[LSF/MM TOPIC] HMM status upstream user what's next, mmu_notifier

[LSF/MM TOPIC] HMM status upstream user what's next, mmu_notifier

[Jerome Glisse]

I want to talk about status of HMM and respective upstream user for
it and also talk about what's next in term of features/improvement
plan (generic page write protection, mmu_notifier, ...). Most likely
short 15-30minutes if mmu_notifier is split into its own topic.

I want to talk about mmu_notifier, specificaly adding more context
information to mmu_notifier callback (why a notification is happening
reclaim, munmap, migrate, ...). Maybe we can grow this into its own
topic and talk about mmu_notifier and issue with it like OOM or being
able to sleep/take lock ... and improving mitigation.

People (mmu_notifier probably interest a larger set):
    "Anshuman Khandual" <khandual@linux.vnet.ibm.com>
    "Balbir Singh" <bsingharora@gmail.com>
    "David Rientjes" <rientjes@google.com>
    "John Hubbard" <jhubbard@nvidia.com>
    "Michal Hocko" <mhocko@suse.com>

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [LSF/MM TOPIC] HMM status upstream user what's next, mmu_notifier

[Liubo(OS Lab)]

On 2018/1/17 5:30, Jerome Glisse wrote:
> I want to talk about status of HMM and respective upstream user for
> it and also talk about what's next in term of features/improvement
> plan (generic page write protection, mmu_notifier, ...). Most likely

I don't think we should consider to push more code to upstream for a nobody-use feature.

AFAIR, Michal also mentioned that HMM need a real user/driver before upstream.
But I haven't seen a workable user/driver version.

Looks like HMM is a custom framework for Nvidia, and Nvidia would not like to open source its driver.
Even if nvidia really use HMM and open sourced its driver, it's probably the only user.
But the HMM framework touched too much core mm code.

Cheers,
Liubo

> short 15-30minutes if mmu_notifier is split into its own topic.
> 
> I want to talk about mmu_notifier, specificaly adding more context
> information to mmu_notifier callback (why a notification is happening
> reclaim, munmap, migrate, ...). Maybe we can grow this into its own
> topic and talk about mmu_notifier and issue with it like OOM or being
> able to sleep/take lock ... and improving mitigation.
> 
> People (mmu_notifier probably interest a larger set):
>     "Anshuman Khandual" <khandual@linux.vnet.ibm.com>
>     "Balbir Singh" <bsingharora@gmail.com>
>     "David Rientjes" <rientjes@google.com>
>     "John Hubbard" <jhubbard@nvidia.com>
>     "Michal Hocko" <mhocko@suse.com>
> 


--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [LSF/MM TOPIC] HMM status upstream user what's next, mmu_notifier

[Figo.zhang]

[-- Attachment #1: Type: text/plain, Size: 1874 bytes --]

2018-01-17 9:19 GMT+08:00 Liubo(OS Lab) <liubo95@huawei.com>:

> On 2018/1/17 5:30, Jerome Glisse wrote:
> > I want to talk about status of HMM and respective upstream user for
> > it and also talk about what's next in term of features/improvement
> > plan (generic page write protection, mmu_notifier, ...). Most likely
>
> I don't think we should consider to push more code to upstream for a
> nobody-use feature.
>
> AFAIR, Michal also mentioned that HMM need a real user/driver before
> upstream.
> But I haven't seen a workable user/driver version.
>
> Looks like HMM is a custom framework for Nvidia, and Nvidia would not like
> to open source its driver.
> Even if nvidia really use HMM and open sourced its driver, it's probably
> the only user.
> But the HMM framework touched too much core mm code.
>

