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

Groups > linux.kernel > #1300902 > unrolled thread

Re: [RFD] CAT user space interface revisited

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: [RFD] CAT user space interface revisited Marcelo Tosatti <mtosatti@redhat.com> - 2016-01-04 18:30 +0100
    Re: [RFD] CAT user space interface revisited Marcelo Tosatti <mtosatti@redhat.com> - 2016-01-04 18:50 +0100
    Re: [RFD] CAT user space interface revisited Thomas Gleixner <tglx@linutronix.de> - 2016-01-06 00:20 +0100
      Re: [RFD] CAT user space interface revisited Marcelo Tosatti <mtosatti@redhat.com> - 2016-01-06 13:50 +0100
        Re: [RFD] CAT user space interface revisited Tejun Heo <tj@kernel.org> - 2016-01-06 14:20 +0100
      Re: [RFD] CAT user space interface revisited Thomas Gleixner <tglx@linutronix.de> - 2016-01-08 21:30 +0100

#1300902 — Re: [RFD] CAT user space interface revisited

On Thu, Dec 31, 2015 at 11:30:57PM +0100, Thomas Gleixner wrote:
> Marcelo,
> 
> On Thu, 31 Dec 2015, Marcelo Tosatti wrote:
> 
> First of all thanks for the explanation.
> 
> > There is one directory structure in this topic, CAT. That is the
> > directory structure which is exposed to userspace to control the 
> > CAT HW. 
> > 
> > With the current patchset posted by Intel ("Subject: [PATCH V16 00/11]
> > x86: Intel Cache Allocation Technology Support"), the directory
> > structure there (the files and directories exposed by that patchset)
> > (*1) does not allow one to configure different CBM masks on each socket
> > (that is, it forces the user to configure the same mask CBM on every
> > socket). This is a blocker for us, and it is one of the points in your
> > proposal.
> > 
> > There was a call between Red Hat and Intel where it was communicated
> > to Intel, and Intel agreed, that it was necessary to fix this (fix this
> > == allow different CBM masks on different sockets).
> > 
> > Now, that is one change to the current directory structure (*1).
> 
> I don't have an idea how that would look like. The current structure is a
> cgroups based hierarchy oriented approach, which does not allow simple things
> like
> 
> T1	00001111
> T2	00111100
> 
> at least not in a way which is natural to the problem at hand.



	cgroupA/

		cbm_mask  (if set, set for all CPUs)

		socket1/cbm_mask
		socket2/cbm_mask
		...
		socketN/cbm_mask (if set, overrides global
		cbm_mask).

Something along those lines.

Do you see any problem with it?

> > (*1) modified to allow for different CBM masks on different sockets, 
> > lets say (*2), is what we have been waiting for Intel to post. 
> > It would handle our usecase, and all use-cases which the current
> > patchset from Intel already handles (Vikas posted emails mentioning
> > there are happy users of the current interface, feel free to ask 
> > him for more details).
> 
> I cannot imagine how that modification to the current interface would solve
> that. Not to talk about per CPU associations which are not related to tasks at
> all.

Not sure what you mean by per CPU associations.

If you fix a cbmmask on a given pCPU, say CPU1, and control which tasks
run on that pCPU, then you control the cbmmask for all tasks (say
tasklist-1) on that CPU, fine.

Can achieve the same by putting all tasks from tasklist-1 into a
cgroup.

> > What i have asked you, and you replied "to go Google read my previous
> > post" is this:
> > What are the advantages over you proposal (which is a completely
> > different directory structure, requiring a complete rewrite),
> > over (*2) ?
> > 
> > (what is my reason behind this: the reason is that if you, with
> > maintainer veto power, forces your proposal to be accepted, it will be
> > necessary to wait for another rewrite (a new set of problems, fully
> > think through your proposal, test it, ...) rather than simply modify an
> > already known, reviewed, already used directory structure.
> > 
> > And functionally, your proposal adds nothing to (*2) (other than, well,
> > being a different directory structure).
> 
> Sorry. I cannot see at all how a modification to the existing interface would
> cover all the sensible use cases I described in a coherent way. I really want
> to see a proper description of the interface before people start hacking on it
> in a frenzy. What you described is: "let's say (*2)" modification. That's
> pretty meager.
> 
> > If Fenghua or you post a patchset, say in 2 weeks, with your proposal,
> > i am fine with that. But i since i doubt that will be the case, i am 
> > pushing for the interface which requires the least amount of changes
> > (and therefore the least amount of time) to be integrated.
> > 
> > >From your email:
> > 
> > "It would even be sufficient for particular use cases to just associate
> > a piece of cache to a given CPU and do not bother with tasks at all.
> > 
> > We really need to make this as configurable as possible from userspace
> > without imposing random restrictions to it. I played around with it on
> > my new intel toy and the restriction to 16 COS ids (that's 8 with CDP
> > enabled) makes it really useless if we force the ids to have the same
> > meaning on all sockets and restrict it to per task partitioning."
> > 
> > Yes, thats the issue we hit, that is the modification that was agreed
> > with Intel, and thats what we are waiting for them to post.
> 
> How do you implement the above - especially that part:
> 
>  "It would even be sufficient for particular use cases to just associate a
>   piece of cache to a given CPU and do not bother with tasks at all."
> 
> as a "simple" modification to (*1) ?

