V2 [PATCH] x86: Adjust tst-cpu-features-supports.c for GCC 11
Commit Message
On Fri, Dec 4, 2020 at 12:12 AM Florian Weimer <fweimer@redhat.com> wrote:
>
> * H. J. Lu via Libc-alpha:
>
> > Set RDRAND usable if CPU supports RDRAND. Also check HAS_CPU_FEATURE
> > instead of CPU_FEATURE_USABLE for FSGSBASE, LM and XSAVES since
> > FSGSBASE/LM require kernel support and XSAVES is supervisor-mode only.
>
> I think the commit subject is misleading because the RDRAND part is a
> glibc bug fix. Rest looks okay to me.
>
Here is the V2 patch. It changed IBT and SHSTK to check HAS_CPU_FEATURE
since they need OS support.
Comments
On Fri, Dec 4, 2020 at 5:07 AM H.J. Lu <hjl.tools@gmail.com> wrote:
>
> On Fri, Dec 4, 2020 at 12:12 AM Florian Weimer <fweimer@redhat.com> wrote:
> >
> > * H. J. Lu via Libc-alpha:
> >
> > > Set RDRAND usable if CPU supports RDRAND. Also check HAS_CPU_FEATURE
> > > instead of CPU_FEATURE_USABLE for FSGSBASE, LM and XSAVES since
> > > FSGSBASE/LM require kernel support and XSAVES is supervisor-mode only.
> >
> > I think the commit subject is misleading because the RDRAND part is a
> > glibc bug fix. Rest looks okay to me.
> >
>
> Here is the V2 patch. It changed IBT and SHSTK to check HAS_CPU_FEATURE
> since they need OS support.
>
I am checking in this.
* H. J. Lu:
> On Fri, Dec 4, 2020 at 12:12 AM Florian Weimer <fweimer@redhat.com> wrote:
>>
>> * H. J. Lu via Libc-alpha:
>>
>> > Set RDRAND usable if CPU supports RDRAND. Also check HAS_CPU_FEATURE
>> > instead of CPU_FEATURE_USABLE for FSGSBASE, LM and XSAVES since
>> > FSGSBASE/LM require kernel support and XSAVES is supervisor-mode only.
>>
>> I think the commit subject is misleading because the RDRAND part is a
>> glibc bug fix. Rest looks okay to me.
>>
>
> Here is the V2 patch. It changed IBT and SHSTK to check HAS_CPU_FEATURE
> since they need OS support.
Patch is okay. But I do wonder why __builtin_cpu_supports doesn't
handle this. Based on the GCC documention, I would say it should
perform what you call a “usable” check.
Thanks,
Florian
On Fri, Dec 4, 2020 at 5:14 AM Florian Weimer <fweimer@redhat.com> wrote:
>
> * H. J. Lu:
>
> > On Fri, Dec 4, 2020 at 12:12 AM Florian Weimer <fweimer@redhat.com> wrote:
> >>
> >> * H. J. Lu via Libc-alpha:
> >>
> >> > Set RDRAND usable if CPU supports RDRAND. Also check HAS_CPU_FEATURE
> >> > instead of CPU_FEATURE_USABLE for FSGSBASE, LM and XSAVES since
> >> > FSGSBASE/LM require kernel support and XSAVES is supervisor-mode only.
> >>
> >> I think the commit subject is misleading because the RDRAND part is a
> >> glibc bug fix. Rest looks okay to me.
> >>
> >
> > Here is the V2 patch. It changed IBT and SHSTK to check HAS_CPU_FEATURE
> > since they need OS support.
>
> Patch is okay. But I do wonder why __builtin_cpu_supports doesn't
> handle this. Based on the GCC documention, I would say it should
> perform what you call a “usable” check.
>
There are no easy and OS independent ways to check if they are usable. Glibc
doesn't set the usable bit for them.
From e7f391efefb919a9c5ad856b504874150b82fc01 Mon Sep 17 00:00:00 2001
From: "H.J. Lu" <hjl.tools@gmail.com>
Date: Thu, 3 Dec 2020 15:02:44 -0800
Subject: [PATCH] x86: Adjust tst-cpu-features-supports.c for GCC 11
Check HAS_CPU_FEATURE instead of CPU_FEATURE_USABLE for FSGSBASE, IBT,
LM, SHSTK and XSAVES since FSGSBASE/IBT/SHSTK/LM require kernel support
and XSAVES is supervisor-mode only.
---
sysdeps/x86/tst-cpu-features-supports.c | 15 ++++++++++-----
1 file changed, 10 insertions(+), 5 deletions(-)
@@ -40,6 +40,11 @@ check_supports (int supports, int usable, const char *supports_name,
#define CHECK_SUPPORTS(str, name) \
check_supports (__builtin_cpu_supports (#str), \
CPU_FEATURE_USABLE (name), \
+ #str, "CPU_FEATURE_USABLE (" #name ")");
+
+#define CHECK_CPU_SUPPORTS(str, name) \
+ check_supports (__builtin_cpu_supports (#str), \
+ HAS_CPU_FEATURE (name), \
#str, "HAS_CPU_FEATURE (" #name ")");
static int
@@ -118,7 +123,7 @@ do_test (int argc, char **argv)
fails += CHECK_SUPPORTS (fma4, FMA4);
#endif
#if __GNUC_PREREQ (11, 0)
- fails += CHECK_SUPPORTS (fsgsbase, FSGSBASE);
+ fails += CHECK_CPU_SUPPORTS (fsgsbase, FSGSBASE);
fails += CHECK_SUPPORTS (fxsave, FXSR);
#endif
#if __GNUC_PREREQ (8, 0)
@@ -126,9 +131,9 @@ do_test (int argc, char **argv)
#endif
#if __GNUC_PREREQ (11, 0)
fails += CHECK_SUPPORTS (hle, HLE);
- fails += CHECK_SUPPORTS (ibt, IBT);
+ fails += CHECK_CPU_SUPPORTS (ibt, IBT);
fails += CHECK_SUPPORTS (lahf_lm, LAHF64_SAHF64);
- fails += CHECK_SUPPORTS (lm, LM);
+ fails += CHECK_CPU_SUPPORTS (lm, LM);
fails += CHECK_SUPPORTS (lwp, LWP);
fails += CHECK_SUPPORTS (lzcnt, LZCNT);
#endif
@@ -150,7 +155,7 @@ do_test (int argc, char **argv)
fails += CHECK_SUPPORTS (rtm, RTM);
fails += CHECK_SUPPORTS (serialize, SERIALIZE);
fails += CHECK_SUPPORTS (sha, SHA);
- fails += CHECK_SUPPORTS (shstk, SHSTK);
+ fails += CHECK_CPU_SUPPORTS (shstk, SHSTK);
#endif
fails += CHECK_SUPPORTS (sse, SSE);
fails += CHECK_SUPPORTS (sse2, SSE2);
@@ -180,7 +185,7 @@ do_test (int argc, char **argv)
fails += CHECK_SUPPORTS (xsave, XSAVE);
fails += CHECK_SUPPORTS (xsavec, XSAVEC);
fails += CHECK_SUPPORTS (xsaveopt, XSAVEOPT);
- fails += CHECK_SUPPORTS (xsaves, XSAVES);
+ fails += CHECK_CPU_SUPPORTS (xsaves, XSAVES);
#endif
printf ("%d differences between __builtin_cpu_supports and glibc code.\n",
--
2.28.0