Skip to contentSkip to conversation

perf: arm64: Tools support for Dwarf unwinding through SVE functions

PATCH Superseded James Clark 4 participants 17 messages May 17, 2022 – May 20, 2022

Subsystems (2)

v2 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

v2 cover perf: arm64: Tools support for Dwarf unwinding through SVE functions

Changes since v1:

* Split patchset into kernel side and Perf tool changes

When SVE registers are pushed onto the stack the VG register is required to unwind because the stack offsets would vary by the SVE register width at the time when the sample was taken.

The patches ("[PATCH v2 0/2] perf: arm64: Kernel support for Dwarf unwinding through SVE functions") add support for sampling the VG register to the kernel and the docs. This is the patchset to add support to userspace perf.

A small change is also required to libunwind or libdw depending on which unwinder is used, and these will be published later. Without these changes Perf continues to work with both libraries, although the VG register is still not used for unwinding.

Thanks
James

James Clark (4):
perf tools: arm64: Copy perf_regs.h from the kernel perf tools: Use dynamic register set for Dwarf unwind perf tools: arm64: Decouple Libunwind register names from Perf perf tools: arm64: Add support for VG register

 tools/arch/arm64/include/uapi/asm/perf_regs.h |  7 +-
 tools/perf/arch/arm64/util/perf_regs.c        | 34 +++++++++
 tools/perf/arch/arm64/util/unwind-libunwind.c | 73 +------------------
 tools/perf/util/evsel.c                       |  2 +-
 tools/perf/util/perf_regs.c                   |  2 +
 5 files changed, 45 insertions(+), 73 deletions(-)
Signature · 7 lines
2.28.0


_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
#1 lore raw

PATCH v2 1/4 perf tools: arm64: Copy perf_regs.h from the kernel

Get the updated header for the newly added VG register.

---
 tools/arch/arm64/include/uapi/asm/perf_regs.h | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)
tools/arch/arm64/include/uapi/asm/perf_regs.h +6 -1
index d54daafa89e3..fd157f46727e 100644
@@ -36,6 +36,11 @@ enum perf_event_arm_regs {
 	PERF_REG_ARM64_LR,
 	PERF_REG_ARM64_SP,
 	PERF_REG_ARM64_PC,
-	PERF_REG_ARM64_MAX,
+
+	/* Extended/pseudo registers */
+	PERF_REG_ARM64_VG = 46, // SVE Vector Granule
+
+	PERF_REG_ARM64_MAX = PERF_REG_ARM64_PC + 1,
+	PERF_REG_ARM64_EXTENDED_MAX = PERF_REG_ARM64_VG + 1
 };
 #endif /* _ASM_ARM64_PERF_REGS_H */
Signature · 7 lines
2.28.0


_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
Reviewed-by on v2 1/4
#2 lore raw

Re: [PATCH v2 1/4] perf tools: arm64: Copy perf_regs.h from the kernel

Quoted 3 lines · James Clark

Get the updated header for the newly added VG register.

Quoted 24 lines
---
 tools/arch/arm64/include/uapi/asm/perf_regs.h | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)
tools/arch/arm64/include/uapi/asm/perf_regs.h +6 -1
index d54daafa89e3..fd157f46727e 100644
@@ -36,6 +36,11 @@ enum perf_event_arm_regs {
 	PERF_REG_ARM64_LR,
 	PERF_REG_ARM64_SP,
 	PERF_REG_ARM64_PC,
-	PERF_REG_ARM64_MAX,
+
+	/* Extended/pseudo registers */
+	PERF_REG_ARM64_VG = 46, // SVE Vector Granule
+
+	PERF_REG_ARM64_MAX = PERF_REG_ARM64_PC + 1,
+	PERF_REG_ARM64_EXTENDED_MAX = PERF_REG_ARM64_VG + 1
 };
 #endif /* _ASM_ARM64_PERF_REGS_H */
Signature · 1 line
2.28.0
#3 lore raw

PATCH v2 2/4 perf tools: Use dynamic register set for Dwarf unwind

In reply to James Clark

Architectures can detect availability of extra registers at runtime so use this more complete set for unwinding. This will include the VG register on arm64 in a later commit.

If the function isn't implemented then PERF_REGS_MASK is returned and there is no change.