As noted above.
>  
> > > I described a directory structure for that qos/cat stuff in my proposal and
> > > that's complete AFAICT.
> > 
> > Ok, lets make the job for the submitter easier. You are the maintainer,
> > so you decide.
> > 
> > Is it enough for you to have (*2) (which was agreed with Intel), or 
> > would you rather prefer to integrate the directory structure at 
> > "[RFD] CAT user space interface revisited" ?
> 
> The only thing I care about as a maintainer is, that we merge something which
> actually reflects the properties of the hardware and gives the admin the
> required flexibility to utilize it fully. I don't care at all if it's my
> proposal or something else which allows to do the same.
> 
> Let me copy the relevant bits from my proposal here once more and let me ask
> questions to the various points so you can tell me how that modification to
> (*1) is going to deal with that.
> 
> >> At top level:
> >>
> >>  xxxxxxx/cat/max_cosids <- Assume that all CPUs are the same
> >>  xxxxxxx/cat/max_maskbits <- Assume that all CPUs are the same

This can be exposed to userspace via a file.

> >>  xxxxxxx/cat/cdp_enable <- Depends on CDP availability
> 
>  Where is that information in (*2) and how is that related to (*1)? If you
>  think it's not required, please explain why.

Intel has come up with a scheme to implement CDP. I'll go read 
that and reply to this email afterwards.

> >> Per socket data:
> >>
> >>  xxxxxxx/cat/socket-0/
> >>  ...
> >>  xxxxxxx/cat/socket-N/l3_size
> >>  xxxxxxx/cat/socket-N/hwsharedbits
> 
>  Where is that information in (*2) and how is that related to (*1)? If you
>  think it's not required, please explain why.

l3_size: userspace can figure that by itself (exposed somewhere in
sysfs).

hwsharedbits: All userspace needs to know is
which bits are shared with HW, to decide whether or not to use that
region of a given socket for a given cbmmask.

So expose that userspace, fine. Can do that in cgroups.

> >> Per socket mask data:
> >>
> >>  xxxxxxx/cat/socket-N/cos-id-0/
> >>  ...
> >>  xxxxxxx/cat/socket-N/cos-id-N/inuse
> >>                               /cat_mask
> >>                               /cdp_mask <- Data mask if CDP enabled
> 
>  Where is that information in (*2) and how is that related to (*1)? If you
>  think it's not required, please explain why.

Unsure - will reply in next email (but per-socket information seems
independent of that).

> 
> >> Per cpu default cos id for the cpus on that socket:
> >> 
> >>  xxxxxxx/cat/socket-N/cpu-x/default_cosid
> >>  ...
> >>  xxxxxxx/cat/socket-N/cpu-N/default_cosid
> >>
> >> The above allows a simple cpu based partitioning. All tasks which do
> >> not have a cache partition assigned on a particular socket use the
> >> default one of the cpu they are running on.
> 
>  Where is that information in (*2) and how is that related to (*1)? If you
>  think it's not required, please explain why.

Not required because with current Intel patchset you'd do:


# mount | grep rdt
cgroup on /sys/fs/cgroup/intel_rdt type cgroup
(rw,nosuid,nodev,noexec,relatime,intel_rdt)
# cd /sys/fs/cgroup/intel_rdt
# ls
cgroupALL              cgroup.procs  cgroup.sane_behavior
notify_on_release  tasks
cgroup.clone_children  cgroupRSVD    intel_rdt.l3_cbm	   release_agent
# cat tasks
1042
1066
1067
1069
...
# cd cgroupALL/
ps auxw  | while read i; do echo $i ; done
| cut -f 2 -d " "   | grep -v PID | while read x ; do echo $x > tasks;
done
-bash: echo: write error: No such process
-bash: echo: write error: No such process
-bash: echo: write error: No such process
-bash: echo: write error: No such process

# cat ../tasks | while read i; do echo $i > tasks; done
# cat ../tasks  | wc -l

0
(no tasks on root cgroup)

# cd ../cgroupRSVD
# cgroupRSVD]# cat tasks
# ps auxw | grep postfix
root	   1942  0.0  0.0  91136  4860 ?        Ss   Nov25   0:00
/usr/libexec/postfix/master -w
postfix    1981  0.0  0.0  91308  6520 ?        S    Nov25   0:00 qmgr
-l -t unix -u
postfix    4416  0.0  0.0  91240  6296 ?        S    17:05   0:00 pickup
-l -t unix -u
root	   4486  0.0  0.0 112652  2304 pts/0    S+   17:31   0:00 grep
--color=auto postfix
# echo 4416 > tasks
# cat intel_rdt.l3_cbm
000ffff0
# cat ../cgroupALL/intel_rdt.l3_cbm
000000ff

Bits f0 are shared between cgroupRSVD and cgroupALL. Lets change:
# echo 0xf > ../cgroupALL/intel_rdt.l3_cbm
# cat ../cgroupALL/intel_rdt.l3_cbm
0000000f

Now they share none.

------

So i have created "cgroupALL" (what you call default_cosid) and
"cgroupRSVD".

> 
> >> Now for the task(s) partitioning:
> >>
> >>  xxxxxxx/cat/partitions/
> >>
> >> Under that directory one can create partitions
> >> 
> >>  xxxxxxx/cat/partitions/p1/tasks
> >>                           /socket-0/cosid
> >>                           ...
> >>                           /socket-n/cosid
> >> 
> >> The default value for the per socket cosid is COSID_DEFAULT, which
> >> causes the task(s) to use the per cpu default id. 
> 
>  Where is that information in (*2) and how is that related to (*1)? If you
>  think it's not required, please explain why.
> 
> Yes. I ask the same question several times and I really want to see the
> directory/interface structure which solves all of the above before anyone
> starts to implement it. 

I don't see the problem, have a sequence of commands above which shows
to set a directory structure which is useful and does what the HW 
interface is supposed to do.

> We already have a completely useless interface (*1)
> and there is no point to implement another one based on it (*2) just because
> it solves your particular issue and is the fastest way forward. User space
> interfaces are hard and we really do not need some half baken solution which
> we have to support forever.

Fine. Can you please tell me what i can't do with the current interface?
AFAICS everything can be done (except missing support for (*2)).

