From: "Metzger, Markus T" <markus.t.metzger@intel.com>
To: Jan Beulich <JBeulich@suse.com>, GDB <gdb-patches@sourceware.org>
Cc: "Wiederhake, Tim" <tim.wiederhake@intel.com>,
Simon Marchi <simon.marchi@ericsson.com>
Subject: RE: [PATCH] x86-64: fix ZMM register state tracking
Date: Mon, 24 Sep 2018 17:19:00 -0000 [thread overview]
Message-ID: <A78C989F6D9628469189715575E55B236B35D702@IRSMSX104.ger.corp.intel.com> (raw)
In-Reply-To: <5B8FD8B302000078001E5940@prv1-mh.provo.novell.com>
Hello Jan,
> The three AVX512 state components are entirely independent - one being in its "init
> state" has no implication whatsoever on either of the other two. Fully separate
> X86_XSTATE_ZMM_H and X86_XSTATE_ZMM handling, to prevent upper halves of
> the upper 16 ZMM registers to display as if they were zero (when they aren't) after
> e.g. VZEROALL/VZEROUPPER.
>
> gdb/
> 2018-09-05 Jan Beulich <jbeulich@suse.com>
>
> * i387-tdep.c (i387_supply_xsave): Split handling of
> X86_XSTATE_ZMM_H and X86_XSTATE_ZMM.
> (i387_collect_xsave): Likewise.
>
> --- a/gdb/i387-tdep.c
> +++ b/gdb/i387-tdep.c
> @@ -923,7 +923,8 @@ i387_supply_xsave (struct regcache *regc
> enum bfd_endian byte_order = gdbarch_byte_order (gdbarch);
> struct gdbarch_tdep *tdep = gdbarch_tdep (gdbarch);
> const gdb_byte *regs = (const gdb_byte *) xsave;
> - int i;
> + int i, zmm_endlo_regnum = I387_ZMM0H_REGNUM (tdep)
> + + std::min (tdep->num_zmm_regs, 16);
It would be nice to comment on this magic 16 and the min operation.
It's how XSAVE organizes things but it isn't entirely intuitive.
> ULONGEST clear_bv;
> static const gdb_byte zero[I386_MAX_REGISTER_SIZE] = { 0 };
> enum
> @@ -1002,7 +1003,8 @@ i387_supply_xsave (struct regcache *regc
> return;
>
> case avx512_zmm_h:
> - if ((clear_bv & (X86_XSTATE_ZMM_H | X86_XSTATE_ZMM)))
> + if ((clear_bv & (regnum < zmm_endlo_regnum ? X86_XSTATE_ZMM_H
> + : X86_XSTATE_ZMM)))
A comment that XSAVE stores the lower 16 registers in a different place
than the higher 16 registers and also guards them by different XCR0 bits
would be nice.
We hid the different places behind those XSAVE_AVX512_ZMM_H_ADDR
macros but there's nothing similar for the guard bits. Maybe add macros
for the guard check, as well?
> regcache->raw_supply (regnum, zero);
> else
> regcache->raw_supply (regnum,
> @@ -1080,21 +1082,17 @@ i387_supply_xsave (struct regcache *regc
> }
> }
>
> - /* Handle the upper ZMM registers. */
> - if ((tdep->xcr0 & (X86_XSTATE_ZMM_H | X86_XSTATE_ZMM)))
> + /* Handle the upper halves of the low 8/16 ZMM registers. */
> + if ((tdep->xcr0 & X86_XSTATE_ZMM_H))
> {
> - if ((clear_bv & (X86_XSTATE_ZMM_H | X86_XSTATE_ZMM)))
> + if ((clear_bv & X86_XSTATE_ZMM_H))
> {
> - for (i = I387_ZMM0H_REGNUM (tdep);
> - i < I387_ZMMENDH_REGNUM (tdep);
> - i++)
> + for (i = I387_ZMM0H_REGNUM (tdep); i < zmm_endlo_regnum; i++)
> regcache->raw_supply (i, zero);
> }
> else
> {
> - for (i = I387_ZMM0H_REGNUM (tdep);
> - i < I387_ZMMENDH_REGNUM (tdep);
> - i++)
> + for (i = I387_ZMM0H_REGNUM (tdep); i < zmm_endlo_regnum; i++)
> regcache->raw_supply (i,
> XSAVE_AVX512_ZMM_H_ADDR (tdep, regs, i));
> }
> @@ -1119,11 +1117,13 @@ i387_supply_xsave (struct regcache *regc
> }
> }
>
> - /* Handle the YMM_AVX512 registers. */
> + /* Handle the upper 16 ZMM/YMM/XMM registers (if any). */
> if ((tdep->xcr0 & X86_XSTATE_ZMM))
> {
> if ((clear_bv & X86_XSTATE_ZMM))
> {
> + for (i = zmm_endlo_regnum; i < I387_ZMMENDH_REGNUM (tdep); i++)
> + regcache->raw_supply (i, zero);
> for (i = I387_YMM16H_REGNUM (tdep);
> i < I387_YMMH_AVX512_END_REGNUM (tdep);
> i++)
> @@ -1135,6 +1135,9 @@ i387_supply_xsave (struct regcache *regc
> }
> else
> {
> + for (i = zmm_endlo_regnum; i < I387_ZMMENDH_REGNUM (tdep); i++)
> + regcache->raw_supply (i,
> + XSAVE_AVX512_ZMM_H_ADDR (tdep, regs, i));
This covers the upper halves of zmm16 to zmm31. Looking at the function it looks
like the lower halves are covered in separate cases avx512_ymmh_avx512 and
avx512_xmmh_avx512. Maybe reflect this in the comment? It currently suggests
that it handles the entire upper 16 registers.
