Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1680263 > unrolled thread

Re: [PATCH] irqchip: gicv3-its: Use NUMA aware memory allocation for ITS tables

Back to article view | Back to linux.kernel

This discussion starts older than the indexed window; earlier articles aren't shown. The article labeled Started by below is the oldest one visible, not the original post.

Contents

  Re: [PATCH] irqchip: gicv3-its: Use NUMA aware memory allocation for  ITS tables Shanker Donthineni <shankerd@codeaurora.org> - 2017-07-03 16:30 +0200
    Re: [PATCH] irqchip: gicv3-its: Use NUMA aware memory allocation for  ITS tables Marc Zyngier <marc.zyngier@arm.com> - 2017-07-03 17:00 +0200
      Re: [PATCH] irqchip: gicv3-its: Use NUMA aware memory allocation for  ITS tables Shanker Donthineni <shankerd@codeaurora.org> - 2017-07-03 17:20 +0200
      Re: [PATCH] irqchip: gicv3-its: Use NUMA aware memory allocation for  ITS tables Ganapatrao Kulkarni <gpkulkarni@gmail.com> - 2017-07-10 11:00 +0200
        Re: [PATCH] irqchip: gicv3-its: Use NUMA aware memory allocation for  ITS tables Ganapatrao Kulkarni <gpkulkarni@gmail.com> - 2017-07-10 11:10 +0200
          Re: [PATCH] irqchip: gicv3-its: Use NUMA aware memory allocation for  ITS tables Marc Zyngier <marc.zyngier@arm.com> - 2017-07-10 11:30 +0200
            Re: [PATCH] irqchip: gicv3-its: Use NUMA aware memory allocation for  ITS tables Ganapatrao Kulkarni <gpkulkarni@gmail.com> - 2017-07-10 12:30 +0200
              Re: [PATCH] irqchip: gicv3-its: Use NUMA aware memory allocation for  ITS tables Shanker Donthineni <shankerd@codeaurora.org> - 2017-07-10 14:40 +0200
                Re: [PATCH] irqchip: gicv3-its: Use NUMA aware memory allocation for  ITS tables Marc Zyngier <marc.zyngier@arm.com> - 2017-07-10 16:00 +0200
              Re: [PATCH] irqchip: gicv3-its: Use NUMA aware memory allocation for  ITS tables Marc Zyngier <marc.zyngier@arm.com> - 2017-07-10 16:00 +0200
                Re: [PATCH] irqchip: gicv3-its: Use NUMA aware memory allocation for  ITS tables Shanker Donthineni <shankerd@codeaurora.org> - 2017-07-10 17:00 +0200
                  Re: [PATCH] irqchip: gicv3-its: Use NUMA aware memory allocation for  ITS tables Marc Zyngier <marc.zyngier@arm.com> - 2017-07-10 17:20 +0200
        Re: [PATCH] irqchip: gicv3-its: Use NUMA aware memory allocation for  ITS tables Marc Zyngier <marc.zyngier@arm.com> - 2017-07-10 11:10 +0200

#1680263 — Re: [PATCH] irqchip: gicv3-its: Use NUMA aware memory allocation for ITS tables

Hi Marc,

On 06/30/2017 03:51 AM, Marc Zyngier wrote:
> On 30/06/17 04:01, Ganapatrao Kulkarni wrote:
>> On Fri, Jun 30, 2017 at 8:04 AM, Ganapatrao Kulkarni
>> <gpkulkarni@gmail.com> wrote:
>>> Hi Shanker,
>>>
>>> On Sun, Jun 25, 2017 at 9:16 PM, Shanker Donthineni
>>> <shankerd@codeaurora.org> wrote:
>>>> The NUMA node information is visible to ITS driver but not being used
>>>> other than handling errata. This patch allocates the memory for ITS
>>>> tables from the corresponding NUMA node using the appropriate NUMA
>>>> aware functions.
>>
>> IMHO, the description would have been more constructive?
>>
>> "All ITS tables are mapped by default to NODE 0 memory.
>> Adding changes to allocate memory from respective NUMA NODES of ITS devices.
>> This will optimize tables access and avoids unnecessary inter-node traffic."
> 
> But more importantly, I'd like to see figures showing the actual benefit
> of this per-node allocation. Given that both of you guys have access to
> such platforms, please show me the numbers!
> 

I'll share the actual results which shows the improvement whenever 
available on our next chips. Current version of Qualcomm qdf2400 doesn't 
support multi socket configuration to capture results and share with you. 

Do you see any other issues with this patch apart from the performance 
improvements. I strongly believe this brings the noticeable improvement 
in numbers on systems where it has multi node memory/CPU configuration.


> Thanks,
> 
> 	M.
> 

-- 
Shanker Donthineni
Qualcomm Datacenter Technologies, Inc. as an affiliate of Qualcomm Technologies, Inc.
Qualcomm Technologies, Inc. is a member of the Code Aurora Forum, a Linux Foundation Collaborative Project.

[toc] | [next] | [standalone]

#1680354

Hi Shanker,

On 03/07/17 15:24, Shanker Donthineni wrote:
> Hi Marc,
> 
> On 06/30/2017 03:51 AM, Marc Zyngier wrote:
>> On 30/06/17 04:01, Ganapatrao Kulkarni wrote:
>>> On Fri, Jun 30, 2017 at 8:04 AM, Ganapatrao Kulkarni
>>> <gpkulkarni@gmail.com> wrote:
>>>> Hi Shanker,
>>>>
>>>> On Sun, Jun 25, 2017 at 9:16 PM, Shanker Donthineni
>>>> <shankerd@codeaurora.org> wrote:
>>>>> The NUMA node information is visible to ITS driver but not being used
>>>>> other than handling errata. This patch allocates the memory for ITS
>>>>> tables from the corresponding NUMA node using the appropriate NUMA
>>>>> aware functions.
>>>
>>> IMHO, the description would have been more constructive?
>>>
>>> "All ITS tables are mapped by default to NODE 0 memory.
>>> Adding changes to allocate memory from respective NUMA NODES of ITS devices.
>>> This will optimize tables access and avoids unnecessary inter-node traffic."
>>
>> But more importantly, I'd like to see figures showing the actual benefit
>> of this per-node allocation. Given that both of you guys have access to
>> such platforms, please show me the numbers!
>>
> 
> I'll share the actual results which shows the improvement whenever 
> available on our next chips. Current version of Qualcomm qdf2400 doesn't 
> support multi socket configuration to capture results and share with you. 
> 
> Do you see any other issues with this patch apart from the performance 
> improvements. I strongly believe this brings the noticeable improvement 
> in numbers on systems where it has multi node memory/CPU configuration.

