From: Andrew Cooper <andrew.cooper3@citrix.com>
To: Xen-devel <xen-devel@lists.xen.org>
Cc: Andrew Cooper <andrew.cooper3@citrix.com>,
Jan Beulich <JBeulich@suse.com>
Subject: [PATCH v2 04/30] xen/x86: Improvements to pv_cpuid()
Date: Fri, 5 Feb 2016 13:41:57 +0000 [thread overview]
Message-ID: <1454679743-18133-5-git-send-email-andrew.cooper3@citrix.com> (raw)
In-Reply-To: <1454679743-18133-1-git-send-email-andrew.cooper3@citrix.com>
pv_cpuid() has two completely separate paths inside it depending on whether
current is dom0 or a domU. This causes unnecessary divergence, and
complicates future improvements. Take steps to undo it.
Changes:
* Create leaf and subleaf variables and use them consistently, instead of a
mix of {a,c} and regs->e{a,c}x as the input parameters.
* Combine the dom0 and domU hypervisor leaf handling, with an early exit.
* Apply sanity checks to domU as well. This brings PV domU cpuid handling in
line with HVM domains and PV dom0.
* Perform a real cpuid instruction for calculating CPUID.0xD[ECX=0].EBX. The
correct xcr0 is in context, and this avoids the O(M*N) loop over the domain
cpuid policy list which exists currently.
Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
---
CC: Jan Beulich <JBeulich@suse.com>
New in v2
---
xen/arch/x86/traps.c | 74 ++++++++++++++++++++--------------------------------
1 file changed, 29 insertions(+), 45 deletions(-)
diff --git a/xen/arch/x86/traps.c b/xen/arch/x86/traps.c
index e105b95..6a181bb 100644
--- a/xen/arch/x86/traps.c
+++ b/xen/arch/x86/traps.c
@@ -824,51 +824,24 @@ int cpuid_hypervisor_leaves( uint32_t idx, uint32_t sub_idx,
void pv_cpuid(struct cpu_user_regs *regs)
{
- uint32_t a, b, c, d;
+ uint32_t leaf, subleaf, a, b, c, d;
struct vcpu *curr = current;
struct domain *currd = curr->domain;
- a = regs->eax;
+ leaf = a = regs->eax;
b = regs->ebx;
- c = regs->ecx;
+ subleaf = c = regs->ecx;
d = regs->edx;
- if ( !is_control_domain(currd) && !is_hardware_domain(currd) )
- {
- unsigned int cpuid_leaf = a, sub_leaf = c;
-
- if ( !cpuid_hypervisor_leaves(a, c, &a, &b, &c, &d) )
- domain_cpuid(currd, a, c, &a, &b, &c, &d);
-
- switch ( cpuid_leaf )
- {
- case XSTATE_CPUID:
- {
- unsigned int _eax, _ebx, _ecx, _edx;
- /* EBX value of main leaf 0 depends on enabled xsave features */
- if ( sub_leaf == 0 && curr->arch.xcr0 )
- {
- /* reset EBX to default value first */
- b = XSTATE_AREA_MIN_SIZE;
- for ( sub_leaf = 2; sub_leaf < 63; sub_leaf++ )
- {
- if ( !(curr->arch.xcr0 & (1ULL << sub_leaf)) )
- continue;
- domain_cpuid(currd, cpuid_leaf, sub_leaf,
- &_eax, &_ebx, &_ecx, &_edx);
- if ( (_eax + _ebx) > b )
- b = _eax + _ebx;
- }
- }
- goto xstate;
- }
- }
+ if ( cpuid_hypervisor_leaves(leaf, subleaf, &a, &b, &c, &d) )
goto out;
- }
- cpuid_count(a, c, &a, &b, &c, &d);
+ if ( !is_control_domain(currd) && !is_hardware_domain(currd) )
+ domain_cpuid(currd, leaf, subleaf, &a, &b, &c, &d);
+ else
+ cpuid_count(leaf, subleaf, &a, &b, &c, &d);
- if ( (regs->eax & 0x7fffffff) == 0x00000001 )
+ if ( (leaf & 0x7fffffff) == 0x00000001 )
{
/* Modify Feature Information. */
if ( !cpu_has_apic )
@@ -883,7 +856,7 @@ void pv_cpuid(struct cpu_user_regs *regs)
}
}
- switch ( regs->_eax )
+ switch ( leaf )
{
case 0x00000001:
/* Modify Feature Information. */
@@ -918,7 +891,7 @@ void pv_cpuid(struct cpu_user_regs *regs)
break;
case 0x00000007:
- if ( regs->_ecx == 0 )
+ if ( subleaf == 0 )
b &= (cpufeat_mask(X86_FEATURE_BMI1) |
cpufeat_mask(X86_FEATURE_HLE) |
cpufeat_mask(X86_FEATURE_AVX2) |
@@ -934,14 +907,29 @@ void pv_cpuid(struct cpu_user_regs *regs)
break;
case XSTATE_CPUID:
- xstate:
if ( !cpu_has_xsave )
goto unsupported;
- if ( regs->_ecx == 1 )
+ switch ( subleaf )
+ {
+ case 0:
{
+ uint32_t tmp;
+
+ /*
+ * Always read CPUID.0xD[ECX=0].EBX from hardware, rather than
+ * domain policy. It varies with enabled xstate, and the correct
+ * xcr0 is in context.
