1 files changed, 53 insertions, 8 deletions

diff --git a/arch/arm64/kernel/perf_regs.c b/arch/arm64/kernel/perf_regs.c
index bd1b74c2436f..b4eece3eb17d 100644
--- a/arch/arm64/kernel/perf_regs.c
+++ b/arch/arm64/kernel/perf_regs.c
@@ -1,28 +1,66 @@
+// SPDX-License-Identifier: GPL-2.0
+#include <linux/compat.h>
 #include <linux/errno.h>
 #include <linux/kernel.h>
 #include <linux/perf_event.h>
 #include <linux/bug.h>
 #include <linux/sched/task_stack.h>
 
-#include <asm/compat.h>
 #include <asm/perf_regs.h>
 #include <asm/ptrace.h>
 
+static u64 perf_ext_regs_value(int idx)
+{
+	switch (idx) {
+	case PERF_REG_ARM64_VG:
+		if (WARN_ON_ONCE(!system_supports_sve()))
+			return 0;
+
+		/*
+		 * Vector granule is current length in bits of SVE registers
+		 * divided by 64.
+		 */
+		return (task_get_sve_vl(current) * 8) / 64;
+	default:
+		WARN_ON_ONCE(true);
+		return 0;
+	}
+}
+
 u64 perf_reg_value(struct pt_regs *regs, int idx)
 {
-	if (WARN_ON_ONCE((u32)idx >= PERF_REG_ARM64_MAX))
+	if (WARN_ON_ONCE((u32)idx >= PERF_REG_ARM64_EXTENDED_MAX))
 		return 0;
 
 	/*
-	 * Compat (i.e. 32 bit) mode:
-	 * - PC has been set in the pt_regs struct in kernel_entry,
-	 * - Handle SP and LR here.
+	 * Our handling of compat tasks (PERF_SAMPLE_REGS_ABI_32) is weird, but
+	 * we're stuck with it for ABI compatibility reasons.
+	 *
+	 * For a 32-bit consumer inspecting a 32-bit task, then it will look at
+	 * the first 16 registers (see arch/arm/include/uapi/asm/perf_regs.h).
+	 * These correspond directly to a prefix of the registers saved in our
+	 * 'struct pt_regs', with the exception of the PC, so we copy that down
+	 * (x15 corresponds to SP_hyp in the architecture).
+	 *
+	 * So far, so good.
+	 *
+	 * The oddity arises when a 64-bit consumer looks at a 32-bit task and
+	 * asks for registers beyond PERF_REG_ARM_MAX. In this case, we return
+	 * SP_usr, LR_usr and PC in the positions where the AArch64 SP, LR and
+	 * PC registers would normally live. The initial idea was to allow a
+	 * 64-bit unwinder to unwind a 32-bit task and, although it's not clear
+	 * how well that works in practice, somebody might be relying on it.
+	 *
+	 * At the time we make a sample, we don't know whether the consumer is
+	 * 32-bit or 64-bit, so we have to cater for both possibilities.
 	 */
 	if (compat_user_mode(regs)) {
 		if ((u32)idx == PERF_REG_ARM64_SP)
 			return regs->compat_sp;
 		if ((u32)idx == PERF_REG_ARM64_LR)
 			return regs->compat_lr;
+		if (idx == 15)
+			return regs->pc;
 	}
 
 	if ((u32)idx == PERF_REG_ARM64_SP)
@@ -31,6 +69,9 @@ u64 perf_reg_value(struct pt_regs *regs, int idx)
 	if ((u32)idx == PERF_REG_ARM64_PC)
 		return regs->pc;
 
+	if ((u32)idx >= PERF_REG_ARM64_MAX)
+		return perf_ext_regs_value(idx);
+
 	return regs->regs[idx];
 }
 
@@ -38,7 +79,12 @@ u64 perf_reg_value(struct pt_regs *regs, int idx)
 
 int perf_reg_validate(u64 mask)
 {
-	if (!mask || mask & REG_RESERVED)
+	u64 reserved_mask = REG_RESERVED;
+
+	if (system_supports_sve())
+		reserved_mask &= ~(1ULL << PERF_REG_ARM64_VG);
+
+	if (!mask || mask & reserved_mask)
 		return -EINVAL;
 
 	return 0;
@@ -53,8 +99,7 @@ u64 perf_reg_abi(struct task_struct *task)
 }
 
 void perf_get_regs_user(struct perf_regs *regs_user,
-			struct pt_regs *regs,
-			struct pt_regs *regs_user_copy)
+			struct pt_regs *regs)
 {
 	regs_user->regs = task_pt_regs(current);
 	regs_user->abi = perf_reg_abi(current);