---
 tools/perf/util/evsel.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/perf/util/evsel.c +1 -1
index 5fd7924f8eb3..787bbcbcd2ae 100644
@@ -896,7 +896,7 @@ static void __evsel__config_callchain(struct evsel *evsel, struct record_opts *o
 					   "specifying a subset with --user-regs may render DWARF unwinding unreliable, "
 					   "so the minimal registers set (IP, SP) is explicitly forced.\n");
 			} else {
-				attr->sample_regs_user |= PERF_REGS_MASK;
+				attr->sample_regs_user |= arch__user_reg_mask();
 			}
 			attr->sample_stack_user = param->dump_size;
 			attr->exclude_callchain_user = 1;
Signature · 7 lines
2.28.0


_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
Reviewed-by on v2 2/4
#4 lore raw

Re: [PATCH v2 2/4] perf tools: Use dynamic register set for Dwarf unwind

Quoted 8 lines · James Clark

Architectures can detect availability of extra registers at runtime so use this more complete set for unwinding. This will include the VG register on arm64 in a later commit.

If the function isn't implemented then PERF_REGS_MASK is returned and there is no change.

This patch looks good to me:

Just curious, do you think should update the test (e.g. arch/arm64/tests/dwarf-unwind.c) to use arch__user_reg_mask()?

Thanks,
Leo

Quoted 20 lines
---
 tools/perf/util/evsel.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/perf/util/evsel.c +1 -1
index 5fd7924f8eb3..787bbcbcd2ae 100644
@@ -896,7 +896,7 @@ static void __evsel__config_callchain(struct evsel *evsel, struct record_opts *o
 					   "specifying a subset with --user-regs may render DWARF unwinding unreliable, "
 					   "so the minimal registers set (IP, SP) is explicitly forced.\n");
 			} else {
-				attr->sample_regs_user |= PERF_REGS_MASK;
+				attr->sample_regs_user |= arch__user_reg_mask();
 			}
 			attr->sample_stack_user = param->dump_size;
 			attr->exclude_callchain_user = 1;
Signature · 1 line
2.28.0
#5 lore raw

Re: [PATCH v2 2/4] perf tools: Use dynamic register set for Dwarf unwind

Quoted 8 lines · Leo Yan
Quoted 7 lines · James Clark

Architectures can detect availability of extra registers at runtime so use this more complete set for unwinding. This will include the VG register on arm64 in a later commit.

If the function isn't implemented then PERF_REGS_MASK is returned and there is no change.

Quoted 7 lines
Quoted 1 line

This patch looks good to me:

Just curious, do you think should update the test (e.g. arch/arm64/tests/dwarf-unwind.c) to use arch__user_reg_mask()?

I don't think so because the normal set of registers is manually loaded in tools/perf/arch/arm64/tests/regs_load.S so it wouldn't include this pseudo register. Also there is no SVE in the call chain of the test so it would never have an effect.

I could add a new test for SVE, but it depends on getting the libunwind changes through first so will have to come later.

Thanks,
James

Quoted 24 lines

Thanks,
Leo

Quoted 20 lines
---
 tools/perf/util/evsel.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/perf/util/evsel.c +1 -1
index 5fd7924f8eb3..787bbcbcd2ae 100644
@@ -896,7 +896,7 @@ static void __evsel__config_callchain(struct evsel *evsel, struct record_opts *o
 					   "specifying a subset with --user-regs may render DWARF unwinding unreliable, "
 					   "so the minimal registers set (IP, SP) is explicitly forced.\n");
 			} else {
-				attr->sample_regs_user |= PERF_REGS_MASK;
+				attr->sample_regs_user |= arch__user_reg_mask();
 			}
 			attr->sample_stack_user = param->dump_size;
 			attr->exclude_callchain_user = 1;
Signature · 1 line
2.28.0
#6 lore raw

Re: [PATCH v2 2/4] perf tools: Use dynamic register set for Dwarf unwind

Quoted 19 lines · James Clark
Quoted 8 lines · Leo Yan
Quoted 7 lines · James Clark

Architectures can detect availability of extra registers at runtime so use this more complete set for unwinding. This will include the VG register on arm64 in a later commit.

If the function isn't implemented then PERF_REGS_MASK is returned and there is no change.

Quoted 7 lines
Quoted 1 line

This patch looks good to me:

Just curious, do you think should update the test (e.g. arch/arm64/tests/dwarf-unwind.c) to use arch__user_reg_mask()?

