From: Lancelot SIX <lsix@lancelotsix.com>
To: John Baldwin <jhb@FreeBSD.org>
Cc: gdb-patches@sourceware.org, Willgerodt@sourceware.org,
Felix <felix.willgerodt@intel.com>,
George@sourceware.org, Jini Susan <JiniSusan.George@amd.com>,
Simon Marchi <simon.marchi@polymtl.ca>
Subject: Re: [RFC 02/13] i387-tdep: Add function to read XSAVE layout from NT_X86_CPUID
Date: Mon, 16 Oct 2023 10:17:32 +0100 [thread overview]
Message-ID: <20231016091732.ybyjym67r2l7e3ol@octopus> (raw)
In-Reply-To: <20231009183617.24862-3-jhb@FreeBSD.org>
Hi,
I am not familiar with XSAVE details, but I have pure c++ style comments
below.
On Mon, Oct 09, 2023 at 11:36:04AM -0700, John Baldwin wrote:
> This can be used by x86 arches to determine the XSAVE layout instead
> of guessing based on the XCR0 mask and XSAVE register note size.
> ---
> gdb/i387-tdep.c | 132 ++++++++++++++++++++++++++++++++++++++++++++++++
> gdb/i387-tdep.h | 8 +++
> 2 files changed, 140 insertions(+)
>
> diff --git a/gdb/i387-tdep.c b/gdb/i387-tdep.c
> index 47667da21c7..1eac2b6bd2a 100644
> --- a/gdb/i387-tdep.c
> +++ b/gdb/i387-tdep.c
> @@ -26,6 +26,8 @@
> #include "target-float.h"
> #include "value.h"
>
> +#include <stdexcept>
> +
> #include "i386-tdep.h"
> #include "i387-tdep.h"
> #include "gdbsupport/x86-xstate.h"
> @@ -987,6 +989,136 @@ i387_guess_xsave_layout (uint64_t xcr0, size_t xsave_size,
> return true;
> }
>
> +/* Parse a reg-x86-cpuid pseudo section building a hash table mapping
> + cpuid leaves to their results. */
> +
> +struct cpuid_key
> +{
> + cpuid_key (uint32_t _leaf, uint32_t _subleaf)
> + : leaf(_leaf), subleaf(_subleaf)
> + {}
> +
> + uint32_t leaf;
> + uint32_t subleaf;
> +
> + constexpr bool operator== (const cpuid_key &other) const
> + { return (leaf == other.leaf && subleaf == other.subleaf); }
> +};
> +
> +namespace std
> +{
> +template<>
> +struct hash<cpuid_key>
> +{
> + size_t operator() (const cpuid_key &key) const
> + {
> + return key.leaf ^ (key.subleaf << 1);
> + }
> +};
> +}
I think there was a discussion not long ago regarding opening std, and
it seems that the prefered approach is to use:
template<>
struct std::hash<cpuid_key>
{
...
};
See
https://sourceware.org/pipermail/gdb-patches/2023-September/202336.html
for the discussion.
> +
> +struct cpuid_values
> +{
> + cpuid_values (uint32_t _eax, uint32_t _ebx, uint32_t _ecx, uint32_t _edx)
> + : eax(_eax), ebx(_ebx), ecx(_ecx), edx(_edx)
> + {}
> +
> + uint32_t eax;
> + uint32_t ebx;
> + uint32_t ecx;
> + uint32_t edx;
> +};
> +
> +typedef std::unordered_map<cpuid_key, cpuid_values> cpuid_map;
> +
> +static cpuid_map
> +i387_parse_cpuid_from_core (bfd *bfd)
> +{
> + asection *section = bfd_get_section_by_name (bfd, ".reg-x86-cpuid");
> + if (section == nullptr)
> + return {};
> +
> + size_t size = bfd_section_size (section);
> + if (size == 0 || (size % (6 * 4)) != 0)
> + return {};
> +
> + char contents[size];
If I remember correctly, VLAs are not a C++ feature (but are supported
as a GCC extension
https://gcc.gnu.org/onlinedocs/gcc/Variable-Length.html). I am unsure
if GDB has a policy regarding the use of extensions, so maybe this is
fine. Otherwise, you could use a std::vector instead (it comes with a
dynamic allocation, but I am not too concerned at this is hardly on a
performance critical path)
std::vector<char> contents (size);
> + if (!bfd_get_section_contents (bfd, section, contents, 0, size))
> + {
> + warning (_("Couldn't read `.reg-x86-cpuid' section in core file."));
> + return {};
> + }
> +