+ */
+ if ( !is_control_domain(currd) && !is_hardware_domain(currd) )
+ cpuid_count(leaf, subleaf, &tmp, &b, &tmp, &tmp);
+ break;
+ }
+
+ case 1:
a &= (boot_cpu_data.x86_capability[cpufeat_word(X86_FEATURE_XSAVEOPT)] &
~cpufeat_mask(X86_FEATURE_XSAVES));
b = c = d = 0;
+ break;
}
break;
@@ -983,15 +971,11 @@ void pv_cpuid(struct cpu_user_regs *regs)
unsupported:
a = b = c = d = 0;
break;
-
- default:
- (void)cpuid_hypervisor_leaves(regs->eax, 0, &a, &b, &c, &d);
- break;
}
out:
/* VPMU may decide to modify some of the leaves */
- vpmu_do_cpuid(regs->eax, &a, &b, &c, &d);
+ vpmu_do_cpuid(leaf, &a, &b, &c, &d);
regs->eax = a;
regs->ebx = b;
--
2.1.4
next prev parent reply other threads:[~2016-02-05 13:41 UTC|newest]
Thread overview: 139+ messages / expand[flat|nested] mbox.gz Atom feed top
2016-02-05 13:41 [PATCH RFC v2 00/30] x86: Improvements to cpuid handling for guests Andrew Cooper
2016-02-05 13:41 ` [PATCH v2 01/30] xen/x86: Drop X86_FEATURE_3DNOW_ALT Andrew Cooper
2016-02-05 13:41 ` [PATCH v2 02/30] xen/x86: Do not store VIA/Cyrix/Centaur CPU features Andrew Cooper
2016-02-05 13:41 ` [PATCH v2 03/30] xen/x86: Drop cpuinfo_x86.x86_power Andrew Cooper
2016-02-05 13:41 ` Andrew Cooper [this message]
2016-02-05 13:41 ` [PATCH v2 05/30] xen/public: Export cpu featureset information in the public API Andrew Cooper
2016-02-12 16:27 ` Jan Beulich
2016-02-17 13:08 ` Andrew Cooper
2016-02-17 13:34 ` Jan Beulich
2016-02-19 17:29 ` Joao Martins
2016-02-19 17:55 ` Andrew Cooper
2016-02-19 22:03 ` Joao Martins
2016-02-20 16:17 ` Andrew Cooper
2016-02-20 17:39 ` Joao Martins
2016-02-20 19:17 ` Andrew Cooper
2016-02-22 18:50 ` Joao Martins
2016-02-05 13:41 ` [PATCH v2 06/30] xen/x86: Script to automatically process featureset information Andrew Cooper
2016-02-12 16:36 ` Jan Beulich
2016-02-12 16:43 ` Andrew Cooper
2016-02-05 13:42 ` [PATCH v2 07/30] xen/x86: Collect more cpuid feature leaves Andrew Cooper
2016-02-12 16:38 ` Jan Beulich
2016-02-05 13:42 ` [PATCH v2 08/30] xen/x86: Mask out unknown features from Xen's capabilities Andrew Cooper
2016-02-12 16:43 ` Jan Beulich
2016-02-12 16:48 ` Andrew Cooper
2016-02-12 17:14 ` Jan Beulich
2016-02-17 13:12 ` Andrew Cooper
2016-02-05 13:42 ` [PATCH v2 09/30] xen/x86: Store antifeatures inverted in a featureset Andrew Cooper
2016-02-12 16:47 ` Jan Beulich
2016-02-12 16:50 ` Andrew Cooper
2016-02-12 17:15 ` Jan Beulich
2016-02-05 13:42 ` [PATCH v2 10/30] xen/x86: Annotate VM applicability in featureset Andrew Cooper
2016-02-12 17:05 ` Jan Beulich
2016-02-12 17:42 ` Andrew Cooper
2016-02-15 9:20 ` Jan Beulich
2016-02-15 14:38 ` Andrew Cooper
2016-02-15 14:50 ` Jan Beulich
2016-02-15 14:53 ` Andrew Cooper
2016-02-15 15:02 ` Jan Beulich
2016-02-15 15:41 ` Andrew Cooper
2016-02-17 19:02 ` Is: PVH dom0 - MWAIT detection logic to get deeper C-states exposed in ACPI AML code. Was:Re: " Konrad Rzeszutek Wilk
2016-02-17 19:58 ` Boris Ostrovsky
2016-02-18 15:02 ` Roger Pau Monné
2016-02-18 15:12 ` Andrew Cooper
2016-02-18 16:24 ` Boris Ostrovsky
2016-02-18 16:48 ` Andrew Cooper
2016-02-18 17:03 ` Roger Pau Monné
2016-02-18 22:08 ` Konrad Rzeszutek Wilk
2016-02-18 15:16 ` David Vrabel
2016-02-05 13:42 ` [PATCH v2 11/30] xen/x86: Calculate maximum host and guest featuresets Andrew Cooper
2016-02-15 13:37 ` Jan Beulich
2016-02-15 14:57 ` Andrew Cooper
2016-02-15 15:07 ` Jan Beulich
2016-02-15 15:52 ` Andrew Cooper
2016-02-05 13:42 ` [PATCH v2 12/30] xen/x86: Generate deep dependencies of features Andrew Cooper
2016-02-15 14:06 ` Jan Beulich
2016-02-15 15:28 ` Andrew Cooper
2016-02-15 15:52 ` Jan Beulich
2016-02-15 16:09 ` Andrew Cooper
2016-02-15 16:27 ` Jan Beulich
2016-02-15 19:07 ` Andrew Cooper
2016-02-16 9:54 ` Jan Beulich
2016-02-17 10:25 ` Andrew Cooper
2016-02-17 10:42 ` Jan Beulich
2016-02-05 13:42 ` [PATCH v2 13/30] xen/x86: Clear dependent features when clearing a cpu cap Andrew Cooper
2016-02-15 14:53 ` Jan Beulich
2016-02-15 15:33 ` Andrew Cooper
2016-02-15 14:56 ` Jan Beulich
2016-02-05 13:42 ` [PATCH v2 14/30] xen/x86: Improve disabling of features which have dependencies Andrew Cooper
2016-02-05 13:42 ` [PATCH v2 15/30] xen/x86: Improvements to in-hypervisor cpuid sanity checks Andrew Cooper
2016-02-15 15:43 ` Jan Beulich
2016-02-15 17:12 ` Andrew Cooper
2016-02-16 10:06 ` Jan Beulich
2016-02-17 10:43 ` Andrew Cooper
2016-02-17 10:55 ` Jan Beulich
2016-02-17 14:02 ` Andrew Cooper
2016-02-17 14:45 ` Jan Beulich
2016-02-18 12:17 ` Andrew Cooper
2016-02-18 13:23 ` Jan Beulich
2016-02-05 13:42 ` [PATCH v2 16/30] x86/cpu: Move set_cpumask() calls into c_early_init() Andrew Cooper
2016-02-16 14:10 ` Jan Beulich
2016-02-17 10:45 ` Andrew Cooper
2016-02-17 10:58 ` Jan Beulich
2016-02-18 12:41 ` Andrew Cooper
2016-02-05 13:42 ` [PATCH v2 17/30] x86/cpu: Common infrastructure for levelling context switching Andrew Cooper
2016-02-16 14:15 ` Jan Beulich
2016-02-17 8:15 ` Jan Beulich
2016-02-17 10:46 ` Andrew Cooper
2016-02-17 19:06 ` Konrad Rzeszutek Wilk
2016-02-05 13:42 ` [PATCH v2 18/30] x86/cpu: Rework AMD masking MSR setup Andrew Cooper
2016-02-17 7:40 ` Jan Beulich
2016-02-17 10:56 ` Andrew Cooper
2016-02-05 13:42 ` [PATCH v2 19/30] x86/cpu: Rework Intel masking/faulting setup Andrew Cooper
2016-02-17 7:57 ` Jan Beulich
2016-02-17 10:59 ` Andrew Cooper
2016-02-05 13:42 ` [PATCH v2 20/30] x86/cpu: Context switch cpuid masks and faulting state in context_switch() Andrew Cooper
2016-02-17 8:06 ` Jan Beulich
2016-02-05 13:42 ` [PATCH v2 21/30] x86/pv: Provide custom cpumasks for PV domains Andrew Cooper
2016-02-17 8:13 ` Jan Beulich
2016-02-17 11:03 ` Andrew Cooper
2016-02-17 11:14 ` Jan Beulich
2016-02-18 12:48 ` Andrew Cooper
2016-02-05 13:42 ` [PATCH v2 22/30] x86/domctl: Update PV domain cpumasks when setting cpuid policy Andrew Cooper
2016-02-17 8:22 ` Jan Beulich
2016-02-17 12:13 ` Andrew Cooper
2016-02-05 13:42 ` [PATCH v2 23/30] xen+tools: Export maximum host and guest cpu featuresets via SYSCTL Andrew Cooper
2016-02-05 16:12 ` Wei Liu
2016-02-17 8:30 ` Jan Beulich
2016-02-17 12:17 ` Andrew Cooper
2016-02-17 12:23 ` Jan Beulich
2016-02-05 13:42 ` [PATCH v2 24/30] tools/libxc: Modify bitmap operations to take void pointers Andrew Cooper
2016-02-05 16:12 ` Wei Liu
2016-02-08 11:40 ` Andrew Cooper
2016-02-08 16:23 ` Tim Deegan
2016-02-08 16:36 ` Ian Campbell
2016-02-10 10:07 ` Andrew Cooper
2016-02-10 10:18 ` Ian Campbell
2016-02-18 13:37 ` Andrew Cooper
2016-02-17 20:06 ` Konrad Rzeszutek Wilk
2016-02-05 13:42 ` [PATCH v2 25/30] tools/libxc: Use public/featureset.h for cpuid policy generation Andrew Cooper
2016-02-05 16:12 ` Wei Liu
2016-02-05 13:42 ` [PATCH v2 26/30] tools/libxc: Expose the automatically generated cpu featuremask information Andrew Cooper
2016-02-05 16:12 ` Wei Liu
2016-02-05 16:15 ` Wei Liu
2016-02-05 13:42 ` [PATCH v2 27/30] tools: Utility for dealing with featuresets Andrew Cooper
2016-02-05 16:13 ` Wei Liu
2016-02-05 13:42 ` [PATCH v2 28/30] tools/libxc: Wire a featureset through to cpuid policy logic Andrew Cooper
2016-02-05 16:13 ` Wei Liu
2016-02-05 13:42 ` [PATCH v2 29/30] tools/libxc: Use featuresets rather than guesswork Andrew Cooper
2016-02-05 16:13 ` Wei Liu
2016-02-17 8:55 ` Jan Beulich
2016-02-17 13:03 ` Andrew Cooper
2016-02-17 13:19 ` Jan Beulich
2016-02-05 13:42 ` [PATCH v2 30/30] tools/libxc: Calculate xstate cpuid leaf from guest information Andrew Cooper
2016-02-05 14:28 ` Jan Beulich
2016-02-05 15:22 ` Andrew Cooper
2016-02-08 17:26 ` [PATCH v2.5 31/30] Fix PV guest XSAVE handling with levelling Andrew Cooper
2016-02-17 9:02 ` Jan Beulich
2016-02-17 13:06 ` Andrew Cooper
2016-02-17 13:36 ` Jan Beulich
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=1454679743-18133-5-git-send-email-andrew.cooper3@citrix.com \
--to=andrew.cooper3@citrix.com \
--cc=JBeulich@suse.com \
--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).