This patch series add sysfs interface to control CPU's hardware
prefetch behavior for performance tuning from userspace for the
processor A64FX and x86 (on supported CPU).

Changes from v4:
  - remove core driver
  - fix to use attribute_group instead of attribute (patch 1, 4)
  - fix to store a pointer of CPU number in driver_data when the
    initialization of sysfs (patch 1, 4)
  - move #if in .c file to .h file (patch 3)
  - fix drivers to be loaded automatically (patch 1, 4)
    - for x86, it is tied to the discovery of a specific CPU model
    - for A64FX, it is tied to the discovery of a specific ACPI device
  - add Kconfig description (patch 2, 5)
https://lore.kernel.org/lkml/20220518063032.2377351-1-tarumizu.kohei@fujitsu.com/

[Background]
============
A64FX and some Intel processors have implementation-dependent register
for controlling CPU's hardware prefetch behavior. A64FX has
IMP_PF_STREAM_DETECT_CTRL_EL0[1], and Intel processors have MSR 0x1a4
(MSR_MISC_FEATURE_CONTROL)[2]. These registers cannot be accessed from
userspace.

[1]https://github.com/fujitsu/A64FX/tree/master/doc/
   A64FX_Specification_HPC_Extension_v1_EN.pdf

[2]https://www.intel.com/content/www/us/en/developer/articles/technical/intel-sdm.html
    Volume 4

The advantage of using these is improved performance. As an example of
performance improvements, the results of running the Stream benchmark
on the A64FX are described in section [Merit].

For MSR 0x1a4, it is also possible to change the value from userspace
via the MSR driver. However, using MSR driver is not recommended, so
it needs a proper kernel interface[3].

[3]https://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git/about/

For these reasons, we provide a new proper kernel interface to control
both IMP_PF_STREAM_DETECT_CTRL_EL0 and MSR 0x1a4.

[Overall design]
================
This driver creates "prefetch_control" directory and some attributes
in every CPU's cache/indexX directory, if CPU supports hardware
prefetch control behavior.

Detailed description of this sysfs interface is in
Documentation/ABI/testing/sysfs-devices-system-cpu (patch6).

This driver needs cache sysfs directory and cache level/type
information. In ARM processor, these information can be obtained
from registers even without ACPI PPTT.
We add processing to create a cache/index directory using only the
information from the register if the machine does not support ACPI
PPTT and CONFIG_ALLOW_INCOMPLETE_CACHE_SYSFS is true in patch3.
This action caused a problem and is described in [Known problem].

[Examples]
==========
This section provides an example of using this sysfs interface at the
x86's model of INTEL_FAM6_BROADWELL_X.

This model has the following register specifications:

[0]    L2 Hardware Prefetcher Disable (R/W)
[1]    L2 Adjacent Cache Line Prefetcher Disable (R/W)
[2]    DCU Hardware Prefetcher Disable (R/W)
[3]    DCU IP Prefetcher Disable (R/W)
[63:4] Reserved

In this case, index0 (L1d cache) corresponds to bit[2,3] and index2
(L2 cache) corresponds to bit [0,1]. A list of attribute files of
index0 and index2 in CPU1 at BROADWELL_X is following:

```
# ls /sys/devices/system/cpu/cpu1/cache/index0/prefetch_control/

hardware_prefetcher_enable
ip_prefetcher_enable

# ls /sys/devices/system/cpu/cpu1/cache/index2/prefetch_control/

adjacent_cache_line_prefetcher_enable
hardware_prefetcher_enable
```

If user would like to disable the setting of "L2 Adjacent Cache Line
Prefetcher Disable (R/W)" in CPU1, do the following:

```
# echo 0 >
# /sys/devices/system/cpu/cpu1/cache/index2/prefetch_control/adjacent_cache_line_prefetcher_enable
```

In another example, a list of index0 at A64FX is following:

```
# ls /sys/devices/system/cpu/cpu1/cache/index0/prefetch_control/

stream_detect_prefetcher_dist
stream_detect_prefetcher_enable
stream_detect_prefetcher_strength
stream_detect_prefetcher_strength_available
```

[Patch organizations]
=====================
This patch series add hardware prefetch control driver for A64FX
and x86.