Quoted 7 lines

I don't think so because the normal set of registers is manually loaded in tools/perf/arch/arm64/tests/regs_load.S so it wouldn't include this pseudo register. Also there is no SVE in the call chain of the test so it would never have an effect.

I could add a new test for SVE, but it depends on getting the libunwind changes through first so will have to come later.

It makes sense to add new SVE test later, thanks for clarification.

So this patch series looks good for me.

Leo

#7 lore raw

PATCH v2 3/4 perf tools: arm64: Decouple Libunwind register names from Perf

In reply to James Clark

Dwarf register numbers and real register numbers on aarch64 are equivalent. Remove the references to the register names from Libunwind so that new registers are supported without having to add build time feature checks for each new register.

The unwinder won't ask for a register that it doesn't know about and Perf will already report an error for an unknown or unrecorded register in the perf_reg_value() function so extra validation isn't needed.

After this change the new VG register can be read by libunwind.

---
 tools/perf/arch/arm64/util/unwind-libunwind.c | 73 +------------------
 1 file changed, 2 insertions(+), 71 deletions(-)
tools/perf/arch/arm64/util/unwind-libunwind.c +2 -71
index 5aecf88e3de6..871af5992298 100644
@@ -10,77 +10,8 @@
 
 int LIBUNWIND__ARCH_REG_ID(int regnum)
 {
-	switch (regnum) {
-	case UNW_AARCH64_X0:
-		return PERF_REG_ARM64_X0;
-	case UNW_AARCH64_X1:
-		return PERF_REG_ARM64_X1;
-	case UNW_AARCH64_X2:
-		return PERF_REG_ARM64_X2;
-	case UNW_AARCH64_X3:
-		return PERF_REG_ARM64_X3;
-	case UNW_AARCH64_X4:
-		return PERF_REG_ARM64_X4;
-	case UNW_AARCH64_X5:
-		return PERF_REG_ARM64_X5;
-	case UNW_AARCH64_X6:
-		return PERF_REG_ARM64_X6;
-	case UNW_AARCH64_X7:
-		return PERF_REG_ARM64_X7;
-	case UNW_AARCH64_X8:
-		return PERF_REG_ARM64_X8;
-	case UNW_AARCH64_X9:
-		return PERF_REG_ARM64_X9;
-	case UNW_AARCH64_X10:
-		return PERF_REG_ARM64_X10;
-	case UNW_AARCH64_X11:
-		return PERF_REG_ARM64_X11;
-	case UNW_AARCH64_X12:
-		return PERF_REG_ARM64_X12;
-	case UNW_AARCH64_X13:
-		return PERF_REG_ARM64_X13;
-	case UNW_AARCH64_X14:
-		return PERF_REG_ARM64_X14;
-	case UNW_AARCH64_X15:
-		return PERF_REG_ARM64_X15;
-	case UNW_AARCH64_X16:
-		return PERF_REG_ARM64_X16;
-	case UNW_AARCH64_X17:
-		return PERF_REG_ARM64_X17;
-	case UNW_AARCH64_X18:
-		return PERF_REG_ARM64_X18;
-	case UNW_AARCH64_X19:
-		return PERF_REG_ARM64_X19;
-	case UNW_AARCH64_X20:
-		return PERF_REG_ARM64_X20;
-	case UNW_AARCH64_X21:
-		return PERF_REG_ARM64_X21;
-	case UNW_AARCH64_X22:
-		return PERF_REG_ARM64_X22;
-	case UNW_AARCH64_X23:
-		return PERF_REG_ARM64_X23;
-	case UNW_AARCH64_X24:
-		return PERF_REG_ARM64_X24;
-	case UNW_AARCH64_X25:
-		return PERF_REG_ARM64_X25;
-	case UNW_AARCH64_X26:
-		return PERF_REG_ARM64_X26;
-	case UNW_AARCH64_X27:
-		return PERF_REG_ARM64_X27;
-	case UNW_AARCH64_X28:
-		return PERF_REG_ARM64_X28;
-	case UNW_AARCH64_X29:
-		return PERF_REG_ARM64_X29;
-	case UNW_AARCH64_X30:
-		return PERF_REG_ARM64_LR;
-	case UNW_AARCH64_SP:
-		return PERF_REG_ARM64_SP;
-	case UNW_AARCH64_PC:
-		return PERF_REG_ARM64_PC;
-	default:
-		pr_err("unwind: invalid reg id %d\n", regnum);
+	if (regnum < 0 || regnum >= PERF_REG_ARM64_EXTENDED_MAX)
 		return -EINVAL;
-	}
 
