Current topics for LSF10/MM Summit 8-9 August in Boston

Current topics for LSF10/MM Summit 8-9 August in Boston

[James Bottomley]

Given that we're under two months out, I thought it would be time to
post a summary of the topics we've collected so far (Nick will post the
MM summit ones later).  Look this over, and if there's anything missing,
propose it ... or if you have cross Storage/FS/MM topics, post them too.

Oh, and since we're not the most organised bunch, if you posted a topic
and don't see it in the list, please resend ... we probably lost it in
an email shuffle.

Current Filesystem Topics:

Alex Elder	Upstream maintainer for XFS, general discussion on FS/IO
Aneesh Kumar	Rich-acl patches which work better with NFSv4 acl and CIFS acl
Anshul Madan	reflink for NFS
Chuck Lever	NFS/IPV6 and NFS O_DIRECT, Wu's read-ahead work, vitro perf tools
Eric Sandeen	Advances in testing, TRIM/DISCARD/Alignment, writeback sanity
James Lentini	reflink for NFS
Jan Kara	Discuss/drive sanity review of writeback and general ext*/jbd 
Michael Rubin	Writeback scaling
Sage Weil	Statlite, generic interface for describing file striping for distributed FS, VFS scalability
Al Viro	Sorting out d_revalidate and other dcache issues
Coly Li		directory/large file scalability
Sorin Faibish	Cache writeback discussion

Current Storage Topics:

Eric Seppanen	Next generation SSDs, performance implications on Linux I/O
Boaz Harrosh	PNFS performance considerations, bio_list based/async raidN for generic use; stable pages for I/O
FUJITA Tomonori	SCSI target mode, iSCSI, block layer SG (bsg), sg, IOMMU, DMA issues
Hannes Reinecke	libfc/multipath/error handing
James Smart	FCOE proposal for rework of the FC sysfs tree, work with Hannes on other transport/SCSI subsystem topics
Jeff Moyer	IO scheduler
Joel Becker	SAN management plugin
Martin Petersen	Updates on DIF/DIX, TRIM/DISCARD/UNMAP, generic support for WRITE_SAME

Plus some MM summit ones which Nick will summarise.

For the benefit of those who've forgotten here's the original Call for
topics and attendees:

This year we'll hold the Linux Storage and Filesystems summit jointly
with the VM summit on the two days before LinuxCon in Boston (that's
Sunday and Monday) at the Renaissance Hotel:

http://events.linuxfoundation.org/events/linuxcon

We're planning to hold some sessions jointly and split into three tracks
(Filesystems, Storage and VM) for others, so we're encouraging proposals
for discussion that cover areas relevant to all three groups as well as
more specific technical topics.

Suggestions for agenda topics should be sent to

lsf10-pc@lists.linuxfoundation.org

and optionally cc the Linux list which would be most interested in it:

SCSI: linux-scsi@vger.kernel.org
FS: linux-fsdevel@vger.kernel.org (plus relevant fs specific list)
MM: linux-mm@kvack.org

Please tag your subject with [LSF/VM TOPIC] so those of us who're not
very organised can find them easily in our inboxes.  The agenda topics
and attendees will be selected by the programme committee, but the final
agenda will be by formed by consensus of the attendees on the day.

We'll try to cap attendance at around 20 per track to facilitate
discussions although the final numbers will depend on the room sizes
at the venue.

Requests to attend should be sent to:

lsf10-pc@lists.linuxfoundation.org

please summarise what you'll bring to the meeting, and what you'd like
to discuss.  please also tag your email with [ATTEND] so there's less
chance of it getting lost in the large mail pile.

Presentations are allowed to guide discussion, but are strongly
discouraged.  There will be no recording or audio bridge, however
written minutes will be published as in previous years:

2009:
http://lwn.net/Articles/327601/
http://lwn.net/Articles/327740/
http://lwn.net/Articles/328347/

Prior years:
http://www.usenix.org/events/lsf08/tech/lsf08sums.pdf
http://www.usenix.org/publications/login/2007-06/openpdfs/lsf07sums.pdf

If you have feedback on last year's meeting that we can use to improve
this year's, please also send that to:

lsf10-pc@lists.linuxfoundation.org

Thanks,

James Bottomley


--
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: Current topics for LSF10/MM Summit 8-9 August in Boston

[Nicholas A. Bellinger]

On Wed, 2010-06-16 at 15:50 -0500, James Bottomley wrote:
> Given that we're under two months out, I thought it would be time to
> post a summary of the topics we've collected so far (Nick will post the
> MM summit ones later).  Look this over, and if there's anything missing,
> propose it ... or if you have cross Storage/FS/MM topics, post them too.
> 
> Oh, and since we're not the most organised bunch, if you posted a topic
> and don't see it in the list, please resend ... we probably lost it in
> an email shuffle.
> 
> Current Filesystem Topics:
> 
> Alex Elder	Upstream maintainer for XFS, general discussion on FS/IO
> Aneesh Kumar	Rich-acl patches which work better with NFSv4 acl and CIFS acl
> Anshul Madan	reflink for NFS
> Chuck Lever	NFS/IPV6 and NFS O_DIRECT, Wu's read-ahead work, vitro perf tools
> Eric Sandeen	Advances in testing, TRIM/DISCARD/Alignment, writeback sanity
> James Lentini	reflink for NFS
> Jan Kara	Discuss/drive sanity review of writeback and general ext*/jbd 
> Michael Rubin	Writeback scaling
> Sage Weil	Statlite, generic interface for describing file striping for distributed FS, VFS scalability
> Al Viro	Sorting out d_revalidate and other dcache issues
> Coly Li		directory/large file scalability
> Sorin Faibish	Cache writeback discussion
> 
> Current Storage Topics:
> 
> Eric Seppanen	Next generation SSDs, performance implications on Linux I/O
> Boaz Harrosh	PNFS performance considerations, bio_list based/async raidN for generic use; stable pages for I/O
> FUJITA Tomonori	SCSI target mode, iSCSI, block layer SG (bsg), sg, IOMMU, DMA issues
> Hannes Reinecke	libfc/multipath/error handing
> James Smart	FCOE proposal for rework of the FC sysfs tree, work with Hannes on other transport/SCSI subsystem topics
> Jeff Moyer	IO scheduler
> Joel Becker	SAN management plugin
> Martin Petersen	Updates on DIF/DIX, TRIM/DISCARD/UNMAP, generic support for WRITE_SAME
> 
> Plus some MM summit ones which Nick will summarise.
> 

Greeting James and co,

I noticed that the bit wrt to the kernel level target mode fabric
independent configfs infrastructure is not mentioned in the above..

http://marc.info/?l=linux-scsi&m=127010303618447&w=2

Where would this best fit in..?

"A virtual filesystem driven by userspace syscalls to represent a target
HBA/DEV model on top upstream Linux storage subsystems for fabric
modules using a generic set of configfs struct config_groups to
represent target mode fabric endpoints (WWN+TPG+LUN) designed to allow
each their own set of fabric dependent attributes on top of a generic
kernel infrastructure.

The model is to allow the Linux VFS to handle the TCM core HBA/DEV logic
and both fabric independent and dependent data structure dependencies
between LKMs in order to simplify the conversion of existing and
creation of new target mode fabric code."

Best,

--nab


--
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: [Lsf10-pc] Current topics for LSF10/MM Summit 8-9 August in Boston

[James Bottomley]

On Wed, 2010-06-16 at 23:35 -0700, Nicholas A. Bellinger wrote:
> I noticed that the bit wrt to the kernel level target mode fabric
> independent configfs infrastructure is not mentioned in the above..
> 
> http://marc.info/?l=linux-scsi&m=127010303618447&w=2

It's missing because the person collating the spreadsheet forgot it ...
I've added it in.

