Re: arm-smmu-v3 high cpu usage for NVMe

From: Marc Zyngier <maz@kernel.org>
To: John Garry <john.garry@huawei.com>
Cc: Jean-Philippe Brucker <jean-philippe@linaro.org>,
	Robin Murphy <robin.murphy@arm.com>,
	Ming Lei <ming.lei@redhat.com>,
	iommu@lists.linux-foundation.org, Will Deacon <will@kernel.org>
Subject: Re: arm-smmu-v3 high cpu usage for NVMe
Date: Mon, 23 Mar 2020 09:16:35 +0000	[thread overview]
Message-ID: <dd375cf6bffacd82174c84a4c7d46283@kernel.org> (raw)
In-Reply-To: <cca67355-672d-81c5-3d37-72004eb8f14f@huawei.com>

On 2020-03-23 09:03, John Garry wrote:
> On 20/03/2020 16:33, Marc Zyngier wrote:
>>> JFYI, I've been playing for "perf annotate" today and it's giving
>>> strange results for my NVMe testing. So "report" looks somewhat sane,
>>> if not a worryingly high % for arm_smmu_cmdq_issue_cmdlist():
>>> 
>>> 
>>>     55.39%  irq/342-nvme0q1  [kernel.kallsyms]  [k] 
>>> arm_smmu_cmdq_issue_cmdlist
>>>      9.74%  irq/342-nvme0q1  [kernel.kallsyms]  [k] 
>>> _raw_spin_unlock_irqrestore
>>>      2.02%  irq/342-nvme0q1  [kernel.kallsyms]  [k] nvme_irq
>>>      1.86%  irq/342-nvme0q1  [kernel.kallsyms]  [k] fput_many
>>>      1.73%  irq/342-nvme0q1  [kernel.kallsyms]  [k]
>>> arm_smmu_atc_inv_domain.constprop.42
>>>      1.67%  irq/342-nvme0q1  [kernel.kallsyms]  [k] __arm_lpae_unmap
>>>      1.49%  irq/342-nvme0q1  [kernel.kallsyms]  [k] aio_complete_rw
>>> 
>>> But "annotate" consistently tells me that a specific instruction
>>> consumes ~99% of the load for the enqueue function:
>>> 
>>>          :                      /* 5. If we are inserting a CMD_SYNC,
>>> we must wait for it to complete */
>>>          :                      if (sync) {
>>>     0.00 :   ffff80001071c948:       ldr     w0, [x29, #108]
>>>          :                      int ret = 0;
>>>     0.00 :   ffff80001071c94c:       mov     w24, #0x0      // #0
>>>          :                      if (sync) {
>>>     0.00 :   ffff80001071c950:       cbnz    w0, ffff80001071c990
>>> <arm_smmu_cmdq_issue_cmdlist+0x420>
>>>          :                      arch_local_irq_restore():
>>>     0.00 :   ffff80001071c954:       msr     daif, x21
>>>          :                      arm_smmu_cmdq_issue_cmdlist():
>>>          :                      }
>>>          :                      }
>>>          :
>>>          :                      local_irq_restore(flags);
>>>          :                      return ret;
>>>          :                      }
>>>    99.51 :   ffff80001071c958:       adrp    x0, ffff800011909000
>>> <page_wait_table+0x14c0>
>> 
> 
> Hi Marc,
> 
>> This is likely the side effect of the re-enabling of interrupts (msr 
>> daif, x21)
>> on the previous instruction which causes the perf interrupt to fire 
>> right after.
> 
> ok, makes sense.
> 
>> 
>> Time to enable pseudo-NMIs in the PMUv3 driver...
>> 
> 
> Do you know if there is any plan for this?

There was. Julien Thierry has a bunch of patches for that [1], but they 
needs
reviving.

> 
> In the meantime, maybe I can do some trickery by putting the
> local_irq_restore() in a separate function, outside
> arm_smmu_cmdq_issue_cmdlist(), to get a fair profile for that same
> function.

I don't see how you can improve the profiling without compromising
the locking in this case...

Thanks,

         M.

[1] https://patchwork.kernel.org/cover/11047407/
-- 
Jazz is not dead. It just smells funny...
_______________________________________________
iommu mailing list
iommu@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/iommu