> 
> Let me enumerate the required points again:
> 
>     1) Information about the hardware properties

Fine. Intel should expose that information.

>     2) Integration of CAT and CDP 

Fine, Intel has comeup with a directory structure for that, 
let me read the patchset again and i'll reply to you.

>     3) Per socket cos-id partitioning

(*2) as listed in the beginning of this e-mail.

>     4) Per cpu default cos-id association

This already exists, and as noted in the command sequence above,
works just fine. Please explain what problem are you seeing.

>     5) Task association to cos-id

Not sure what that means. Please explain.

> 
> Can you please explain in a simple directory based scheme, like the one I gave
> you how all of these points are going to be solved with a modification to (*1)?
> 
> Thanks,
> 
> 	tglx

Thanks Thomas, this style discussion is quite useful.

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

[toc] | [next] | [standalone]

#1300904

On Mon, Jan 04, 2016 at 03:20:54PM -0200, Marcelo Tosatti wrote:
> On Thu, Dec 31, 2015 at 11:30:57PM +0100, Thomas Gleixner wrote:
> > Marcelo,
> > 
> > On Thu, 31 Dec 2015, Marcelo Tosatti wrote:
> > 
> > First of all thanks for the explanation.
> > 
> > > There is one directory structure in this topic, CAT. That is the
> > > directory structure which is exposed to userspace to control the 
> > > CAT HW. 
> > > 
> > > With the current patchset posted by Intel ("Subject: [PATCH V16 00/11]
> > > x86: Intel Cache Allocation Technology Support"), the directory
> > > structure there (the files and directories exposed by that patchset)
> > > (*1) does not allow one to configure different CBM masks on each socket
> > > (that is, it forces the user to configure the same mask CBM on every
> > > socket). This is a blocker for us, and it is one of the points in your
> > > proposal.
> > > 
> > > There was a call between Red Hat and Intel where it was communicated
> > > to Intel, and Intel agreed, that it was necessary to fix this (fix this
> > > == allow different CBM masks on different sockets).
> > > 
> > > Now, that is one change to the current directory structure (*1).
> > 
> > I don't have an idea how that would look like. The current structure is a
> > cgroups based hierarchy oriented approach, which does not allow simple things
> > like
> > 
> > T1	00001111
> > T2	00111100
> > 
> > at least not in a way which is natural to the problem at hand.
> 
> 
> 
> 	cgroupA/
> 
> 		cbm_mask  (if set, set for all CPUs)
> 
> 		socket1/cbm_mask
> 		socket2/cbm_mask
> 		...
> 		socketN/cbm_mask (if set, overrides global
> 		cbm_mask).
> 
> Something along those lines.
> 
> Do you see any problem with it?
> 
> > > (*1) modified to allow for different CBM masks on different sockets, 
> > > lets say (*2), is what we have been waiting for Intel to post. 
> > > It would handle our usecase, and all use-cases which the current
> > > patchset from Intel already handles (Vikas posted emails mentioning
> > > there are happy users of the current interface, feel free to ask 
> > > him for more details).
> > 
> > I cannot imagine how that modification to the current interface would solve
> > that. Not to talk about per CPU associations which are not related to tasks at
> > all.
> 
> Not sure what you mean by per CPU associations.
> 
> If you fix a cbmmask on a given pCPU, say CPU1, and control which tasks
> run on that pCPU, then you control the cbmmask for all tasks (say
> tasklist-1) on that CPU, fine.
> 
> Can achieve the same by putting all tasks from tasklist-1 into a
> cgroup.
> 
> > > What i have asked you, and you replied "to go Google read my previous
> > > post" is this:
> > > What are the advantages over you proposal (which is a completely
> > > different directory structure, requiring a complete rewrite),
> > > over (*2) ?
> > > 
> > > (what is my reason behind this: the reason is that if you, with
> > > maintainer veto power, forces your proposal to be accepted, it will be
> > > necessary to wait for another rewrite (a new set of problems, fully
> > > think through your proposal, test it, ...) rather than simply modify an
> > > already known, reviewed, already used directory structure.
> > > 
> > > And functionally, your proposal adds nothing to (*2) (other than, well,
> > > being a different directory structure).
> > 
> > Sorry. I cannot see at all how a modification to the existing interface would
> > cover all the sensible use cases I described in a coherent way. I really want
> > to see a proper description of the interface before people start hacking on it
> > in a frenzy. What you described is: "let's say (*2)" modification. That's
> > pretty meager.
> > 
> > > If Fenghua or you post a patchset, say in 2 weeks, with your proposal,
> > > i am fine with that. But i since i doubt that will be the case, i am 
> > > pushing for the interface which requires the least amount of changes
> > > (and therefore the least amount of time) to be integrated.
> > > 
> > > >From your email:
> > > 
> > > "It would even be sufficient for particular use cases to just associate
> > > a piece of cache to a given CPU and do not bother with tasks at all.
> > > 
> > > We really need to make this as configurable as possible from userspace
> > > without imposing random restrictions to it. I played around with it on
> > > my new intel toy and the restriction to 16 COS ids (that's 8 with CDP
> > > enabled) makes it really useless if we force the ids to have the same
> > > meaning on all sockets and restrict it to per task partitioning."
> > > 
> > > Yes, thats the issue we hit, that is the modification that was agreed
> > > with Intel, and thats what we are waiting for them to post.
> > 
> > How do you implement the above - especially that part:
> > 
> >  "It would even be sufficient for particular use cases to just associate a
> >   piece of cache to a given CPU and do not bother with tasks at all."
> > 
> > as a "simple" modification to (*1) ?
> 
> As noted above.
> >  
> > > > I described a directory structure for that qos/cat stuff in my proposal and
> > > > that's complete AFAICT.
> > > 
> > > Ok, lets make the job for the submitter easier. You are the maintainer,
> > > so you decide.
> > > 
> > > Is it enough for you to have (*2) (which was agreed with Intel), or 
> > > would you rather prefer to integrate the directory structure at 
> > > "[RFD] CAT user space interface revisited" ?
> > 
> > The only thing I care about as a maintainer is, that we merge something which
> > actually reflects the properties of the hardware and gives the admin the
> > required flexibility to utilize it fully. I don't care at all if it's my
> > proposal or something else which allows to do the same.
> > 
> > Let me copy the relevant bits from my proposal here once more and let me ask
> > questions to the various points so you can tell me how that modification to
> > (*1) is going to deal with that.
> > 
> > >> At top level:
> > >>
> > >>  xxxxxxx/cat/max_cosids <- Assume that all CPUs are the same
> > >>  xxxxxxx/cat/max_maskbits <- Assume that all CPUs are the same
> 
> This can be exposed to userspace via a file.
> 
> > >>  xxxxxxx/cat/cdp_enable <- Depends on CDP availability
> > 
> >  Where is that information in (*2) and how is that related to (*1)? If you
> >  think it's not required, please explain why.
> 
> Intel has come up with a scheme to implement CDP. I'll go read 
> that and reply to this email afterwards.

