On 02/01/2021 06:43 PM, Thomas Bogendoerfer wrote:

On Tue, Dec 29, 2020 at 08:55:59PM +0800, Tiezhu Yang wrote:

quoted

+++ b/arch/mips/include/uapi/asm/perf_regs.h

@@ -0,0 +1,42 @@
+/* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note */
+#ifndef _ASM_MIPS_PERF_REGS_H
+#define _ASM_MIPS_PERF_REGS_H
+
+enum perf_event_mips_regs {
+	PERF_REG_MIPS_PC,
+	PERF_REG_MIPS_R1,
+	PERF_REG_MIPS_R2,
+	PERF_REG_MIPS_R3,
+	PERF_REG_MIPS_R4,
+	PERF_REG_MIPS_R5,
+	PERF_REG_MIPS_R6,
+	PERF_REG_MIPS_R7,
+	PERF_REG_MIPS_R8,
+	PERF_REG_MIPS_R9,
+	PERF_REG_MIPS_R10,
+	PERF_REG_MIPS_R11,
+	PERF_REG_MIPS_R12,
+	PERF_REG_MIPS_R13,
+	PERF_REG_MIPS_R14,
+	PERF_REG_MIPS_R15,
+	PERF_REG_MIPS_R16,
+	PERF_REG_MIPS_R17,
+	PERF_REG_MIPS_R18,
+	PERF_REG_MIPS_R19,
+	PERF_REG_MIPS_R20,
+	PERF_REG_MIPS_R21,
+	PERF_REG_MIPS_R22,
+	PERF_REG_MIPS_R23,
+	PERF_REG_MIPS_R24,
+	PERF_REG_MIPS_R25,
+	/*
+	 * 26 and 27 are k0 and k1, they are always clobbered thus not
+	 * stored.
+	 */

haveing this hole here make all code more complicated. Does it hurt
to have R26 and R27 in the list ?

I think there is no effect if have R26 and R27 in the list.

In the perf_reg_value(), PERF_REG_MIPS_R{26,27} are default case.

+u64 perf_reg_value(struct pt_regs *regs, int idx)
+{
+ long v;
+
+ switch (idx) {
+ case PERF_REG_MIPS_PC:
+ v = regs->cp0_epc;
+ break;
+ case PERF_REG_MIPS_R1 ... PERF_REG_MIPS_R25:
+ v = regs->regs[idx - PERF_REG_MIPS_R1 + 1];
+ break;
+ case PERF_REG_MIPS_R28 ... PERF_REG_MIPS_R31:
+ v = regs->regs[idx - PERF_REG_MIPS_R28 + 28];
+ break;
+
+ default:
+ WARN_ON_ONCE(1);
+ return 0;
+ }
+
+ return (s64)v; /* Sign extend if 32-bit. */
+} Should I modify enum perf_event_mips_regs to add R26 and R27,
and then send v2?

Thanks,
Tiezhu

Thomas.