linux-kernel - [PATCH v4 2/4] arm64/fpsimd: Discover maximum vector length implemented by any CPU

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-Id: <20240704-kvm-arm64-fix-pkvm-sve-vl-v4-2-b6898ab23dc4@kernel.org>
Date: Thu, 04 Jul 2024 18:28:17 +0100
From: Mark Brown <broonie@...nel.org>
To: Catalin Marinas <catalin.marinas@....com>, 
 Will Deacon <will@...nel.org>, Marc Zyngier <maz@...nel.org>, 
 Oliver Upton <oliver.upton@...ux.dev>, James Morse <james.morse@....com>, 
 Suzuki K Poulose <suzuki.poulose@....com>, Fuad Tabba <tabba@...gle.com>
Cc: linux-arm-kernel@...ts.infradead.org, linux-kernel@...r.kernel.org, 
 kvmarm@...ts.linux.dev, Mark Brown <broonie@...nel.org>
Subject: [PATCH v4 2/4] arm64/fpsimd: Discover maximum vector length
 implemented by any CPU

When discovering the vector lengths for SVE and SME we do not currently
record the maximum VL supported on any individual CPU.  This is expected
to be the same for all CPUs but the architecture allows asymmetry, if we
do encounter an asymmetric system then some CPUs may support VLs higher
than the maximum Linux will use.  Since the pKVM hypervisor needs to
support saving and restoring anything the host can physically set it
needs to know the maximum value any CPU could have, add support for
enumerating it and validation for late CPUs.

Acked-by: Catalin Marinas <catalin.marinas@....com>
Signed-off-by: Mark Brown <broonie@...nel.org>
---
 arch/arm64/include/asm/fpsimd.h | 13 +++++++++++++
 arch/arm64/kernel/fpsimd.c      | 26 +++++++++++++++++++++++++-
 2 files changed, 38 insertions(+), 1 deletion(-)

diff --git a/arch/arm64/include/asm/fpsimd.h b/arch/arm64/include/asm/fpsimd.h
index 51c21265b4fa..cd19713c9deb 100644
--- a/arch/arm64/include/asm/fpsimd.h
+++ b/arch/arm64/include/asm/fpsimd.h
@@ -188,6 +188,9 @@ struct vl_info {
 	int max_vl;
 	int max_virtualisable_vl;
 
+	/* Maximum vector length observed on any CPU */
+	int max_cpu_vl;
+
 	/*
 	 * Set of available vector lengths,
 	 * where length vq encoded as bit __vq_to_bit(vq):
@@ -278,6 +281,11 @@ static inline int vec_max_virtualisable_vl(enum vec_type type)
 	return vl_info[type].max_virtualisable_vl;
 }
 
+static inline int vec_max_cpu_vl(enum vec_type type)
+{
+	return vl_info[type].max_cpu_vl;
+}
+
 static inline int sve_max_vl(void)
 {
 	return vec_max_vl(ARM64_VEC_SVE);
@@ -288,6 +296,11 @@ static inline int sve_max_virtualisable_vl(void)
 	return vec_max_virtualisable_vl(ARM64_VEC_SVE);
 }
 
+static inline int sve_max_cpu_vl(void)
+{
+	return vec_max_cpu_vl(ARM64_VEC_SVE);
+}
+
 /* Ensure vq >= SVE_VQ_MIN && vq <= SVE_VQ_MAX before calling this function */
 static inline bool vq_available(enum vec_type type, unsigned int vq)
 {
diff --git a/arch/arm64/kernel/fpsimd.c b/arch/arm64/kernel/fpsimd.c
index 22542fb81812..83984cb3f21a 100644
--- a/arch/arm64/kernel/fpsimd.c
+++ b/arch/arm64/kernel/fpsimd.c
@@ -129,6 +129,7 @@ __ro_after_init struct vl_info vl_info[ARM64_VEC_MAX] = {
 		.min_vl			= SVE_VL_MIN,
 		.max_vl			= SVE_VL_MIN,
 		.max_virtualisable_vl	= SVE_VL_MIN,
+		.max_cpu_vl		= SVE_VL_MIN,
 	},
 #endif
 #ifdef CONFIG_ARM64_SME
@@ -1041,8 +1042,13 @@ static void vec_probe_vqs(struct vl_info *info,
 void __init vec_init_vq_map(enum vec_type type)
 {
 	struct vl_info *info = &vl_info[type];
+	unsigned long b;
+
 	vec_probe_vqs(info, info->vq_map);
 	bitmap_copy(info->vq_partial_map, info->vq_map, SVE_VQ_MAX);
+
+	b = find_first_bit(info->vq_map, SVE_VQ_MAX);
+	info->max_cpu_vl = __bit_to_vl(b);
 }
 
 /*
@@ -1054,11 +1060,16 @@ void vec_update_vq_map(enum vec_type type)
 {
 	struct vl_info *info = &vl_info[type];
 	DECLARE_BITMAP(tmp_map, SVE_VQ_MAX);
+	unsigned long b;
 
 	vec_probe_vqs(info, tmp_map);
 	bitmap_and(info->vq_map, info->vq_map, tmp_map, SVE_VQ_MAX);
 	bitmap_or(info->vq_partial_map, info->vq_partial_map, tmp_map,
 		  SVE_VQ_MAX);
+
+	b = find_first_bit(tmp_map, SVE_VQ_MAX);
+	if (__bit_to_vl(b) > info->max_cpu_vl)
+		info->max_cpu_vl = __bit_to_vl(b);
 }
 
 /*
@@ -1069,10 +1080,23 @@ int vec_verify_vq_map(enum vec_type type)
 {
 	struct vl_info *info = &vl_info[type];
 	DECLARE_BITMAP(tmp_map, SVE_VQ_MAX);
-	unsigned long b;
+	unsigned long b, max_vl;
 
 	vec_probe_vqs(info, tmp_map);
 
+	/*
+	 * Currently the maximum VL is only used for pKVM which
+	 * doesn't allow late CPUs but we don't expect asymmetry and
+	 * if we encounter any then future users will need handling so
+	 * warn if we see anything.
+	 */
+	max_vl = __bit_to_vl(find_first_bit(tmp_map, SVE_VQ_MAX));
+	if (max_vl > info->max_cpu_vl) {
+		pr_warn("%s: cpu%d: increases maximum VL to %lu\n",
+			info->name, smp_processor_id(), max_vl);
+		info->max_cpu_vl = max_vl;
+	}
+
 	bitmap_complement(tmp_map, tmp_map, SVE_VQ_MAX);
 	if (bitmap_intersects(tmp_map, info->vq_map, SVE_VQ_MAX)) {
 		pr_warn("%s: cpu%d: Required vector length(s) missing\n",

-- 
2.39.2