Pasting relevant parts of the patchset submission. 
Looks fine to me, two files, one for data cache cbmmask, another
for instruction cache cbm mask. 
Those two files would be moved to "socket-N" directories.

(will review the CDP patchset...).

Subject: [PATCH V2 0/5] x86: Intel Code Data Prioritization Support

This patch set supports Intel code data prioritization which is an
extension of cache allocation and allows to allocate code and data cache
seperately. It also includes cgroup interface for the user as seperate
patches. The cgroup interface for cache alloc is also resent.

This patch adds enumeration support for Code Data Prioritization(CDP)
feature found in future Intel Xeon processors. It includes CPUID
enumeration routines for CDP.

CDP is an extension to Cache Allocation and lets threads allocate subset
of L3 cache for code and data separately. The allocation is represented
by the code or data cache capacity bit mask(cbm) MSRs
IA32_L3_QOS_MASK_n. Each Class of service would be associated with one
dcache_cbm and one icache_cbm MSR and hence the number of available
CLOSids is halved with CDP. The association for a CLOSid 'n' is shown
below :

data_cbm_address (n) = base + (n <<1)
code_cbm_address (n) = base + (n <<1) +1.
During scheduling the kernel writes the CLOSid
of the thread to IA32_PQR_ASSOC_MSR.

Adds two files to the intel_rdt cgroup 'dcache_cbm' and 'icache_cbm'
when code data prioritization(cdp) support is present. The files
represent the data capacity bit mask(cbm) and instruction cbm for L3
cache. User can specify the data and code cbm and the threads belonging
to the cgroup would get to fill the l3 cache represented by the cbm with
the data or code.

For ex: Consider a scenario where the max cbm bits is 10 and L3 cache
size is 10MB:
then specifying a dcache_cbm = 0x3 and icache_cbm = 0xc would give 2MB
of exclusive cache for data and code for the tasks to fill in.

This feature is an extension to cache allocation and lets user specify a
capacity for code and data separately. Initially these cbms would have
the same value as the l3_cbm(which represents the common cbm for code
and data). Once the user tries to write to either the dcache_cbm or
icache_cbm, the kernel tries to enable the cdp mode in hardware by
writing to the IA32_PQOS_CFG MSR. The switch is only possible if the
number of Class of service IDs(CLOSids) used is < half of total CLOSids
available at the time of switch. This is because the CLOSIds are halved
once CDP is enabled and each CLOSid now maps to a data IA32_L3_QOS_n MSR
and a code IA32_L3_QOS_n MSR.
Once the CDP is enabled user can use the dcache_cbm and icache_cbm just
like the l3_cbm. The CLOSids are not exposed to the user and maintained
by the kernel internally.

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

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

#1302263

Marcelo,

On Mon, 4 Jan 2016, Marcelo Tosatti wrote:
> On Thu, Dec 31, 2015 at 11:30:57PM +0100, Thomas Gleixner wrote:
> > I don't have an idea how that would look like. The current structure is a
> > cgroups based hierarchy oriented approach, which does not allow simple things
> > like
> > 
> > T1	00001111
> > T2	00111100
> > 
> > at least not in a way which is natural to the problem at hand.
> 
> 
> 
> 	cgroupA/
> 
> 		cbm_mask  (if set, set for all CPUs)

You mean sockets, right?

> 
> 		socket1/cbm_mask
> 		socket2/cbm_mask
> 		...
> 		socketN/cbm_mask (if set, overrides global
> 		cbm_mask).
> 
> Something along those lines.
> 
> Do you see any problem with it?

So for that case:

task1:	 cbm_mask 00001111
task2:	 cbm_mask 00111100

i.e. task1 and task2 share bit 2/3 of the mask. 

I need to have two cgroups: cgroup1 and cgroup2, task1 is member of cgroup1
and task2 is member of cgroup2, right?

So now add some more of this and then figure out, which cbm_masks are in use
on which socket. That means I need to go through all cgroups and find the
cbm_masks there.

With my proposed directory structure you get a very clear view about the
in-use closids and the associated cbm_masks. That view represents the hardware
in the best way. With the cgroups stuff we get an artificial representation
which does not tell us anything about the in-use closids and the associated
cbm_masks.
 
> > I cannot imagine how that modification to the current interface would solve
> > that. Not to talk about per CPU associations which are not related to tasks at
> > all.
> 
> Not sure what you mean by per CPU associations.

As I wrote before:

 "It would even be sufficient for particular use cases to just associate
  a piece of cache to a given CPU and do not bother with tasks at all."