> + cpuid_map map;
> + size_t index = 0;
> + while (index < size)
> + {
> + uint32_t leaf = bfd_get_32 (bfd, contents + index);
> + uint32_t count = bfd_get_32 (bfd, contents + index + 4);
> + uint32_t eax = bfd_get_32 (bfd, contents + index + 8);
> + uint32_t ebx = bfd_get_32 (bfd, contents + index + 12);
> + uint32_t ecx = bfd_get_32 (bfd, contents + index + 16);
> + uint32_t edx = bfd_get_32 (bfd, contents + index + 20);
> +
> + if (map.count (cpuid_key (leaf, count)) != 0)
> + {
> + warning (_("Duplicate cpuid leaf %#x,%#x"), leaf, count);
> + return {};
> + }
> + map.emplace (cpuid_key (leaf, count),
> + cpuid_values (eax, ebx, ecx, edx));
As Simon pointed out, there are two lookups here, where you can get away
with just one. However, this is C++17 only which is not [yet] available
in GDB. Instead, you can use the value returned by emplace to know if
an insertation has been done or not:
auto emplace_result = map.emplace (cpuid_key (leaf, count),
cpuid_values (eax, ebx, ecx, edx));
if (!emplace_result.second)
{
warning (_("Duplicate cpuid leaf %#x,%#x"), leaf, count);
return {};
}
> +
> + index += 6 * 4;
> + }
> +
> + return map;
> +}
> +
> +/* Fetch the offset of a specific XSAVE extended region. */
> +
> +static int
I think it is worth returning uint32_t here as int is (in theory) target
dependent.
> +xsave_feature_offset (cpuid_map &map, uint64_t xcr0, int feature)
I think that the MAP parameter could be `const` here.
> +{
> + if ((xcr0 & (1ULL << feature)) == 0)
> + return 0;
> +
> + return map.at (cpuid_key (0xd, feature)).ebx;
> +}
> +
> +/* See i387-tdep.h. */
> +
> +bool
> +i387_read_xsave_layout_from_core (bfd *bfd, uint64_t xcr0, size_t xsave_size,
> + x86_xsave_layout &layout)
> +{
> + cpuid_map map = i387_parse_cpuid_from_core (bfd);
> + if (map.empty ())
> + return false;
> +
> + try
> + {
> + layout.sizeof_xsave = xsave_size;
> + layout.avx_offset = xsave_feature_offset (map, xcr0,
> + X86_XSTATE_AVX_ID);
> + layout.bndregs_offset = xsave_feature_offset (map, xcr0,
> + X86_XSTATE_BNDREGS_ID);
> + layout.bndcfg_offset = xsave_feature_offset (map, xcr0,
> + X86_XSTATE_BNDCFG_ID);
> + layout.k_offset = xsave_feature_offset (map, xcr0,
> + X86_XSTATE_K_ID);
> + layout.zmm_h_offset = xsave_feature_offset (map, xcr0,
> + X86_XSTATE_ZMM_H_ID);
> + layout.zmm_offset = xsave_feature_offset (map, xcr0, X86_XSTATE_ZMM_ID);
> + layout.pkru_offset = xsave_feature_offset (map, xcr0, X86_XSTATE_PKRU_ID);
> + }
> + catch (const std::out_of_range &)
> + {
> + return false;
> + }
> +
> + return true;
> +}
> +
> /* Extract from XSAVE a bitset of the features that are available on the
> target, but which have not yet been enabled. */
>
> diff --git a/gdb/i387-tdep.h b/gdb/i387-tdep.h
> index e149e30e52e..b16b9a60b67 100644
> --- a/gdb/i387-tdep.h
> +++ b/gdb/i387-tdep.h
> @@ -147,6 +147,14 @@ extern void i387_supply_fxsave (struct regcache *regcache, int regnum,
> extern bool i387_guess_xsave_layout (uint64_t xcr0, size_t xsave_size,
> x86_xsave_layout &layout);
>
> +/* Determine the XSAVE layout from the `reg-x86-cpuid` section in a
> + core dump. Returns true on sucess, or false if a layout can not be
s/sucess/success/
> + read. */
> +
> +extern bool i387_read_xsave_layout_from_core (bfd *bfd, uint64_t xcr0,
> + size_t xsave_size,
> + x86_xsave_layout &layout);
> +
> /* Similar to i387_supply_fxsave, but use XSAVE extended state. */
>
> extern void i387_supply_xsave (struct regcache *regcache, int regnum,
> --
> 2.41.0
>
Best,
Lancelot.