> for (i = I387_YMM16H_REGNUM (tdep);
> i < I387_YMMH_AVX512_END_REGNUM (tdep);
> i++)
> @@ -1340,7 +1343,8 @@ i387_collect_xsave (const struct regcach
> gdb_byte *p, *regs = (gdb_byte *) xsave;
> gdb_byte raw[I386_MAX_REGISTER_SIZE];
> ULONGEST initial_xstate_bv, clear_bv, xstate_bv = 0;
> - unsigned int i;
> + unsigned int i, zmm_endlo_regnum = I387_ZMM0H_REGNUM (tdep)
> + + std::min (tdep->num_zmm_regs, 16);
> enum
> {
> x87_ctrl_or_mxcsr = 0x1,
> @@ -1441,9 +1445,8 @@ i387_collect_xsave (const struct regcach
> i < I387_MPXEND_REGNUM (tdep); i++)
> memset (XSAVE_MPX_ADDR (tdep, regs, i), 0, 8);
>
> - if ((clear_bv & (X86_XSTATE_ZMM_H | X86_XSTATE_ZMM)))
> - for (i = I387_ZMM0H_REGNUM (tdep);
> - i < I387_ZMMENDH_REGNUM (tdep); i++)
> + if ((clear_bv & X86_XSTATE_ZMM_H))
> + for (i = I387_ZMM0H_REGNUM (tdep); i < zmm_endlo_regnum; i++)
> memset (XSAVE_AVX512_ZMM_H_ADDR (tdep, regs, i), 0, 32);
>
> if ((clear_bv & X86_XSTATE_K))
> @@ -1453,6 +1456,8 @@ i387_collect_xsave (const struct regcach
>
> if ((clear_bv & X86_XSTATE_ZMM))
> {
> + for (i = zmm_endlo_regnum; i < I387_ZMMENDH_REGNUM (tdep); i++)
> + memset (XSAVE_AVX512_ZMM_H_ADDR (tdep, regs, i), 0, 32);
> for (i = I387_YMM16H_REGNUM (tdep);
> i < I387_YMMH_AVX512_END_REGNUM (tdep); i++)
> memset (XSAVE_YMM_AVX512_ADDR (tdep, regs, i), 0, 16);
Looks OK to me.
Regards,
Markus.
Intel Deutschland GmbH
Registered Address: Am Campeon 10-12, 85579 Neubiberg, Germany
Tel: +49 89 99 8853-0, www.intel.de
Managing Directors: Christin Eisenschmid, Christian Lamprechter
Chairperson of the Supervisory Board: Nicole Lau
Registered Office: Munich
Commercial Register: Amtsgericht Muenchen HRB 186928
next prev parent reply other threads:[~2018-09-24 17:19 UTC|newest]
Thread overview: 22+ messages / expand[flat|nested] mbox.gz Atom feed top
2018-09-05 13:23 Jan Beulich
2018-09-07 23:13 ` Simon Marchi
2018-09-10 6:25 ` Jan Beulich
2018-09-10 13:01 ` Simon Marchi
2018-09-18 13:38 ` Jan Beulich
2018-09-25 3:29 ` Simon Marchi
2018-09-25 15:04 ` Jan Beulich
2018-10-02 19:20 ` Simon Marchi
2018-10-03 14:30 ` Jan Beulich
2018-10-03 14:41 ` Simon Marchi
2018-09-11 10:34 ` Pedro Alves
2018-09-11 11:34 ` Jan Beulich
2018-09-24 17:19 ` Metzger, Markus T [this message]
2018-09-25 7:47 ` Jan Beulich
2018-10-10 15:12 ` [PATCH v2] " Jan Beulich
2018-10-29 10:31 ` Ping: " Jan Beulich
[not found] ` <8bafd220c79c3936de8ce3d6421a9c01@polymtl.ca>
2018-11-07 9:07 ` Metzger, Markus T
2018-11-07 9:12 ` Jan Beulich
2018-11-07 13:18 ` Metzger, Markus T
2018-11-07 13:25 ` Jan Beulich
2018-11-07 14:18 ` Simon Marchi
[not found] ` <5BE41AE902000078001F993A@prv1-mh.provo.novell.com>
2018-11-08 16:24 ` Simon Marchi
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=A78C989F6D9628469189715575E55B236B35D702@IRSMSX104.ger.corp.intel.com \
--to=markus.t.metzger@intel.com \
--cc=JBeulich@suse.com \
--cc=gdb-patches@sourceware.org \
--cc=simon.marchi@ericsson.com \
--cc=tim.wiederhake@intel.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox