FEAT_LSFE is optional from v9.5, it adds new instructions for atomic memory operations with floating point values. We have no immediate use for it in kernel, provide a hwcap so userspace can discover it and allow the ID register field to be exposed to KVM guests.
Signed-off-by: Mark Brown broonie@kernel.org --- Changes in v4: - Rebase onto arm64/for-next/cpufeature, note that both patches have build dependencies on this. - Drop unneeded cc clobber in hwcap. - Use STRFADD as the instruction probed in hwcap. - Link to v3: https://lore.kernel.org/r/20250818-arm64-lsfe-v3-0-af6f4d66eb39@kernel.org
Changes in v3: - Rebase onto v6.17-rc1. - Link to v2: https://lore.kernel.org/r/20250703-arm64-lsfe-v2-0-eced80999cb4@kernel.org
Changes in v2: - Fix result of vi dropping in hwcap test. - Link to v1: https://lore.kernel.org/r/20250627-arm64-lsfe-v1-0-68351c4bf741@kernel.org
--- Mark Brown (2): KVM: arm64: Expose FEAT_LSFE to guests kselftest/arm64: Add lsfe to the hwcaps test
arch/arm64/kvm/sys_regs.c | 4 +++- tools/testing/selftests/arm64/abi/hwcap.c | 21 +++++++++++++++++++++ 2 files changed, 24 insertions(+), 1 deletion(-) --- base-commit: 220928e52cb03d223b3acad3888baf0687486d21 change-id: 20250625-arm64-lsfe-0810cf98adc2
Best regards, -- Mark Brown broonie@kernel.org
FEAT_LSFE (Large System Float Extension), providing atomic floating point memory operations, is optional from v9.5. This feature adds no new architectural state, expose the relevant ID register field to guests so they can discover it.
Signed-off-by: Mark Brown broonie@kernel.org --- arch/arm64/kvm/sys_regs.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)
diff --git a/arch/arm64/kvm/sys_regs.c b/arch/arm64/kvm/sys_regs.c index 82ffb3b3b3cf..08e2cb94bff3 100644 --- a/arch/arm64/kvm/sys_regs.c +++ b/arch/arm64/kvm/sys_regs.c @@ -1642,7 +1642,8 @@ static u64 __kvm_read_sanitised_id_reg(const struct kvm_vcpu *vcpu, val &= ~ARM64_FEATURE_MASK(ID_AA64ISAR2_EL1_WFxT); break; case SYS_ID_AA64ISAR3_EL1: - val &= ID_AA64ISAR3_EL1_FPRCVT | ID_AA64ISAR3_EL1_FAMINMAX; + val &= ID_AA64ISAR3_EL1_FPRCVT | ID_AA64ISAR3_EL1_LSFE | + ID_AA64ISAR3_EL1_FAMINMAX; break; case SYS_ID_AA64MMFR2_EL1: val &= ~ID_AA64MMFR2_EL1_CCIDX_MASK; @@ -2991,6 +2992,7 @@ static const struct sys_reg_desc sys_reg_descs[] = { ID_AA64ISAR2_EL1_APA3 | ID_AA64ISAR2_EL1_GPA3)), ID_WRITABLE(ID_AA64ISAR3_EL1, (ID_AA64ISAR3_EL1_FPRCVT | + ID_AA64ISAR3_EL1_LSFE | ID_AA64ISAR3_EL1_FAMINMAX)), ID_UNALLOCATED(6,4), ID_UNALLOCATED(6,5),
On Thu, Sep 18, 2025 at 08:42:06PM +0100, Mark Brown wrote:
FEAT_LSFE (Large System Float Extension), providing atomic floating point memory operations, is optional from v9.5. This feature adds no new architectural state, expose the relevant ID register field to guests so they can discover it.
Signed-off-by: Mark Brown broonie@kernel.org
We also need a test in set_id_regs.c selftest for the writability of this new feature field.
Otherwise:
Reviewed-by: Oliver Upton oliver.upton@linux.dev
Thanks, Oliver
On Thu, Sep 18, 2025 at 01:57:13PM -0700, Oliver Upton wrote:
We also need a test in set_id_regs.c selftest for the writability of this new feature field.
Sure. We seem to be missing some existing fields from that register too.
On Thu, 18 Sep 2025 20:42:06 +0100, Mark Brown wrote:
FEAT_LSFE (Large System Float Extension), providing atomic floating point memory operations, is optional from v9.5. This feature adds no new architectural state, expose the relevant ID register field to guests so they can discover it.
Applied to next, thanks!
[1/2] KVM: arm64: Expose FEAT_LSFE to guests commit: 5d20605c8e7930254f7bee41336e421be247181c
Cheers,
M.
This feature has no traps associated with it so the SIGILL is not reliable.
Signed-off-by: Mark Brown broonie@kernel.org --- tools/testing/selftests/arm64/abi/hwcap.c | 21 +++++++++++++++++++++ 1 file changed, 21 insertions(+)
diff --git a/tools/testing/selftests/arm64/abi/hwcap.c b/tools/testing/selftests/arm64/abi/hwcap.c index 002ec38a8bbb..0f9a64932481 100644 --- a/tools/testing/selftests/arm64/abi/hwcap.c +++ b/tools/testing/selftests/arm64/abi/hwcap.c @@ -17,6 +17,8 @@ #include <asm/sigcontext.h> #include <asm/unistd.h>
+#include <linux/auxvec.h> + #include "../../kselftest.h"
#define TESTS_PER_HWCAP 3 @@ -169,6 +171,18 @@ static void lse128_sigill(void) : "cc", "memory"); }
+static void lsfe_sigill(void) +{ + float __attribute__ ((aligned (16))) mem; + register float *memp asm ("x0") = &mem; + + /* STFADD H0, [X0] */ + asm volatile(".inst 0x7c20801f" + : "+r" (memp) + : + : "memory"); +} + static void lut_sigill(void) { /* LUTI2 V0.16B, { V0.16B }, V[0] */ @@ -762,6 +776,13 @@ static const struct hwcap_data { .cpuinfo = "lse128", .sigill_fn = lse128_sigill, }, + { + .name = "LSFE", + .at_hwcap = AT_HWCAP3, + .hwcap_bit = HWCAP3_LSFE, + .cpuinfo = "lsfe", + .sigill_fn = lsfe_sigill, + }, { .name = "LUT", .at_hwcap = AT_HWCAP2,
On Thu, 18 Sep 2025 20:42:05 +0100, Mark Brown wrote:
FEAT_LSFE is optional from v9.5, it adds new instructions for atomic memory operations with floating point values. We have no immediate use for it in kernel, provide a hwcap so userspace can discover it and allow the ID register field to be exposed to KVM guests.
Applied selftest update to arm64 (for-next/selftests), thanks!
[2/2] kselftest/arm64: Add lsfe to the hwcaps test https://git.kernel.org/arm64/c/777fb19ed8d6
Cheers,
linux-kselftest-mirror@lists.linaro.org
-
Marc Zyngier -
Mark Brown -
Oliver Upton -
Will Deacon