These being controlled by XCR0, enabling support is relatively
straightforward. Note however that there won't be any use of them until
their dependent ISA extension CPUID flags are exposed, not the least due
to recalculate_xstate() handling the dependencies in kind of a reverse
manner.
Signed-off-by: Jan Beulich <jbeulich@suse.com>
---
Intended to replace part of (and go beyond) "x86/cpufeatures: Add new
cpuid features in SPR to featureset".
---
v4: Add AMX-FP16 and AMX-COMPLEX. Add dependency on LM. Re-base. Split
off of AMX series.
v3: Add new states to XSTATE_NONLAZY.
v2: New.
--- a/xen/arch/x86/cpu-policy.c
+++ b/xen/arch/x86/cpu-policy.c
@@ -216,6 +216,9 @@ static void recalculate_xstate(struct cp
if ( p->feat.pku )
xstates |= X86_XCR0_PKRU;
+ if ( p->feat.amx_tile )
+ xstates |= X86_XCR0_TILE_CFG | X86_XCR0_TILE_DATA;
+
/* Subleaf 0 */
p->xstate.max_size =
xstate_uncompressed_size(xstates & ~XSTATE_XSAVES_ONLY);
--- a/xen/arch/x86/include/asm/xstate.h
+++ b/xen/arch/x86/include/asm/xstate.h
@@ -35,7 +35,8 @@ extern uint32_t mxcsr_mask;
XSTATE_NONLAZY)
#define XSTATE_ALL (~(1ULL << 63))
-#define XSTATE_NONLAZY (X86_XCR0_BNDREGS | X86_XCR0_BNDCSR | X86_XCR0_PKRU)
+#define XSTATE_NONLAZY (X86_XCR0_BNDREGS | X86_XCR0_BNDCSR | X86_XCR0_PKRU | \
+ X86_XCR0_TILE_CFG | X86_XCR0_TILE_DATA)
#define XSTATE_LAZY (XSTATE_ALL & ~XSTATE_NONLAZY)
#define XSTATE_XSAVES_ONLY 0
#define XSTATE_COMPACTION_ENABLED (1ULL << 63)
--- a/xen/include/public/arch-x86/cpufeatureset.h
+++ b/xen/include/public/arch-x86/cpufeatureset.h
@@ -278,8 +278,10 @@ XEN_CPUFEATURE(HYBRID, 9*32+15) /
XEN_CPUFEATURE(TSXLDTRK, 9*32+16) /*a TSX load tracking suspend/resume insns */
XEN_CPUFEATURE(ARCH_LBR, 9*32+19) /* Architectural Last Branch Record */
XEN_CPUFEATURE(CET_IBT, 9*32+20) /* CET - Indirect Branch Tracking */
+XEN_CPUFEATURE(AMX_BF16, 9*32+22) /* AMX BFloat16 instruction */
XEN_CPUFEATURE(AVX512_FP16, 9*32+23) /*A AVX512 FP16 instructions */
XEN_CPUFEATURE(AMX_TILE, 9*32+24) /* AMX Tile architecture */
+XEN_CPUFEATURE(AMX_INT8, 9*32+25) /* AMX 8-bit integer instructions */
XEN_CPUFEATURE(IBRSB, 9*32+26) /*A IBRS and IBPB support (used by Intel) */
XEN_CPUFEATURE(STIBP, 9*32+27) /*A STIBP */
XEN_CPUFEATURE(L1D_FLUSH, 9*32+28) /*S MSR_FLUSH_CMD and L1D flush. */
@@ -297,6 +299,7 @@ XEN_CPUFEATURE(FZRM, 10*32+10) /
XEN_CPUFEATURE(FSRS, 10*32+11) /*A Fast Short REP STOSB */
XEN_CPUFEATURE(FSRCS, 10*32+12) /*A Fast Short REP CMPSB/SCASB */
XEN_CPUFEATURE(WRMSRNS, 10*32+19) /*S WRMSR Non-Serialising */
+XEN_CPUFEATURE(AMX_FP16, 10*32+21) /* AMX FP16 instruction */
XEN_CPUFEATURE(AVX_IFMA, 10*32+23) /*A AVX-IFMA Instructions */
/* AMD-defined CPU features, CPUID level 0x80000021.eax, word 11 */
@@ -331,6 +334,7 @@ XEN_CPUFEATURE(MCDT_NO, 13*32
/* Intel-defined CPU features, CPUID level 0x00000007:1.edx, word 15 */
XEN_CPUFEATURE(AVX_VNNI_INT8, 15*32+ 4) /*A AVX-VNNI-INT8 Instructions */
XEN_CPUFEATURE(AVX_NE_CONVERT, 15*32+ 5) /*A AVX-NE-CONVERT Instructions */
+XEN_CPUFEATURE(AMX_COMPLEX, 15*32+ 8) /* AMX Complex Instructions */
XEN_CPUFEATURE(AVX_VNNI_INT16, 15*32+10) /*A AVX-VNNI-INT16 Instructions */
XEN_CPUFEATURE(PREFETCHI, 15*32+14) /*A PREFETCHIT{0,1} Instructions */
XEN_CPUFEATURE(CET_SSS, 15*32+18) /* CET Supervisor Shadow Stacks safe to use */
--- a/xen/tools/gen-cpuid.py
+++ b/xen/tools/gen-cpuid.py
@@ -252,7 +252,7 @@ def crunch_numbers(state):