next prev parent reply other threads:[~2023-10-16 9:17 UTC|newest]
Thread overview: 37+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-10-09 18:36 [RFC 00/13] Proposal for a new NT_X86_CPUID core dump note John Baldwin
2023-10-09 18:36 ` [RFC 01/13] binutils: Support for the " John Baldwin
2023-10-16 9:23 ` Lancelot SIX
2023-10-16 23:23 ` John Baldwin
2023-10-09 18:36 ` [RFC 02/13] i387-tdep: Add function to read XSAVE layout from NT_X86_CPUID John Baldwin
2023-10-12 4:27 ` Simon Marchi
2023-10-16 23:52 ` John Baldwin
2023-10-16 9:17 ` Lancelot SIX [this message]
2023-10-17 0:04 ` John Baldwin
2023-10-09 18:36 ` [RFC 03/13] gdb: Use NT_X86_CPUID in x86 FreeBSD architectures to read XSAVE layouts John Baldwin
2023-10-09 18:36 ` [RFC 04/13] " John Baldwin
2023-10-12 4:28 ` Simon Marchi
2023-10-17 0:07 ` John Baldwin
2023-10-09 18:36 ` [RFC 05/13] nat/x86-cpuid.h: Remove non-x86 fallbacks John Baldwin
2023-10-12 4:29 ` Simon Marchi
2023-10-09 18:36 ` [RFC 06/13] nat/x86-cpuid: Add a function to build the contents of a NT_X86_CPUID note John Baldwin
2023-10-12 4:41 ` Simon Marchi
2023-10-17 0:22 ` John Baldwin
2023-10-09 18:36 ` [RFC 07/13] x86_elf_make_cpuid_note: Helper routine to build NT_X86_CPUID ELF note John Baldwin
2023-10-09 18:36 ` [RFC 08/13] x86-fbsd-nat: Support fetching TARGET_OBJECT_X86_CPUID objects John Baldwin
2023-10-09 18:36 ` [RFC 09/13] fbsd-tdep: Export fbsd_make_corefile_notes John Baldwin
2023-10-09 18:36 ` [RFC 10/13] {amd64,i386}-fbsd-tdep: Include NT_X86_CPUID notes in core dumps from gcore John Baldwin
2023-10-16 9:31 ` [RFC 10/13] {amd64, i386}-fbsd-tdep: " Lancelot SIX
2023-10-17 0:26 ` John Baldwin
2023-10-09 18:36 ` [RFC 11/13] x86-linux-nat: Support fetching TARGET_OBJECT_X86_CPUID objects John Baldwin
2023-10-09 18:36 ` [RFC 12/13] linux-tdep: Export linux_make_corefile_notes John Baldwin
2023-10-09 18:36 ` [RFC 13/13] {amd64,i386}-linux-tdep: Include NT_X86_CPUID notes in core dumps from gcore John Baldwin
2023-10-10 16:30 ` [RFC 00/13] Proposal for a new NT_X86_CPUID core dump note George, Jini Susan
2023-10-12 4:01 ` Simon Marchi
2023-10-12 14:33 ` Simon Marchi
2023-10-12 17:18 ` John Baldwin
2023-10-13 9:38 ` George, Jini Susan
2023-10-17 0:36 ` John Baldwin
2023-10-26 16:18 ` George, Jini Susan
2023-10-27 2:53 ` John Baldwin
2023-10-27 11:11 ` George, Jini Susan
2023-10-31 16:41 ` John Baldwin
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=20231016091732.ybyjym67r2l7e3ol@octopus \
--to=lsix@lancelotsix.com \
--cc=George@sourceware.org \
--cc=JiniSusan.George@amd.com \
--cc=Willgerodt@sourceware.org \
--cc=felix.willgerodt@intel.com \
--cc=gdb-patches@sourceware.org \
--cc=jhb@FreeBSD.org \
--cc=simon.marchi@polymtl.ca \
/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;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).