I agree that it *could* show an improvement, but it very much depends on
how often the ITS misses in its caches. For this kind of patches, I want
to see two things:

1) It brings a measurable benefit on NUMA platforms
2) it doesn't adversely impact non-NUMA systems

I can deal with (2), but I have no way of evaluating (1), mostly for the
lack of an infrastructure exercising multiple ITSs at the same time.

Thanks,

	M.
-- 
Jazz is not dead. It just smells funny...

[toc] | [prev] | [next] | [standalone]

#1680440

Hi Marc,

On 07/03/2017 09:53 AM, Marc Zyngier wrote:
> Hi Shanker,
> 
> On 03/07/17 15:24, Shanker Donthineni wrote:
>> Hi Marc,
>>
>> On 06/30/2017 03:51 AM, Marc Zyngier wrote:
>>> On 30/06/17 04:01, Ganapatrao Kulkarni wrote:
>>>> On Fri, Jun 30, 2017 at 8:04 AM, Ganapatrao Kulkarni
>>>> <gpkulkarni@gmail.com> wrote:
>>>>> Hi Shanker,
>>>>>
>>>>> On Sun, Jun 25, 2017 at 9:16 PM, Shanker Donthineni
>>>>> <shankerd@codeaurora.org> wrote:
>>>>>> The NUMA node information is visible to ITS driver but not being used
>>>>>> other than handling errata. This patch allocates the memory for ITS
>>>>>> tables from the corresponding NUMA node using the appropriate NUMA
>>>>>> aware functions.
>>>>
>>>> IMHO, the description would have been more constructive?
>>>>
>>>> "All ITS tables are mapped by default to NODE 0 memory.
>>>> Adding changes to allocate memory from respective NUMA NODES of ITS devices.
>>>> This will optimize tables access and avoids unnecessary inter-node traffic."
>>>
>>> But more importantly, I'd like to see figures showing the actual benefit
>>> of this per-node allocation. Given that both of you guys have access to
>>> such platforms, please show me the numbers!
>>>
>>
>> I'll share the actual results which shows the improvement whenever 
>> available on our next chips. Current version of Qualcomm qdf2400 doesn't 
>> support multi socket configuration to capture results and share with you. 
>>
>> Do you see any other issues with this patch apart from the performance 
>> improvements. I strongly believe this brings the noticeable improvement 
>> in numbers on systems where it has multi node memory/CPU configuration.
> 
> I agree that it *could* show an improvement, but it very much depends on
> how often the ITS misses in its caches. For this kind of patches, I want
> to see two things:
> 

Just imagine systems with hundreds of PCI-SRIOV virtual functions and
assigning some of them to virtual machines, and systems with GICv4 feature. 
There should be a lot of cache misses on ITS VCPU, DEVICE and COLLECTION 
lookups. And also VLPI patches that you've posted for comments are forcing 
to use VLPI feature for each VM irrespective of pass-through device assignment. 


> 1) It brings a measurable benefit on NUMA platforms
> 2) it doesn't adversely impact non-NUMA systems
> 
It should not affect the ITS hardware behavior non-NUMA based system since 
software always allocate memory from a single (default) NUMA node.

> I can deal with (2), but I have no way of evaluating (1), mostly for the
> lack of an infrastructure exercising multiple ITSs at the same time.
> 
Agree with you, but it takes some time for me to provide the test results. 

> Thanks,
> 
> 	M.
> 

-- 
Shanker Donthineni
Qualcomm Datacenter Technologies, Inc. as an affiliate of Qualcomm Technologies, Inc.
Qualcomm Technologies, Inc. is a member of the Code Aurora Forum, a Linux Foundation Collaborative Project.

[toc] | [prev] | [next] | [standalone]

#1684067

Hi Marc,

On Mon, Jul 3, 2017 at 8:23 PM, Marc Zyngier <marc.zyngier@arm.com> wrote:
> Hi Shanker,
>
> On 03/07/17 15:24, Shanker Donthineni wrote:
>> Hi Marc,
>>
>> On 06/30/2017 03:51 AM, Marc Zyngier wrote:
>>> On 30/06/17 04:01, Ganapatrao Kulkarni wrote:
>>>> On Fri, Jun 30, 2017 at 8:04 AM, Ganapatrao Kulkarni
>>>> <gpkulkarni@gmail.com> wrote:
>>>>> Hi Shanker,
>>>>>
>>>>> On Sun, Jun 25, 2017 at 9:16 PM, Shanker Donthineni
>>>>> <shankerd@codeaurora.org> wrote:
>>>>>> The NUMA node information is visible to ITS driver but not being used
>>>>>> other than handling errata. This patch allocates the memory for ITS
>>>>>> tables from the corresponding NUMA node using the appropriate NUMA
>>>>>> aware functions.
>>>>
>>>> IMHO, the description would have been more constructive?
>>>>
>>>> "All ITS tables are mapped by default to NODE 0 memory.
>>>> Adding changes to allocate memory from respective NUMA NODES of ITS devices.
>>>> This will optimize tables access and avoids unnecessary inter-node traffic."
>>>
>>> But more importantly, I'd like to see figures showing the actual benefit
>>> of this per-node allocation. Given that both of you guys have access to
>>> such platforms, please show me the numbers!
>>>
>>
>> I'll share the actual results which shows the improvement whenever
>> available on our next chips. Current version of Qualcomm qdf2400 doesn't
>> support multi socket configuration to capture results and share with you.
>>
>> Do you see any other issues with this patch apart from the performance
>> improvements. I strongly believe this brings the noticeable improvement
>> in numbers on systems where it has multi node memory/CPU configuration.
>
> I agree that it *could* show an improvement, but it very much depends on
> how often the ITS misses in its caches. For this kind of patches, I want
> to see two things:
>
> 1) It brings a measurable benefit on NUMA platforms

Did some measurement of interrupt response time for LPIs and we don't
see any major
improvement due to caching of Tables. However, we have seen
improvements of around 5%.
IMO, we should merge this patch to have NUMA aware allocations and to
avoid unwanted inter-node transactions.

