diff mbox

Error booting Xen

Message ID 56A8D03E02000078000CB8B8@prv-mh.provo.novell.com (mailing list archive)
State New, archived
Headers show

Commit Message

Jan Beulich Jan. 27, 2016, 1:12 p.m. UTC
>>> On 26.01.16 at 19:02, <write.harmandeep@gmail.com> wrote:
> Last time, I did absolutely nothing. System was idle
> and it crashed just after the login. Now, I booted the
> system again and this time, there is no reset. But,
> performance of the system is very slow. Browser
> (Mozilla Firefox) freezes a lot. Also, before applying
> patches, when I used to disabe xsave it resulted in
> same kind of performance issues. And the following
> is still present in the log.
> 
> (XEN) traps.c:3290: GPF (0000): ffff82d0801c1cea -> ffff82d080252e5c
> (XEN) d1v1 fault#1: mxcsr=00001f80
> (XEN) d1v1 xs=0000000000000003 xc=8000000000000000
> (XEN) d1v1 r0=0000000000000000 r1=0000000000000000
> (XEN) d1v1 r2=0000000000000000 r3=0000000000000000
> (XEN) d1v1 r4=0000000000000000 r5=0000000000000000
> (XEN) traps.c:3290: GPF (0000): ffff82d0801c1cea -> ffff82d080252e5c
> (XEN) d1v1 fault#2: mxcsr=00001f80
> (XEN) d1v1 xs=0000000000000000 xc=0000000000000000
> (XEN) d1v1 r0=0000000000000000 r1=0000000000000000
> (XEN) d1v1 r2=0000000000000000 r3=0000000000000000
> (XEN) d1v1 r4=0000000000000000 r5=0000000000000000
> 
> Full log here: http://paste2.org/C8WpyKOg 

This together with ...

> On Tue, Jan 26, 2016 at 10:53 PM, Jan Beulich <JBeulich@suse.com> wrote:
>>>>> On 26.01.16 at 18:01, <write.harmandeep@gmail.com> wrote:
>>> I tried 3rd patch together with earlier two. I'm
>>> afraid the problem is not solved completely.
>>> Full log goes here, http://paste2.org/KEAetMHb 

... this, and both being apparently the same build makes me suspect
uninitialized data to get passed in from the tool stack. But that's a
secondary issue for now. For the immediate problem here are four
patches replacing the three earlier ones (I think only one of them is
unchanged, so be sure to remove the old ones first).

Their intended ordering is:
x86-xsaves-init.patch
x86-xstate-align.patch
x86-xrstors-fault.patch
x86-xstate-validate.patch

Jan
x86/xstate: fix fault behavior on XRSTORS

XRSTORS unconditionally faults when xcomp_bv has bit 63 clear. Instead
of just fixing this issue, overhaul the fault recovery code, which -
one of the many mistakes made when xstate support got introduced - was
blindly mirroring that accompanying FXRSTOR, neglecting the fact that
XRSTOR{,S} aren't all-or-nothing instructions. The new code, first of
all, does all the recovery actions in C, simplifying the inline
assembly used. And it does its work in a multi-stage fashion: Upon
first seeing a fault, state fixups get applied strictly based on what
architecturally may cause #GP. When seeing another fault despite the
fixups done, state gets fully reset. A third fault would then lead to
crashing the domain (instead of hanging the hypervisor in an infinite
loop of recurring faults).

Reported-by: Harmandeep Kaur <write.harmandeep@gmail.com>
Signed-off-by: Jan Beulich <jbeulich@suse.com>
x86/xstate: fix xcomp_bv initialization

We must not clear the compaction bit when using XSAVES/XRSTORS. And
we need to guarantee that xcomp_bv never has any bits clear which
are set in xstate_bv (which requires partly undoing commit 83ae0bb226
["x86/xsave: simplify xcomp_bv initialization"]). Split initialization
of xcomp_bv from the other FPU/SSE/AVX related state setup in
arch_set_info_guest() and hvm_load_cpu_ctxt().

Reported-by: Harmandeep Kaur <write.harmandeep@gmail.com>
Signed-off-by: Jan Beulich <jbeulich@suse.com>

--- unstable.orig/xen/arch/x86/domain.c	2016-01-27 09:29:50.000000000 +0100
+++ unstable/xen/arch/x86/domain.c	2016-01-27 09:52:37.000000000 +0100
@@ -922,15 +922,10 @@ int arch_set_info_guest(
     {
         memcpy(v->arch.fpu_ctxt, &c.nat->fpu_ctxt, sizeof(c.nat->fpu_ctxt));
         if ( v->arch.xsave_area )
-        {
             v->arch.xsave_area->xsave_hdr.xstate_bv = XSTATE_FP_SSE;
-            v->arch.xsave_area->xsave_hdr.xcomp_bv =
-                cpu_has_xsaves ? XSTATE_COMPACTION_ENABLED : 0;
-        }
     }
     else if ( v->arch.xsave_area )
-        memset(&v->arch.xsave_area->xsave_hdr, 0,
-               sizeof(v->arch.xsave_area->xsave_hdr));
+        v->arch.xsave_area->xsave_hdr.xstate_bv = 0;
     else
     {
         typeof(v->arch.xsave_area->fpu_sse) *fpu_sse = v->arch.fpu_ctxt;
@@ -939,6 +934,14 @@ int arch_set_info_guest(
         fpu_sse->fcw = FCW_DEFAULT;
         fpu_sse->mxcsr = MXCSR_DEFAULT;
     }
+    if ( cpu_has_xsaves )
+    {
+        ASSERT(v->arch.xsave_area);
+        v->arch.xsave_area->xsave_hdr.xcomp_bv = XSTATE_COMPACTION_ENABLED |
+            v->arch.xsave_area->xsave_hdr.xstate_bv;
+    }
+    else if ( v->arch.xsave_area )
+        v->arch.xsave_area->xsave_hdr.xcomp_bv = 0;
 
     if ( !compat )
     {
--- unstable.orig/xen/arch/x86/hvm/hvm.c	2015-12-18 12:22:20.000000000 +0100
+++ unstable/xen/arch/x86/hvm/hvm.c	2016-01-27 09:52:26.000000000 +0100
@@ -2094,11 +2094,17 @@ static int hvm_load_cpu_ctxt(struct doma
 
         memcpy(v->arch.xsave_area, ctxt.fpu_regs, sizeof(ctxt.fpu_regs));
         xsave_area->xsave_hdr.xstate_bv = XSTATE_FP_SSE;
-        xsave_area->xsave_hdr.xcomp_bv =
-            cpu_has_xsaves ? XSTATE_COMPACTION_ENABLED : 0;
     }
     else
         memcpy(v->arch.fpu_ctxt, ctxt.fpu_regs, sizeof(ctxt.fpu_regs));
+    if ( cpu_has_xsaves )
+    {
+        ASSERT(v->arch.xsave_area);
+        v->arch.xsave_area->xsave_hdr.xcomp_bv = XSTATE_COMPACTION_ENABLED |
+            v->arch.xsave_area->xsave_hdr.xstate_bv;
+    }
+    else if ( v->arch.xsave_area )
+        v->arch.xsave_area->xsave_hdr.xcomp_bv = 0;
 
     v->arch.user_regs.eax = ctxt.rax;
     v->arch.user_regs.ebx = ctxt.rbx;
@@ -5488,8 +5494,8 @@ void hvm_vcpu_reset_state(struct vcpu *v
     if ( v->arch.xsave_area )
     {
         v->arch.xsave_area->xsave_hdr.xstate_bv = XSTATE_FP;
-        v->arch.xsave_area->xsave_hdr.xcomp_bv =
-            cpu_has_xsaves ? XSTATE_COMPACTION_ENABLED : 0;
+        v->arch.xsave_area->xsave_hdr.xcomp_bv = cpu_has_xsaves
+            ? XSTATE_COMPACTION_ENABLED | XSTATE_FP : 0;
     }
 
     v->arch.vgc_flags = VGCF_online;
x86: adjust xsave structure attributes

The packed attribute was pointlessly used here - there are no
misaligned fields, and hence even if the attribute took effect, it
would at best lead to the compiler generating worse code.

At the same time specify the required alignment of the fpu_sse sub-
structure, such that the various typeof() uses on that field obtain
pointers to properly aligned memory (knowledge which a compiler may
want to make use of).

Also add suitable build-time checks.

Signed-off-by: Jan Beulich <jbeulich@suse.com>

--- unstable.orig/xen/arch/x86/i387.c	2016-01-25 11:30:11.000000000 +0100
+++ unstable/xen/arch/x86/i387.c	2016-01-25 09:35:36.000000000 +0100
@@ -277,7 +277,9 @@ int vcpu_init_fpu(struct vcpu *v)
     }
     else
     {
-        v->arch.fpu_ctxt = _xzalloc(sizeof(v->arch.xsave_area->fpu_sse), 16);
+        BUILD_BUG_ON(__alignof(v->arch.xsave_area->fpu_sse) < 16);
+        v->arch.fpu_ctxt = _xzalloc(sizeof(v->arch.xsave_area->fpu_sse),
+                                    __alignof(v->arch.xsave_area->fpu_sse));
         if ( v->arch.fpu_ctxt )
         {
             typeof(v->arch.xsave_area->fpu_sse) *fpu_sse = v->arch.fpu_ctxt;
--- unstable.orig/xen/arch/x86/xstate.c	2016-01-25 11:30:11.000000000 +0100
+++ unstable/xen/arch/x86/xstate.c	2016-01-25 09:35:12.000000000 +0100
@@ -414,7 +414,8 @@ int xstate_alloc_save_area(struct vcpu *
     BUG_ON(xsave_cntxt_size < XSTATE_AREA_MIN_SIZE);
 
     /* XSAVE/XRSTOR requires the save area be 64-byte-boundary aligned. */
-    save_area = _xzalloc(xsave_cntxt_size, 64);
+    BUILD_BUG_ON(__alignof(*save_area) < 64);
+    save_area = _xzalloc(xsave_cntxt_size, __alignof(*save_area));
     if ( save_area == NULL )
         return -ENOMEM;
 
--- unstable.orig/xen/include/asm-x86/xstate.h	2016-01-25 11:30:11.000000000 +0100
+++ unstable/xen/include/asm-x86/xstate.h	2016-01-25 11:33:20.000000000 +0100
@@ -48,9 +48,9 @@ extern u64 xfeature_mask;
 extern unsigned int *xstate_sizes;
 
 /* extended state save area */
-struct __packed __attribute__((aligned (64))) xsave_struct
+struct __attribute__((aligned (64))) xsave_struct
 {
-    union {                                  /* FPU/MMX, SSE */
+    union __attribute__((aligned(16))) {     /* FPU/MMX, SSE */
         char x[512];
         struct {
             uint16_t fcw;
x86/xstate: extend validation to cover full header

Since we never hand out compacted state, at least for now we're also
not going to accept such.

Reported-by: Harmandeep Kaur <write.harmandeep@gmail.com>
Signed-off-by: Jan Beulich <jbeulich@suse.com>

--- unstable.orig/xen/arch/x86/domctl.c	2016-01-27 10:54:16.000000000 +0100
+++ unstable/xen/arch/x86/domctl.c	2016-01-27 10:44:52.000000000 +0100
@@ -958,7 +958,7 @@ long arch_do_domctl(
             {
                 if ( evc->size >= 2 * sizeof(uint64_t) + XSTATE_AREA_MIN_SIZE )
                     ret = validate_xstate(_xcr0, _xcr0_accum,
-                                          _xsave_area->xsave_hdr.xstate_bv);
+                                          &_xsave_area->xsave_hdr);
             }
             else if ( !_xcr0 )
                 ret = 0;
--- unstable.orig/xen/arch/x86/hvm/hvm.c	2016-01-27 09:52:26.000000000 +0100
+++ unstable/xen/arch/x86/hvm/hvm.c	2016-01-27 11:09:44.000000000 +0100
@@ -2178,6 +2178,19 @@ static int hvm_save_cpu_xsave_states(str
     return 0;
 }
 
+/*
+ * Structure layout conformity checks, documenting correctness of the cast in
+ * the invocation of validate_xstate() below.
+ * Leverage CONFIG_COMPAT machinery to perform this.
+ */
+#define xen_xsave_hdr xsave_hdr
+#define compat_xsave_hdr hvm_hw_cpu_xsave_hdr
+CHECK_FIELD_(struct, xsave_hdr, xstate_bv);
+CHECK_FIELD_(struct, xsave_hdr, xcomp_bv);
+CHECK_FIELD_(struct, xsave_hdr, reserved);
+#undef compat_xsave_hdr
+#undef xen_xsave_hdr
+
 static int hvm_load_cpu_xsave_states(struct domain *d, hvm_domain_context_t *h)
 {
     unsigned int vcpuid, size;
@@ -2233,7 +2246,7 @@ static int hvm_load_cpu_xsave_states(str
     h->cur += desc->length;
 
     err = validate_xstate(ctxt->xcr0, ctxt->xcr0_accum,
-                          ctxt->save_area.xsave_hdr.xstate_bv);
+                          (const void *)&ctxt->save_area.xsave_hdr);
     if ( err )
     {
         printk(XENLOG_G_WARNING
--- unstable.orig/xen/arch/x86/xstate.c	2016-01-27 10:23:06.000000000 +0100
+++ unstable/xen/arch/x86/xstate.c	2016-01-27 10:48:22.000000000 +0100
@@ -614,17 +614,24 @@ static bool_t valid_xcr0(u64 xcr0)
     return !(xcr0 & XSTATE_BNDREGS) == !(xcr0 & XSTATE_BNDCSR);
 }
 
-int validate_xstate(u64 xcr0, u64 xcr0_accum, u64 xstate_bv)
+int validate_xstate(u64 xcr0, u64 xcr0_accum, const struct xsave_hdr *hdr)
 {
-    if ( (xstate_bv & ~xcr0_accum) ||
+    unsigned int i;
+
+    if ( (hdr->xstate_bv & ~xcr0_accum) ||
          (xcr0 & ~xcr0_accum) ||
          !valid_xcr0(xcr0) ||
          !valid_xcr0(xcr0_accum) )
         return -EINVAL;
 
-    if ( xcr0_accum & ~xfeature_mask )
+    if ( (xcr0_accum & ~xfeature_mask) ||
+         hdr->xcomp_bv )
         return -EOPNOTSUPP;
 
+    for ( i = 0; i < ARRAY_SIZE(hdr->reserved); ++i )
+        if ( hdr->reserved[i] )
+            return -EIO;
+
     return 0;
 }
 
--- unstable.orig/xen/include/asm-x86/xstate.h	2016-01-25 11:33:20.000000000 +0100
+++ unstable/xen/include/asm-x86/xstate.h	2016-01-27 10:57:54.000000000 +0100
@@ -72,14 +72,13 @@ struct __attribute__((aligned (64))) xsa
         };
     } fpu_sse;
 
-    struct {
+    struct xsave_hdr {
         u64 xstate_bv;
         u64 xcomp_bv;
         u64 reserved[6];
     } xsave_hdr;                             /* The 64-byte header */
 
-    struct { char x[XSTATE_YMM_SIZE]; } ymm; /* YMM */
-    char   data[];                           /* Future new states */
+    char data[];                             /* Variable layout states */
 };
 
