On 23.05.2024 13:16, Andrew Cooper wrote:
> Right now, xstate_ctxt_size() performs a cross-check of size with CPUID in for
> every call. This is expensive, being used for domain create/migrate, as well
> as to service certain guest CPUID instructions.
>
> Instead, arrange to check the sizes once at boot. See the code comments for
> details. Right now, it just checks hardware against the algorithm
> expectations. Later patches will add further cross-checking.
>
> Introduce the missing X86_XCR0_* and X86_XSS_* constants, and a couple of
> missing CPUID bits. This is to maximise coverage in the sanity check, even if
> we don't expect to use/virtualise some of these features any time soon. Leave
> HDC and HWP alone for now. We don't have CPUID bits from them stored nicely.
Since you say "the missing", ...
> --- a/xen/arch/x86/include/asm/x86-defns.h
> +++ b/xen/arch/x86/include/asm/x86-defns.h
> @@ -77,7 +77,7 @@
> #define X86_CR4_PKS 0x01000000 /* Protection Key Supervisor */
>
> /*
> - * XSTATE component flags in XCR0
> + * XSTATE component flags in XCR0 | MSR_XSS
> */
> #define X86_XCR0_FP_POS 0
> #define X86_XCR0_FP (1ULL << X86_XCR0_FP_POS)
> @@ -95,11 +95,34 @@
> #define X86_XCR0_ZMM (1ULL << X86_XCR0_ZMM_POS)
> #define X86_XCR0_HI_ZMM_POS 7
> #define X86_XCR0_HI_ZMM (1ULL << X86_XCR0_HI_ZMM_POS)
> +#define X86_XSS_PROC_TRACE (_AC(1, ULL) << 8)
> #define X86_XCR0_PKRU_POS 9
> #define X86_XCR0_PKRU (1ULL << X86_XCR0_PKRU_POS)
> +#define X86_XSS_PASID (_AC(1, ULL) << 10)
> +#define X86_XSS_CET_U (_AC(1, ULL) << 11)
> +#define X86_XSS_CET_S (_AC(1, ULL) << 12)
> +#define X86_XSS_HDC (_AC(1, ULL) << 13)
> +#define X86_XSS_UINTR (_AC(1, ULL) << 14)
> +#define X86_XSS_LBR (_AC(1, ULL) << 15)
> +#define X86_XSS_HWP (_AC(1, ULL) << 16)
> +#define X86_XCR0_TILE_CFG (_AC(1, ULL) << 17)
> +#define X86_XCR0_TILE_DATA (_AC(1, ULL) << 18)
... I'm wondering if you deliberately left out APX (bit 19).
Since you're re-doing some of what I have long had in patches already,
I'd also like to ask whether the last underscores each in the two AMX
names really are useful in your opinion. While rebasing isn't going
to be difficult either way, it would be yet simpler with
X86_XCR0_TILECFG and X86_XCR0_TILEDATA, as I've had it in my patches
for over 3 years.
> --- a/xen/arch/x86/xstate.c
> +++ b/xen/arch/x86/xstate.c
> @@ -604,9 +604,156 @@ static bool valid_xcr0(uint64_t xcr0)
> if ( !(xcr0 & X86_XCR0_BNDREGS) != !(xcr0 & X86_XCR0_BNDCSR) )
> return false;
>
> + /* TILE_CFG and TILE_DATA must be the same. */
> + if ( !(xcr0 & X86_XCR0_TILE_CFG) != !(xcr0 & X86_XCR0_TILE_DATA) )
> + return false;
> +
> return true;
> }
>
> +struct xcheck_state {
> + uint64_t states;
> + uint32_t uncomp_size;
> + uint32_t comp_size;
> +};
> +
> +static void __init check_new_xstate(struct xcheck_state *s, uint64_t new)
> +{
> + uint32_t hw_size;
> +
> + BUILD_BUG_ON(X86_XCR0_STATES & X86_XSS_STATES);
> +
> + BUG_ON(s->states & new); /* States only increase. */
> + BUG_ON(!valid_xcr0(s->states | new)); /* Xen thinks it's a good value. */
> + BUG_ON(new & ~(X86_XCR0_STATES | X86_XSS_STATES)); /* Known state. */
> + BUG_ON((new & X86_XCR0_STATES) &&
> + (new & X86_XSS_STATES)); /* User or supervisor, not both. */
> +
> + s->states |= new;
> + if ( new & X86_XCR0_STATES )
> + {
> + if ( !set_xcr0(s->states & X86_XCR0_STATES) )
> + BUG();
> + }
> + else
> + set_msr_xss(s->states & X86_XSS_STATES);
> +
> + /*
> + * Check the uncompressed size. Some XSTATEs are out-of-order and fill
> in
> + * prior holes in the state area, so we check that the size doesn't
> + * decrease.
> + */
> + hw_size = cpuid_count_ebx(0xd, 0);
> +
> + if ( hw_size < s->uncomp_size )
> + panic("XSTATE 0x%016"PRIx64", new bits {%63pbl}, uncompressed hw
> size %#x < prev size %#x\n",
> + s->states, &new, hw_size, s->uncomp_size);
> +
> + s->uncomp_size = hw_size;
> +
> + /*
> + * Check the compressed size, if available. All components strictly
> + * appear in index order. In principle there are no holes, but some
> + * components have their base address 64-byte aligned for efficiency
> + * reasons (e.g. AMX-TILE) and there are other components small enough to
> + * fit in the gap (e.g. PKRU) without increasing the overall length.
> + */
> + hw_size = cpuid_count_ebx(0xd, 1);
> +
> + if ( cpu_has_xsavec )
> + {
> + if ( hw_size < s->comp_size )
> + panic("XSTATE 0x%016"PRIx64", new bits {%63pbl}, compressed hw
> size %#x < prev size %#x\n",
> + s->states, &new, hw_size, s->comp_size);
> +
> + s->comp_size = hw_size;
> + }
> + else
> + BUG_ON(hw_size); /* Compressed size reported, but no XSAVEC ? */
I'm not quite happy with this being fatal to booting. Maybe just WARN_ON()?
Jan
