From: Zhao Liu <zhao1.liu@intel.com>
To: John Snow <jsnow@redhat.com>
Cc: qemu-devel@nongnu.org, Paolo Bonzini <pbonzini@redhat.com>,
kvm@vger.kernel.org, Marcelo Tosatti <mtosatti@redhat.com>,
Zhao Liu <zhao1.liu@intel.com>
Subject: Re: [PATCH] i386/cpu: mark CPU properties as const
Date: Fri, 20 Nov 2026 17:58:23 +0800 [thread overview]
Message-ID: <awAZv3AsmNdJ7oUW@intel.com> (raw)
In-Reply-To: <20261119180859.1044425-1-jsnow@redhat.com>
Hi John,
On Thu, Nov 19, 2026 at 01:08:59PM -0500, John Snow wrote:
> Date: Thu, 19 Nov 2026 13:08:59 -0500
> From: John Snow <jsnow@redhat.com>
> Subject: [PATCH] i386/cpu: mark CPU properties as const
>
> Currently, these large property tables are loaded in the read-write
> `.data` segment; by marking them as const they can be loaded as part
> of `.rodata` instead, where these tables can be shared between multiple
> instances of QEMU as RssFile memory instead of RssAnon memory.
>
> Signed-off-by: John Snow <jsnow@redhat.com>
> ---
> target/i386/cpu-internal.h | 2 +-
> target/i386/cpu-system.c | 2 +-
> target/i386/cpu.c | 30 +++++++++++++++---------------
> target/i386/kvm/tdx.c | 2 +-
> 4 files changed, 18 insertions(+), 18 deletions(-)
I'm fine with the "const" change, but it seems only
cpuid2_cache_descriptors can move to .rodata:
$ objdump -t build/libqemu-x86_64-softmmu.a.p/target_i386_cpu.c.o | grep -wE 'feature_word_info|cpuid2_cache_descriptors|epyc_v5_cache_info|epyc_v4_cache_info'
0000000000005fa0 l O .data.rel.ro.local 0000000000000020 epyc_v4_cache_info
0000000000005f80 l O .data.rel.ro.local 0000000000000020 epyc_v5_cache_info
0000000000000d20 g O .rodata 0000000000001284 cpuid2_cache_descriptors
00000000000060a0 g O .data.rel.ro.local 0000000000006a80 feature_word_info
I think it's because FeatureWordInfo and CPUCaches have pointer fields.
So with the commit message fixed:
Reviewed-by: Zhao Liu <zhao1.liu@intel.com>
...
Futhermore, for CPUCaches, it could embed the cache info by value:
typedef struct CPUCaches {
CPUCacheInfo l1d_cache;
CPUCacheInfo l1i_cache;
CPUCacheInfo l2_cache;
CPUCacheInfo l3_cache;
} CPUCaches;
although the cache info will be stored in .rodata, this would also give
each vCPU its own copy, so smp-cache would no longer modify the global
tables. The cost is that CPUX86State grows per vCPU (from 4 pointers to
4 full CPUCacheInfo). So, maybe there’s room for further optimization,
but that’s a separate issue, not a blocker for this patch.
For FeatureWordInfo, what I can think of right now is to make the
feature name array a fixed length like:
#define FEAT_NAME_SIZE 36 /* strlen("vmx-invept-single-context-noglobals") + 1 */
typedef struct FeatureWordInfo {
FeatureWordType type;
char feat_names[64][FEAT_NAME_SIZE];
...
} FeatureWordInfo;
Thanks,
Zhao
prev parent reply other threads:[~2026-11-20 9:58 UTC|newest]
Thread overview: 4+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-11-19 18:08 [PATCH] i386/cpu: mark CPU properties as const John Snow
2026-11-19 18:40 ` Peter Maydell
2026-11-19 18:55 ` Richard Henderson
2026-11-20 9:58 ` Zhao Liu [this message]
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=awAZv3AsmNdJ7oUW@intel.com \
--to=zhao1.liu@intel.com \
--cc=jsnow@redhat.com \
--cc=kvm@vger.kernel.org \
--cc=mtosatti@redhat.com \
--cc=pbonzini@redhat.com \
--cc=qemu-devel@nongnu.org \
/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 an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.