 /* extended state operations */
@@ -90,7 +89,8 @@ uint64_t get_msr_xss(void);
 void xsave(struct vcpu *v, uint64_t mask);
 void xrstor(struct vcpu *v, uint64_t mask);
 bool_t xsave_enabled(const struct vcpu *v);
-int __must_check validate_xstate(u64 xcr0, u64 xcr0_accum, u64 xstate_bv);
+int __must_check validate_xstate(u64 xcr0, u64 xcr0_accum,
+                                 const struct xsave_hdr *);
 int __must_check handle_xsetbv(u32 index, u64 new_bv);
 void expand_xsave_states(struct vcpu *v, void *dest, unsigned int size);
 void compress_xsave_states(struct vcpu *v, const void *src, unsigned int size);
--- unstable.orig/xen/include/public/arch-x86/hvm/save.h	2016-01-13 07:56:27.000000000 +0100
+++ unstable/xen/include/public/arch-x86/hvm/save.h	2016-01-27 11:09:20.000000000 +0100
@@ -550,12 +550,11 @@ struct hvm_hw_cpu_xsave {
     struct {
         struct { char x[512]; } fpu_sse;
 
-        struct {
+        struct hvm_hw_cpu_xsave_hdr {
             uint64_t xstate_bv;         /* Updated by XRSTOR */
-            uint64_t reserved[7];
+            uint64_t xcomp_bv;          /* Updated by XRSTOR{C,S} */
+            uint64_t reserved[6];
         } xsave_hdr;                    /* The 64-byte header */
-
-        struct { char x[0]; } ymm;    /* YMM */
     } save_area;
 };

Comments

Harmandeep Kaur Jan. 27, 2016, 1:28 p.m. UTC | #1
I tried to apply your patches but it seems
to have some merge conflicts with latest
staging branch.

~/xen$ git apply ~/Downloads/x86-xsaves-init.patch
error: patch failed: xen/arch/x86/hvm/hvm.c:2094
error: xen/arch/x86/hvm/hvm.c: patch does not apply

Do you mind having a look ?

Regards,
Harmandeep

On Wed, Jan 27, 2016 at 6:42 PM, Jan Beulich <JBeulich@suse.com> wrote:
>>>> On 26.01.16 at 19:02, <write.harmandeep@gmail.com> wrote:
>> Last time, I did absolutely nothing. System was idle
>> and it crashed just after the login. Now, I booted the
>> system again and this time, there is no reset. But,
>> performance of the system is very slow. Browser
>> (Mozilla Firefox) freezes a lot. Also, before applying
>> patches, when I used to disabe xsave it resulted in
>> same kind of performance issues. And the following
>> is still present in the log.
>>
>> (XEN) traps.c:3290: GPF (0000): ffff82d0801c1cea -> ffff82d080252e5c
>> (XEN) d1v1 fault#1: mxcsr=00001f80
>> (XEN) d1v1 xs=0000000000000003 xc=8000000000000000
>> (XEN) d1v1 r0=0000000000000000 r1=0000000000000000
>> (XEN) d1v1 r2=0000000000000000 r3=0000000000000000
>> (XEN) d1v1 r4=0000000000000000 r5=0000000000000000
>> (XEN) traps.c:3290: GPF (0000): ffff82d0801c1cea -> ffff82d080252e5c
>> (XEN) d1v1 fault#2: mxcsr=00001f80
>> (XEN) d1v1 xs=0000000000000000 xc=0000000000000000
>> (XEN) d1v1 r0=0000000000000000 r1=0000000000000000
>> (XEN) d1v1 r2=0000000000000000 r3=0000000000000000
>> (XEN) d1v1 r4=0000000000000000 r5=0000000000000000
>>
>> Full log here: http://paste2.org/C8WpyKOg
>
> This together with ...
>
>> On Tue, Jan 26, 2016 at 10:53 PM, Jan Beulich <JBeulich@suse.com> wrote:
>>>>>> On 26.01.16 at 18:01, <write.harmandeep@gmail.com> wrote:
>>>> I tried 3rd patch together with earlier two. I'm
>>>> afraid the problem is not solved completely.
>>>> Full log goes here, http://paste2.org/KEAetMHb
>
> ... this, and both being apparently the same build makes me suspect
> uninitialized data to get passed in from the tool stack. But that's a
> secondary issue for now. For the immediate problem here are four
> patches replacing the three earlier ones (I think only one of them is
> unchanged, so be sure to remove the old ones first).
>
> Their intended ordering is:
> x86-xsaves-init.patch
> x86-xstate-align.patch
> x86-xrstors-fault.patch
> x86-xstate-validate.patch
>
> Jan
>
diff mbox