James

Re: Current topics for LSF10/MM Summit 8-9 August in Boston

[Christof Schmitt]

On Wed, Jun 16, 2010 at 03:50:59PM -0500, James Bottomley wrote:
> Given that we're under two months out, I thought it would be time to
> post a summary of the topics we've collected so far (Nick will post the
> MM summit ones later).  Look this over, and if there's anything missing,
> propose it ... or if you have cross Storage/FS/MM topics, post them too.
> 
> Oh, and since we're not the most organised bunch, if you posted a topic
> and don't see it in the list, please resend ... we probably lost it in
> an email shuffle.
> 
> Current Filesystem Topics:
> 
> Alex Elder	Upstream maintainer for XFS, general discussion on FS/IO
> Aneesh Kumar	Rich-acl patches which work better with NFSv4 acl and CIFS acl
> Anshul Madan	reflink for NFS
> Chuck Lever	NFS/IPV6 and NFS O_DIRECT, Wu's read-ahead work, vitro perf tools
> Eric Sandeen	Advances in testing, TRIM/DISCARD/Alignment, writeback sanity
> James Lentini	reflink for NFS
> Jan Kara	Discuss/drive sanity review of writeback and general ext*/jbd 
> Michael Rubin	Writeback scaling
> Sage Weil	Statlite, generic interface for describing file striping for distributed FS, VFS scalability
> Al Viro	Sorting out d_revalidate and other dcache issues
> Coly Li		directory/large file scalability
> Sorin Faibish	Cache writeback discussion
> 
> Current Storage Topics:
> 
> Eric Seppanen	Next generation SSDs, performance implications on Linux I/O
> Boaz Harrosh	PNFS performance considerations, bio_list based/async raidN for generic use; stable pages for I/O
> FUJITA Tomonori	SCSI target mode, iSCSI, block layer SG (bsg), sg, IOMMU, DMA issues
> Hannes Reinecke	libfc/multipath/error handing
> James Smart	FCOE proposal for rework of the FC sysfs tree, work with Hannes on other transport/SCSI subsystem topics
> Jeff Moyer	IO scheduler
> Joel Becker	SAN management plugin
> Martin Petersen	Updates on DIF/DIX, TRIM/DISCARD/UNMAP, generic support for WRITE_SAME
> 
> Plus some MM summit ones which Nick will summarise.
[...]

What about the topic "Stable pages while IO"?
http://www.spinics.net/lists/linux-scsi/msg44074.html

Was it lost during the e-mail shuffle or will it be part of the MM topics?

Christof

Re: Current topics for LSF10/MM Summit 8-9 August in Boston

[James Bottomley]

On Thu, 2010-06-17 at 18:00 +0200, Christof Schmitt wrote:
> On Wed, Jun 16, 2010 at 03:50:59PM -0500, James Bottomley wrote:
> > Given that we're under two months out, I thought it would be time to
> > post a summary of the topics we've collected so far (Nick will post the
> > MM summit ones later).  Look this over, and if there's anything missing,
> > propose it ... or if you have cross Storage/FS/MM topics, post them too.
> > 
> > Oh, and since we're not the most organised bunch, if you posted a topic
> > and don't see it in the list, please resend ... we probably lost it in
> > an email shuffle.
> > 
> > Current Filesystem Topics:
> > 
> > Alex Elder	Upstream maintainer for XFS, general discussion on FS/IO
> > Aneesh Kumar	Rich-acl patches which work better with NFSv4 acl and CIFS acl
> > Anshul Madan	reflink for NFS
> > Chuck Lever	NFS/IPV6 and NFS O_DIRECT, Wu's read-ahead work, vitro perf tools
> > Eric Sandeen	Advances in testing, TRIM/DISCARD/Alignment, writeback sanity
> > James Lentini	reflink for NFS
> > Jan Kara	Discuss/drive sanity review of writeback and general ext*/jbd 
> > Michael Rubin	Writeback scaling
> > Sage Weil	Statlite, generic interface for describing file striping for distributed FS, VFS scalability
> > Al Viro	Sorting out d_revalidate and other dcache issues
> > Coly Li		directory/large file scalability
> > Sorin Faibish	Cache writeback discussion
> > 
> > Current Storage Topics:
> > 
> > Eric Seppanen	Next generation SSDs, performance implications on Linux I/O
> > Boaz Harrosh	PNFS performance considerations, bio_list based/async raidN for generic use; stable pages for I/O
> > FUJITA Tomonori	SCSI target mode, iSCSI, block layer SG (bsg), sg, IOMMU, DMA issues
> > Hannes Reinecke	libfc/multipath/error handing
> > James Smart	FCOE proposal for rework of the FC sysfs tree, work with Hannes on other transport/SCSI subsystem topics
> > Jeff Moyer	IO scheduler
> > Joel Becker	SAN management plugin
> > Martin Petersen	Updates on DIF/DIX, TRIM/DISCARD/UNMAP, generic support for WRITE_SAME
> > 
> > Plus some MM summit ones which Nick will summarise.
> [...]
> 
> What about the topic "Stable pages while IO"?
> http://www.spinics.net/lists/linux-scsi/msg44074.html
> 
> Was it lost during the e-mail shuffle or will it be part of the MM topics?

It's actually listed under 'dma issues' ... but there's really been no
satisfactory resolution or discussion of how one might be achieved.
Most filesystems rely on modifications to in-flight pages for efficiency
and copying every fs I/O page would be horrendous both for performance
and memory consumption.  Nor has there really been an indication that
it's a serious issue.  The two sufferers are DIF and iSCSI checksum.
The latter generates the checksum late enough that it can just discard
incorrect pages ... the former might need simply to turn off DIF for
everything other than DIRECT IO.

James

Re: Current topics for LSF10/MM Summit 8-9 August in Boston

[Boaz Harrosh]

On 06/17/2010 12:07 PM, James Bottomley wrote:
> On Thu, 2010-06-17 at 18:00 +0200, Christof Schmitt wrote:
>> On Wed, Jun 16, 2010 at 03:50:59PM -0500, James Bottomley wrote:
>>> Given that we're under two months out, I thought it would be time to
>>> post a summary of the topics we've collected so far (Nick will post the
>>> MM summit ones later).  Look this over, and if there's anything missing,
>>> propose it ... or if you have cross Storage/FS/MM topics, post them too.
>>>
>>> Oh, and since we're not the most organised bunch, if you posted a topic
>>> and don't see it in the list, please resend ... we probably lost it in
>>> an email shuffle.
>>>
>>> Current Filesystem Topics:
>>>
>>> Alex Elder	Upstream maintainer for XFS, general discussion on FS/IO
>>> Aneesh Kumar	Rich-acl patches which work better with NFSv4 acl and CIFS acl
>>> Anshul Madan	reflink for NFS
>>> Chuck Lever	NFS/IPV6 and NFS O_DIRECT, Wu's read-ahead work, vitro perf tools
>>> Eric Sandeen	Advances in testing, TRIM/DISCARD/Alignment, writeback sanity
>>> James Lentini	reflink for NFS
>>> Jan Kara	Discuss/drive sanity review of writeback and general ext*/jbd 
>>> Michael Rubin	Writeback scaling
>>> Sage Weil	Statlite, generic interface for describing file striping for distributed FS, VFS scalability
>>> Al Viro	Sorting out d_revalidate and other dcache issues
>>> Coly Li		directory/large file scalability
>>> Sorin Faibish	Cache writeback discussion
>>>
>>> Current Storage Topics:
>>>
>>> Eric Seppanen	Next generation SSDs, performance implications on Linux I/O
>>> Boaz Harrosh	PNFS performance considerations, bio_list based/async raidN for generic use; stable pages for I/O
>>> FUJITA Tomonori	SCSI target mode, iSCSI, block layer SG (bsg), sg, IOMMU, DMA issues
>>> Hannes Reinecke	libfc/multipath/error handing
>>> James Smart	FCOE proposal for rework of the FC sysfs tree, work with Hannes on other transport/SCSI subsystem topics
>>> Jeff Moyer	IO scheduler
>>> Joel Becker	SAN management plugin
>>> Martin Petersen	Updates on DIF/DIX, TRIM/DISCARD/UNMAP, generic support for WRITE_SAME
>>>
>>> Plus some MM summit ones which Nick will summarise.
>> [...]
>>
>> What about the topic "Stable pages while IO"?
>> http://www.spinics.net/lists/linux-scsi/msg44074.html
>>
>> Was it lost during the e-mail shuffle or will it be part of the MM topics?
> 
> It's actually listed under 'dma issues' ... but there's really been no
> satisfactory resolution or discussion of how one might be achieved.
> Most filesystems rely on modifications to in-flight pages for efficiency
> and copying every fs I/O page would be horrendous both for performance
> and memory consumption.  Nor has there really been an indication that
> it's a serious issue.  The two sufferers are DIF and iSCSI checksum.
> The latter generates the checksum late enough that it can just discard
> incorrect pages ... the former might need simply to turn off DIF for
> everything other than DIRECT IO.

What about raid and mirror that do copy the complete IO load
just because of that? They gave up long ago but wouldn't they
gain if that was revisited? (And so would DIF/checksum)

> 
> James
> 

Boaz

Re: Current topics for LSF10/MM Summit 8-9 August in Boston

[Vladislav Bolkhovitin]

James Bottomley, on 06/17/2010 08:07 PM wrote:
> On Thu, 2010-06-17 at 18:00 +0200, Christof Schmitt wrote:
>> On Wed, Jun 16, 2010 at 03:50:59PM -0500, James Bottomley wrote:
>>> Given that we're under two months out, I thought it would be time to
>>> post a summary of the topics we've collected so far (Nick will post the
>>> MM summit ones later).  Look this over, and if there's anything missing,
>>> propose it ... or if you have cross Storage/FS/MM topics, post them too.
>>>
>>> Oh, and since we're not the most organised bunch, if you posted a topic
>>> and don't see it in the list, please resend ... we probably lost it in
>>> an email shuffle.
>>>
>>> Current Filesystem Topics:
>>>
>>> Alex Elder	Upstream maintainer for XFS, general discussion on FS/IO
>>> Aneesh Kumar	Rich-acl patches which work better with NFSv4 acl and CIFS acl
>>> Anshul Madan	reflink for NFS
>>> Chuck Lever	NFS/IPV6 and NFS O_DIRECT, Wu's read-ahead work, vitro perf tools
>>> Eric Sandeen	Advances in testing, TRIM/DISCARD/Alignment, writeback sanity
>>> James Lentini	reflink for NFS
>>> Jan Kara	Discuss/drive sanity review of writeback and general ext*/jbd 
>>> Michael Rubin	Writeback scaling
>>> Sage Weil	Statlite, generic interface for describing file striping for distributed FS, VFS scalability
>>> Al Viro	Sorting out d_revalidate and other dcache issues
>>> Coly Li		directory/large file scalability
>>> Sorin Faibish	Cache writeback discussion
>>>
>>> Current Storage Topics:
>>>
>>> Eric Seppanen	Next generation SSDs, performance implications on Linux I/O
>>> Boaz Harrosh	PNFS performance considerations, bio_list based/async raidN for generic use; stable pages for I/O
>>> FUJITA Tomonori	SCSI target mode, iSCSI, block layer SG (bsg), sg, IOMMU, DMA issues
>>> Hannes Reinecke	libfc/multipath/error handing
>>> James Smart	FCOE proposal for rework of the FC sysfs tree, work with Hannes on other transport/SCSI subsystem topics
>>> Jeff Moyer	IO scheduler
>>> Joel Becker	SAN management plugin
>>> Martin Petersen	Updates on DIF/DIX, TRIM/DISCARD/UNMAP, generic support for WRITE_SAME
>>>
>>> Plus some MM summit ones which Nick will summarise.
>> [...]
>>
>> What about the topic "Stable pages while IO"?
>> http://www.spinics.net/lists/linux-scsi/msg44074.html
>>
>> Was it lost during the e-mail shuffle or will it be part of the MM topics?
> 
> It's actually listed under 'dma issues' ... but there's really been no
> satisfactory resolution or discussion of how one might be achieved.
> Most filesystems rely on modifications to in-flight pages for efficiency
> and copying every fs I/O page would be horrendous both for performance
> and memory consumption.  Nor has there really been an indication that
> it's a serious issue.  The two sufferers are DIF and iSCSI checksum.

You forgot the third: advanced storage, including MPIO clusters, where 
retry of the write of the modified in-flight pages while the original 
write for them not yet completed might cause out of the expected order 
execution of the writes and data corruption (old data written instead of 
new).

Vlad

Re: Current topics for LSF10/MM Summit 8-9 August in Boston

[James Bottomley]

On Thu, 2010-06-17 at 20:34 +0400, Vladislav Bolkhovitin wrote:
> James Bottomley, on 06/17/2010 08:07 PM wrote:
> > On Thu, 2010-06-17 at 18:00 +0200, Christof Schmitt wrote:
> >> On Wed, Jun 16, 2010 at 03:50:59PM -0500, James Bottomley wrote:
> >>> Given that we're under two months out, I thought it would be time to
> >>> post a summary of the topics we've collected so far (Nick will post the
> >>> MM summit ones later).  Look this over, and if there's anything missing,
> >>> propose it ... or if you have cross Storage/FS/MM topics, post them too.
> >>>
> >>> Oh, and since we're not the most organised bunch, if you posted a topic
> >>> and don't see it in the list, please resend ... we probably lost it in
> >>> an email shuffle.
> >>>
> >>> Current Filesystem Topics:
> >>>
> >>> Alex Elder	Upstream maintainer for XFS, general discussion on FS/IO
> >>> Aneesh Kumar	Rich-acl patches which work better with NFSv4 acl and CIFS acl
> >>> Anshul Madan	reflink for NFS
> >>> Chuck Lever	NFS/IPV6 and NFS O_DIRECT, Wu's read-ahead work, vitro perf tools
> >>> Eric Sandeen	Advances in testing, TRIM/DISCARD/Alignment, writeback sanity
> >>> James Lentini	reflink for NFS
> >>> Jan Kara	Discuss/drive sanity review of writeback and general ext*/jbd 
> >>> Michael Rubin	Writeback scaling
> >>> Sage Weil	Statlite, generic interface for describing file striping for distributed FS, VFS scalability
> >>> Al Viro	Sorting out d_revalidate and other dcache issues
> >>> Coly Li		directory/large file scalability
> >>> Sorin Faibish	Cache writeback discussion
> >>>
> >>> Current Storage Topics:
> >>>
> >>> Eric Seppanen	Next generation SSDs, performance implications on Linux I/O
> >>> Boaz Harrosh	PNFS performance considerations, bio_list based/async raidN for generic use; stable pages for I/O
> >>> FUJITA Tomonori	SCSI target mode, iSCSI, block layer SG (bsg), sg, IOMMU, DMA issues
> >>> Hannes Reinecke	libfc/multipath/error handing
> >>> James Smart	FCOE proposal for rework of the FC sysfs tree, work with Hannes on other transport/SCSI subsystem topics
> >>> Jeff Moyer	IO scheduler
> >>> Joel Becker	SAN management plugin
> >>> Martin Petersen	Updates on DIF/DIX, TRIM/DISCARD/UNMAP, generic support for WRITE_SAME
> >>>
> >>> Plus some MM summit ones which Nick will summarise.
> >> [...]
> >>
> >> What about the topic "Stable pages while IO"?
> >> http://www.spinics.net/lists/linux-scsi/msg44074.html
> >>
> >> Was it lost during the e-mail shuffle or will it be part of the MM topics?
> > 
> > It's actually listed under 'dma issues' ... but there's really been no
> > satisfactory resolution or discussion of how one might be achieved.
> > Most filesystems rely on modifications to in-flight pages for efficiency
> > and copying every fs I/O page would be horrendous both for performance
> > and memory consumption.  Nor has there really been an indication that
> > it's a serious issue.  The two sufferers are DIF and iSCSI checksum.
> 
> You forgot the third: advanced storage, including MPIO clusters, where 
> retry of the write of the modified in-flight pages while the original 
> write for them not yet completed might cause out of the expected order 
> execution of the writes and data corruption (old data written instead of 
> new).

I don't think that's a problem. Multiple commands in flight to the same
I/O region can get reordered because we only use simple tagging
regardless of advanced or otherwise storage.  The VM seems to wait for
one write to complete before starting another because of the way the
flush threads work.

James

Re: Current topics for LSF10/MM Summit 8-9 August in Boston

[Vladislav Bolkhovitin]

James Bottomley, on 06/17/2010 08:42 PM wrote:
> On Thu, 2010-06-17 at 20:34 +0400, Vladislav Bolkhovitin wrote:
>> James Bottomley, on 06/17/2010 08:07 PM wrote:
>>> On Thu, 2010-06-17 at 18:00 +0200, Christof Schmitt wrote:
>>>> On Wed, Jun 16, 2010 at 03:50:59PM -0500, James Bottomley wrote:
>>>>> Given that we're under two months out, I thought it would be time to
>>>>> post a summary of the topics we've collected so far (Nick will post the
>>>>> MM summit ones later).  Look this over, and if there's anything missing,
>>>>> propose it ... or if you have cross Storage/FS/MM topics, post them too.
>>>>>
>>>>> Oh, and since we're not the most organised bunch, if you posted a topic
>>>>> and don't see it in the list, please resend ... we probably lost it in
>>>>> an email shuffle.
>>>>>
>>>>> Current Filesystem Topics:
>>>>>
>>>>> Alex Elder	Upstream maintainer for XFS, general discussion on FS/IO
>>>>> Aneesh Kumar	Rich-acl patches which work better with NFSv4 acl and CIFS acl
>>>>> Anshul Madan	reflink for NFS
>>>>> Chuck Lever	NFS/IPV6 and NFS O_DIRECT, Wu's read-ahead work, vitro perf tools
>>>>> Eric Sandeen	Advances in testing, TRIM/DISCARD/Alignment, writeback sanity
>>>>> James Lentini	reflink for NFS
>>>>> Jan Kara	Discuss/drive sanity review of writeback and general ext*/jbd 
>>>>> Michael Rubin	Writeback scaling
>>>>> Sage Weil	Statlite, generic interface for describing file striping for distributed FS, VFS scalability
>>>>> Al Viro	Sorting out d_revalidate and other dcache issues
>>>>> Coly Li		directory/large file scalability
>>>>> Sorin Faibish	Cache writeback discussion
>>>>>
>>>>> Current Storage Topics:
>>>>>
>>>>> Eric Seppanen	Next generation SSDs, performance implications on Linux I/O
>>>>> Boaz Harrosh	PNFS performance considerations, bio_list based/async raidN for generic use; stable pages for I/O
>>>>> FUJITA Tomonori	SCSI target mode, iSCSI, block layer SG (bsg), sg, IOMMU, DMA issues
>>>>> Hannes Reinecke	libfc/multipath/error handing
>>>>> James Smart	FCOE proposal for rework of the FC sysfs tree, work with Hannes on other transport/SCSI subsystem topics
>>>>> Jeff Moyer	IO scheduler
>>>>> Joel Becker	SAN management plugin
>>>>> Martin Petersen	Updates on DIF/DIX, TRIM/DISCARD/UNMAP, generic support for WRITE_SAME
>>>>>
>>>>> Plus some MM summit ones which Nick will summarise.
>>>> [...]
>>>>
>>>> What about the topic "Stable pages while IO"?
>>>> http://www.spinics.net/lists/linux-scsi/msg44074.html
>>>>
>>>> Was it lost during the e-mail shuffle or will it be part of the MM topics?
>>> It's actually listed under 'dma issues' ... but there's really been no
>>> satisfactory resolution or discussion of how one might be achieved.
>>> Most filesystems rely on modifications to in-flight pages for efficiency
>>> and copying every fs I/O page would be horrendous both for performance
>>> and memory consumption.  Nor has there really been an indication that
>>> it's a serious issue.  The two sufferers are DIF and iSCSI checksum.
>> You forgot the third: advanced storage, including MPIO clusters, where 
>> retry of the write of the modified in-flight pages while the original 
>> write for them not yet completed might cause out of the expected order 
>> execution of the writes and data corruption (old data written instead of 
>> new).
> 
> I don't think that's a problem. Multiple commands in flight to the same
> I/O region can get reordered because we only use simple tagging
> regardless of advanced or otherwise storage. The VM seems to wait for
> one write to complete before starting another because of the way the
> flush threads work.

I hope so, but: (1) we can see such writes (see 
http://lists.linbit.com/pipermail/drbd-user/2009-April/011891.html, for 
instance) and (2) Boaz said it's possible. From the "seems" you wrote 
looks like your are also not too sure. So, if it isn't possible, it 
would be good if someone familar with VM internals confirmed this.

Gennadiy,

If possible, can you recheck in your setup with a real Linux as 
initiator to confirm if Linux is suffers from the concurrent writes 
you've seen or not, please?

Vlad

Re: Current topics for LSF10/MM Summit 8-9 August in Boston

[James Bottomley]