HMM looks suitable for FPGA user case, FPGA and CPU need coherency. ~_~


>
> Cheers,
> Liubo
>
> > short 15-30minutes if mmu_notifier is split into its own topic.
> >
> > I want to talk about mmu_notifier, specificaly adding more context
> > information to mmu_notifier callback (why a notification is happening
> > reclaim, munmap, migrate, ...). Maybe we can grow this into its own
> > topic and talk about mmu_notifier and issue with it like OOM or being
> > able to sleep/take lock ... and improving mitigation.
> >
> > People (mmu_notifier probably interest a larger set):
> >     "Anshuman Khandual" <khandual@linux.vnet.ibm.com>
> >     "Balbir Singh" <bsingharora@gmail.com>
> >     "David Rientjes" <rientjes@google.com>
> >     "John Hubbard" <jhubbard@nvidia.com>
> >     "Michal Hocko" <mhocko@suse.com>
> >
>
>
> --
> To unsubscribe, send a message with 'unsubscribe linux-mm' in
> the body to majordomo@kvack.org.  For more info on Linux MM,
> see: http://www.linux-mm.org/ .
> Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>
>

[-- Attachment #2: Type: text/html, Size: 3103 bytes --]

Re: [LSF/MM TOPIC] HMM status upstream user what's next, mmu_notifier

[Jerome Glisse]

On Wed, Jan 17, 2018 at 09:19:56AM +0800, Liubo(OS Lab) wrote:
> On 2018/1/17 5:30, Jerome Glisse wrote:
> > I want to talk about status of HMM and respective upstream user for
> > it and also talk about what's next in term of features/improvement
> > plan (generic page write protection, mmu_notifier, ...). Most likely
> 
> I don't think we should consider to push more code to upstream for a nobody-use feature.
> 
> AFAIR, Michal also mentioned that HMM need a real user/driver before upstream.
> But I haven't seen a workable user/driver version.
> 
> Looks like HMM is a custom framework for Nvidia, and Nvidia would not like to open source its driver.
> Even if nvidia really use HMM and open sourced its driver, it's probably the only user.
> But the HMM framework touched too much core mm code.

So it is not NVidia only, they are other GPU from different company that
intend to use HMM, i can't comment any further on timeline.

Regarding NVidia hardware i intend to have the patchset to use HMM inside
nouveau ready before summit and post as RFC. Per drm sub-system guideline
we can not upstream it until we also have a functional userspace stack and
we are also working on that.

Cheers,
Jerome

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [LSF/MM TOPIC] HMM status upstream user what's next, mmu_notifier

[Michal Hocko]

On Tue 16-01-18 21:05:02, Jerome Glisse wrote:
> On Wed, Jan 17, 2018 at 09:19:56AM +0800, Liubo(OS Lab) wrote:
> > On 2018/1/17 5:30, Jerome Glisse wrote:
> > > I want to talk about status of HMM and respective upstream user for
> > > it and also talk about what's next in term of features/improvement
> > > plan (generic page write protection, mmu_notifier, ...). Most likely
> > 
> > I don't think we should consider to push more code to upstream for a nobody-use feature.
> > 
> > AFAIR, Michal also mentioned that HMM need a real user/driver before upstream.
> > But I haven't seen a workable user/driver version.
> > 
> > Looks like HMM is a custom framework for Nvidia, and Nvidia would not like to open source its driver.
> > Even if nvidia really use HMM and open sourced its driver, it's probably the only user.
> > But the HMM framework touched too much core mm code.
> 
> So it is not NVidia only, they are other GPU from different company that
> intend to use HMM, i can't comment any further on timeline.
> 
> Regarding NVidia hardware i intend to have the patchset to use HMM inside
> nouveau ready before summit and post as RFC. Per drm sub-system guideline
> we can not upstream it until we also have a functional userspace stack and
> we are also working on that.