- patch1: Add hardware prefetch control driver for A64FX

  Adds module init/exit code to create sysfs attributes for A64FX with
  "stream_detect_prefetcher_enable", "stream_detect_prefetcher_strong"
  and "stream_detect_prefetcher_dist".

- patch2: Add Kconfig/Makefile to build driver for A64FX

- patch3: Create cache sysfs directory without ACPI PPTT for hardware
  prefetch control

  Hardware Prefetch control driver needs cache sysfs directory and cache
  level/type information. In ARM processor, these information can be
  obtained from register(CLIDR_EL1) even without PPTT. Therefore, we
  set the cpu_map_populated to true to create cache sysfs directory, if
  the machine doesn't have PPTT.

- patch4: Add hardware prefetch control driver for x86

  Adds module init/exit code to create sysfs attributes for x86 with
  "hardware_prefetcher_enable", "ip_prefetcher_enable" and
  "adjacent_cache_line_prefetcher_enable".

- patch5: Add Kconfig/Makefile to build driver for x86

- patch6: Add documentation for the new sysfs interface

[Known problem]
===============
- `lscpu` command terminates with -ENOENT because cache/index directory
  is exists but shared_cpu_map file does not exist. This is due to
  patch5, which creates a cache/index directory containing only level
  and type without ACPI PPTT.

[Merit]
=======
For reference, here is the result of STREAM Triad when tuning with
the "s file in L1 and L2 cache on A64FX.

| dist combination  | Pattern A   | Pattern B   |
|-------------------|-------------|-------------|
| L1:256,  L2:1024  | 234505.2144 | 114600.0801 |
| L1:1536, L2:1024  | 279172.8742 | 118979.4542 |
| L1:256,  L2:10240 | 247716.7757 | 127364.1533 |
| L1:1536, L2:10240 | 283675.6625 | 125950.6847 |

In pattern A, we set the size of the array to 174720, which is about
half the size of the L1d cache. In pattern B, we set the size of the
array to 10485120, which is about twice the size of the L2 cache.

In pattern A, a change of dist at L1 has a larger effect. On the other
hand, in pattern B, the change of dist at L2 has a larger effect.
As described above, the optimal dist combination depends on the
characteristics of the application. Therefore, such a sysfs interface
is useful for performance tuning.

Best regards,
Kohei Tarumizu

Kohei Tarumizu (6):
  soc: fujitsu: Add hardware prefetch control driver for A64FX
  soc: fujitsu: Add Kconfig/Makefile to build hardware prefetch control
    driver
  arm64: Create cache sysfs directory without ACPI PPTT for hardware
    prefetch control
  x86: Add hardware prefetch control driver for x86
  x86: Add Kconfig/Makefile to build hardware prefetch control driver
  docs: ABI: Add sysfs documentation interface of hardware prefetch
    control driver

 .../ABI/testing/sysfs-devices-system-cpu      |  98 ++++
 MAINTAINERS                                   |   6 +
 arch/arm64/kernel/cacheinfo.c                 |  27 +
 arch/x86/Kconfig                              |  17 +
 arch/x86/kernel/cpu/Makefile                  |   2 +
 arch/x86/kernel/cpu/x86-pfctl.c               | 363 +++++++++++++
 drivers/acpi/pptt.c                           |  18 +
 drivers/soc/Kconfig                           |   1 +
 drivers/soc/Makefile                          |   1 +
 drivers/soc/fujitsu/Kconfig                   |  26 +
 drivers/soc/fujitsu/Makefile                  |   2 +
 drivers/soc/fujitsu/a64fx-pfctl.c             | 484 ++++++++++++++++++
 include/linux/acpi.h                          |   5 +
 include/linux/cacheinfo.h                     |  10 +
 14 files changed, 1060 insertions(+)
 create mode 100644 arch/x86/kernel/cpu/x86-pfctl.c
 create mode 100644 drivers/soc/fujitsu/Kconfig
 create mode 100644 drivers/soc/fujitsu/Makefile
 create mode 100644 drivers/soc/fujitsu/a64fx-pfctl.c