On Thu, 2010-06-17 at 21:11 +0400, Vladislav Bolkhovitin wrote:
> James Bottomley, on 06/17/2010 08:42 PM wrote:
> > On Thu, 2010-06-17 at 20:34 +0400, Vladislav Bolkhovitin wrote:
> >> James Bottomley, on 06/17/2010 08:07 PM wrote:
> >>> On Thu, 2010-06-17 at 18:00 +0200, Christof Schmitt wrote:
> >>>> On Wed, Jun 16, 2010 at 03:50:59PM -0500, James Bottomley wrote:
> >>>>> Given that we're under two months out, I thought it would be time to
> >>>>> post a summary of the topics we've collected so far (Nick will post the
> >>>>> MM summit ones later).  Look this over, and if there's anything missing,
> >>>>> propose it ... or if you have cross Storage/FS/MM topics, post them too.
> >>>>>
> >>>>> Oh, and since we're not the most organised bunch, if you posted a topic
> >>>>> and don't see it in the list, please resend ... we probably lost it in
> >>>>> an email shuffle.
> >>>>>
> >>>>> Current Filesystem Topics:
> >>>>>
> >>>>> Alex Elder	Upstream maintainer for XFS, general discussion on FS/IO
> >>>>> Aneesh Kumar	Rich-acl patches which work better with NFSv4 acl and CIFS acl
> >>>>> Anshul Madan	reflink for NFS
> >>>>> Chuck Lever	NFS/IPV6 and NFS O_DIRECT, Wu's read-ahead work, vitro perf tools
> >>>>> Eric Sandeen	Advances in testing, TRIM/DISCARD/Alignment, writeback sanity
> >>>>> James Lentini	reflink for NFS
> >>>>> Jan Kara	Discuss/drive sanity review of writeback and general ext*/jbd 
> >>>>> Michael Rubin	Writeback scaling
> >>>>> Sage Weil	Statlite, generic interface for describing file striping for distributed FS, VFS scalability
> >>>>> Al Viro	Sorting out d_revalidate and other dcache issues
> >>>>> Coly Li		directory/large file scalability
> >>>>> Sorin Faibish	Cache writeback discussion
> >>>>>
> >>>>> Current Storage Topics:
> >>>>>
> >>>>> Eric Seppanen	Next generation SSDs, performance implications on Linux I/O
> >>>>> Boaz Harrosh	PNFS performance considerations, bio_list based/async raidN for generic use; stable pages for I/O
> >>>>> FUJITA Tomonori	SCSI target mode, iSCSI, block layer SG (bsg), sg, IOMMU, DMA issues
> >>>>> Hannes Reinecke	libfc/multipath/error handing
> >>>>> James Smart	FCOE proposal for rework of the FC sysfs tree, work with Hannes on other transport/SCSI subsystem topics
> >>>>> Jeff Moyer	IO scheduler
> >>>>> Joel Becker	SAN management plugin
> >>>>> Martin Petersen	Updates on DIF/DIX, TRIM/DISCARD/UNMAP, generic support for WRITE_SAME
> >>>>>
> >>>>> Plus some MM summit ones which Nick will summarise.
> >>>> [...]
> >>>>
> >>>> What about the topic "Stable pages while IO"?
> >>>> http://www.spinics.net/lists/linux-scsi/msg44074.html
> >>>>
> >>>> Was it lost during the e-mail shuffle or will it be part of the MM topics?
> >>> It's actually listed under 'dma issues' ... but there's really been no
> >>> satisfactory resolution or discussion of how one might be achieved.
> >>> Most filesystems rely on modifications to in-flight pages for efficiency
> >>> and copying every fs I/O page would be horrendous both for performance
> >>> and memory consumption.  Nor has there really been an indication that
> >>> it's a serious issue.  The two sufferers are DIF and iSCSI checksum.
> >> You forgot the third: advanced storage, including MPIO clusters, where 
> >> retry of the write of the modified in-flight pages while the original 
> >> write for them not yet completed might cause out of the expected order 
> >> execution of the writes and data corruption (old data written instead of 
> >> new).
> > 
> > I don't think that's a problem. Multiple commands in flight to the same
> > I/O region can get reordered because we only use simple tagging
> > regardless of advanced or otherwise storage. The VM seems to wait for
> > one write to complete before starting another because of the way the
> > flush threads work.
> 
> I hope so, but: (1) we can see such writes (see 
> http://lists.linbit.com/pipermail/drbd-user/2009-April/011891.html, for 
> instance)

So the email says blockio mode ... which I take it isn't through the
pagecache cleaning?  All bets are off if the user initiates the
writeback ... and certainly you can get two blocks in flight for the
same destination using DIRECT IO ... but that's up to the applications
to fix ... we don't guarantee ordering in that case.

James


--
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: Current topics for LSF10/MM Summit 8-9 August in Boston

[Vladislav Bolkhovitin]

James Bottomley, on 06/17/2010 09:37 PM wrote:
> On Thu, 2010-06-17 at 21:11 +0400, Vladislav Bolkhovitin wrote:
>> James Bottomley, on 06/17/2010 08:42 PM wrote:
>>> On Thu, 2010-06-17 at 20:34 +0400, Vladislav Bolkhovitin wrote:
>>>> James Bottomley, on 06/17/2010 08:07 PM wrote:
>>>>> On Thu, 2010-06-17 at 18:00 +0200, Christof Schmitt wrote:
>>>>>> On Wed, Jun 16, 2010 at 03:50:59PM -0500, James Bottomley wrote:
>>>>>>> Given that we're under two months out, I thought it would be time to
>>>>>>> post a summary of the topics we've collected so far (Nick will post the
>>>>>>> MM summit ones later).  Look this over, and if there's anything missing,
>>>>>>> propose it ... or if you have cross Storage/FS/MM topics, post them too.
>>>>>>>
>>>>>>> Oh, and since we're not the most organised bunch, if you posted a topic
>>>>>>> and don't see it in the list, please resend ... we probably lost it in
>>>>>>> an email shuffle.
>>>>>>>
>>>>>>> Current Filesystem Topics:
>>>>>>>
>>>>>>> Alex Elder	Upstream maintainer for XFS, general discussion on FS/IO
>>>>>>> Aneesh Kumar	Rich-acl patches which work better with NFSv4 acl and CIFS acl
>>>>>>> Anshul Madan	reflink for NFS
>>>>>>> Chuck Lever	NFS/IPV6 and NFS O_DIRECT, Wu's read-ahead work, vitro perf tools
>>>>>>> Eric Sandeen	Advances in testing, TRIM/DISCARD/Alignment, writeback sanity
>>>>>>> James Lentini	reflink for NFS
>>>>>>> Jan Kara	Discuss/drive sanity review of writeback and general ext*/jbd 
>>>>>>> Michael Rubin	Writeback scaling
>>>>>>> Sage Weil	Statlite, generic interface for describing file striping for distributed FS, VFS scalability
>>>>>>> Al Viro	Sorting out d_revalidate and other dcache issues
>>>>>>> Coly Li		directory/large file scalability
>>>>>>> Sorin Faibish	Cache writeback discussion
>>>>>>>
>>>>>>> Current Storage Topics:
>>>>>>>
>>>>>>> Eric Seppanen	Next generation SSDs, performance implications on Linux I/O
>>>>>>> Boaz Harrosh	PNFS performance considerations, bio_list based/async raidN for generic use; stable pages for I/O
>>>>>>> FUJITA Tomonori	SCSI target mode, iSCSI, block layer SG (bsg), sg, IOMMU, DMA issues
>>>>>>> Hannes Reinecke	libfc/multipath/error handing
>>>>>>> James Smart	FCOE proposal for rework of the FC sysfs tree, work with Hannes on other transport/SCSI subsystem topics
>>>>>>> Jeff Moyer	IO scheduler
>>>>>>> Joel Becker	SAN management plugin
>>>>>>> Martin Petersen	Updates on DIF/DIX, TRIM/DISCARD/UNMAP, generic support for WRITE_SAME
>>>>>>>
>>>>>>> Plus some MM summit ones which Nick will summarise.
>>>>>> [...]
>>>>>>
>>>>>> What about the topic "Stable pages while IO"?
>>>>>> http://www.spinics.net/lists/linux-scsi/msg44074.html
>>>>>>
>>>>>> Was it lost during the e-mail shuffle or will it be part of the MM topics?
>>>>> It's actually listed under 'dma issues' ... but there's really been no
>>>>> satisfactory resolution or discussion of how one might be achieved.
>>>>> Most filesystems rely on modifications to in-flight pages for efficiency
>>>>> and copying every fs I/O page would be horrendous both for performance
>>>>> and memory consumption.  Nor has there really been an indication that
>>>>> it's a serious issue.  The two sufferers are DIF and iSCSI checksum.
>>>> You forgot the third: advanced storage, including MPIO clusters, where 
>>>> retry of the write of the modified in-flight pages while the original 
>>>> write for them not yet completed might cause out of the expected order 
>>>> execution of the writes and data corruption (old data written instead of 
>>>> new).
>>> I don't think that's a problem. Multiple commands in flight to the same
>>> I/O region can get reordered because we only use simple tagging
>>> regardless of advanced or otherwise storage. The VM seems to wait for
>>> one write to complete before starting another because of the way the
>>> flush threads work.
>> I hope so, but: (1) we can see such writes (see 
>> http://lists.linbit.com/pipermail/drbd-user/2009-April/011891.html, for 
>> instance)
> 
> So the email says blockio mode ... which I take it isn't through the
> pagecache cleaning?  All bets are off if the user initiates the
> writeback ... and certainly you can get two blocks in flight for the
> same destination using DIRECT IO ... but that's up to the applications
> to fix ... we don't guarantee ordering in that case.