I would really appreciate to at least see some form of RFC work. It
would be a pity to have a HMM discussion anniversary without an actual
user in sight.
-- 
Michal Hocko
SUSE Labs

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [LSF/MM TOPIC] HMM status upstream user what's next, mmu_notifier

[John Hubbard]

On 01/16/2018 01:30 PM, Jerome Glisse wrote:
> I want to talk about status of HMM and respective upstream user for
> it and also talk about what's next in term of features/improvement
> plan (generic page write protection, mmu_notifier, ...). Most likely
> short 15-30minutes if mmu_notifier is split into its own topic.
> 
> I want to talk about mmu_notifier, specificaly adding more context
> information to mmu_notifier callback (why a notification is happening
> reclaim, munmap, migrate, ...). Maybe we can grow this into its own
> topic and talk about mmu_notifier and issue with it like OOM or being
> able to sleep/take lock ... and improving mitigation.
> 
> People (mmu_notifier probably interest a larger set):
>     "Anshuman Khandual" <khandual@linux.vnet.ibm.com>
>     "Balbir Singh" <bsingharora@gmail.com>
>     "David Rientjes" <rientjes@google.com>
>     "John Hubbard" <jhubbard@nvidia.com>
>     "Michal Hocko" <mhocko@suse.com>
> 

Hi Jerome,

Thanks again for including me here. I am very interesting in discussing this,
seeing as how we're busy adding HMM support to our driver(s).

The teardown cases still concern us, as you know (today's HMM lacks
a callback for when the struct mm disappears, for example), but that's really
just a subset of the mmu_notifier discussion that you list above.

Hope to see you there.

thanks,
-- 
John Hubbard
NVIDIA

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [LSF/MM TOPIC] HMM status upstream user what's next, mmu_notifier

[Michal Hocko]

On Tue 16-01-18 16:30:08, Jerome Glisse wrote:
> I want to talk about status of HMM and respective upstream user for
> it and also talk about what's next in term of features/improvement
> plan (generic page write protection, mmu_notifier, ...). Most likely
> short 15-30minutes if mmu_notifier is split into its own topic.
> 
> I want to talk about mmu_notifier, specificaly adding more context
> information to mmu_notifier callback (why a notification is happening
> reclaim, munmap, migrate, ...). Maybe we can grow this into its own
> topic and talk about mmu_notifier and issue with it like OOM or being
> able to sleep/take lock ... and improving mitigation.
> 
> People (mmu_notifier probably interest a larger set):
>     "Anshuman Khandual" <khandual@linux.vnet.ibm.com>
>     "Balbir Singh" <bsingharora@gmail.com>
>     "David Rientjes" <rientjes@google.com>
>     "John Hubbard" <jhubbard@nvidia.com>
>     "Michal Hocko" <mhocko@suse.com>

I am definitely interested.

-- 
Michal Hocko
SUSE Labs

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [LSF/MM TOPIC] HMM status upstream user what's next, mmu_notifier

[Balbir Singh]

On Fri, Jan 26, 2018 at 2:37 AM, Michal Hocko <mhocko@suse.com> wrote:
> On Tue 16-01-18 16:30:08, Jerome Glisse wrote:
>> I want to talk about status of HMM and respective upstream user for
>> it and also talk about what's next in term of features/improvement
>> plan (generic page write protection, mmu_notifier, ...). Most likely
>> short 15-30minutes if mmu_notifier is split into its own topic.
>>
>> I want to talk about mmu_notifier, specificaly adding more context
>> information to mmu_notifier callback (why a notification is happening
>> reclaim, munmap, migrate, ...). Maybe we can grow this into its own
>> topic and talk about mmu_notifier and issue with it like OOM or being
>> able to sleep/take lock ... and improving mitigation.
>>
>> People (mmu_notifier probably interest a larger set):
>>     "Anshuman Khandual" <khandual@linux.vnet.ibm.com>
>>     "Balbir Singh" <bsingharora@gmail.com>
>>     "David Rientjes" <rientjes@google.com>
>>     "John Hubbard" <jhubbard@nvidia.com>
>>     "Michal Hocko" <mhocko@suse.com>
>
> I am definitely interested.

me too and as I said in the previous email, we may have other examples
of coherent memory like openCAPI and lots of experience with both NUMA
and HMM/HMM-CDM. I'd like to share them and discuss the path forward

Balbir Singh.

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>