-- 
2.27.0

On Tue, Jun 7, 2022 at 2:07 PM Kohei Tarumizu
<tarumizu.kohei@fujitsu.com> wrote:

> This patch series add sysfs interface to control CPU's hardware
> prefetch behavior for performance tuning from userspace for the
> processor A64FX and x86 (on supported CPU).

OK

> A64FX and some Intel processors have implementation-dependent register
> for controlling CPU's hardware prefetch behavior. A64FX has
> IMP_PF_STREAM_DETECT_CTRL_EL0[1], and Intel processors have MSR 0x1a4
> (MSR_MISC_FEATURE_CONTROL)[2].

Hardware prefetch (I guess of memory contents) is a memory hierarchy feature.

Linux has a memory hierarchy manager, conveniently named "mm",
developed by some of the smartest people I know. The main problem
addressed by that is paging, but prefetching into the CPU from the
next lowest level in the memory hierarchy is just another memory
hierarchy hardware feature, such as hard disks, primary RAM etc.

> These registers cannot be accessed from userspace.

Good. The kernel managed hardware. If the memory hierarchy people have
userspace now doing stuff behind their back, through some special
interface, that makes their world more complicated.

This looks like it needs information from the generic memory manager,
from the scheduler, and possibly all the way down from the block
layer to do the right thing, so it has no business in userspace.
Have you seen mm/damon for example? Access to statistics for
memory access patterns seems really useful for tuning the behaviour
of this hardware. Just my €0.01.

If it does interact with userspace I suppose it should be using control
groups, like everything else of this type, see e.g. mm/memcontrol.c,
not custom sysfs files.

Just an example from one of the patches:

+                       - "* Adjacent Cache Line Prefetcher Disable (R/W)"
+                           corresponds to the
"adjacent_cache_line_prefetcher_enable"

I might only be on "a little knowledge is dangerous" on the memory
manager topics, but I know for sure that they at times adjust the members
of structs to fit nicely on cache lines. And now this? It looks really useful
for kernel machinery that know very well what needs to go into the cache
line next and when.

Talk to the people on linux-mm and memory maintainer Andrew Morton on
how to do this right, it's a really interesting feature! Also given
that people say
that the memory hierarchy is an important part in the performance of the Apple
M1 (M2) silicon, I expect that machine to have this too?

Yours,
Linus Walleij

Hi Linus,

Thanks for the comment.

> OK
> 
> > A64FX and some Intel processors have implementation-dependent register
> > for controlling CPU's hardware prefetch behavior. A64FX has
> > IMP_PF_STREAM_DETECT_CTRL_EL0[1], and Intel processors have MSR
> 0x1a4
> > (MSR_MISC_FEATURE_CONTROL)[2].
> 
> Hardware prefetch (I guess of memory contents) is a memory hierarchy feature.
> 
> Linux has a memory hierarchy manager, conveniently named "mm", developed
> by some of the smartest people I know. The main problem addressed by that is
> paging, but prefetching into the CPU from the next lowest level in the memory
> hierarchy is just another memory hierarchy hardware feature, such as hard
> disks, primary RAM etc.
> 
> > These registers cannot be accessed from userspace.
> 
> Good. The kernel managed hardware. If the memory hierarchy people have
> userspace now doing stuff behind their back, through some special interface,
> that makes their world more complicated.
> 
> This looks like it needs information from the generic memory manager, from the
> scheduler, and possibly all the way down from the block layer to do the right
> thing, so it has no business in userspace.
> Have you seen mm/damon for example? Access to statistics for memory
> access patterns seems really useful for tuning the behaviour of this hardware.
> Just my €0.01.

