[Bug dynamic-link/29953] New: Invalid x86_non_temporal_threshold without cache info

[Bug dynamic-link/29953] New: Invalid x86_non_temporal_threshold without cache info

[engelke at in dot tum.de]

https://sourceware.org/bugzilla/show_bug.cgi?id=29953

            Bug ID: 29953
           Summary: Invalid x86_non_temporal_threshold without cache info
           Product: glibc
           Version: 2.36
            Status: UNCONFIRMED
          Severity: minor
          Priority: P2
         Component: dynamic-link
          Assignee: unassigned at sourceware dot org
          Reporter: engelke at in dot tum.de
                CC: hjl.tools at gmail dot com
  Target Milestone: ---

When cpuid reports no information on a shared cache (e.g., in an emulator), the
x86_non_temporal_threshold will be set to zero, causing memcpy/memset to behave
wrong for mid-sized operations. sysdeps/x86/dl-cacheinfo.h indicates that the
minimum value must be 0x4040, but this is not enforced for the default value
(see also commit b446822b6ae).

In the emulator, `/lib64/ld-linux-x86-64.so.2 --list-tunables` reports:

    glibc.cpu.x86_non_temporal_threshold: 0x0 (min: 0x0, max:
0xffffffffffffffff)

Setting the tunable explicitly (env
GLIBC_TUNABLES=glibc.cpu.x86_non_temporal_threshold=0x4040) avoids the problem.

As a side note, ld.so won't load any DSO on x86-64 ("CPU ISA level is lower
than required") when the cpuid vendor not listed in sysdeps/x86/cpu-features.c,
as the feature bits in cpuid leaf 1 are not decoded in get_common_indices
(family is NULL).

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug dynamic-link/29953] Invalid x86_non_temporal_threshold without cache info

[goldstein.w.n at gmail dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=29953

Noah Goldstein <goldstein.w.n at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |goldstein.w.n at gmail dot com

--- Comment #1 from Noah Goldstein <goldstein.w.n at gmail dot com> ---
Created attachment 14545
  --> https://sourceware.org/bugzilla/attachment.cgi?id=14545&action=edit
Add minval

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug dynamic-link/29953] Invalid x86_non_temporal_threshold without cache info

[goldstein.w.n at gmail dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=29953

--- Comment #2 from Noah Goldstein <goldstein.w.n at gmail dot com> ---
(In reply to Noah Goldstein from comment #1)
> Created attachment 14545 [details]
> Add minval

Sorry wasn't clear, does that fix the issue?

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug dynamic-link/29953] Invalid x86_non_temporal_threshold without cache info

[engelke at in dot tum.de]

https://sourceware.org/bugzilla/show_bug.cgi?id=29953

--- Comment #3 from Alexis Engelke <engelke at in dot tum.de> ---
Created attachment 14547
  --> https://sourceware.org/bugzilla/attachment.cgi?id=14547&action=edit
Ensure minimum value

(In reply to Noah Goldstein from comment #2)
> Sorry wasn't clear, does that fix the issue?

No. Now --list-tunables reports:

    glibc.cpu.x86_non_temporal_threshold: 0x0 (min: 0x4040, max:
0xffffffffffffffff)

The attached patch fixes the problem that the default value is never checked
against the limits.

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug dynamic-link/29953] Invalid x86_non_temporal_threshold without cache info

[hjl.tools at gmail dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=29953

H.J. Lu <hjl.tools at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
  Attachment #14545|0                           |1
        is obsolete|                            |
  Attachment #14547|0                           |1
        is obsolete|                            |

--- Comment #4 from H.J. Lu <hjl.tools at gmail dot com> ---
Created attachment 14551
  --> https://sourceware.org/bugzilla/attachment.cgi?id=14551&action=edit
A patch

Try this.

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug dynamic-link/29953] Invalid x86_non_temporal_threshold without cache info

[goldstein.w.n at gmail dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=29953

--- Comment #5 from Noah Goldstein <goldstein.w.n at gmail dot com> ---
(In reply to H.J. Lu from comment #4)
> Created attachment 14551 [details]
> A patch
> 
> Try this.

Oh wow, I had always assumed `TUNABLE_SET_WITH_BOUNDS` would just clamp the
value.

We also have a bug with `x86_rep_movsb_threshold` which requires at least 128
(2x AVX512 aligning vectors).

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug dynamic-link/29953] Invalid x86_non_temporal_threshold without cache info

[hjl.tools at gmail dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=29953

--- Comment #6 from H.J. Lu <hjl.tools at gmail dot com> ---
(In reply to Noah Goldstein from comment #5)
> We also have a bug with `x86_rep_movsb_threshold` which requires at least
> 128 (2x AVX512 aligning vectors).

x86_rep_movsb_threshold should be fine.  minimum_rep_movsb_threshold should
be set and checked properly.