-	return -EINVAL;
+	return regnum;
 }
Signature · 7 lines
2.28.0


_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
Reviewed-by on v2 3/4
#8 lore raw

Re: [PATCH v2 3/4] perf tools: arm64: Decouple Libunwind register names from Perf

Quoted 5 lines · James Clark

Dwarf register numbers and real register numbers on aarch64 are equivalent. Remove the references to the register names from Libunwind so that new registers are supported without having to add build time feature checks for each new register.

Quoted 8 lines

The unwinder won't ask for a register that it doesn't know about and Perf will already report an error for an unknown or unrecorded register in the perf_reg_value() function so extra validation isn't needed.

After this change the new VG register can be read by libunwind.

Looks good to me:

#9 lore raw

PATCH v2 4/4 perf tools: arm64: Add support for VG register

In reply to James Clark

Add the name of the VG register so it can be used in --user-regs

The event will fail to open if the register is requested but not available so only add it to the mask if the kernel supports sve and also if it supports that specific register.

---
 tools/perf/arch/arm64/util/perf_regs.c | 34 ++++++++++++++++++++++++++
 tools/perf/util/perf_regs.c            |  2 ++
 2 files changed, 36 insertions(+)
tools/perf/arch/arm64/util/perf_regs.c +34 -0
index 476b037eea1c..c0a921512a90 100644
@@ -2,9 +2,11 @@
 #include <errno.h>
 #include <regex.h>
 #include <string.h>
+#include <sys/auxv.h>
 #include <linux/kernel.h>
 #include <linux/zalloc.h>
 
+#include "../../../perf-sys.h"
 #include "../../../util/debug.h"
 #include "../../../util/event.h"
 #include "../../../util/perf_regs.h"
@@ -43,6 +45,7 @@ const struct sample_reg sample_reg_masks[] = {
 	SMPL_REG(lr, PERF_REG_ARM64_LR),
 	SMPL_REG(sp, PERF_REG_ARM64_SP),
 	SMPL_REG(pc, PERF_REG_ARM64_PC),
+	SMPL_REG(vg, PERF_REG_ARM64_VG),
 	SMPL_REG_END
 };
 
@@ -131,3 +134,34 @@ int arch_sdt_arg_parse_op(char *old_op, char **new_op)
 
 	return SDT_ARG_VALID;
 }
+
+uint64_t arch__user_reg_mask(void)
+{
+	struct perf_event_attr attr = {
+		.type                   = PERF_TYPE_HARDWARE,
+		.config                 = PERF_COUNT_HW_CPU_CYCLES,
+		.sample_type            = PERF_SAMPLE_REGS_USER,
+		.disabled               = 1,
+		.exclude_kernel         = 1,
+		.sample_period		= 1,
+		.sample_regs_user	= PERF_REGS_MASK
+	};
+	int fd;
+
+	if (getauxval(AT_HWCAP) & HWCAP_SVE)
+		attr.sample_regs_user |= SMPL_REG_MASK(PERF_REG_ARM64_VG);
+
+	/*
+	 * Check if the pmu supports perf extended regs, before
+	 * returning the register mask to sample.
+	 */
+	if (attr.sample_regs_user != PERF_REGS_MASK) {
+		event_attr_init(&attr);
+		fd = sys_perf_event_open(&attr, 0, -1, -1, 0);
+		if (fd != -1) {
+			close(fd);
+			return attr.sample_regs_user;
+		}
+	}
+	return PERF_REGS_MASK;
+}
tools/perf/util/perf_regs.c +2 -0
index a982e40ee5a9..872dd3d38782 100644
@@ -103,6 +103,8 @@ static const char *__perf_reg_name_arm64(int id)
 		return "lr";
 	case PERF_REG_ARM64_PC:
 		return "pc";
+	case PERF_REG_ARM64_VG:
+		return "vg";
 	default:
 		return NULL;
 	}
Signature · 1 line
2.28.0
#10 lore raw

Re: [PATCH v2 4/4] perf tools: arm64: Add support for VG register

Quoted 59 lines · James Clark

Add the name of the VG register so it can be used in --user-regs

The event will fail to open if the register is requested but not available so only add it to the mask if the kernel supports sve and also if it supports that specific register.