> If you fix a cbmmask on a given pCPU, say CPU1, and control which tasks
> run on that pCPU, then you control the cbmmask for all tasks (say
> tasklist-1) on that CPU, fine.
> 
> Can achieve the same by putting all tasks from tasklist-1 into a
> cgroup.

Which means, that I need to go and find everything including kernel threads
and put them into a particular cgroup. That's really not useful and it simply
does not work:

To which cgroup belongs a dynamically created per cpu worker thread? To the
cgroup of the parent. But is the parent necessarily in the proper cgroup? No,
there is no guarantee. So it ends up in some random cgroup unless I start
chasing every new thread, instead of letting it use the default cosid of the
CPU.

Having a per cpu default cos-id which is used when the task does not have a
cos-id associated makes a lot of sense and makes it simpler to utilize that
facility.

> > >> Per cpu default cos id for the cpus on that socket:
> > >> 
> > >>  xxxxxxx/cat/socket-N/cpu-x/default_cosid
> > >>  ...
> > >>  xxxxxxx/cat/socket-N/cpu-N/default_cosid
> > >>
> > >> The above allows a simple cpu based partitioning. All tasks which do
> > >> not have a cache partition assigned on a particular socket use the
> > >> default one of the cpu they are running on.
> > 
> >  Where is that information in (*2) and how is that related to (*1)? If you
> >  think it's not required, please explain why.
> 
> Not required because with current Intel patchset you'd do:

<SNIP>
...
</SNIP>

> # cat intel_rdt.l3_cbm
> 000ffff0
> # cat ../cgroupALL/intel_rdt.l3_cbm
> 000000ff
> 
> Bits f0 are shared between cgroupRSVD and cgroupALL. Lets change:
> # echo 0xf > ../cgroupALL/intel_rdt.l3_cbm
> # cat ../cgroupALL/intel_rdt.l3_cbm
> 0000000f
> 
> Now they share none.

Well, you changed ALL and everything, but you still did not assign a
particular cos-id to a particular CPU as their default.
 
> > >> Now for the task(s) partitioning:
> > >>
> > >>  xxxxxxx/cat/partitions/
> > >>
> > >> Under that directory one can create partitions
> > >> 
> > >>  xxxxxxx/cat/partitions/p1/tasks
> > >>                           /socket-0/cosid
> > >>                           ...
> > >>                           /socket-n/cosid
> > >> 
> > >> The default value for the per socket cosid is COSID_DEFAULT, which
> > >> causes the task(s) to use the per cpu default id. 
> > 
> >  Where is that information in (*2) and how is that related to (*1)? If you
> >  think it's not required, please explain why.
> > 
> > Yes. I ask the same question several times and I really want to see the
> > directory/interface structure which solves all of the above before anyone
> > starts to implement it. 
> 
> I don't see the problem, have a sequence of commands above which shows
> to set a directory structure which is useful and does what the HW 
> interface is supposed to do.

Well, you have a sequence of commands, which gives you the result which you
need for your particular problem.

> > We already have a completely useless interface (*1)
> > and there is no point to implement another one based on it (*2) just because
> > it solves your particular issue and is the fastest way forward. User space
> > interfaces are hard and we really do not need some half baken solution which
> > we have to support forever.
> 
> Fine. Can you please tell me what i can't do with the current interface?
> AFAICS everything can be done (except missing support for (*2)).

1) There is no consistent view of the facility. Is a sysadmin supposed to add
   printks to the kernel to figure that out or should he keep track of that
   information on a piece of paper? Neither option is useful if you have to
   analyze a system which was set up 3 month ago.

2) Simple and easy default settings for tasks and CPUs

   Rather than forcing the admin to find everything which might be related,
   it's way better to have configurable defaults.

   The cgroup interface allows you to set a task default, because everything
   which is in the root group is going to use that, but that default is
   useless. See #3

   That still does not give me the an simple and easy to use way to set a per
   cpu default.
 
3) Non hierarchical setup

   The current interface has a rdt_root_group, which is set up at init. That
   group uses a closid (one of the few we have). And you cannot use that thing
   for anything else than having all bits set in the mask because all groups
   you create underneath must be a subset of the parent group.

   That is simply crap.

   We force something which is entirely not hierarchical into a structure
   which is designed for hierarchical problems and thereby waste one of the
   scarce and pretious resources.

   That technology is straight forward partitioning and has nothing
   hierarchical at all.

The decision to use cgroups was wrong in the beginning and it does not become
better by pretending that it solves some particular use cases and by repeating
that everything can be solved with it.

If all I have is a hammer I certainly can pretend that everything is a
nail. We all know how well that works ....

> >     4) Per cpu default cos-id association
> 
> This already exists, and as noted in the command sequence above,
> works just fine. Please explain what problem are you seeing.

No it does not exist. You emulate it by forcing stuff into cgroups which is
not at all required if you have a proper and well thought out interface.

> >     5) Task association to cos-id
> 
> Not sure what that means. Please explain.

> > >>  xxxxxxx/cat/partitions/p1/tasks
> > >>                           /socket-0/cosid
> > >>                           ...
> > >>                           /socket-n/cosid

Yes, I agree, that this is very similar to the cgroup mechanism, but it is not
in a pointless hierarchy. It's just the last step of the mechanism which I
proposed to represent the hardware in the best way and give the admin the
required flexibility. Again:

This is general information:

xxxxxxx/cat/max_cosids
xxxxxxx/cat/max_maskbits
xxxxxxx/cat/cdp_enable

This is per socket information and per socket cos-id management

xxxxxxx/cat/socket-0/...
xxxxxxx/cat/socket-N/hwsharedbits
		    /cos-id-0/...
		    /cos-id-N/in-use
			     /cat_mask
			     /cdp_mask

This is per cpu default cos-id association

xxxxxxx/cat/socket-0/...
xxxxxxx/cat/socket-N/cpu-x/default_cosid
xxxxxxx/cat/socket-N/cpu-N/default_cosid

This is partitioning, where tasks are associated.

xxxxxxx/cat/partitions/
xxxxxxx/cat/partitions/p1/tasks
			 /socket-0/cosid
			 /socket-N/cosid

That's the part which can be expressed with cgroups somehow, but for the price
of losing a cosid and having a pointless hierarchy. Again, there is nothing
hierarchical in RDT/CAT/CDP. It's all about partitioning and unfortunately the
number of possible partitions is massively limited.