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug dynamic-link/29953] Invalid x86_non_temporal_threshold without cache info

[goldstein.w.n at gmail dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=29953

--- Comment #7 from Noah Goldstein <goldstein.w.n at gmail dot com> ---
(In reply to H.J. Lu from comment #6)
> (In reply to Noah Goldstein from comment #5)
> > We also have a bug with `x86_rep_movsb_threshold` which requires at least
> > 128 (2x AVX512 aligning vectors).
> 
> x86_rep_movsb_threshold should be fine.  minimum_rep_movsb_threshold should
> be set and checked properly.

Seems like the exact same issue no?

```
  tunable_size = TUNABLE_GET (x86_rep_movsb_threshold, long int, NULL);
  if (tunable_size > minimum_rep_movsb_threshold)
    rep_movsb_threshold = tunable_size;

  // if tunable_size was zero, rep_movsb_threshold == 0

  ...

  // Won't clamp, will leave x86_rep_movsb_threshold at zero.
  TUNABLE_SET_WITH_BOUNDS (x86_rep_movsb_threshold, rep_movsb_threshold,
                           minimum_rep_movsb_threshold, SIZE_MAX);




```

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug dynamic-link/29953] Invalid x86_non_temporal_threshold without cache info

[hjl.tools at gmail dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=29953

--- Comment #8 from H.J. Lu <hjl.tools at gmail dot com> ---
(In reply to Noah Goldstein from comment #7)
> (In reply to H.J. Lu from comment #6)
> > (In reply to Noah Goldstein from comment #5)
> > > We also have a bug with `x86_rep_movsb_threshold` which requires at least
> > > 128 (2x AVX512 aligning vectors).
> > 
> > x86_rep_movsb_threshold should be fine.  minimum_rep_movsb_threshold should
> > be set and checked properly.
> 
> Seems like the exact same issue no?
> 
> ```
>   tunable_size = TUNABLE_GET (x86_rep_movsb_threshold, long int, NULL);
>   if (tunable_size > minimum_rep_movsb_threshold)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
>     rep_movsb_threshold = tunable_size;
> 
>   // if tunable_size was zero, rep_movsb_threshold == 0

Since minimum_rep_movsb_threshold is > 0, rep_movsb_threshold won't be zero.

>  
>   ...
> 
>   // Won't clamp, will leave x86_rep_movsb_threshold at zero.
>   TUNABLE_SET_WITH_BOUNDS (x86_rep_movsb_threshold, rep_movsb_threshold,
> 			   minimum_rep_movsb_threshold, SIZE_MAX);
> 
> 
> 
> 
> ```

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug dynamic-link/29953] Invalid x86_non_temporal_threshold without cache info

[goldstein.w.n at gmail dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=29953

--- Comment #9 from Noah Goldstein <goldstein.w.n at gmail dot com> ---
Created attachment 14552
  --> https://sourceware.org/bugzilla/attachment.cgi?id=14552&action=edit
Fix with new API for clamping tunable value

Personally more infavor of an approach like this.

Seems a bit insane that when we have a minval > 0, we default to potentially
leaving the tunable at zero.

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug dynamic-link/29953] Invalid x86_non_temporal_threshold without cache info

[hjl.tools at gmail dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=29953

--- Comment #10 from H.J. Lu <hjl.tools at gmail dot com> ---
(In reply to Noah Goldstein from comment #9)
> Created attachment 14552 [details]
> Fix with new API for clamping tunable value
> 
> Personally more infavor of an approach like this.
> 
> Seems a bit insane that when we have a minval > 0, we default to potentially
> leaving the tunable at zero.

TUNABLE_SET_WITH_BOUNDS is used to update dynamic bounds for display purpose.
The tunable value may be retrieved before TUNABLE_SET_WITH_BOUNDS.

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug dynamic-link/29953] Invalid x86_non_temporal_threshold without cache info

[engelke at in dot tum.de]

https://sourceware.org/bugzilla/show_bug.cgi?id=29953

--- Comment #11 from Alexis Engelke <engelke at in dot tum.de> ---
(In reply to H.J. Lu from comment #4)
> Try this.
This fixes the problem, thanks.

(In reply to Noah Goldstein from comment #9)
> Created attachment 14552 [details]
> Fix with new API for clamping tunable value
This also fixes the bug when tunables are enabled, but has the problem that no
clamping will happen when tunables are disabled.

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug dynamic-link/29953] Invalid x86_non_temporal_threshold without cache info

[goldstein.w.n at gmail dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=29953

--- Comment #12 from Noah Goldstein <goldstein.w.n at gmail dot com> ---
(In reply to Alexis Engelke from comment #11)
> (In reply to H.J. Lu from comment #4)
> > Try this.
> This fixes the problem, thanks.
> 
> (In reply to Noah Goldstein from comment #9)
> > Created attachment 14552 [details]
> > Fix with new API for clamping tunable value
> This also fixes the bug when tunables are enabled, but has the problem that
> no clamping will happen when tunables are disabled.