That's blockio on the target. The target stack just passed down to its 
backstorage incoming from the initiator SCSI commands as a block 
requests in 1:1 mapping. But the SCSI commands were sent by the 
initiator. So, the concurrent writes came from the initiator.

Gennadiy,

Which application did you use on the initiator to generate load when you 
had the concurrent writes?

Vlad

Re: Current topics for LSF10/MM Summit 8-9 August in Boston

[Christof Schmitt]

On Thu, Jun 17, 2010 at 11:07:30AM -0500, James Bottomley wrote:
> On Thu, 2010-06-17 at 18:00 +0200, Christof Schmitt wrote:
> > On Wed, Jun 16, 2010 at 03:50:59PM -0500, James Bottomley wrote:
> > > Given that we're under two months out, I thought it would be time to
> > > post a summary of the topics we've collected so far (Nick will post the
> > > MM summit ones later).  Look this over, and if there's anything missing,
> > > propose it ... or if you have cross Storage/FS/MM topics, post them too.
> > > 
> > > Oh, and since we're not the most organised bunch, if you posted a topic
> > > and don't see it in the list, please resend ... we probably lost it in
> > > an email shuffle.
> > > 
> > > Current Filesystem Topics:
> > > 
> > > Alex Elder	Upstream maintainer for XFS, general discussion on FS/IO
> > > Aneesh Kumar	Rich-acl patches which work better with NFSv4 acl and CIFS acl
> > > Anshul Madan	reflink for NFS
> > > Chuck Lever	NFS/IPV6 and NFS O_DIRECT, Wu's read-ahead work, vitro perf tools
> > > Eric Sandeen	Advances in testing, TRIM/DISCARD/Alignment, writeback sanity
> > > James Lentini	reflink for NFS
> > > Jan Kara	Discuss/drive sanity review of writeback and general ext*/jbd 
> > > Michael Rubin	Writeback scaling
> > > Sage Weil	Statlite, generic interface for describing file striping for distributed FS, VFS scalability
> > > Al Viro	Sorting out d_revalidate and other dcache issues
> > > Coly Li		directory/large file scalability
> > > Sorin Faibish	Cache writeback discussion
> > > 
> > > Current Storage Topics:
> > > 
> > > Eric Seppanen	Next generation SSDs, performance implications on Linux I/O
> > > Boaz Harrosh	PNFS performance considerations, bio_list based/async raidN for generic use; stable pages for I/O
> > > FUJITA Tomonori	SCSI target mode, iSCSI, block layer SG (bsg), sg, IOMMU, DMA issues
> > > Hannes Reinecke	libfc/multipath/error handing
> > > James Smart	FCOE proposal for rework of the FC sysfs tree, work with Hannes on other transport/SCSI subsystem topics
> > > Jeff Moyer	IO scheduler
> > > Joel Becker	SAN management plugin
> > > Martin Petersen	Updates on DIF/DIX, TRIM/DISCARD/UNMAP, generic support for WRITE_SAME
> > > 
> > > Plus some MM summit ones which Nick will summarise.
> > [...]
> > 
> > What about the topic "Stable pages while IO"?
> > http://www.spinics.net/lists/linux-scsi/msg44074.html
> > 
> > Was it lost during the e-mail shuffle or will it be part of the MM topics?
> 
> It's actually listed under 'dma issues' ... but there's really been no
> satisfactory resolution or discussion of how one might be achieved.
> Most filesystems rely on modifications to in-flight pages for efficiency
> and copying every fs I/O page would be horrendous both for performance
> and memory consumption.  Nor has there really been an indication that
> it's a serious issue.  The two sufferers are DIF and iSCSI checksum.
> The latter generates the checksum late enough that it can just discard
> incorrect pages ... the former might need simply to turn off DIF for
> everything other than DIRECT IO.

It is a serious problem when using DIF, so turning off this feature or
only using XFS and direct i/o does not sound very satisfying. But then
i also see the points that have been discussed and that there is no
simple solution.

Christof

Re: [Lsf10-pc] Current topics for LSF10/MM Summit 8-9 August in Boston

[J. Bruce Fields]

On Thu, Jun 17, 2010 at 11:07:30AM -0500, James Bottomley wrote:
> It's actually listed under 'dma issues' ... but there's really been no
> satisfactory resolution or discussion of how one might be achieved.
> Most filesystems rely on modifications to in-flight pages for efficiency
> and copying every fs I/O page would be horrendous both for performance
> and memory consumption.  Nor has there really been an indication that
> it's a serious issue.  The two sufferers are DIF and iSCSI checksum.

And, again, NFS (both client (on writes) and server (on reads)), when
using sec=krb5i.  Haven't tried to reproduce the problem, but I believe
it would result in spurious IO errors.

--b.

Current MM topics for LSF10/MM Summit 8-9 August in Boston

[Nick Piggin]

Apologies for the delay, there haven't been a lot of explicit proposals
for topics, but I think there is a lot to talk about, so I would like to
hear more ideas but I have suggested a few of my own.

Rik van Riel		Memory management under virtualization (with KVM)
Andrea Arcangeli	Transparent hugepages
KOSAKI Motohiro		get_user_pages vs COW problem
Boaz Harrosh		Stable page contents under writeback
Michael Rubin, Sorin Faibish, Jan Kara	Writeback issues (several subtopics)

I propose some other possible topics for agenda
- Zone aware slab reclaim (eg. dentry, inode)
- mmap_sem scalability, again
- Page allocator scalability
- OOM killer work
- Direct reclaim, direct writeback problems