Tested-by: Ganapatrao Kulkarni <ganapatrao.kulkarni@cavium.com>

> 2) it doesn't adversely impact non-NUMA systems
AFAIK, no impact on non-NUMA and on single node NUMA systems.

>
> I can deal with (2), but I have no way of evaluating (1), mostly for the
> lack of an infrastructure exercising multiple ITSs at the same time.
>
> Thanks,
>
>         M.
> --
> Jazz is not dead. It just smells funny...

thanks
Ganapat

[toc] | [prev] | [next] | [standalone]

#1684072

On Mon, Jul 10, 2017 at 2:36 PM, Marc Zyngier <marc.zyngier@arm.com> wrote:
> On 10/07/17 09:48, Ganapatrao Kulkarni wrote:
>> Hi Marc,
>>
>> On Mon, Jul 3, 2017 at 8:23 PM, Marc Zyngier <marc.zyngier@arm.com> wrote:
>>> Hi Shanker,
>>>
>>> On 03/07/17 15:24, Shanker Donthineni wrote:
>>>> Hi Marc,
>>>>
>>>> On 06/30/2017 03:51 AM, Marc Zyngier wrote:
>>>>> On 30/06/17 04:01, Ganapatrao Kulkarni wrote:
>>>>>> On Fri, Jun 30, 2017 at 8:04 AM, Ganapatrao Kulkarni
>>>>>> <gpkulkarni@gmail.com> wrote:
>>>>>>> Hi Shanker,
>>>>>>>
>>>>>>> On Sun, Jun 25, 2017 at 9:16 PM, Shanker Donthineni
>>>>>>> <shankerd@codeaurora.org> wrote:
>>>>>>>> The NUMA node information is visible to ITS driver but not being used
>>>>>>>> other than handling errata. This patch allocates the memory for ITS
>>>>>>>> tables from the corresponding NUMA node using the appropriate NUMA
>>>>>>>> aware functions.
>>>>>>
>>>>>> IMHO, the description would have been more constructive?
>>>>>>
>>>>>> "All ITS tables are mapped by default to NODE 0 memory.
>>>>>> Adding changes to allocate memory from respective NUMA NODES of ITS devices.
>>>>>> This will optimize tables access and avoids unnecessary inter-node traffic."
>>>>>
>>>>> But more importantly, I'd like to see figures showing the actual benefit
>>>>> of this per-node allocation. Given that both of you guys have access to
>>>>> such platforms, please show me the numbers!
>>>>>
>>>>
>>>> I'll share the actual results which shows the improvement whenever
>>>> available on our next chips. Current version of Qualcomm qdf2400 doesn't
>>>> support multi socket configuration to capture results and share with you.
>>>>
>>>> Do you see any other issues with this patch apart from the performance
>>>> improvements. I strongly believe this brings the noticeable improvement
>>>> in numbers on systems where it has multi node memory/CPU configuration.
>>>
>>> I agree that it *could* show an improvement, but it very much depends on
>>> how often the ITS misses in its caches. For this kind of patches, I want
>>> to see two things:
>>>
>>> 1) It brings a measurable benefit on NUMA platforms
>>
>> Did some measurement of interrupt response time for LPIs and we don't
>> see any major
>> improvement due to caching of Tables. However, we have seen
>> improvements of around 5%.
>
> An improvement of what exactly?

interrupt response time.
>
>         M.
> --
> Jazz is not dead. It just smells funny...

thanks
Ganapat

[toc] | [prev] | [next] | [standalone]

#1684081

On 10/07/17 10:08, Ganapatrao Kulkarni wrote:
> On Mon, Jul 10, 2017 at 2:36 PM, Marc Zyngier <marc.zyngier@arm.com> wrote:
>> On 10/07/17 09:48, Ganapatrao Kulkarni wrote:
>>> Hi Marc,
>>>
>>> On Mon, Jul 3, 2017 at 8:23 PM, Marc Zyngier <marc.zyngier@arm.com> wrote:
>>>> Hi Shanker,
>>>>
>>>> On 03/07/17 15:24, Shanker Donthineni wrote:
>>>>> Hi Marc,
>>>>>
>>>>> On 06/30/2017 03:51 AM, Marc Zyngier wrote:
>>>>>> On 30/06/17 04:01, Ganapatrao Kulkarni wrote:
>>>>>>> On Fri, Jun 30, 2017 at 8:04 AM, Ganapatrao Kulkarni
>>>>>>> <gpkulkarni@gmail.com> wrote:
>>>>>>>> Hi Shanker,
>>>>>>>>
>>>>>>>> On Sun, Jun 25, 2017 at 9:16 PM, Shanker Donthineni
>>>>>>>> <shankerd@codeaurora.org> wrote:
>>>>>>>>> The NUMA node information is visible to ITS driver but not being used
>>>>>>>>> other than handling errata. This patch allocates the memory for ITS
>>>>>>>>> tables from the corresponding NUMA node using the appropriate NUMA
>>>>>>>>> aware functions.
>>>>>>>
>>>>>>> IMHO, the description would have been more constructive?
>>>>>>>
>>>>>>> "All ITS tables are mapped by default to NODE 0 memory.
>>>>>>> Adding changes to allocate memory from respective NUMA NODES of ITS devices.
>>>>>>> This will optimize tables access and avoids unnecessary inter-node traffic."
>>>>>>
>>>>>> But more importantly, I'd like to see figures showing the actual benefit
>>>>>> of this per-node allocation. Given that both of you guys have access to
>>>>>> such platforms, please show me the numbers!
>>>>>>
>>>>>
>>>>> I'll share the actual results which shows the improvement whenever
>>>>> available on our next chips. Current version of Qualcomm qdf2400 doesn't
>>>>> support multi socket configuration to capture results and share with you.
>>>>>
>>>>> Do you see any other issues with this patch apart from the performance
>>>>> improvements. I strongly believe this brings the noticeable improvement
>>>>> in numbers on systems where it has multi node memory/CPU configuration.
>>>>
>>>> I agree that it *could* show an improvement, but it very much depends on
>>>> how often the ITS misses in its caches. For this kind of patches, I want
>>>> to see two things:
>>>>
>>>> 1) It brings a measurable benefit on NUMA platforms
>>>
>>> Did some measurement of interrupt response time for LPIs and we don't
>>> see any major
>>> improvement due to caching of Tables. However, we have seen
>>> improvements of around 5%.
>>
>> An improvement of what exactly?
> 
> interrupt response time.