We can go with H.J's fix, making the tunable setter clamping can be an
independent QOL change later on.

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug dynamic-link/29953] Invalid x86_non_temporal_threshold without cache info

[cvs-commit at gcc dot gnu.org]

https://sourceware.org/bugzilla/show_bug.cgi?id=29953

--- Comment #13 from cvs-commit at gcc dot gnu.org <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by H.J. Lu <hjl@sourceware.org>:

https://sourceware.org/git/gitweb.cgi?p=glibc.git;h=48b74865c63840b288bd85b4d8743533b73b339b

commit 48b74865c63840b288bd85b4d8743533b73b339b
Author: H.J. Lu <hjl.tools@gmail.com>
Date:   Tue Jan 3 13:06:48 2023 -0800

    x86: Check minimum/maximum of non_temporal_threshold [BZ #29953]

    The minimum non_temporal_threshold is 0x4040.  non_temporal_threshold may
    be set to less than the minimum value when the shared cache size isn't
    available (e.g., in an emulator) or by the tunable.  Add checks for
    minimum and maximum of non_temporal_threshold.

    This fixes BZ #29953.

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug dynamic-link/29953] Invalid x86_non_temporal_threshold without cache info

[decui at microsoft dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=29953

Dexuan Cui <decui at microsoft dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |decui at microsoft dot com

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug dynamic-link/29953] Invalid x86_non_temporal_threshold without cache info

[fweimer at redhat dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=29953

--- Comment #14 from Florian Weimer <fweimer at redhat dot com> ---
*** Bug 30037 has been marked as a duplicate of this bug. ***

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug dynamic-link/29953] Invalid x86_non_temporal_threshold without cache info

[fweimer at redhat dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=29953

Florian Weimer <fweimer at redhat dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |fweimer at redhat dot com

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug dynamic-link/29953] Invalid x86_non_temporal_threshold without cache info

[fweimer at redhat dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=29953

--- Comment #15 from Florian Weimer <fweimer at redhat dot com> ---
(In reply to cvs-commit@gcc.gnu.org from comment #13)
> The master branch has been updated by H.J. Lu <hjl@sourceware.org>:
> 
> https://sourceware.org/git/gitweb.cgi?p=glibc.git;
> h=48b74865c63840b288bd85b4d8743533b73b339b
> 
> commit 48b74865c63840b288bd85b4d8743533b73b339b
> Author: H.J. Lu <hjl.tools@gmail.com>
> Date:   Tue Jan 3 13:06:48 2023 -0800
> 
>     x86: Check minimum/maximum of non_temporal_threshold [BZ #29953]
>     
>     The minimum non_temporal_threshold is 0x4040.  non_temporal_threshold may
>     be set to less than the minimum value when the shared cache size isn't
>     available (e.g., in an emulator) or by the tunable.  Add checks for
>     minimum and maximum of non_temporal_threshold.
>     
>     This fixes BZ #29953.

H.J., do you intend to backport this? Can we close this bug?

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug dynamic-link/29953] Invalid x86_non_temporal_threshold without cache info

[hjl.tools at gmail dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=29953

--- Comment #16 from H.J. Lu <hjl.tools at gmail dot com> ---
(In reply to Florian Weimer from comment #15)
> (In reply to cvs-commit@gcc.gnu.org from comment #13)
> > The master branch has been updated by H.J. Lu <hjl@sourceware.org>:
> > 
> > https://sourceware.org/git/gitweb.cgi?p=glibc.git;
> > h=48b74865c63840b288bd85b4d8743533b73b339b
> > 
> > commit 48b74865c63840b288bd85b4d8743533b73b339b
> > Author: H.J. Lu <hjl.tools@gmail.com>
> > Date:   Tue Jan 3 13:06:48 2023 -0800
> > 
> >     x86: Check minimum/maximum of non_temporal_threshold [BZ #29953]
> >     
> >     The minimum non_temporal_threshold is 0x4040.  non_temporal_threshold may
> >     be set to less than the minimum value when the shared cache size isn't
> >     available (e.g., in an emulator) or by the tunable.  Add checks for
> >     minimum and maximum of non_temporal_threshold.
> >     
> >     This fixes BZ #29953.
> 
> H.J., do you intend to backport this? Can we close this bug?

I will backport it.