Thanks,
Nick

On Wed, Jun 16, 2010 at 03:50:59PM -0500, James Bottomley wrote:
> Given that we're under two months out, I thought it would be time to
> post a summary of the topics we've collected so far (Nick will post the
> MM summit ones later).  Look this over, and if there's anything missing,
> propose it ... or if you have cross Storage/FS/MM topics, post them too.
> 
> Oh, and since we're not the most organised bunch, if you posted a topic
> and don't see it in the list, please resend ... we probably lost it in
> an email shuffle.
> 
> Current Filesystem Topics:
> 
> Alex Elder	Upstream maintainer for XFS, general discussion on FS/IO
> Aneesh Kumar	Rich-acl patches which work better with NFSv4 acl and CIFS acl
> Anshul Madan	reflink for NFS
> Chuck Lever	NFS/IPV6 and NFS O_DIRECT, Wu's read-ahead work, vitro perf tools
> Eric Sandeen	Advances in testing, TRIM/DISCARD/Alignment, writeback sanity
> James Lentini	reflink for NFS
> Jan Kara	Discuss/drive sanity review of writeback and general ext*/jbd 
> Michael Rubin	Writeback scaling
> Sage Weil	Statlite, generic interface for describing file striping for distributed FS, VFS scalability
> Al Viro	Sorting out d_revalidate and other dcache issues
> Coly Li		directory/large file scalability
> Sorin Faibish	Cache writeback discussion
> 
> Current Storage Topics:
> 
> Eric Seppanen	Next generation SSDs, performance implications on Linux I/O
> Boaz Harrosh	PNFS performance considerations, bio_list based/async raidN for generic use; stable pages for I/O
> FUJITA Tomonori	SCSI target mode, iSCSI, block layer SG (bsg), sg, IOMMU, DMA issues
> Hannes Reinecke	libfc/multipath/error handing
> James Smart	FCOE proposal for rework of the FC sysfs tree, work with Hannes on other transport/SCSI subsystem topics
> Jeff Moyer	IO scheduler
> Joel Becker	SAN management plugin
> Martin Petersen	Updates on DIF/DIX, TRIM/DISCARD/UNMAP, generic support for WRITE_SAME
> 
> Plus some MM summit ones which Nick will summarise.
> 
> For the benefit of those who've forgotten here's the original Call for
> topics and attendees:
> 
> This year we'll hold the Linux Storage and Filesystems summit jointly
> with the VM summit on the two days before LinuxCon in Boston (that's
> Sunday and Monday) at the Renaissance Hotel:
> 
> http://events.linuxfoundation.org/events/linuxcon
> 
> We're planning to hold some sessions jointly and split into three tracks
> (Filesystems, Storage and VM) for others, so we're encouraging proposals
> for discussion that cover areas relevant to all three groups as well as
> more specific technical topics.
> 
> Suggestions for agenda topics should be sent to
> 
> lsf10-pc@lists.linuxfoundation.org
> 
> and optionally cc the Linux list which would be most interested in it:
> 
> SCSI: linux-scsi@vger.kernel.org
> FS: linux-fsdevel@vger.kernel.org (plus relevant fs specific list)
> MM: linux-mm@kvack.org
> 
> Please tag your subject with [LSF/VM TOPIC] so those of us who're not
> very organised can find them easily in our inboxes.  The agenda topics
> and attendees will be selected by the programme committee, but the final
> agenda will be by formed by consensus of the attendees on the day.
> 
> We'll try to cap attendance at around 20 per track to facilitate
> discussions although the final numbers will depend on the room sizes
> at the venue.
> 
> Requests to attend should be sent to:
> 
> lsf10-pc@lists.linuxfoundation.org
> 
> please summarise what you'll bring to the meeting, and what you'd like
> to discuss.  please also tag your email with [ATTEND] so there's less
> chance of it getting lost in the large mail pile.
> 
> Presentations are allowed to guide discussion, but are strongly
> discouraged.  There will be no recording or audio bridge, however
> written minutes will be published as in previous years:
> 
> 2009:
> http://lwn.net/Articles/327601/
> http://lwn.net/Articles/327740/
> http://lwn.net/Articles/328347/
> 
> Prior years:
> http://www.usenix.org/events/lsf08/tech/lsf08sums.pdf
> http://www.usenix.org/publications/login/2007-06/openpdfs/lsf07sums.pdf
> 
> If you have feedback on last year's meeting that we can use to improve
> this year's, please also send that to:
> 
> lsf10-pc@lists.linuxfoundation.org

Re: [Lsf10-pc] Current MM topics for LSF10/MM Summit 8-9 August in Boston

[Andrea Arcangeli]

On Mon, Jun 21, 2010 at 10:05:26PM +1000, Nick Piggin wrote:
> Andrea Arcangeli	Transparent hugepages

Sure fine on my side. I got a proposal accepted for presentation at
KVM Forum 2010 about it too the days after the VM summit too.

> KOSAKI Motohiro		get_user_pages vs COW problem

Just a side note, not sure exactly what is meant to be discussed about
this bug, considering the fact this is still unsolved isn't technical
problem as there were plenty of fixes available, and the one that seem
to had better chance to get included was the worst one in my view, as
it tried to fix it in a couple of gup caller (but failed, also because
finding all put_page pin release is kind of a pain as they're spread
all over the place and not identified as gup_put_page, and in addition
to the instability and lack of completeness of the fix, it was also
the most inefficient as it added unnecessary and coarse locking) plus
all gup callers are affected, not just a few. I normally call it gup
vs fork race. Luckily not all threaded apps uses O_DIRECT and fork and
pretend to do the direct-io in different sub-page chunks of the same
page from different threads (KVM would probably be affected if it
didn't use MADV_DONTFORK on the O_DIRECT memory, as it might run fork
to execute some network script when adding an hotplug pci net device
for example). But surely we can discuss the fix we prefer for this
bug, or at least we can agree it needs fixing.

Other topics looks very interesting too!

Thanks,
Andrea

--
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: [Lsf10-pc] Current MM topics for LSF10/MM Summit 8-9 August in Boston

[Gleb Natapov]

On Mon, Jun 21, 2010 at 03:16:08PM +0200, Andrea Arcangeli wrote:
> > KOSAKI Motohiro		get_user_pages vs COW problem
> 
> Just a side note, not sure exactly what is meant to be discussed about
> this bug, considering the fact this is still unsolved isn't technical
> problem as there were plenty of fixes available, and the one that seem
> to had better chance to get included was the worst one in my view, as
> it tried to fix it in a couple of gup caller (but failed, also because
> finding all put_page pin release is kind of a pain as they're spread
> all over the place and not identified as gup_put_page, and in addition
> to the instability and lack of completeness of the fix, it was also
> the most inefficient as it added unnecessary and coarse locking) plus
> all gup callers are affected, not just a few. I normally call it gup
> vs fork race. Luckily not all threaded apps uses O_DIRECT and fork and
> pretend to do the direct-io in different sub-page chunks of the same
> page from different threads (KVM would probably be affected if it
> didn't use MADV_DONTFORK on the O_DIRECT memory, as it might run fork
> to execute some network script when adding an hotplug pci net device
> for example). But surely we can discuss the fix we prefer for this
> bug, or at least we can agree it needs fixing.
> 
KVM is actually affected by the bug. The fix was posted today:
http://www.mail-archive.com/kvm@vger.kernel.org/msg36759.html

--
			Gleb.

--
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: [Lsf10-pc] Current MM topics for LSF10/MM Summit 8-9 August in Boston

[Andrea Arcangeli]

On Mon, Jun 21, 2010 at 04:22:38PM +0300, Gleb Natapov wrote:
> On Mon, Jun 21, 2010 at 03:16:08PM +0200, Andrea Arcangeli wrote:
> > > KOSAKI Motohiro		get_user_pages vs COW problem
> > 
> > Just a side note, not sure exactly what is meant to be discussed about
> > this bug, considering the fact this is still unsolved isn't technical
> > problem as there were plenty of fixes available, and the one that seem
> > to had better chance to get included was the worst one in my view, as
> > it tried to fix it in a couple of gup caller (but failed, also because
> > finding all put_page pin release is kind of a pain as they're spread
> > all over the place and not identified as gup_put_page, and in addition
> > to the instability and lack of completeness of the fix, it was also
> > the most inefficient as it added unnecessary and coarse locking) plus
> > all gup callers are affected, not just a few. I normally call it gup
> > vs fork race. Luckily not all threaded apps uses O_DIRECT and fork and
> > pretend to do the direct-io in different sub-page chunks of the same
> > page from different threads (KVM would probably be affected if it
> > didn't use MADV_DONTFORK on the O_DIRECT memory, as it might run fork
> > to execute some network script when adding an hotplug pci net device
> > for example). But surely we can discuss the fix we prefer for this
> > bug, or at least we can agree it needs fixing.
> > 
> KVM is actually affected by the bug. The fix was posted today:
> http://www.mail-archive.com/kvm@vger.kernel.org/msg36759.html

Interesting... so this is the page returned by gup that doesn't match
anymore the page after an user write into qemu context after
fork. Clearly any of the fixes proposed would have prevented this bug
in the first place as they would assign a copy to the child, so yes
it's likely this same bug. It's quite sad to have this workload that
is superfluous if gup would behave as supposed by the caller. Also I'd
prefer if you would use MADV_DONTFORK for the fix, as that will at
least optimize fork and it would still be ok to keep even after we fix
the VM while this workaround of using tmpfs should be backed out.