Thank you for the information. I will see if mm/damon statistics can
be used for tuning.

> If it does interact with userspace I suppose it should be using control groups,
> like everything else of this type, see e.g. mm/memcontrol.c, not custom sysfs
> files.

Hardware prefetch registers exist for each core, and the settings are
independent for each cache. Therefore, currently, I created it under
/sys/devices/system/cpu/cpu*/cache/index*.
However, when user actually configure it for an application, they may
want to set it on a per-process basis. Considering that, I think
control groups is suitable for this usage.

For example, is your idea of interface like the following?

```
  /sys/fs/cgroup/memory/memory.hardware_prefetcher.enable
```

Cpuset controller has information about which CPU a process belonging
to a group is bound to, so maybe cpuset controller is more appropriate.

Control groups has hierarchical structure, so it is necessary to
consider whether they can map hardware prefetch behavior well.
Currentry I have two concerns.
First, upper hierarchy contains the same CPU as the lower hierarchy.
In this case, it may not be possible to configure independent setting
in each hierarchy.
Next, context switch considerations. This function rewrites the
value of the register that exists for each core. Therefore, the
register value must be changed at the timing of the context switch
with a process belonging to a different group.

> Just an example from one of the patches:
> 
> +                       - "* Adjacent Cache Line Prefetcher Disable (R/W)"
> +                           corresponds to the
> "adjacent_cache_line_prefetcher_enable"
> 
> I might only be on "a little knowledge is dangerous" on the memory manager
> topics, but I know for sure that they at times adjust the members of structs to fit
> nicely on cache lines. And now this? It looks really useful for kernel machinery
> that know very well what needs to go into the cache line next and when.
> 
> Talk to the people on linux-mm and memory maintainer Andrew Morton on how
> to do this right, it's a really interesting feature! Also given that people say that
> the memory hierarchy is an important part in the performance of the Apple
> M1 (M2) silicon, I expect that machine to have this too?

I think this proposal will be useful for users, so I will proceed
with concrete studies and talk to the MM people.

On Tue, Jun 07, 2022 at 09:05:24PM +0900, Kohei Tarumizu wrote:
> This patch series add sysfs interface to control CPU's hardware
> prefetch behavior for performance tuning from userspace for the
> processor A64FX and x86 (on supported CPU).

Why does userspace want to even do this?

How will they do this?

What programs will do this?

And why isn't just automatic and why does this hardware require manual
intervention to work properly?

thanks,

greg k-h

Thanks for the comment.

> Why does userspace want to even do this?

This is because the optimal settings may differ from application to
application.

Examples of performance improvements for applications with simple
memory access characteristics are described in [merit] section.
However, some applications have complex characteristics, so it is
difficult to predict if an application will improve without actually
trying it out.

This is not necessary for all applications. However, I want to provide
as a minimal interface that can be used by those who want to improve
their application even a little.

> How will they do this?

I assume to be used to tune a specific core and execute an application
on that core. The execution example is as follows.

1) The user tunes the parameters of a specific core before executing
   the program.

```
# echo 1024 > /sys/devices/system/cpu/cpu12/cache/index0/prefetch_control/stream_detect_prefetcher_dist
# echo 1024 > /sys/devices/system/cpu/cpu12/cache/index2/prefetch_control/stream_detect_prefetcher_dist
# echo 1024 > /sys/devices/system/cpu/cpu13/cache/index0/prefetch_control/stream_detect_prefetcher_dist
# echo 1024 > /sys/devices/system/cpu/cpu13/cache/index2/prefetch_control/stream_detect_prefetcher_dist
```

2) Execute the program bound to the target core.

```
# taskset -c 12-13 a.out
```

If the interface is exposed, the user can develop a library to execute
1) and 2) operation instead.

> What programs will do this?

It is assumed to be used by programs that execute many continuous
memory access. It may be useful for other applications, but I can't
explain them in detail right away.

> And why isn't just automatic and why does this hardware require manual
> intervention to work properly?