Patch

--- unstable.orig/xen/arch/x86/xstate.c	2016-01-25 09:35:12.000000000 +0100
+++ unstable/xen/arch/x86/xstate.c	2016-01-27 10:23:06.000000000 +0100
@@ -29,6 +29,8 @@  unsigned int *__read_mostly xstate_sizes
 static unsigned int __read_mostly xstate_features;
 static unsigned int __read_mostly xstate_comp_offsets[sizeof(xfeature_mask)*8];
 
+static uint32_t __read_mostly mxcsr_mask = MXCSR_DEFAULT;
+
 /* Cached xcr0 for fast read */
 static DEFINE_PER_CPU(uint64_t, xcr0);
 
@@ -342,6 +344,7 @@  void xrstor(struct vcpu *v, uint64_t mas
     uint32_t hmask = mask >> 32;
     uint32_t lmask = mask;
     struct xsave_struct *ptr = v->arch.xsave_area;
+    unsigned int faults, prev_faults;
 
     /*
      * AMD CPUs don't save/restore FDP/FIP/FOP unless an exception
@@ -361,35 +364,85 @@  void xrstor(struct vcpu *v, uint64_t mas
     /*
      * XRSTOR can fault if passed a corrupted data block. We handle this
      * possibility, which may occur if the block was passed to us by control
-     * tools or through VCPUOP_initialise, by silently clearing the block.
+     * tools or through VCPUOP_initialise, by silently adjusting state.
      */
-    switch ( __builtin_expect(ptr->fpu_sse.x[FPU_WORD_SIZE_OFFSET], 8) )
+    for ( prev_faults = faults = 0; ; prev_faults = faults )
     {
+        switch ( __builtin_expect(ptr->fpu_sse.x[FPU_WORD_SIZE_OFFSET], 8) )
+        {
 #define XRSTOR(pfx) \
         alternative_io("1: .byte " pfx "0x0f,0xae,0x2f\n" \
+                       "3:\n" \
                        "   .section .fixup,\"ax\"\n" \
-                       "2: mov %[size],%%ecx\n" \
-                       "   xor %[lmask_out],%[lmask_out]\n" \
-                       "   rep stosb\n" \
-                       "   lea %[mem],%[ptr]\n" \
-                       "   mov %[lmask_in],%[lmask_out]\n" \
-                       "   jmp 1b\n" \
+                       "2: inc%z[faults] %[faults]\n" \
+                       "   jmp 3b\n" \
                        "   .previous\n" \
                        _ASM_EXTABLE(1b, 2b), \
                        ".byte " pfx "0x0f,0xc7,0x1f\n", \
                        X86_FEATURE_XSAVES, \
-                       ASM_OUTPUT2([ptr] "+&D" (ptr), [lmask_out] "+&a" (lmask)), \
-                       [mem] "m" (*ptr), [lmask_in] "g" (lmask), \
-                       [hmask] "d" (hmask), [size] "m" (xsave_cntxt_size) \
-                       : "ecx")
-
-    default:
-        XRSTOR("0x48,");
-        break;
-    case 4: case 2:
-        XRSTOR("");
-        break;
+                       ASM_OUTPUT2([mem] "+m" (*ptr), [faults] "+g" (faults)), \
+                       [lmask] "a" (lmask), [hmask] "d" (hmask), \
+                       [ptr] "D" (ptr))
+
+        default:
+            XRSTOR("0x48,");
+            break;
+        case 4: case 2:
+            XRSTOR("");
+            break;
 #undef XRSTOR
+        }