# instruction groups which are specified to require XSAVE for state
# management.
XSAVE: [XSAVEOPT, XSAVEC, XGETBV1, XSAVES,
- AVX, MPX, PKU, LWP],
+ AVX, MPX, PKU, AMX_TILE, LWP],
# AVX is taken to mean hardware support for 256bit registers (which in
# practice depends on the VEX prefix to encode), and the instructions
@@ -274,7 +274,7 @@ def crunch_numbers(state):
# superpages, PCID and PKU are only available in 4 level paging.
# NO_LMSL indicates the absense of Long Mode Segment Limits, which
# have been dropped in hardware.
- LM: [CX16, PCID, LAHF_LM, PAGE1GB, PKU, NO_LMSL],
+ LM: [CX16, PCID, LAHF_LM, PAGE1GB, PKU, NO_LMSL, AMX_TILE],
# AMD K6-2+ and K6-III processors shipped with 3DNow+, beyond the
# standard 3DNow in the earlier K6 processors.
@@ -338,6 +338,11 @@ def crunch_numbers(state):
# The behaviour described by RRSBA depend on eIBRS being active.
EIBRS: [RRSBA],
+
+ # AMX-TILE means hardware support for tile registers and general non-
+ # computational instructions. All further AMX features are built on top
+ # of AMX-TILE.
+ AMX_TILE: [AMX_BF16, AMX_INT8, AMX_FP16, AMX_COMPLEX],
}
deep_features = tuple(sorted(deps.keys()))On 05/09/2024 12:36 pm, Jan Beulich wrote: > These being controlled by XCR0, enabling support is relatively > straightforward. Note however that there won't be any use of them until > their dependent ISA extension CPUID flags are exposed, not the least due > to recalculate_xstate() handling the dependencies in kind of a reverse > manner. > > Signed-off-by: Jan Beulich <jbeulich@suse.com> Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com> It might be worth noting that xstate_check_sizes() already probes AMX_TILE? For future xstates, I'd expect matching updates to recalculate_xstate() and xstate_check_sizes(). > --- > Intended to replace part of (and go beyond) "x86/cpufeatures: Add new > cpuid features in SPR to featureset". Matt is OoO for a while. Given this interaction was the only concern, I'll probably rebase the feature patch and commit it.
On 05.09.2024 14:21, Andrew Cooper wrote: > On 05/09/2024 12:36 pm, Jan Beulich wrote: >> These being controlled by XCR0, enabling support is relatively >> straightforward. Note however that there won't be any use of them until >> their dependent ISA extension CPUID flags are exposed, not the least due >> to recalculate_xstate() handling the dependencies in kind of a reverse >> manner. >> >> Signed-off-by: Jan Beulich <jbeulich@suse.com> > > Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com> Thanks. > It might be worth noting that xstate_check_sizes() already probes AMX_TILE? I've added a sentence. > For future xstates, I'd expect matching updates to recalculate_xstate() > and xstate_check_sizes(). Indeed that's how the somewhat similar APX patch has it that's yet to be posted. >> --- >> Intended to replace part of (and go beyond) "x86/cpufeatures: Add new >> cpuid features in SPR to featureset". > > Matt is OoO for a while. Given this interaction was the only concern, > I'll probably rebase the feature patch and commit it. Fine with me. Jan
© 2016 - 2024 Red Hat, Inc.