This patch was meant to address that checking cpu_has_xsave is not enough.
Since it only checks the availability of the feature but it does not check
whether memory has allocated properly or not. It is possible that xsave can be
used without memory being properly allocated and results in clobbering of
memory. We have already encountered two random boot failures caused by xsave
patch in the past due to this so we want to put some safeguard to ensure this
will not happen again.
Maybe the proper thing to do is to have a new function call xsave_enabled(),
this function then checks for whether memory has allocated properly in addition
to checking cpu_has_xsave.
What do you think or do you have a better suggestion?
Allen
-----Original Message-----
From: xen-devel-bounces@xxxxxxxxxxxxxxxxxxx
[mailto:xen-devel-bounces@xxxxxxxxxxxxxxxxxxx] On Behalf Of Keir Fraser
Sent: Thursday, February 17, 2011 11:13 PM
To: Wei, Gang; xen-devel@xxxxxxxxxxxxxxxxxxx
Cc: wei.huang2@xxxxxxx
Subject: [Xen-devel] Re: [PATCH] x86: add strictly sanity check for XSAVE/XRSTOR
On 18/02/2011 02:45, "Wei, Gang" <gang.wei@xxxxxxxxx> wrote:
> This patch is trying to make issues around XSAVE/XRSTOR induced in future easy
> to be exposed.
The fact that xsave_alloc_save_area() is called unconditionally on the vcpu
allocation path suffices I think. It's pretty easy to eyeball that no
successfully initialised non-idle vcpu can have an xsave area smaller than
min_size.
I like assertions a lot, but not carpet bombed all over the code.
-- Keir
> Jimmy
>
> x86: add strictly sanity check for XSAVE/XRSTOR
>
> Signed-off-by: Wei Gang <gang.wei@xxxxxxxxx>
>
> diff -r 137ad3347504 xen/arch/x86/domctl.c
> --- a/xen/arch/x86/domctl.c Mon Feb 14 17:02:55 2011 +0000
> +++ b/xen/arch/x86/domctl.c Fri Feb 18 16:00:41 2011 +0800
> @@ -1604,8 +1604,13 @@ void arch_get_info_guest(struct vcpu *v,
>
> /* Fill legacy context from xsave area first */
> if ( cpu_has_xsave )
> + {
> + ASSERT(xsave_cntxt_size >= XSAVE_AREA_MIN_SIZE);
> + ASSERT(v->arch.xsave_area);
> +
> memcpy(v->arch.xsave_area, &v->arch.guest_context.fpu_ctxt,
> sizeof(v->arch.guest_context.fpu_ctxt));
> + }
>
> if ( !is_pv_32on64_domain(v->domain) )
> memcpy(c.nat, &v->arch.guest_context, sizeof(*c.nat)); diff -r
> 137ad3347504 xen/arch/x86/hvm/hvm.c
> --- a/xen/arch/x86/hvm/hvm.c Mon Feb 14 17:02:55 2011 +0000
> +++ b/xen/arch/x86/hvm/hvm.c Fri Feb 18 16:03:23 2011 +0800
> @@ -777,6 +777,9 @@ static int hvm_load_cpu_ctxt(struct doma
> {
> struct xsave_struct *xsave_area = v->arch.xsave_area;
>
> + ASSERT(xsave_cntxt_size >= XSAVE_AREA_MIN_SIZE);
> + ASSERT(v->arch.xsave_area);
> +
> memcpy(v->arch.xsave_area, ctxt.fpu_regs, sizeof(ctxt.fpu_regs));
> xsave_area->xsave_hdr.xstate_bv = XSTATE_FP_SSE;
> v->arch.xcr0_accum = XSTATE_FP_SSE; @@ -834,6 +837,7 @@ static int
> hvm_save_cpu_xsave_states(str
> if ( !cpu_has_xsave )
> return 0; /* do nothing */
>
> + ASSERT(xsave_cntxt_size >= XSAVE_AREA_MIN_SIZE);
> for_each_vcpu ( d, v )
> {
> if ( _hvm_init_entry(h, CPU_XSAVE_CODE, v->vcpu_id,
> HVM_CPU_XSAVE_SIZE) ) @@ -846,8 +850,12 @@ static int
> hvm_save_cpu_xsave_states(str
> ctxt->xcr0 = v->arch.xcr0;
> ctxt->xcr0_accum = v->arch.xcr0_accum;
> if ( v->fpu_initialised )
> + {
> + ASSERT(v->arch.xsave_area);
> +
> memcpy(&ctxt->save_area,
> v->arch.xsave_area, xsave_cntxt_size);
> + }
> }
>
> return 0;
> @@ -873,6 +881,9 @@ static int hvm_load_cpu_xsave_states(str
> gdprintk(XENLOG_ERR, "HVM restore: domain has no vcpu %u\n", vcpuid);
> return -EINVAL;
> }
> +
> + ASSERT(xsave_cntxt_size >= XSAVE_AREA_MIN_SIZE);
> + ASSERT(v->arch.xsave_area);
>
> /* Customized checking for entry since our entry is of variable length */
> desc = (struct hvm_save_descriptor *)&h->data[h->cur]; diff -r
> 137ad3347504 xen/arch/x86/i387.c
> --- a/xen/arch/x86/i387.c Mon Feb 14 17:02:55 2011 +0000
> +++ b/xen/arch/x86/i387.c Fri Feb 18 16:00:41 2011 +0800
> @@ -71,6 +71,9 @@ void setup_fpu(struct vcpu *v)
>
> if ( cpu_has_xsave )
> {
> + ASSERT(xsave_cntxt_size >= XSAVE_AREA_MIN_SIZE);
> + ASSERT(v->arch.xsave_area);
> +
> /*
> * XCR0 normally represents what guest OS set. In case of Xen itself,
> * we set all supported feature mask before doing save/restore.
> @@ -118,6 +121,9 @@ void save_init_fpu(struct vcpu *v)
>
> if ( cpu_has_xsave )
> {
> + ASSERT(xsave_cntxt_size >= XSAVE_AREA_MIN_SIZE);
> + ASSERT(v->arch.xsave_area);
> +
> /* XCR0 normally represents what guest OS set. In case of Xen itself,
> * we set all accumulated feature mask before doing save/restore.
> */
_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel
_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel
|
Home