It is difficult for the hardware to determine the optimal parameters
in advance. Therefore, I think that the register is provided to change
the behavior of the hardware.

On Tue, Jun 14, 2022 at 11:55:39AM +0000, tarumizu.kohei@fujitsu.com wrote:
> Thanks for the comment.
> 
> > Why does userspace want to even do this?
> 
> This is because the optimal settings may differ from application to
> application.

That's not ok.  Linux is a "general purpose" operating system and needs
to work well for all applications.  Doing application-specific-tuning
based on the specific hardware like this is a nightmare for users, and
will be for you as you will now have to support this specific model to
work correctly on all future kernel releases for the next 20+ years.
Are you willing to do that?

> Examples of performance improvements for applications with simple
> memory access characteristics are described in [merit] section.
> However, some applications have complex characteristics, so it is
> difficult to predict if an application will improve without actually
> trying it out.

Then perhaps it isn't anything that they should try out :)

Shouldn't the kernel know how the application works (based on the
resources it asks for) and tune itself based on that automatically?

If not, how is a user supposed to know how to do this?

> This is not necessary for all applications. However, I want to provide
> as a minimal interface that can be used by those who want to improve
> their application even a little.
> 
> > How will they do this?
> 
> I assume to be used to tune a specific core and execute an application
> on that core. The execution example is as follows.
> 
> 1) The user tunes the parameters of a specific core before executing
>    the program.
> 
> ```
> # echo 1024 > /sys/devices/system/cpu/cpu12/cache/index0/prefetch_control/stream_detect_prefetcher_dist
> # echo 1024 > /sys/devices/system/cpu/cpu12/cache/index2/prefetch_control/stream_detect_prefetcher_dist
> # echo 1024 > /sys/devices/system/cpu/cpu13/cache/index0/prefetch_control/stream_detect_prefetcher_dist
> # echo 1024 > /sys/devices/system/cpu/cpu13/cache/index2/prefetch_control/stream_detect_prefetcher_dist
> ```

What is "1024" here?  Where is any of this documented?  And why these
specific sysfs files and not others?

> 2) Execute the program bound to the target core.
> 
> ```
> # taskset -c 12-13 a.out
> ```
> 
> If the interface is exposed, the user can develop a library to execute
> 1) and 2) operation instead.

If you have no such user today, nor a library, how do you know any of
this works well?  And again, how will you support this going forward?
Or is this specific api only going to be for one specific piece of
hardware and never any future ones?

> > What programs will do this?
> 
> It is assumed to be used by programs that execute many continuous
> memory access. It may be useful for other applications, but I can't
> explain them in detail right away.

So you haven't tested this on any real applications?  We need real users
before being able to add new apis.  Otherwise we can just remove the
apis :)

> > And why isn't just automatic and why does this hardware require manual
> > intervention to work properly?
> 
> It is difficult for the hardware to determine the optimal parameters
> in advance. Therefore, I think that the register is provided to change
> the behavior of the hardware.

Kernel programming for a general purpose operating system is hard, but
it is possible :)

good luck!

greg k-h

Hi Greg,

> That's not ok.  Linux is a "general purpose" operating system and needs to
> work well for all applications.  Doing application-specific-tuning based on the
> specific hardware like this is a nightmare for users,

Hardware prefetch behavior is enabled by default in x86 and A64FX.
Many applications can perform well without changing the register
setting. Use this feature for some applications that want to be
improved performance.

In particular, A64FX's hardware prefetch behavior is used for HPC
applications. The user running HPC applications needs to improve
performance as much as possible. This feature is useful for such
users. Therefore, some A64FX machines have their own drivers that
control hardware prefetch behavior. It is built into the software
products for A64FX and cannot be used without purchase.

I want to make this feature available to people who want to improve
performance without purchase the product. This is limited in use and
depends on the characteristics of the application. Isn't this match
with "general purpose"?

> and will be for you as you
> will now have to support this specific model to work correctly on all future
> kernel releases for the next 20+ years.
> Are you willing to do that?