I asked you last time already, but you just gave me random shell commands to
show that it can be done. I ask again:

Can you please explain in a simple directory based scheme, like the one I
gave you above how all of these points are going to be solved with "some
modifications" to the existing cgroup thingy.

And just for completeness, lets look at a practical real world use case:

   1  Socket
  18  CPUs
   4  COSIds (Yes, that's all that hardware gives me)
  32  mask bits
   2  hw shared bits at position 30 and 31

  Have 4 CPU partitions:
  
  CPU  0 -  5   general tasks
  CPU  6 -  9   RT1
  CPU 10 - 13   RT2
  CPU 14 - 17   RT3

, Let each CPU partition have 1/4 of the cache.

  Here is my solution:

  # echo 0xff000000 > xxxx/cat/socket-0/cosid-0
  # echo 0x00ff0000 > xxxx/cat/socket-0/cosid-1
  # echo 0x0000ff00 > xxxx/cat/socket-0/cosid-2
  # echo 0x000000ff > xxxx/cat/socket-0/cosid-3

  # for CPU in 0 1 2 3 4 5; do
  #  echo 0 > xxxx/cat/socket-0/cpu-$CPU/default_cosid;
  # done

  # CPU=6
  # while [ $I -lt 18 ]; do
  #   let "ID = 1 + (($CPU - 6) / 4)"
  #   echo $ID > xxxx/cat/socket-0/cpu-$CPU/default_cosid;
  #   let "CPU += 1"
  # done

That's it. Simple, right?

The real interesting thing here is, that you can't do that at all with that
current cgroup thingy. Simply because you run out of cosids.

Even if you have enough COSids on a newer CPU then your solution will be way
more complex and you still have not solved the issues of chasing kernel
threads etc.

Thanks,

	tglx
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

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

#1302734

On Wed, Jan 06, 2016 at 12:09:50AM +0100, Thomas Gleixner wrote:
> Marcelo,
> 
> On Mon, 4 Jan 2016, Marcelo Tosatti wrote:
> > On Thu, Dec 31, 2015 at 11:30:57PM +0100, Thomas Gleixner wrote:
> > > I don't have an idea how that would look like. The current structure is a
> > > cgroups based hierarchy oriented approach, which does not allow simple things
> > > like
> > > 
> > > T1	00001111
> > > T2	00111100
> > > 
> > > at least not in a way which is natural to the problem at hand.
> > 
> > 
> > 
> > 	cgroupA/
> > 
> > 		cbm_mask  (if set, set for all CPUs)
> 
> You mean sockets, right?
> 
> > 
> > 		socket1/cbm_mask
> > 		socket2/cbm_mask
> > 		...
> > 		socketN/cbm_mask (if set, overrides global
> > 		cbm_mask).
> > 
> > Something along those lines.
> > 
> > Do you see any problem with it?
> 
> So for that case:
> 
> task1:	 cbm_mask 00001111
> task2:	 cbm_mask 00111100
> 
> i.e. task1 and task2 share bit 2/3 of the mask. 
> 
> I need to have two cgroups: cgroup1 and cgroup2, task1 is member of cgroup1
> and task2 is member of cgroup2, right?
> 
> So now add some more of this and then figure out, which cbm_masks are in use
> on which socket. That means I need to go through all cgroups and find the
> cbm_masks there.

Yes.

> With my proposed directory structure you get a very clear view about the
> in-use closids and the associated cbm_masks. That view represents the hardware
> in the best way. With the cgroups stuff we get an artificial representation
> which does not tell us anything about the in-use closids and the associated
> cbm_masks.

Because you expose cos-ID ---> cbm / cdp masks.

Fine, agree thats nice.

> > > I cannot imagine how that modification to the current interface would solve
> > > that. Not to talk about per CPU associations which are not related to tasks at
> > > all.
> > 
> > Not sure what you mean by per CPU associations.
> 
> As I wrote before:
> 
>  "It would even be sufficient for particular use cases to just associate
>   a piece of cache to a given CPU and do not bother with tasks at all."
> 
> > If you fix a cbmmask on a given pCPU, say CPU1, and control which tasks
> > run on that pCPU, then you control the cbmmask for all tasks (say
> > tasklist-1) on that CPU, fine.
> > 
> > Can achieve the same by putting all tasks from tasklist-1 into a
> > cgroup.
> 
> Which means, that I need to go and find everything including kernel threads
> and put them into a particular cgroup. That's really not useful and it simply
> does not work:
> 
> To which cgroup belongs a dynamically created per cpu worker thread? To the
> cgroup of the parent. But is the parent necessarily in the proper cgroup? No,
> there is no guarantee. So it ends up in some random cgroup unless I start
> chasing every new thread, instead of letting it use the default cosid of the
> CPU.

Well, i suppose cgroups has facilities to handle this? That is, what is
required is:

On task creation, move the new task to a particular cgroup, based on
some visible characteristic of the task: (process name matching OR explicit
kernel thread creator specification OR ...).

Because there are two cases. Consider a kernel thread T, which contains 
code that is timing sensitive therefore requires to use a COSID (which
means use a reserved portion of cache).