---
 tools/perf/arch/arm64/util/perf_regs.c | 34 ++++++++++++++++++++++++++
 tools/perf/util/perf_regs.c            |  2 ++
 2 files changed, 36 insertions(+)
tools/perf/arch/arm64/util/perf_regs.c +22 -0
index 476b037eea1c..c0a921512a90 100644
@@ -2,9 +2,11 @@
 #include <errno.h>
 #include <regex.h>
 #include <string.h>
+#include <sys/auxv.h>
 #include <linux/kernel.h>
 #include <linux/zalloc.h>
 
+#include "../../../perf-sys.h"
 #include "../../../util/debug.h"
 #include "../../../util/event.h"
 #include "../../../util/perf_regs.h"
@@ -43,6 +45,7 @@ const struct sample_reg sample_reg_masks[] = {
 	SMPL_REG(lr, PERF_REG_ARM64_LR),
 	SMPL_REG(sp, PERF_REG_ARM64_SP),
 	SMPL_REG(pc, PERF_REG_ARM64_PC),
+	SMPL_REG(vg, PERF_REG_ARM64_VG),
 	SMPL_REG_END
 };
 
@@ -131,3 +134,34 @@ int arch_sdt_arg_parse_op(char *old_op, char **new_op)
 
 	return SDT_ARG_VALID;
 }
+
+uint64_t arch__user_reg_mask(void)
+{
+	struct perf_event_attr attr = {
+		.type                   = PERF_TYPE_HARDWARE,
+		.config                 = PERF_COUNT_HW_CPU_CYCLES,
+		.sample_type            = PERF_SAMPLE_REGS_USER,
+		.disabled               = 1,
+		.exclude_kernel         = 1,
+		.sample_period		= 1,
+		.sample_regs_user	= PERF_REGS_MASK
+	};
+	int fd;
+
+	if (getauxval(AT_HWCAP) & HWCAP_SVE)
+		attr.sample_regs_user |= SMPL_REG_MASK(PERF_REG_ARM64_VG);
+
+	/*
+	 * Check if the pmu supports perf extended regs, before

tools/perf/arch/arm64/util/perf_regs.c int arch_sdt_arg_parse_op(char *old_op, char **new_op) · line 165

Quoted 10 lines
tools/perf/arch/arm64/util/perf_regs.c +10 -0
+	 * returning the register mask to sample.
+	 */
+	if (attr.sample_regs_user != PERF_REGS_MASK) {
+		event_attr_init(&attr);
+		fd = sys_perf_event_open(&attr, 0, -1, -1, 0);
+		if (fd != -1) {
+			close(fd);
+			return attr.sample_regs_user;
+		}
+	}

Just curious, since we can know SVE is supported from reading auxiliary value, can we directly return the register mask as below?

  PERF_REGS_MASK | SMPL_REG_MASK(PERF_REG_ARM64_VG);

Except this question, this patch looks good to me.

Thanks,
Leo

Quoted 18 lines
tools/perf/arch/arm64/util/perf_regs.c +2 -0
+	return PERF_REGS_MASK;
+}
tools/perf/util/perf_regs.c +2 -0
index a982e40ee5a9..872dd3d38782 100644
@@ -103,6 +103,8 @@ static const char *__perf_reg_name_arm64(int id)
 		return "lr";
 	case PERF_REG_ARM64_PC:
 		return "pc";
+	case PERF_REG_ARM64_VG:
+		return "vg";
 	default:
 		return NULL;
 	}
Signature · 1 line
2.28.0
#11 lore raw

Re: [PATCH v2 4/4] perf tools: arm64: Add support for VG register

Quoted 74 lines · Leo Yan
Quoted 59 lines · James Clark

Add the name of the VG register so it can be used in --user-regs

The event will fail to open if the register is requested but not available so only add it to the mask if the kernel supports sve and also if it supports that specific register.