--
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: [Lsf10-pc] Current MM topics for LSF10/MM Summit 8-9 August in Boston

[Gleb Natapov]

CCing Avi,

On Mon, Jun 21, 2010 at 04:09:39PM +0200, Andrea Arcangeli wrote:
> On Mon, Jun 21, 2010 at 04:22:38PM +0300, Gleb Natapov wrote:
> > On Mon, Jun 21, 2010 at 03:16:08PM +0200, Andrea Arcangeli wrote:
> > > > KOSAKI Motohiro		get_user_pages vs COW problem
> > > 
> > > Just a side note, not sure exactly what is meant to be discussed about
> > > this bug, considering the fact this is still unsolved isn't technical
> > > problem as there were plenty of fixes available, and the one that seem
> > > to had better chance to get included was the worst one in my view, as
> > > it tried to fix it in a couple of gup caller (but failed, also because
> > > finding all put_page pin release is kind of a pain as they're spread
> > > all over the place and not identified as gup_put_page, and in addition
> > > to the instability and lack of completeness of the fix, it was also
> > > the most inefficient as it added unnecessary and coarse locking) plus
> > > all gup callers are affected, not just a few. I normally call it gup
> > > vs fork race. Luckily not all threaded apps uses O_DIRECT and fork and
> > > pretend to do the direct-io in different sub-page chunks of the same
> > > page from different threads (KVM would probably be affected if it
> > > didn't use MADV_DONTFORK on the O_DIRECT memory, as it might run fork
> > > to execute some network script when adding an hotplug pci net device
> > > for example). But surely we can discuss the fix we prefer for this
> > > bug, or at least we can agree it needs fixing.
> > > 
> > KVM is actually affected by the bug. The fix was posted today:
> > http://www.mail-archive.com/kvm@vger.kernel.org/msg36759.html
> 
> Interesting... so this is the page returned by gup that doesn't match
> anymore the page after an user write into qemu context after
> fork. Clearly any of the fixes proposed would have prevented this bug
> in the first place as they would assign a copy to the child, so yes
> it's likely this same bug. It's quite sad to have this workload that
> is superfluous if gup would behave as supposed by the caller. Also I'd
> prefer if you would use MADV_DONTFORK for the fix, as that will at
> least optimize fork and it would still be ok to keep even after we fix
> the VM while this workaround of using tmpfs should be backed out.
Avi did the fix. We discussed using MADV_DONTFORK for that, but calling
madvise() from kernel deemed to be messy.

--
			Gleb.

--
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: [Lsf10-pc] Current MM topics for LSF10/MM Summit 8-9 August in Boston

[Andrea Arcangeli]

On Mon, Jun 21, 2010 at 05:18:56PM +0300, Gleb Natapov wrote:
> Avi did the fix. We discussed using MADV_DONTFORK for that, but calling
> madvise() from kernel deemed to be messy.

Agree that calling madvise looks messy. It's possible to set
VM_DONTCOPY under mmap_sem write mode and it'll work as well.

But surely we can as well keep this quicker fix until the fork vs gup
race is fixed, and back it out later.

--
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: [Lsf10-pc] Current MM topics for LSF10/MM Summit 8-9 August in Boston

[Avi Kivity]

On 06/21/2010 05:29 PM, Andrea Arcangeli wrote:
> On Mon, Jun 21, 2010 at 05:18:56PM +0300, Gleb Natapov wrote:
>    
>> Avi did the fix. We discussed using MADV_DONTFORK for that, but calling
>> madvise() from kernel deemed to be messy.
>>      
> Agree that calling madvise looks messy. It's possible to set
> VM_DONTCOPY under mmap_sem write mode and it'll work as well.
>    

But we aren't guaranteed to get our own vma, yes?

> But surely we can as well keep this quicker fix until the fork vs gup
> race is fixed, and back it out later.
>    

Right.

Note kvm shouldn't be calling do_mmap() in any case.  I let that in 
because it was simple and because we had a userspace interface relying 
on that, but that's no longer the case, so I'll make that page kernel owned.

-- 
error compiling committee.c: too many arguments to function

Re: [Lsf10-pc] Current MM topics for LSF10/MM Summit 8-9 August in Boston

[Andrea Arcangeli]

On Mon, Jun 21, 2010 at 05:31:41PM +0300, Avi Kivity wrote:
> On 06/21/2010 05:29 PM, Andrea Arcangeli wrote:
> > On Mon, Jun 21, 2010 at 05:18:56PM +0300, Gleb Natapov wrote:
> >    
> >> Avi did the fix. We discussed using MADV_DONTFORK for that, but calling
> >> madvise() from kernel deemed to be messy.
> >>      
> > Agree that calling madvise looks messy. It's possible to set
> > VM_DONTCOPY under mmap_sem write mode and it'll work as well.
> >    
> 
> But we aren't guaranteed to get our own vma, yes?

Correct, one would need to call split_vma like madvise_behavior does
before setting the flag. For sure current fix is simpler ;).

> Note kvm shouldn't be calling do_mmap() in any case.  I let that in 
> because it was simple and because we had a userspace interface relying 
> on that, but that's no longer the case, so I'll make that page kernel owned.

Agree ;).

--
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: [Lsf10-pc] Current MM topics for LSF10/MM Summit 8-9 August in Boston

[KOSAKI Motohiro]

> On Mon, Jun 21, 2010 at 10:05:26PM +1000, Nick Piggin wrote:
> > Andrea Arcangeli	Transparent hugepages
> 
> Sure fine on my side. I got a proposal accepted for presentation at
> KVM Forum 2010 about it too the days after the VM summit too.
> 
> > KOSAKI Motohiro		get_user_pages vs COW problem
> 
> Just a side note, not sure exactly what is meant to be discussed about
> this bug, considering the fact this is still unsolved isn't technical
> problem as there were plenty of fixes available, and the one that seem
> to had better chance to get included was the worst one in my view, as
> it tried to fix it in a couple of gup caller (but failed, also because
> finding all put_page pin release is kind of a pain as they're spread
> all over the place and not identified as gup_put_page, and in addition
> to the instability and lack of completeness of the fix, it was also
> the most inefficient as it added unnecessary and coarse locking) plus
> all gup callers are affected, not just a few. I normally call it gup
> vs fork race. Luckily not all threaded apps uses O_DIRECT and fork and
> pretend to do the direct-io in different sub-page chunks of the same
> page from different threads (KVM would probably be affected if it
> didn't use MADV_DONTFORK on the O_DIRECT memory, as it might run fork
> to execute some network script when adding an hotplug pci net device
> for example). But surely we can discuss the fix we prefer for this
> bug, or at least we can agree it needs fixing.

If people don't want this. I'm ok to drop this. In my personal concern
most important topics are following four topics. (again it's only _my_
concern)


Rik van Riel		Memory management under virtualization (with KVM)
Andrea Arcangeli	Transparent hugepages
- mmap_sem scalability, again
- Direct reclaim, direct writeback problems

Re: Current MM topics for LSF10/MM Summit 8-9 August in Boston

[David Rientjes]

On Mon, 21 Jun 2010, Nick Piggin wrote:

> - OOM killer work

There's a talk about this during the conference which is geared toward 
both developers and sysadmins so it would probably be redundant at the 
summit.

--
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: Current MM topics for LSF10/MM Summit 8-9 August in Boston

[James Bottomley]

On Mon, 2010-06-21 at 13:00 -0700, David Rientjes wrote:
> On Mon, 21 Jun 2010, Nick Piggin wrote:
> 
> > - OOM killer work
> 
> There's a talk about this during the conference which is geared toward 
> both developers and sysadmins so it would probably be redundant at the 
> summit.

I really don't think so; the two things should be orthogonal.  The
Summit is usually about forward looking stuff (and getting good people
together to discuss it).  Even the best tech talk is usually only
telling users what's there, how to use it, what the history is and
possibly what the author is doing to fix any problems.

James