Measured how? On which HW? Using which benchmark?

Give me the actual benchmark results. Don't expect me to accept this
kind of hand-wavy statement.

	M.
-- 
Jazz is not dead. It just smells funny...

[toc] | [prev] | [next] | [standalone]

#1684122

Hi Marc,

On Mon, Jul 10, 2017 at 2:53 PM, Marc Zyngier <marc.zyngier@arm.com> wrote:
> On 10/07/17 10:08, Ganapatrao Kulkarni wrote:
>> On Mon, Jul 10, 2017 at 2:36 PM, Marc Zyngier <marc.zyngier@arm.com> wrote:
>>> On 10/07/17 09:48, Ganapatrao Kulkarni wrote:
>>>> Hi Marc,
>>>>
>>>> On Mon, Jul 3, 2017 at 8:23 PM, Marc Zyngier <marc.zyngier@arm.com> wrote:
>>>>> Hi Shanker,
>>>>>
>>>>> On 03/07/17 15:24, Shanker Donthineni wrote:
>>>>>> Hi Marc,
>>>>>>
>>>>>> On 06/30/2017 03:51 AM, Marc Zyngier wrote:
>>>>>>> On 30/06/17 04:01, Ganapatrao Kulkarni wrote:
>>>>>>>> On Fri, Jun 30, 2017 at 8:04 AM, Ganapatrao Kulkarni
>>>>>>>> <gpkulkarni@gmail.com> wrote:
>>>>>>>>> Hi Shanker,
>>>>>>>>>
>>>>>>>>> On Sun, Jun 25, 2017 at 9:16 PM, Shanker Donthineni
>>>>>>>>> <shankerd@codeaurora.org> wrote:
>>>>>>>>>> The NUMA node information is visible to ITS driver but not being used
>>>>>>>>>> other than handling errata. This patch allocates the memory for ITS
>>>>>>>>>> tables from the corresponding NUMA node using the appropriate NUMA
>>>>>>>>>> aware functions.
>>>>>>>>
>>>>>>>> IMHO, the description would have been more constructive?
>>>>>>>>
>>>>>>>> "All ITS tables are mapped by default to NODE 0 memory.
>>>>>>>> Adding changes to allocate memory from respective NUMA NODES of ITS devices.
>>>>>>>> This will optimize tables access and avoids unnecessary inter-node traffic."
>>>>>>>
>>>>>>> But more importantly, I'd like to see figures showing the actual benefit
>>>>>>> of this per-node allocation. Given that both of you guys have access to
>>>>>>> such platforms, please show me the numbers!
>>>>>>>
>>>>>>
>>>>>> I'll share the actual results which shows the improvement whenever
>>>>>> available on our next chips. Current version of Qualcomm qdf2400 doesn't
>>>>>> support multi socket configuration to capture results and share with you.
>>>>>>
>>>>>> Do you see any other issues with this patch apart from the performance
>>>>>> improvements. I strongly believe this brings the noticeable improvement
>>>>>> in numbers on systems where it has multi node memory/CPU configuration.
>>>>>
>>>>> I agree that it *could* show an improvement, but it very much depends on
>>>>> how often the ITS misses in its caches. For this kind of patches, I want
>>>>> to see two things:
>>>>>
>>>>> 1) It brings a measurable benefit on NUMA platforms
>>>>
>>>> Did some measurement of interrupt response time for LPIs and we don't
>>>> see any major
>>>> improvement due to caching of Tables. However, we have seen
>>>> improvements of around 5%.
>>>
>>> An improvement of what exactly?
>>
>> interrupt response time.
>
> Measured how? On which HW? Using which benchmark?

This has been tested on ThunderX2.
We have instrumented gic-v3-its driver code to create dummy LPI device
with few vectors.
The LPI is induced from dummy device(through sysfs by writing to
TRANSLATOR reg).
The ISR routine(gic_handle_irq) being called to handle the induced LPI.
NODE 1 cpu is used to induce LPI and NODE 1 cpu/collection is mapped
in ITT to route this LPI.

CPU timer counter are sampled at the time LPI is Induced and in ISR
routine to calculate interrupt response time.
the result shown improvement of 5% with this patch.

Do you have any recommended benchmarks to test the same?

>
> Give me the actual benchmark results. Don't expect me to accept this
> kind of hand-wavy statement.
>
>         M.
> --
> Jazz is not dead. It just smells funny...

thanks
Ganapat

[toc] | [prev] | [next] | [standalone]

#1684191

Marc,

Do you have any other concerns taking this patch?