---
 tools/perf/arch/arm64/util/perf_regs.c | 34 ++++++++++++++++++++++++++
 tools/perf/util/perf_regs.c            |  2 ++
 2 files changed, 36 insertions(+)
tools/perf/arch/arm64/util/perf_regs.c +22 -0
index 476b037eea1c..c0a921512a90 100644
@@ -2,9 +2,11 @@
 #include <errno.h>
 #include <regex.h>
 #include <string.h>
+#include <sys/auxv.h>
 #include <linux/kernel.h>
 #include <linux/zalloc.h>
 
+#include "../../../perf-sys.h"
 #include "../../../util/debug.h"
 #include "../../../util/event.h"
 #include "../../../util/perf_regs.h"
@@ -43,6 +45,7 @@ const struct sample_reg sample_reg_masks[] = {
 	SMPL_REG(lr, PERF_REG_ARM64_LR),
 	SMPL_REG(sp, PERF_REG_ARM64_SP),
 	SMPL_REG(pc, PERF_REG_ARM64_PC),
+	SMPL_REG(vg, PERF_REG_ARM64_VG),
 	SMPL_REG_END
 };
 
@@ -131,3 +134,34 @@ int arch_sdt_arg_parse_op(char *old_op, char **new_op)
 
 	return SDT_ARG_VALID;
 }
+
+uint64_t arch__user_reg_mask(void)
+{
+	struct perf_event_attr attr = {
+		.type                   = PERF_TYPE_HARDWARE,
+		.config                 = PERF_COUNT_HW_CPU_CYCLES,
+		.sample_type            = PERF_SAMPLE_REGS_USER,
+		.disabled               = 1,
+		.exclude_kernel         = 1,
+		.sample_period		= 1,
+		.sample_regs_user	= PERF_REGS_MASK
+	};
+	int fd;
+
+	if (getauxval(AT_HWCAP) & HWCAP_SVE)
+		attr.sample_regs_user |= SMPL_REG_MASK(PERF_REG_ARM64_VG);
+
+	/*
+	 * Check if the pmu supports perf extended regs, before

tools/perf/arch/arm64/util/perf_regs.c int arch_sdt_arg_parse_op(char *old_op, char **new_op) · line 165

Quoted 10 lines
tools/perf/arch/arm64/util/perf_regs.c +10 -0
+	 * returning the register mask to sample.
+	 */
+	if (attr.sample_regs_user != PERF_REGS_MASK) {
+		event_attr_init(&attr);
+		fd = sys_perf_event_open(&attr, 0, -1, -1, 0);
+		if (fd != -1) {
+			close(fd);
+			return attr.sample_regs_user;
+		}
+	}

Just curious, since we can know SVE is supported from reading auxiliary value, can we directly return the register mask as below?

Quoted 1 line
  PERF_REGS_MASK | SMPL_REG_MASK(PERF_REG_ARM64_VG);

I was trying to cover the case where the system supports SVE, but the kernel doesn't have my changes to add the VG register yet.

Technically I could just attempt to open the event without checking for SVE first and see if it works or not. But I preferred to be explicit so it's obvious why we're doing that.

James

Quoted 24 lines

Except this question, this patch looks good to me.

Thanks,
Leo

Quoted 18 lines
tools/perf/arch/arm64/util/perf_regs.c +2 -0
+	return PERF_REGS_MASK;
+}
tools/perf/util/perf_regs.c +2 -0
index a982e40ee5a9..872dd3d38782 100644
@@ -103,6 +103,8 @@ static const char *__perf_reg_name_arm64(int id)
 		return "lr";
 	case PERF_REG_ARM64_PC:
 		return "pc";
+	case PERF_REG_ARM64_VG:
+		return "vg";
 	default:
 		return NULL;
 	}
Signature · 1 line
2.28.0
Reviewed-by
#12 lore raw

Re: [PATCH v2 4/4] perf tools: arm64: Add support for VG register

On Wed, May 18, 2022 at 10:44:57AM +0100, James Clark wrote:

[...]

Quoted 19 lines
Quoted 19 lines
Quoted 5 lines
tools/perf/arch/arm64/util/perf_regs.c +5 -0
+	if (getauxval(AT_HWCAP) & HWCAP_SVE)
+		attr.sample_regs_user |= SMPL_REG_MASK(PERF_REG_ARM64_VG);
+
+	/*
+	 * Check if the pmu supports perf extended regs, before

tools/perf/arch/arm64/util/perf_regs.c int arch_sdt_arg_parse_op(char *old_op, char **new_op) · line 165

Quoted 10 lines
tools/perf/arch/arm64/util/perf_regs.c +10 -0
+	 * returning the register mask to sample.
+	 */
+	if (attr.sample_regs_user != PERF_REGS_MASK) {
+		event_attr_init(&attr);
+		fd = sys_perf_event_open(&attr, 0, -1, -1, 0);
+		if (fd != -1) {
+			close(fd);
+			return attr.sample_regs_user;
+		}
+	}

Just curious, since we can know SVE is supported from reading auxiliary value, can we directly return the register mask as below?