Case 1) kernel thread T starts kernel thread R, which is also timing
sensitive (and wants to use the same COSID as kernel thread T). 
In that case, the cgroup's default (inherit cgroup from parent)
behaviour is correct.

Case 2) kernel thread T starts kernel thread X, which is not timing
sensitive, therefore kernel thread X should use "default cosid".
In the case of cgroups, in the example used elsewhere in this thread,
kernel thread X should be moved to "cgroupsALL".

Strictly speaking there is a third case:

Case 3) kernel thread T starts kernel thread Z, which wants to
be moved to a different COSID other than kernel thread T's COSID.

So using the default COSID is not necessarily the correct thing to do
(this should be configurable on a per-case basis).

> Having a per cpu default cos-id which is used when the task does not have a
> cos-id associated makes a lot of sense and makes it simpler to utilize that
> facility.

You would need a facility to switch to "inherit cgroup from parent"
mode, and also to handle case 3 (which i supposed cgroups did, because 
the same problem exists for other cgroup controllers).

> > > >> Per cpu default cos id for the cpus on that socket:
> > > >> 
> > > >>  xxxxxxx/cat/socket-N/cpu-x/default_cosid
> > > >>  ...
> > > >>  xxxxxxx/cat/socket-N/cpu-N/default_cosid
> > > >>
> > > >> The above allows a simple cpu based partitioning. All tasks which do
> > > >> not have a cache partition assigned on a particular socket use the
> > > >> default one of the cpu they are running on.
> > > 
> > >  Where is that information in (*2) and how is that related to (*1)? If you
> > >  think it's not required, please explain why.
> > 
> > Not required because with current Intel patchset you'd do:
> 
> <SNIP>
> ...
> </SNIP>
> 
> > # cat intel_rdt.l3_cbm
> > 000ffff0
> > # cat ../cgroupALL/intel_rdt.l3_cbm
> > 000000ff
> > 
> > Bits f0 are shared between cgroupRSVD and cgroupALL. Lets change:
> > # echo 0xf > ../cgroupALL/intel_rdt.l3_cbm
> > # cat ../cgroupALL/intel_rdt.l3_cbm
> > 0000000f
> > 
> > Now they share none.
> 
> Well, you changed ALL and everything, but you still did not assign a
> particular cos-id to a particular CPU as their default.
>  
> > > >> Now for the task(s) partitioning:
> > > >>
> > > >>  xxxxxxx/cat/partitions/
> > > >>
> > > >> Under that directory one can create partitions
> > > >> 
> > > >>  xxxxxxx/cat/partitions/p1/tasks
> > > >>                           /socket-0/cosid
> > > >>                           ...
> > > >>                           /socket-n/cosid
> > > >> 
> > > >> The default value for the per socket cosid is COSID_DEFAULT, which
> > > >> causes the task(s) to use the per cpu default id. 
> > > 
> > >  Where is that information in (*2) and how is that related to (*1)? If you
> > >  think it's not required, please explain why.
> > > 
> > > Yes. I ask the same question several times and I really want to see the
> > > directory/interface structure which solves all of the above before anyone
> > > starts to implement it. 
> > 
> > I don't see the problem, have a sequence of commands above which shows
> > to set a directory structure which is useful and does what the HW 
> > interface is supposed to do.
> 
> Well, you have a sequence of commands, which gives you the result which you
> need for your particular problem.
> 
> > > We already have a completely useless interface (*1)
> > > and there is no point to implement another one based on it (*2) just because
> > > it solves your particular issue and is the fastest way forward. User space
> > > interfaces are hard and we really do not need some half baken solution which
> > > we have to support forever.
> > 
> > Fine. Can you please tell me what i can't do with the current interface?
> > AFAICS everything can be done (except missing support for (*2)).
> 
> 1) There is no consistent view of the facility. Is a sysadmin supposed to add
>    printks to the kernel to figure that out or should he keep track of that
>    information on a piece of paper? Neither option is useful if you have to
>    analyze a system which was set up 3 month ago.

Parse the cgroups CAT directory.

> 2) Simple and easy default settings for tasks and CPUs
> 
>    Rather than forcing the admin to find everything which might be related,
>    it's way better to have configurable defaults.
> 
>    The cgroup interface allows you to set a task default, because everything
>    which is in the root group is going to use that, but that default is
>    useless. See #3
> 
>    That still does not give me the an simple and easy to use way to set a per
>    cpu default.

I also dislike the cgroups interface, your proposal is indeed nicer.

>  
> 3) Non hierarchical setup
> 
>    The current interface has a rdt_root_group, which is set up at init. That
>    group uses a closid (one of the few we have). And you cannot use that thing
>    for anything else than having all bits set in the mask because all groups
>    you create underneath must be a subset of the parent group.
> 
>    That is simply crap.
> 
>    We force something which is entirely not hierarchical into a structure
>    which is designed for hierarchical problems and thereby waste one of the
>    scarce and pretious resources.

Agree.