On 07/10/2017 05:21 AM, Ganapatrao Kulkarni wrote:
> Hi Marc,
> 
> On Mon, Jul 10, 2017 at 2:53 PM, Marc Zyngier <marc.zyngier@arm.com> wrote:
>> On 10/07/17 10:08, Ganapatrao Kulkarni wrote:
>>> On Mon, Jul 10, 2017 at 2:36 PM, Marc Zyngier <marc.zyngier@arm.com> wrote:
>>>> On 10/07/17 09:48, Ganapatrao Kulkarni wrote:
>>>>> Hi Marc,
>>>>>
>>>>> On Mon, Jul 3, 2017 at 8:23 PM, Marc Zyngier <marc.zyngier@arm.com> wrote:
>>>>>> Hi Shanker,
>>>>>>
>>>>>> On 03/07/17 15:24, Shanker Donthineni wrote:
>>>>>>> Hi Marc,
>>>>>>>
>>>>>>> On 06/30/2017 03:51 AM, Marc Zyngier wrote:
>>>>>>>> On 30/06/17 04:01, Ganapatrao Kulkarni wrote:
>>>>>>>>> On Fri, Jun 30, 2017 at 8:04 AM, Ganapatrao Kulkarni
>>>>>>>>> <gpkulkarni@gmail.com> wrote:
>>>>>>>>>> Hi Shanker,
>>>>>>>>>>
>>>>>>>>>> On Sun, Jun 25, 2017 at 9:16 PM, Shanker Donthineni
>>>>>>>>>> <shankerd@codeaurora.org> wrote:
>>>>>>>>>>> The NUMA node information is visible to ITS driver but not being used
>>>>>>>>>>> other than handling errata. This patch allocates the memory for ITS
>>>>>>>>>>> tables from the corresponding NUMA node using the appropriate NUMA
>>>>>>>>>>> aware functions.
>>>>>>>>>
>>>>>>>>> IMHO, the description would have been more constructive?
>>>>>>>>>
>>>>>>>>> "All ITS tables are mapped by default to NODE 0 memory.
>>>>>>>>> Adding changes to allocate memory from respective NUMA NODES of ITS devices.
>>>>>>>>> This will optimize tables access and avoids unnecessary inter-node traffic."
>>>>>>>>
>>>>>>>> But more importantly, I'd like to see figures showing the actual benefit
>>>>>>>> of this per-node allocation. Given that both of you guys have access to
>>>>>>>> such platforms, please show me the numbers!
>>>>>>>>
>>>>>>>
>>>>>>> I'll share the actual results which shows the improvement whenever
>>>>>>> available on our next chips. Current version of Qualcomm qdf2400 doesn't
>>>>>>> support multi socket configuration to capture results and share with you.
>>>>>>>
>>>>>>> Do you see any other issues with this patch apart from the performance
>>>>>>> improvements. I strongly believe this brings the noticeable improvement
>>>>>>> in numbers on systems where it has multi node memory/CPU configuration.
>>>>>>
>>>>>> I agree that it *could* show an improvement, but it very much depends on
>>>>>> how often the ITS misses in its caches. For this kind of patches, I want
>>>>>> to see two things:
>>>>>>
>>>>>> 1) It brings a measurable benefit on NUMA platforms
>>>>>
>>>>> Did some measurement of interrupt response time for LPIs and we don't
>>>>> see any major
>>>>> improvement due to caching of Tables. However, we have seen
>>>>> improvements of around 5%.
>>>>
>>>> An improvement of what exactly?
>>>
>>> interrupt response time.
>>
>> Measured how? On which HW? Using which benchmark?
> 
> This has been tested on ThunderX2.
> We have instrumented gic-v3-its driver code to create dummy LPI device
> with few vectors.
> The LPI is induced from dummy device(through sysfs by writing to
> TRANSLATOR reg).
> The ISR routine(gic_handle_irq) being called to handle the induced LPI.
> NODE 1 cpu is used to induce LPI and NODE 1 cpu/collection is mapped
> in ITT to route this LPI.
> 
> CPU timer counter are sampled at the time LPI is Induced and in ISR
> routine to calculate interrupt response time.
> the result shown improvement of 5% with this patch.
> 
> Do you have any recommended benchmarks to test the same?
> 

Ganapatrao,

Thanks for your efforts on instrumenting ITS driver code to show interrupt performance
improvement of 5% on the ThunderX2 hardware. Actually the current ITS driver is not 
consistent on allocating memory for ITS/GICR tables, GICR pending tables are allocated
from the corresponding NUMA node based on CPU proximity, but not the other tables.


>>
>> Give me the actual benchmark results. Don't expect me to accept this
>> kind of hand-wavy statement.
>>
>>         M.
>> --
>> Jazz is not dead. It just smells funny...
> 
> thanks
> Ganapat
> 
> _______________________________________________
> linux-arm-kernel mailing list
> linux-arm-kernel@lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
> 

-- 
Shanker Donthineni
Qualcomm Datacenter Technologies, Inc. as an affiliate of Qualcomm Technologies, Inc.
Qualcomm Technologies, Inc. is a member of the Code Aurora Forum, a Linux Foundation Collaborative Project.

[toc] | [prev] | [next] | [standalone]

#1684265

On 10/07/17 13:30, Shanker Donthineni wrote:
> Marc,
> 
> Do you have any other concerns taking this patch?

Plenty. All I have seen so far is a wet finger in the air and the claim
that it reduces "something" by 5%, without any actual figure.

Sorry, I'm not buying it without people putting in the effort to show me
that this is worth it.

Thanks,

	M.
-- 
Jazz is not dead. It just smells funny...

[toc] | [prev] | [next] | [standalone]

#1684259

On 10/07/17 11:21, Ganapatrao Kulkarni wrote:
> Hi Marc,
> 
> On Mon, Jul 10, 2017 at 2:53 PM, Marc Zyngier <marc.zyngier@arm.com> wrote:
>> On 10/07/17 10:08, Ganapatrao Kulkarni wrote:
>>> On Mon, Jul 10, 2017 at 2:36 PM, Marc Zyngier <marc.zyngier@arm.com> wrote:
>>>> On 10/07/17 09:48, Ganapatrao Kulkarni wrote:
>>>>> Hi Marc,
>>>>>
>>>>> On Mon, Jul 3, 2017 at 8:23 PM, Marc Zyngier <marc.zyngier@arm.com> wrote:
>>>>>> Hi Shanker,
>>>>>>
>>>>>> On 03/07/17 15:24, Shanker Donthineni wrote:
>>>>>>> Hi Marc,
>>>>>>>
>>>>>>> On 06/30/2017 03:51 AM, Marc Zyngier wrote:
>>>>>>>> On 30/06/17 04:01, Ganapatrao Kulkarni wrote:
>>>>>>>>> On Fri, Jun 30, 2017 at 8:04 AM, Ganapatrao Kulkarni
>>>>>>>>> <gpkulkarni@gmail.com> wrote:
>>>>>>>>>> Hi Shanker,
>>>>>>>>>>
>>>>>>>>>> On Sun, Jun 25, 2017 at 9:16 PM, Shanker Donthineni
>>>>>>>>>> <shankerd@codeaurora.org> wrote:
>>>>>>>>>>> The NUMA node information is visible to ITS driver but not being used
>>>>>>>>>>> other than handling errata. This patch allocates the memory for ITS
>>>>>>>>>>> tables from the corresponding NUMA node using the appropriate NUMA
>>>>>>>>>>> aware functions.
>>>>>>>>>
>>>>>>>>> IMHO, the description would have been more constructive?
>>>>>>>>>
>>>>>>>>> "All ITS tables are mapped by default to NODE 0 memory.
>>>>>>>>> Adding changes to allocate memory from respective NUMA NODES of ITS devices.
>>>>>>>>> This will optimize tables access and avoids unnecessary inter-node traffic."
>>>>>>>>
>>>>>>>> But more importantly, I'd like to see figures showing the actual benefit
>>>>>>>> of this per-node allocation. Given that both of you guys have access to
>>>>>>>> such platforms, please show me the numbers!
>>>>>>>>
>>>>>>>
>>>>>>> I'll share the actual results which shows the improvement whenever
>>>>>>> available on our next chips. Current version of Qualcomm qdf2400 doesn't
>>>>>>> support multi socket configuration to capture results and share with you.
>>>>>>>
>>>>>>> Do you see any other issues with this patch apart from the performance
>>>>>>> improvements. I strongly believe this brings the noticeable improvement
>>>>>>> in numbers on systems where it has multi node memory/CPU configuration.
>>>>>>
>>>>>> I agree that it *could* show an improvement, but it very much depends on
>>>>>> how often the ITS misses in its caches. For this kind of patches, I want
>>>>>> to see two things:
>>>>>>
>>>>>> 1) It brings a measurable benefit on NUMA platforms
>>>>>
>>>>> Did some measurement of interrupt response time for LPIs and we don't
>>>>> see any major
>>>>> improvement due to caching of Tables. However, we have seen
>>>>> improvements of around 5%.
>>>>
>>>> An improvement of what exactly?
>>>
>>> interrupt response time.
>>
>> Measured how? On which HW? Using which benchmark?
> 
> This has been tested on ThunderX2.
> We have instrumented gic-v3-its driver code to create dummy LPI device
> with few vectors.
> The LPI is induced from dummy device(through sysfs by writing to
> TRANSLATOR reg).
> The ISR routine(gic_handle_irq) being called to handle the induced LPI.
> NODE 1 cpu is used to induce LPI and NODE 1 cpu/collection is mapped
> in ITT to route this LPI.
> 
> CPU timer counter are sampled at the time LPI is Induced and in ISR
> routine to calculate interrupt response time.
> the result shown improvement of 5% with this patch.

And you call that a realistic measurement of the latency? Really? Sorry,
but I cannot take you seriously here.

> Do you have any recommended benchmarks to test the same?

Run a standard benchmark such as netperf, post the result with and
without that patch. The above is just plain ridiculous.

	M.
-- 
Jazz is not dead. It just smells funny...

[toc] | [prev] | [next] | [standalone]

