From: Shuai Ruan <shuai.ruan@linux.intel.com>
To: Jan Beulich <JBeulich@suse.com>
Cc: andrew.cooper3@citrix.com, keir@xen.org, xen-devel@lists.xen.org
Subject: Re: [V5] x86/xsaves: fix overwriting between non-lazy/lazy xsaves
Date: Wed, 23 Mar 2016 10:02:24 +0800 [thread overview]
Message-ID: <45615.7328481369$1458698802@news.gmane.org> (raw)
In-Reply-To: <56F1660902000078000DF38B@prv-mh.provo.novell.com>
On Tue, Mar 22, 2016 at 08:34:33AM -0600, Jan Beulich wrote:
> >>> On 18.03.16 at 04:01, <shuai.ruan@linux.intel.com> wrote:
> > * Copy legacy XSAVE area, to avoid complications with CPUID
> > * leaves 0 and 1 in the loop below.
> > */
> > memcpy(xsave, src, FXSAVE_SIZE);
> >
> > - /* Set XSTATE_BV and XCOMP_BV. */
> > + /* Set XSTATE_BV. */
> > xsave->xsave_hdr.xstate_bv = xstate_bv;
> > - xsave->xsave_hdr.xcomp_bv = v->arch.xcr0_accum | XSTATE_COMPACTION_ENABLED;
> > setup_xstate_comp(xstate_comp_offsets, xstate_bv);
>
> I see you set xcomp_bv (and hence the compaction bit) in xrstor()
> now, but afaict that doesn't allow you to completely drop initializing
> the field here, as the code there looks at the compaction bit.
>
> > + else
> > + XRSTOR("0x48,","0x0f,0xae,0x2f"); /* xrstor */
>
> At this point, what guarantees that xcomp_bv is zero, no matter
> where the state to be loaded originates from? I think at least in
> arch_set_info_guest(), hvm_load_cpu_ctxt(), and
> hvm_vcpu_reset_state() you went too far deleting code, and you
> really need to keep the storing of zero there. Did you draw, just
> for yourself, mentally or on a sheet of paper, a diagram illustrating
> the various state transitions?
>
For above two comments.
The patch is base on [v4]x86/xsaves: calculate the xstate_comp_offsets base
on xstate_bv and I answer your question on why caculate xstate_comp_offset
based on xstate_bv in previous thread. If that is right, drop
initializing xcomp_bv is ok. Now xcomp_bv can guarantee to be zero for
arch_set_info_guest() and hvm_load_cpu_ctxt(). If the drop is wrong
(due to my misunderstand of the SDM), I will change the if () here.
For hvm_vcpu_reset_state(), we should depend on whether xsaves is used
to decide whether to init xcomp_bv or not. And currently we use
xcr0_accum to indicate the use of xsaves, when hvm_vcpu_reset_state()
is called , can vcpu->xcr0_accum indicate using of xsaves ?
I think in hvm_vcpu_reset_state(), we should leave xcomp_bv zero.
> Xen-devel@lists.xen.org
> http://lists.xen.org/xen-devel
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel
next prev parent reply other threads:[~2016-03-23 2:02 UTC|newest]
Thread overview: 6+ messages / expand[flat|nested] mbox.gz Atom feed top
2016-03-18 3:01 [V5] x86/xsaves: fix overwriting between non-lazy/lazy xsaves Shuai Ruan
2016-03-22 14:34 ` Jan Beulich
2016-03-23 2:02 ` Shuai Ruan [this message]
[not found] ` <20160323020224.GB4131@shuai.ruan@linux.intel.com>
2016-03-23 6:14 ` Shuai Ruan
[not found] ` <20160323061455.GA12388@shuai.ruan@linux.intel.com>
2016-03-23 10:21 ` Jan Beulich
2016-03-23 11:01 ` Shuai Ruan
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='45615.7328481369$1458698802@news.gmane.org' \
--to=shuai.ruan@linux.intel.com \
--cc=JBeulich@suse.com \
--cc=andrew.cooper3@citrix.com \
--cc=keir@xen.org \
--cc=xen-devel@lists.xen.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 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 NNTP newsgroup(s).