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug dynamic-link/29953] Invalid x86_non_temporal_threshold without cache info

[cvs-commit at gcc dot gnu.org]

https://sourceware.org/bugzilla/show_bug.cgi?id=29953

--- Comment #17 from cvs-commit at gcc dot gnu.org <cvs-commit at gcc dot gnu.org> ---
The release/2.36/master branch has been updated by H.J. Lu
<hjl@sourceware.org>:

https://sourceware.org/git/gitweb.cgi?p=glibc.git;h=f3991fec8071dbcf3ec9f13a91c738b66fcd4159

commit f3991fec8071dbcf3ec9f13a91c738b66fcd4159
Author: H.J. Lu <hjl.tools@gmail.com>
Date:   Tue Jan 3 13:06:48 2023 -0800

    x86: Check minimum/maximum of non_temporal_threshold [BZ #29953]

    The minimum non_temporal_threshold is 0x4040.  non_temporal_threshold may
    be set to less than the minimum value when the shared cache size isn't
    available (e.g., in an emulator) or by the tunable.  Add checks for
    minimum and maximum of non_temporal_threshold.

    This fixes BZ #29953.

    (cherry picked from commit 48b74865c63840b288bd85b4d8743533b73b339b)

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug dynamic-link/29953] Invalid x86_non_temporal_threshold without cache info

[cvs-commit at gcc dot gnu.org]

https://sourceware.org/bugzilla/show_bug.cgi?id=29953

--- Comment #18 from cvs-commit at gcc dot gnu.org <cvs-commit at gcc dot gnu.org> ---
The release/2.34/master branch has been updated by Florian Weimer
<fw@sourceware.org>:

https://sourceware.org/git/gitweb.cgi?p=glibc.git;h=89c017de2f52d17862bda9a6f8382e913457bfbe

commit 89c017de2f52d17862bda9a6f8382e913457bfbe
Author: H.J. Lu <hjl.tools@gmail.com>
Date:   Tue Jan 3 13:06:48 2023 -0800

    x86: Check minimum/maximum of non_temporal_threshold [BZ #29953]

    The minimum non_temporal_threshold is 0x4040.  non_temporal_threshold may
    be set to less than the minimum value when the shared cache size isn't
    available (e.g., in an emulator) or by the tunable.  Add checks for
    minimum and maximum of non_temporal_threshold.

    This fixes BZ #29953.

    (cherry picked from commit 48b74865c63840b288bd85b4d8743533b73b339b)

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug dynamic-link/29953] Invalid x86_non_temporal_threshold without cache info

[cvs-commit at gcc dot gnu.org]

https://sourceware.org/bugzilla/show_bug.cgi?id=29953

--- Comment #19 from cvs-commit at gcc dot gnu.org <cvs-commit at gcc dot gnu.org> ---
The release/2.35/master branch has been updated by Florian Weimer
<fw@sourceware.org>:

https://sourceware.org/git/gitweb.cgi?p=glibc.git;h=b7cc55a24ea303875d95d0db3f73f9239bba2583

commit b7cc55a24ea303875d95d0db3f73f9239bba2583
Author: H.J. Lu <hjl.tools@gmail.com>
Date:   Tue Jan 3 13:06:48 2023 -0800

    x86: Check minimum/maximum of non_temporal_threshold [BZ #29953]

    The minimum non_temporal_threshold is 0x4040.  non_temporal_threshold may
    be set to less than the minimum value when the shared cache size isn't
    available (e.g., in an emulator) or by the tunable.  Add checks for
    minimum and maximum of non_temporal_threshold.

    This fixes BZ #29953.

    (cherry picked from commit 48b74865c63840b288bd85b4d8743533b73b339b)

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug dynamic-link/29953] Invalid x86_non_temporal_threshold without cache info

[fw at deneb dot enyo.de]

https://sourceware.org/bugzilla/show_bug.cgi?id=29953

Florian Weimer <fw at deneb dot enyo.de> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
                 CC|                            |fw at deneb dot enyo.de
   Target Milestone|---                         |2.37
         Resolution|---                         |FIXED

--- Comment #20 from Florian Weimer <fw at deneb dot enyo.de> ---
Backports completed.

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug dynamic-link/29953] Invalid x86_non_temporal_threshold without cache info

[fweimer at redhat dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=29953

Florian Weimer <fweimer at redhat dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           See Also|                            |https://sourceware.org/bugz
                   |                            |illa/show_bug.cgi?id=30429

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug dynamic-link/29953] Invalid x86_non_temporal_threshold without cache info

[fweimer at redhat dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=29953

Florian Weimer <fweimer at redhat dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           See Also|                            |https://sourceware.org/bugz
                   |                            |illa/show_bug.cgi?id=30428

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug dynamic-link/29953] Invalid x86_non_temporal_threshold without cache info

[fweimer at redhat dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=29953

Florian Weimer <fweimer at redhat dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           See Also|                            |https://sourceware.org/bugz
                   |                            |illa/show_bug.cgi?id=30643

-- 
You are receiving this mail because:
You are on the CC list for the bug.