Rather than relying on a specific model of this API, I want to make it
generally available. However, it may not be so now. I am willing to
support this if I could make it a community-approved interface.

> Then perhaps it isn't anything that they should try out :)
> 
> Shouldn't the kernel know how the application works (based on the resources
> it asks for) and tune itself based on that automatically?
> 
> If not, how is a user supposed to know how to do this?

It is useful for users if it can be done automatically by the kernel.
I will consider if there is anything I can do using statistical
information.

> What is "1024" here?  Where is any of this documented?

This parameter means the difference in bytes between the memory
address the program is currently accessing and the memory address
accessed by the hardware prefetch. My document in
sysfs-devices-system-cpu does not specify what the distance means, so
I will add it.

For reference, the hardware prefetch details are described below.

"https://github.com/fujitsu/A64FX/tree/master/doc"
    A64FX_Microarchitecture_Manual_en_1.7.pdf

> And why these
> specific sysfs files and not others?

I wanted to show an example of changing only the hardware prefetch
distance. There is no special reason not to specify with other sysfs
files.

> If you have no such user today, nor a library, how do you know any of this works
> well?

The prefetch control function included in the software product for
A64FX does the similar operation, and it works well.

> And again, how will you support this going forward?
> Or is this specific api only going to be for one specific piece of hardware and
> never any future ones?

In order to make the interface widely usable in the future, I will
consider a different specification from the current one. For example
control groups which Linus proposaled is one of them.

> So you haven't tested this on any real applications?  We need real users before
> being able to add new apis.  Otherwise we can just remove the apis :)

At least, some A64FX users use this behavior. However, currently, I
don't have which applications and how much performance will be
improved. I will try to get the application actually used and confirm
that it is effective.

> Kernel programming for a general purpose operating system is hard, but it is
> possible :)

I will try to do kernel programming for a general purpose operating
system.

On Fri, Jun 17, 2022 at 11:21 AM tarumizu.kohei@fujitsu.com
<tarumizu.kohei@fujitsu.com> wrote:

Jumping in here.

> Hi Greg,
>
> > That's not ok.  Linux is a "general purpose" operating system and needs to
> > work well for all applications.  Doing application-specific-tuning based on the
> > specific hardware like this is a nightmare for users,
>
> Hardware prefetch behavior is enabled by default in x86 and A64FX.
> Many applications can perform well without changing the register
> setting. Use this feature for some applications that want to be
> improved performance.

The right way to solve this is to make the Linux kernel contain the necessary
heuristics to identify which tasks and thus cores need this to improve
efficiency and then apply it automatically.

Putting it in userspace is making a human do a machines job which isn't
sustainable.

By putting the heuristics in kernelspace Linux will improve performance also
on workloads the human operator didn't think of as the machine will
detect them from
statictical or other behaviour patterns.

Yours,
Linus Walleij

On 6/27/22 02:36, Linus Walleij wrote:
> The right way to solve this is to make the Linux kernel contain the
> necessary heuristics to identify which tasks and thus cores need this
> to improve efficiency and then apply it automatically.

I agree in theory.  But, I also want a pony in theory.

Any suggestions for how to do this in the real world?

Otherwise, I'm inclined to say that this series incrementally makes
things better in the real world by at least moving folks away from wrmsr(1).

On Tue, Jun 28, 2022 at 5:47 PM Dave Hansen <dave.hansen@intel.com> wrote:

> On 6/27/22 02:36, Linus Walleij wrote:

> > The right way to solve this is to make the Linux kernel contain the
> > necessary heuristics to identify which tasks and thus cores need this
> > to improve efficiency and then apply it automatically.
>
> I agree in theory.  But, I also want a pony in theory.
>
> Any suggestions for how to do this in the real world?

Well if the knobs are exposed to userspace, how do people using
these knobs know when to turn them? A profiler? perf? All that
data is available to the kernel too.

The memory access pattern statistics from mm/damon
was what I suggested as a starting point.

