(cpu, &prv->initialized));
@@ -2387,7 +2389,7 @@  static const struct scheduler sched_credit2_def = {
     .alloc_vdata    = csched2_alloc_vdata,
     .free_vdata     = csched2_free_vdata,
     .init_pdata     = csched2_init_pdata,
-    .free_pdata     = csched2_free_pdata,
+    .deinit_pdata   = csched2_deinit_pdata,
     .switch_sched   = csched2_switch_sched,
     .alloc_domdata  = csched2_alloc_domdata,
     .free_domdata   = csched2_free_domdata,
diff --git a/xen/common/schedule.c b/xen/common/schedule.c
index 5546999..1a64521 100644
--- a/xen/common/schedule.c
+++ b/xen/common/schedule.c
@@ -1529,7 +1529,7 @@  static void cpu_schedule_down(unsigned int cpu)
     struct schedule_data *sd = &per_cpu(schedule_data, cpu);
     struct scheduler *sched = per_cpu(scheduler, cpu);
 
-    SCHED_OP(sched, free_pdata, sd->sched_priv, cpu);
+    SCHED_OP(sched, free_pdata, sd->sched_priv);
     SCHED_OP(sched, free_vdata, idle_vcpu[cpu]->sched_priv);
 
     idle_vcpu[cpu]->sched_priv = NULL;
@@ -1554,8 +1554,10 @@  static int cpu_schedule_callback(
     case CPU_UP_PREPARE:
         rc = cpu_schedule_up(cpu);
         break;
-    case CPU_UP_CANCELED:
     case CPU_DEAD:
+        SCHED_OP(sched, deinit_pdata, sd->sched_priv, cpu);
+        /* Fallthrough */
+    case CPU_UP_CANCELED:
         cpu_schedule_down(cpu);
         break;
     default:
@@ -1684,7 +1686,7 @@  int schedule_cpu_switch(unsigned int cpu, struct cpupool *c)
     vpriv = SCHED_OP(new_ops, alloc_vdata, idle, idle->domain->sched_priv);
     if ( vpriv == NULL )
     {
-        SCHED_OP(new_ops, free_pdata, ppriv, cpu);
+        SCHED_OP(new_ops, free_pdata, ppriv);
         return -ENOMEM;
     }
 
@@ -1714,7 +1716,8 @@  int schedule_cpu_switch(unsigned int cpu, struct cpupool *c)
     SCHED_OP(new_ops, tick_resume, cpu);
 
     SCHED_OP(old_ops, free_vdata, vpriv_old);
-    SCHED_OP(old_ops, free_pdata, ppriv_old, cpu);
+    SCHED_OP(old_ops, deinit_pdata, ppriv_old, cpu);
+    SCHED_OP(old_ops, free_pdata, ppriv_old);
 
  out:
     per_cpu(cpupool, cpu) = c;
diff --git a/xen/include/xen/sched-if.h b/xen/include/xen/sched-if.h
index 1db7c8d..240f66c 100644
--- a/xen/include/xen/sched-if.h
+++ b/xen/include/xen/sched-if.h
@@ -135,9 +135,10 @@  struct scheduler {
     void         (*free_vdata)     (const struct scheduler *, void *);
     void *       (*alloc_vdata)    (const struct scheduler *, struct vcpu *,
                                     void *);
-    void         (*free_pdata)     (const struct scheduler *, void *, int);
+    void         (*free_pdata)     (const struct scheduler *, void *);
     void *       (*alloc_pdata)    (const struct scheduler *, int);
     void         (*init_pdata)     (const struct scheduler *, void *, int);
+    void         (*deinit_pdata)   (const struct scheduler *, void *, int);
     void         (*free_domdata)   (const struct scheduler *, void *);
     void *       (*alloc_domdata)  (const struct scheduler *, struct domain *);