Quoted 8 lines
Quoted 1 line
  PERF_REGS_MASK | SMPL_REG_MASK(PERF_REG_ARM64_VG);

I was trying to cover the case where the system supports SVE, but the kernel doesn't have my changes to add the VG register yet.

Technically I could just attempt to open the event without checking for SVE first and see if it works or not. But I preferred to be explicit so it's obvious why we're doing that.

Understand; LGTM.

#13 lore raw
In reply to James Clark

Em Tue, May 17, 2022 at 11:20:01AM +0100, James Clark escreveu:

Quoted 3 lines

Changes since v1:

* Split patchset into kernel side and Perf tool changes

Thanks, now I'll wait for the kernel side to be merged.

- Arnaldo

Quoted 31 lines

When SVE registers are pushed onto the stack the VG register is required to unwind because the stack offsets would vary by the SVE register width at the time when the sample was taken.

The patches ("[PATCH v2 0/2] perf: arm64: Kernel support for Dwarf unwinding through SVE functions") add support for sampling the VG register to the kernel and the docs. This is the patchset to add support to userspace perf.

A small change is also required to libunwind or libdw depending on which unwinder is used, and these will be published later. Without these changes Perf continues to work with both libraries, although the VG register is still not used for unwinding.

Thanks
James

James Clark (4):
perf tools: arm64: Copy perf_regs.h from the kernel perf tools: Use dynamic register set for Dwarf unwind perf tools: arm64: Decouple Libunwind register names from Perf perf tools: arm64: Add support for VG register

 tools/arch/arm64/include/uapi/asm/perf_regs.h |  7 +-
 tools/perf/arch/arm64/util/perf_regs.c        | 34 +++++++++
 tools/perf/arch/arm64/util/unwind-libunwind.c | 73 +------------------
 tools/perf/util/evsel.c                       |  2 +-
 tools/perf/util/perf_regs.c                   |  2 +
 5 files changed, 45 insertions(+), 73 deletions(-)
Signature · 1 line
2.28.0
Signature · 1 line
- Arnaldo
#14 lore raw
Quoted 8 lines · Arnaldo Carvalho de Melo

Em Tue, May 17, 2022 at 11:20:01AM +0100, James Clark escreveu:

Quoted 3 lines

Changes since v1:

* Split patchset into kernel side and Perf tool changes

Thanks, now I'll wait for the kernel side to be merged.

- Arnaldo

Hi,

I think we also need to fix the below import (it's trying to import the perf_regs.h from the kernel, not the tools/ dir).

tools/perf/util/libunwind/arm64.c +0 -0
index 15f60fd09424..014d82159656 100644
@@ -24,7 +24,7 @@

 #include "unwind.h"
 #include "libunwind-aarch64.h"
 #define perf_event_arm_regs perf_event_arm64_regs -#include <../../../../arch/arm64/include/uapi/asm/perf_regs.h> +#include <../../../arch/arm64/include/uapi/asm/perf_regs.h> #undef perf_event_arm_regs
 #include "../../arch/arm64/util/unwind-libunwind.c"

#15 lore raw

Em Fri, May 20, 2022 at 12:46:24PM +0100, German Gomez escreveu:

Quoted 8 lines · Arnaldo Carvalho de Melo
Quoted 6 lines

Em Tue, May 17, 2022 at 11:20:01AM +0100, James Clark escreveu:

Quoted 3 lines

Changes since v1:

* Split patchset into kernel side and Perf tool changes

Thanks, now I'll wait for the kernel side to be merged.

Quoted 7 lines
Quoted 2 lines

- Arnaldo

Hi,

I think we also need to fix the below import (it's trying to import the perf_regs.h from the kernel, not the tools/ dir).