We have pretty elaborate heuristics in the kernel to identify the
behaviour of processes, one example is the BFQ block scheduler
which determines I/O priority weights of processed based on
how interactive they are.

If we can determine things like that I am pretty sure we can determine how
computing intense a task is for example, by using memory access
statistics and scheduler information: if the process is constantly
READY to run over a few context switches and PC also stays in a
certain rage of memory like two adjacent pages then it is probably
running a hard kernel, if that is what we need to know here. It
doesn't seem too far-fetched?

We have the performance counters as well. That should be possible to
utilize to get even more precise heuristics? Maybe that is what userspace
is using to determine this already.

I'm not saying there has to be a simple solution, but maybe there
is something like a really complicated solution? We have academic
researchers that like to look at things like this.

> Otherwise, I'm inclined to say that this series incrementally makes
> things better in the real world by at least moving folks away from wrmsr(1).

I don't know if yet another ABI that needs to be maintained helps
the situation much, it's just a contract that we will have to
maintain for no gain. However if userspace is messing with that
register behind our back and we know better, we
can just overwrite it with the policy we determine is better in the
kernel.

Yours,
Linus Walleij

On 6/28/22 13:20, Linus Walleij wrote:
> On Tue, Jun 28, 2022 at 5:47 PM Dave Hansen <dave.hansen@intel.com> wrote:
>> On 6/27/22 02:36, Linus Walleij wrote:
>>> The right way to solve this is to make the Linux kernel contain the
>>> necessary heuristics to identify which tasks and thus cores need this
>>> to improve efficiency and then apply it automatically.
>>
>> I agree in theory.  But, I also want a pony in theory.
>>
>> Any suggestions for how to do this in the real world?
> 
> Well if the knobs are exposed to userspace, how do people using
> these knobs know when to turn them? A profiler? perf? All that
> data is available to the kernel too.

They run their fortran app.  Change the MSRs.  Run it again.  See if it
simulated the nuclear weapon blast any faster or slower.  Rinse.  Repeat.

One thing that is missing from the changelog and cover letter here: On
x86, there's a 'wrmsr(1)' tool.  That took pokes at Model Specific
Registers (MSRs) via the /dev/cpu/X/msr interface.  That interface is a
very, very thinly-veiled wrapper around the WRMSR (WRite MSR) instruction.

In other words, on x86, our current interface allows userspace programs
to arbitrarily poke at our most sensitive hardware configuration
registers.  One of the most common reasons users have reported doing
this (we have pr_warn()ings about it) is controlling the prefetch hardware.

This interface would take a good chunk of the x86 wrmsr(1) audience and
convert them over to a less dangerous interface.  That's a win on x86.
We don't even *remotely* have line-of-sight for a generic solution for
the kernel to figure out a single "best" value for these registers.

Hi Dave,

> They run their fortran app.  Change the MSRs.  Run it again.  See if it
> simulated the nuclear weapon blast any faster or slower.  Rinse.  Repeat.
> 
> One thing that is missing from the changelog and cover letter here: On x86,
> there's a 'wrmsr(1)' tool.  That took pokes at Model Specific Registers (MSRs)
> via the /dev/cpu/X/msr interface.  That interface is a very, very thinly-veiled
> wrapper around the WRMSR (WRite MSR) instruction.
> 
> In other words, on x86, our current interface allows userspace programs to
> arbitrarily poke at our most sensitive hardware configuration registers.  One of
> the most common reasons users have reported doing this (we have
> pr_warn()ings about it) is controlling the prefetch hardware.
> 
> This interface would take a good chunk of the x86 wrmsr(1) audience and
> convert them over to a less dangerous interface.  That's a win on x86.
> We don't even *remotely* have line-of-sight for a generic solution for the kernel
> to figure out a single "best" value for these registers.

Thank you for mentioning wrmsr tool.

This is one of the reason why I want to add the sysfs interface. I
will add the description that this interface can be used instead of
wrmsr tool (or MSR driver) for hardware prefetch control usage to the
cover letter.