>    That technology is straight forward partitioning and has nothing
>    hierarchical at all.
> 
> The decision to use cgroups was wrong in the beginning and it does not become
> better by pretending that it solves some particular use cases and by repeating
> that everything can be solved with it.
> 
> If all I have is a hammer I certainly can pretend that everything is a
> nail. We all know how well that works ....
> 
> > >     4) Per cpu default cos-id association
> > 
> > This already exists, and as noted in the command sequence above,
> > works just fine. Please explain what problem are you seeing.
> 
> No it does not exist. You emulate it by forcing stuff into cgroups which is
> not at all required if you have a proper and well thought out interface.
> 
> > >     5) Task association to cos-id
> > 
> > Not sure what that means. Please explain.
> 
> > > >>  xxxxxxx/cat/partitions/p1/tasks
> > > >>                           /socket-0/cosid
> > > >>                           ...
> > > >>                           /socket-n/cosid
> 
> Yes, I agree, that this is very similar to the cgroup mechanism, but it is not
> in a pointless hierarchy. It's just the last step of the mechanism which I
> proposed to represent the hardware in the best way and give the admin the
> required flexibility. Again:
> 
> This is general information:
> 
> xxxxxxx/cat/max_cosids
> xxxxxxx/cat/max_maskbits
> xxxxxxx/cat/cdp_enable
> 
> This is per socket information and per socket cos-id management
> 
> xxxxxxx/cat/socket-0/...
> xxxxxxx/cat/socket-N/hwsharedbits
> 		    /cos-id-0/...
> 		    /cos-id-N/in-use
> 			     /cat_mask
> 			     /cdp_mask
> 
> This is per cpu default cos-id association
> 
> xxxxxxx/cat/socket-0/...
> xxxxxxx/cat/socket-N/cpu-x/default_cosid
> xxxxxxx/cat/socket-N/cpu-N/default_cosid
> 
> This is partitioning, where tasks are associated.
> 
> xxxxxxx/cat/partitions/
> xxxxxxx/cat/partitions/p1/tasks
> 			 /socket-0/cosid
> 			 /socket-N/cosid
> 
> That's the part which can be expressed with cgroups somehow, but for the price
> of losing a cosid and having a pointless hierarchy. Again, there is nothing
> hierarchical in RDT/CAT/CDP. It's all about partitioning and unfortunately the
> number of possible partitions is massively limited.
> 
> I asked you last time already, but you just gave me random shell commands to
> show that it can be done. I ask again:
> 
> Can you please explain in a simple directory based scheme, like the one I
> gave you above how all of these points are going to be solved with "some
> modifications" to the existing cgroup thingy.
> 
> And just for completeness, lets look at a practical real world use case:
> 
>    1  Socket
>   18  CPUs
>    4  COSIds (Yes, that's all that hardware gives me)
>   32  mask bits
>    2  hw shared bits at position 30 and 31
> 
>   Have 4 CPU partitions:
>   
>   CPU  0 -  5   general tasks
>   CPU  6 -  9   RT1
>   CPU 10 - 13   RT2
>   CPU 14 - 17   RT3
> 
> , Let each CPU partition have 1/4 of the cache.
> 
>   Here is my solution:
> 
>   # echo 0xff000000 > xxxx/cat/socket-0/cosid-0
>   # echo 0x00ff0000 > xxxx/cat/socket-0/cosid-1
>   # echo 0x0000ff00 > xxxx/cat/socket-0/cosid-2
>   # echo 0x000000ff > xxxx/cat/socket-0/cosid-3
> 
>   # for CPU in 0 1 2 3 4 5; do
>   #  echo 0 > xxxx/cat/socket-0/cpu-$CPU/default_cosid;
>   # done
> 
>   # CPU=6
>   # while [ $I -lt 18 ]; do
>   #   let "ID = 1 + (($CPU - 6) / 4)"
>   #   echo $ID > xxxx/cat/socket-0/cpu-$CPU/default_cosid;
>   #   let "CPU += 1"
>   # done
> 
> That's it. Simple, right?
> 
> The real interesting thing here is, that you can't do that at all with that
> current cgroup thingy. Simply because you run out of cosids.
> 
> Even if you have enough COSids on a newer CPU then your solution will be way
> more complex and you still have not solved the issues of chasing kernel
> threads etc.
> 
> Thanks,
> 
> 	tglx

Fine, i agree. Need to solve the problem of assignment of cosids on
creation of kernel threads as discussed above (the 3 points).

Fenghua what are your thoughts?

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

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

#1302755

Hello, Marcelo.

On Wed, Jan 06, 2016 at 10:46:15AM -0200, Marcelo Tosatti wrote:
> Well, i suppose cgroups has facilities to handle this? That is, what is
> required is:

No, it doesn't.

> On task creation, move the new task to a particular cgroup, based on
> some visible characteristic of the task: (process name matching OR explicit
> kernel thread creator specification OR ...).

cgroup's primary goal is resource tracking and control.  For userland
processes, following fork / clone is enough; however, for a lot of
kthread tasks, task isn't even the right unit.  e.g. Think of CPU
cycles spent on packet reception, spawning per-cgroup kthreads to
handle packet rx separately isn't a realistic option.  The granularity
needs to be higher.  Except for a handful of cases, this pattern
holds.  Another example is IO resources spent during journal write.
Most of in-kernel resource tracking can't be split per-kthread.

While assigning kthreads to specific cgroups can be useful for a few
specific use cases, in terms of in-kernel resource tracking, it's more
of a distraction.

Please stop using cgroup for random task grouping.  Supporting the
level of flexibility to support arbitrary grouping gets in the way of
implementing proper resource control.  You won't be happy because
cgroup's rules get in the way and cgroup won't be happy because your
random stuff gets in the way of proper resource control.

Thomas's proposal obviously works better for the task at hand.  Maybe
there's something which can be extracted out of cgroup and shared for
task group tracking, if nothing else, hooks and synchronization, but
please don't tack it on top of cgroup when it doesn't really fit the
hierarchical resource distribution model.

Thanks.

-- 
tejun
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

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

#1304987

On Wed, 6 Jan 2016, Thomas Gleixner wrote:
> 
> This is general information:
> 
> xxxxxxx/cat/max_cosids
> xxxxxxx/cat/max_maskbits
> xxxxxxx/cat/cdp_enable
> 
> This is per socket information and per socket cos-id management

Just had a discussion with HPA about this and he pointed out that socket might
not be the proper solution. The correct partitioning unit is cache domains. So
we should change socket-X to L3-domain-X. That'll allow us to integrate future
things like CAT on L2.

Thanks,

	tglx

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web