#1684317

Hi Marc,

On 07/10/2017 08:50 AM, Marc Zyngier wrote:
> On 10/07/17 11:21, Ganapatrao Kulkarni wrote:
>> Hi Marc,
>>
>> On Mon, Jul 10, 2017 at 2:53 PM, Marc Zyngier <marc.zyngier@arm.com> wrote:
>>> On 10/07/17 10:08, Ganapatrao Kulkarni wrote:
>>>> On Mon, Jul 10, 2017 at 2:36 PM, Marc Zyngier <marc.zyngier@arm.com> wrote:
>>>>> On 10/07/17 09:48, Ganapatrao Kulkarni wrote:
>>>>>> Hi Marc,
>>>>>>
>>>>>> On Mon, Jul 3, 2017 at 8:23 PM, Marc Zyngier <marc.zyngier@arm.com> wrote:
>>>>>>> Hi Shanker,
>>>>>>>
>>>>>>> On 03/07/17 15:24, Shanker Donthineni wrote:
>>>>>>>> Hi Marc,
>>>>>>>>
>>>>>>>> On 06/30/2017 03:51 AM, Marc Zyngier wrote:
>>>>>>>>> On 30/06/17 04:01, Ganapatrao Kulkarni wrote:
>>>>>>>>>> On Fri, Jun 30, 2017 at 8:04 AM, Ganapatrao Kulkarni
>>>>>>>>>> <gpkulkarni@gmail.com> wrote:
>>>>>>>>>>> Hi Shanker,
>>>>>>>>>>>
>>>>>>>>>>> On Sun, Jun 25, 2017 at 9:16 PM, Shanker Donthineni
>>>>>>>>>>> <shankerd@codeaurora.org> wrote:
>>>>>>>>>>>> The NUMA node information is visible to ITS driver but not being used
>>>>>>>>>>>> other than handling errata. This patch allocates the memory for ITS
>>>>>>>>>>>> tables from the corresponding NUMA node using the appropriate NUMA
>>>>>>>>>>>> aware functions.
>>>>>>>>>>
>>>>>>>>>> IMHO, the description would have been more constructive?
>>>>>>>>>>
>>>>>>>>>> "All ITS tables are mapped by default to NODE 0 memory.
>>>>>>>>>> Adding changes to allocate memory from respective NUMA NODES of ITS devices.
>>>>>>>>>> This will optimize tables access and avoids unnecessary inter-node traffic."
>>>>>>>>>
>>>>>>>>> But more importantly, I'd like to see figures showing the actual benefit
>>>>>>>>> of this per-node allocation. Given that both of you guys have access to
>>>>>>>>> such platforms, please show me the numbers!
>>>>>>>>>
>>>>>>>>
>>>>>>>> I'll share the actual results which shows the improvement whenever
>>>>>>>> available on our next chips. Current version of Qualcomm qdf2400 doesn't
>>>>>>>> support multi socket configuration to capture results and share with you.
>>>>>>>>
>>>>>>>> Do you see any other issues with this patch apart from the performance
>>>>>>>> improvements. I strongly believe this brings the noticeable improvement
>>>>>>>> in numbers on systems where it has multi node memory/CPU configuration.
>>>>>>>
>>>>>>> I agree that it *could* show an improvement, but it very much depends on
>>>>>>> how often the ITS misses in its caches. For this kind of patches, I want
>>>>>>> to see two things:
>>>>>>>
>>>>>>> 1) It brings a measurable benefit on NUMA platforms
>>>>>>
>>>>>> Did some measurement of interrupt response time for LPIs and we don't
>>>>>> see any major
>>>>>> improvement due to caching of Tables. However, we have seen
>>>>>> improvements of around 5%.
>>>>>
>>>>> An improvement of what exactly?
>>>>
>>>> interrupt response time.
>>>
>>> Measured how? On which HW? Using which benchmark?
>>
>> This has been tested on ThunderX2.
>> We have instrumented gic-v3-its driver code to create dummy LPI device
>> with few vectors.
>> The LPI is induced from dummy device(through sysfs by writing to
>> TRANSLATOR reg).
>> The ISR routine(gic_handle_irq) being called to handle the induced LPI.
>> NODE 1 cpu is used to induce LPI and NODE 1 cpu/collection is mapped
>> in ITT to route this LPI.
>>
>> CPU timer counter are sampled at the time LPI is Induced and in ISR
>> routine to calculate interrupt response time.
>> the result shown improvement of 5% with this patch.
> 
> And you call that a realistic measurement of the latency? Really? Sorry,
> but I cannot take you seriously here.
> 
>> Do you have any recommended benchmarks to test the same?
> 
> Run a standard benchmark such as netperf, post the result with and
> without that patch. The above is just plain ridiculous.
>

The whole purpose of ACPI subtable "GIC Interrupt Translation Service (ITS) Affinity structure" 
is to provide the proximity information to OS so that software will take advantage of NUMA
aware allocations to improve the read latency of ITS/GICR tables, not just for implementing 
software workarounds.

 
I believe ITS driver should provide NUMA aware allocations just like x86 Linux drivers. How much
performance improvement we observer is based on the individual SOC implementation, inter NODE
latency, inter node traffic, cache capacity, and type of the test used to measure results.

Please consider this patch irrespective of the test results running on a specific hardware. We
need this patch for upcoming Qualcomm server chips. 


 
> 	M.
> 

-- 
Shanker Donthineni
Qualcomm Datacenter Technologies, Inc. as an affiliate of Qualcomm Technologies, Inc.
Qualcomm Technologies, Inc. is a member of the Code Aurora Forum, a Linux Foundation Collaborative Project.

[toc] | [prev] | [next] | [standalone]