I read below that we should not accesse any MSR directly from
userspace without restriction.
https://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git/about/

On Tue, Jun 28, 2022 at 11:02 PM Dave Hansen <dave.hansen@intel.com> wrote:
> On 6/28/22 13:20, Linus Walleij wrote:
> >
> > Well if the knobs are exposed to userspace, how do people using
> > these knobs know when to turn them? A profiler? perf? All that
> > data is available to the kernel too.
>
> They run their fortran app.  Change the MSRs.  Run it again.  See if it
> simulated the nuclear weapon blast any faster or slower.  Rinse.  Repeat.

That sounds like a schoolbook definition of the trial-and-error method.
https://en.wikipedia.org/wiki/Trial_and_error

That's fair. But these people really need a better hammer.

> This interface would take a good chunk of the x86 wrmsr(1) audience and
> convert them over to a less dangerous interface.  That's a win on x86.
> We don't even *remotely* have line-of-sight for a generic solution for
> the kernel to figure out a single "best" value for these registers.

Maybe less dangerous for them, but maybe more dangerous for the kernel
community who signs up to maintain the behaviour of that interface
perpetually.

Yours,
Linus Walleij

Hi Linus,

> The right way to solve this is to make the Linux kernel contain the necessary
> heuristics to identify which tasks and thus cores need this to improve efficiency
> and then apply it automatically.
> 
> Putting it in userspace is making a human do a machines job which isn't
> sustainable.
> 
> By putting the heuristics in kernelspace Linux will improve performance also on
> workloads the human operator didn't think of as the machine will detect them from
> statictical or other behaviour patterns.

In order to put the heuristics into kernelspace Linux, I think it
necessary to consider the following two points.

1) Which cores are tied with the process?
This is different from the core on which the process can run. It
probably need to combine some CPU resource limit to avoid affecting
non-target processes.

2) How to derive the value to set in the register?
It is necessary to verify whether an appropriate set value can be
derived using statistical information, etc. In addition, to prevent
the cost of automatic derivation from exceeding the value that would
be improved by it.

I don't have a prospect for resolving these issues yet. I will
continue these considerations.

>> The right way to solve this is to make the Linux kernel contain the necessary
>> heuristics to identify which tasks and thus cores need this to improve efficiency
>> and then apply it automatically.
>>
>> Putting it in userspace is making a human do a machines job which isn't
>> sustainable.
>>
>> By putting the heuristics in kernelspace Linux will improve performance also on
>> workloads the human operator didn't think of as the machine will detect them from
>> statictical or other behaviour patterns.
>
>In order to put the heuristics into kernelspace Linux, I think it
>necessary to consider the following two points.
>
>1) Which cores are tied with the process?
>This is different from the core on which the process can run. It
>probably need to combine some CPU resource limit to avoid affecting
>non-target processes.
>
>2) How to derive the value to set in the register?
>It is necessary to verify whether an appropriate set value can be
>derived using statistical information, etc. In addition, to prevent
>the cost of automatic derivation from exceeding the value that would
>be improved by it.
>
>I don't have a prospect for resolving these issues yet. I will
>continue these considerations.

Another approach would be to make the set of prefetch settings
a task attribute. Then set them in the context switch code when
the process is about to run on a CPU.

But that assumes you can cheaply change the attributes. If doing
so requires multiple MSR writes (on x86) it might be a non-starter.

-Tony

Hi Tony,

Thanks for the comment.

> Another approach would be to make the set of prefetch settings a task attribute.
> Then set them in the context switch code when the process is about to run on
> a CPU.
> 
> But that assumes you can cheaply change the attributes. If doing so requires
> multiple MSR writes (on x86) it might be a non-starter.

On the x86 and A64FX, each parameter for controlling hardware prefetch
is contained in one register. The current specification makes each
parameter a separate attribute, so we need to write as many times as
there are parameters to change. However it is possible to change the
attribute with one MSR write per core by changing multiple parameters
before the context switch.