[PATCH v4 0/2] arm64: Support FEAT_LSFE (Large System Float Extension)

[PATCH v4 0/2] arm64: Support FEAT_LSFE (Large System Float Extension)

[Mark Brown]

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>

[PATCH v4 1/2] KVM: arm64: Expose FEAT_LSFE to guests

[Mark Brown]

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),

-- 
2.47.2

[PATCH v4 2/2] kselftest/arm64: Add lsfe to the hwcaps test

[Mark Brown]

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,

-- 
2.47.2

Re: [PATCH v4 1/2] KVM: arm64: Expose FEAT_LSFE to guests

[Oliver Upton]

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

Re: [PATCH v4 1/2] KVM: arm64: Expose FEAT_LSFE to guests

[Mark Brown]

[-- Attachment #1: Type: text/plain, Size: 235 bytes --]

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.

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 488 bytes --]

Re: (subset) [PATCH v4 1/2] KVM: arm64: Expose FEAT_LSFE to guests

[Marc Zyngier]

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.
-- 
Without deviation from the norm, progress is not possible.

Re: [PATCH v4 0/2] arm64: Support FEAT_LSFE (Large System Float Extension)

[Will Deacon]

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,
-- 
Will

https://fixes.arm64.dev
https://next.arm64.dev
https://will.arm64.dev