#1684341

On 10/07/17 15:57, Shanker Donthineni wrote:
> Hi Marc,
> 
> On 07/10/2017 08:50 AM, Marc Zyngier wrote:
>> On 10/07/17 11:21, Ganapatrao Kulkarni wrote:
>>> Hi Marc,
>>>
>>> On Mon, Jul 10, 2017 at 2:53 PM, Marc Zyngier <marc.zyngier@arm.com> wrote:
>>>> On 10/07/17 10:08, Ganapatrao Kulkarni wrote:
>>>>> On Mon, Jul 10, 2017 at 2:36 PM, Marc Zyngier <marc.zyngier@arm.com> wrote:
>>>>>> On 10/07/17 09:48, Ganapatrao Kulkarni wrote:
>>>>>>> Hi Marc,
>>>>>>>
>>>>>>> On Mon, Jul 3, 2017 at 8:23 PM, Marc Zyngier <marc.zyngier@arm.com> wrote:
>>>>>>>> Hi Shanker,
>>>>>>>>
>>>>>>>> On 03/07/17 15:24, Shanker Donthineni wrote:
>>>>>>>>> Hi Marc,
>>>>>>>>>
>>>>>>>>> On 06/30/2017 03:51 AM, Marc Zyngier wrote:
>>>>>>>>>> On 30/06/17 04:01, Ganapatrao Kulkarni wrote:
>>>>>>>>>>> On Fri, Jun 30, 2017 at 8:04 AM, Ganapatrao Kulkarni
>>>>>>>>>>> <gpkulkarni@gmail.com> wrote:
>>>>>>>>>>>> Hi Shanker,
>>>>>>>>>>>>
>>>>>>>>>>>> On Sun, Jun 25, 2017 at 9:16 PM, Shanker Donthineni
>>>>>>>>>>>> <shankerd@codeaurora.org> wrote:
>>>>>>>>>>>>> The NUMA node information is visible to ITS driver but not being used
>>>>>>>>>>>>> other than handling errata. This patch allocates the memory for ITS
>>>>>>>>>>>>> tables from the corresponding NUMA node using the appropriate NUMA
>>>>>>>>>>>>> aware functions.
>>>>>>>>>>>
>>>>>>>>>>> IMHO, the description would have been more constructive?
>>>>>>>>>>>
>>>>>>>>>>> "All ITS tables are mapped by default to NODE 0 memory.
>>>>>>>>>>> Adding changes to allocate memory from respective NUMA NODES of ITS devices.
>>>>>>>>>>> This will optimize tables access and avoids unnecessary inter-node traffic."
>>>>>>>>>>
>>>>>>>>>> But more importantly, I'd like to see figures showing the actual benefit
>>>>>>>>>> of this per-node allocation. Given that both of you guys have access to
>>>>>>>>>> such platforms, please show me the numbers!
>>>>>>>>>>
>>>>>>>>>
>>>>>>>>> I'll share the actual results which shows the improvement whenever
>>>>>>>>> available on our next chips. Current version of Qualcomm qdf2400 doesn't
>>>>>>>>> support multi socket configuration to capture results and share with you.
>>>>>>>>>
>>>>>>>>> Do you see any other issues with this patch apart from the performance
>>>>>>>>> improvements. I strongly believe this brings the noticeable improvement
>>>>>>>>> in numbers on systems where it has multi node memory/CPU configuration.
>>>>>>>>
>>>>>>>> I agree that it *could* show an improvement, but it very much depends on
>>>>>>>> how often the ITS misses in its caches. For this kind of patches, I want
>>>>>>>> to see two things:
>>>>>>>>
>>>>>>>> 1) It brings a measurable benefit on NUMA platforms
>>>>>>>
>>>>>>> Did some measurement of interrupt response time for LPIs and we don't
>>>>>>> see any major
>>>>>>> improvement due to caching of Tables. However, we have seen
>>>>>>> improvements of around 5%.
>>>>>>
>>>>>> An improvement of what exactly?
>>>>>
>>>>> interrupt response time.
>>>>
>>>> Measured how? On which HW? Using which benchmark?
>>>
>>> This has been tested on ThunderX2.
>>> We have instrumented gic-v3-its driver code to create dummy LPI device
>>> with few vectors.
>>> The LPI is induced from dummy device(through sysfs by writing to
>>> TRANSLATOR reg).
>>> The ISR routine(gic_handle_irq) being called to handle the induced LPI.
>>> NODE 1 cpu is used to induce LPI and NODE 1 cpu/collection is mapped
>>> in ITT to route this LPI.
>>>
>>> CPU timer counter are sampled at the time LPI is Induced and in ISR
>>> routine to calculate interrupt response time.
>>> the result shown improvement of 5% with this patch.
>>
>> And you call that a realistic measurement of the latency? Really? Sorry,
>> but I cannot take you seriously here.
>>
>>> Do you have any recommended benchmarks to test the same?
>>
>> Run a standard benchmark such as netperf, post the result with and
>> without that patch. The above is just plain ridiculous.
>>
> 
> The whole purpose of ACPI subtable "GIC Interrupt Translation Service (ITS) Affinity structure" 
> is to provide the proximity information to OS so that software will take advantage of NUMA
> aware allocations to improve the read latency of ITS/GICR tables, not just for implementing 
> software workarounds.
> 
>  
> I believe ITS driver should provide NUMA aware allocations just like x86 Linux drivers. How much
> performance improvement we observer is based on the individual SOC implementation, inter NODE
> latency, inter node traffic, cache capacity, and type of the test used to measure results.
> 
> Please consider this patch irrespective of the test results running on a specific hardware. We
> need this patch for upcoming Qualcomm server chips. 

"I believe" and "We need" are not a proof of the usefulness of this. We
can argue all day, or you can provide a set of convincing results. Your
choice. But I can guarantee you the the latter is a much better method
than the former.

If you (or Cavium) cannot be bothered to provide tangible results that
this is useful, why should I take this at face value? This is just like
any other improvement we make to the kernel. We back it *with data*.

	M.
-- 
Jazz is not dead. It just smells funny...

[toc] | [prev] | [next] | [standalone]

#1684073

On 10/07/17 09:48, Ganapatrao Kulkarni wrote:
> Hi Marc,
> 
> On Mon, Jul 3, 2017 at 8:23 PM, Marc Zyngier <marc.zyngier@arm.com> wrote:
>> Hi Shanker,
>>
>> On 03/07/17 15:24, Shanker Donthineni wrote:
>>> Hi Marc,
>>>
>>> On 06/30/2017 03:51 AM, Marc Zyngier wrote:
>>>> On 30/06/17 04:01, Ganapatrao Kulkarni wrote:
>>>>> On Fri, Jun 30, 2017 at 8:04 AM, Ganapatrao Kulkarni
>>>>> <gpkulkarni@gmail.com> wrote:
>>>>>> Hi Shanker,
>>>>>>
>>>>>> On Sun, Jun 25, 2017 at 9:16 PM, Shanker Donthineni
>>>>>> <shankerd@codeaurora.org> wrote:
>>>>>>> The NUMA node information is visible to ITS driver but not being used
>>>>>>> other than handling errata. This patch allocates the memory for ITS
>>>>>>> tables from the corresponding NUMA node using the appropriate NUMA
>>>>>>> aware functions.
>>>>>
>>>>> IMHO, the description would have been more constructive?
>>>>>
>>>>> "All ITS tables are mapped by default to NODE 0 memory.
>>>>> Adding changes to allocate memory from respective NUMA NODES of ITS devices.
>>>>> This will optimize tables access and avoids unnecessary inter-node traffic."
>>>>
>>>> But more importantly, I'd like to see figures showing the actual benefit
>>>> of this per-node allocation. Given that both of you guys have access to
>>>> such platforms, please show me the numbers!
>>>>
>>>
>>> I'll share the actual results which shows the improvement whenever
>>> available on our next chips. Current version of Qualcomm qdf2400 doesn't
>>> support multi socket configuration to capture results and share with you.
>>>
>>> Do you see any other issues with this patch apart from the performance
>>> improvements. I strongly believe this brings the noticeable improvement
>>> in numbers on systems where it has multi node memory/CPU configuration.
>>
>> I agree that it *could* show an improvement, but it very much depends on
>> how often the ITS misses in its caches. For this kind of patches, I want
>> to see two things:
>>
>> 1) It brings a measurable benefit on NUMA platforms
> 
> Did some measurement of interrupt response time for LPIs and we don't
> see any major
> improvement due to caching of Tables. However, we have seen
> improvements of around 5%.

An improvement of what exactly?

	M.
-- 
Jazz is not dead. It just smells funny...

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web