+        if ( likely(faults == prev_faults) )
+            break;
+#ifndef NDEBUG
+        gprintk(XENLOG_WARNING, "fault#%u: mxcsr=%08x\n",
+                faults, ptr->fpu_sse.mxcsr);
+        gprintk(XENLOG_WARNING, "xs=%016lx xc=%016lx\n",
+                ptr->xsave_hdr.xstate_bv, ptr->xsave_hdr.xcomp_bv);
+        gprintk(XENLOG_WARNING, "r0=%016lx r1=%016lx\n",
+                ptr->xsave_hdr.reserved[0], ptr->xsave_hdr.reserved[1]);
+        gprintk(XENLOG_WARNING, "r2=%016lx r3=%016lx\n",
+                ptr->xsave_hdr.reserved[2], ptr->xsave_hdr.reserved[3]);
+        gprintk(XENLOG_WARNING, "r4=%016lx r5=%016lx\n",
+                ptr->xsave_hdr.reserved[4], ptr->xsave_hdr.reserved[5]);
+#endif
+        switch ( faults )
+        {
+        case 1:
+            /* Stage 1: Reset state to be loaded. */
+            ptr->xsave_hdr.xstate_bv &= ~mask;
+            /*
+             * Also try to eliminate fault reasons, even if this shouldn't be
+             * needed here (other code should ensure the sanity of the data).
+             */
+            if ( ((mask & XSTATE_SSE) ||
+                  ((mask & XSTATE_YMM) &&
+                   !(ptr->xsave_hdr.xcomp_bv & XSTATE_COMPACTION_ENABLED))) )
+                ptr->fpu_sse.mxcsr &= mxcsr_mask;
+            if ( cpu_has_xsaves || cpu_has_xsavec )
+            {
+                ptr->xsave_hdr.xcomp_bv &= this_cpu(xcr0) | this_cpu(xss);
+                ptr->xsave_hdr.xstate_bv &= ptr->xsave_hdr.xcomp_bv;
+                ptr->xsave_hdr.xcomp_bv |= XSTATE_COMPACTION_ENABLED;
+            }
+            else
+            {
+                ptr->xsave_hdr.xstate_bv &= this_cpu(xcr0);
+                ptr->xsave_hdr.xcomp_bv = 0;
+            }
+            memset(ptr->xsave_hdr.reserved, 0, sizeof(ptr->xsave_hdr.reserved));
+            continue;
+        case 2:
+            /* Stage 2: Reset all state. */
+            ptr->fpu_sse.mxcsr = MXCSR_DEFAULT;
+            ptr->xsave_hdr.xstate_bv = 0;
+            ptr->xsave_hdr.xcomp_bv = cpu_has_xsaves
+                                      ? XSTATE_COMPACTION_ENABLED : 0;
+            continue;
+        default:
+            domain_crash(current->domain);
+            break;
+        }
     }
 }
 
@@ -496,6 +549,8 @@  void xstate_init(struct cpuinfo_x86 *c)
 
     if ( bsp )
     {
+        static typeof(current->arch.xsave_area->fpu_sse) __initdata ctxt;
+
         xfeature_mask = feature_mask;
         /*
          * xsave_cntxt_size is the max size required by enabled features.
@@ -504,6 +559,10 @@  void xstate_init(struct cpuinfo_x86 *c)
         xsave_cntxt_size = _xstate_ctxt_size(feature_mask);
         printk("%s: using cntxt_size: %#x and states: %#"PRIx64"\n",
             __func__, xsave_cntxt_size, xfeature_mask);
+
+        asm ( "fxsave %0" : "=m" (ctxt) );
+        if ( ctxt.mxcsr_mask )
+            mxcsr_mask = ctxt.mxcsr_mask;
     }
     else
     {