Sure, that is why it is important to do a:

make -C tools/perf build-test

before posting patches, as it will do, among other tests, a detached tarball test using:

⬢[acme@toolbox perf-urgent]$ make help | grep perf
perf-tar-src-pkg - Build perf-5.18.0-rc7.tar source tarball
perf-targz-src-pkg - Build perf-5.18.0-rc7.tar.gz source tarball perf-tarbz2-src-pkg - Build perf-5.18.0-rc7.tar.bz2 source tarball perf-tarxz-src-pkg - Build perf-5.18.0-rc7.tar.xz source tarball perf-tarzst-src-pkg - Build perf-5.18.0-rc7.tar.zst source tarball
⬢[acme@toolbox perf-urgent]$

That will pick just what is in tools/perf/MANIFEST and that doesn´t include the kernel headers, just what is in tools/ then building with this patchkit would fail.

To recap:

⬢[acme@toolbox perf-urgent]$ make perf-tarxz-src-pkg
TAR
PERF_VERSION = 5.18.rc7.gc645054ef517
⬢[acme@toolbox perf-urgent]$ ls -la perf-5.18.0-rc7.tar.xz -rw-r--r--. 1 acme acme 2364832 May 20 09:32 perf-5.18.0-rc7.tar.xz ⬢[acme@toolbox perf-urgent]$ tar tvf perf-5.18.0-rc7.tar.xz | tail
-rw-rw-r-- root/root 7364 2022-05-20 09:32 perf-5.18.0-rc7/tools/perf/util/vdso.c
-rw-rw-r-- root/root 639 2022-05-20 09:32 perf-5.18.0-rc7/tools/perf/util/vdso.h
-rw-rw-r-- root/root 1793 2022-05-20 09:32 perf-5.18.0-rc7/tools/perf/util/zlib.c
-rw-rw-r-- root/root 2672 2022-05-20 09:32 perf-5.18.0-rc7/tools/perf/util/zstd.c
drwxrwxr-x root/root 0 2022-05-20 09:32 perf-5.18.0-rc7/tools/scripts/

-rw-rw-r-- root/root      1091 2022-05-20 09:32 perf-5.18.0-rc7/tools/scripts/Makefile.arch
-rw-rw-r-- root/root      5905 2022-05-20 09:32 perf-5.18.0-rc7/tools/scripts/Makefile.include
-rw-rw-r-- root/root      5288 2022-05-20 09:32 perf-5.18.0-rc7/tools/scripts/utilities.mak

-rw-r--r-- acme/acme 41 2022-05-20 09:32 perf-5.18.0-rc7/HEAD
-rw-r--r-- acme/acme 46 2022-05-20 09:32 perf-5.18.0-rc7/PERF-VERSION-FILE
⬢[acme@toolbox perf-urgent]$

- Arnaldo

Quoted 13 lines
tools/perf/util/libunwind/arm64.c +1 -1
index 15f60fd09424..014d82159656 100644
@@ -24,7 +24,7 @@
  #include "unwind.h"
  #include "libunwind-aarch64.h"
  #define perf_event_arm_regs perf_event_arm64_regs
-#include <../../../../arch/arm64/include/uapi/asm/perf_regs.h>
+#include <../../../arch/arm64/include/uapi/asm/perf_regs.h>
  #undef perf_event_arm_regs
  #include "../../arch/arm64/util/unwind-libunwind.c"
 
Signature · 1 line
- Arnaldo
#16 lore raw
Quoted 12 lines · Arnaldo Carvalho de Melo

Em Fri, May 20, 2022 at 12:46:24PM +0100, German Gomez escreveu:

Quoted 11 lines · Arnaldo Carvalho de Melo
Quoted 8 lines

Em Tue, May 17, 2022 at 11:20:01AM +0100, James Clark escreveu:

Quoted 3 lines

Changes since v1:

* Split patchset into kernel side and Perf tool changes

Thanks, now I'll wait for the kernel side to be merged.

- Arnaldo

Quoted 9 lines
Quoted 4 lines

Hi,

I think we also need to fix the below import (it's trying to import the perf_regs.h from the kernel, not the tools/ dir).

Sure, that is why it is important to do a:

make -C tools/perf build-test

Oops thanks, I will submit another version with a fix for this included.

James

Quoted 50 lines

before posting patches, as it will do, among other tests, a detached tarball test using:

⬢[acme@toolbox perf-urgent]$ make help | grep perf
perf-tar-src-pkg - Build perf-5.18.0-rc7.tar source tarball
perf-targz-src-pkg - Build perf-5.18.0-rc7.tar.gz source tarball perf-tarbz2-src-pkg - Build perf-5.18.0-rc7.tar.bz2 source tarball perf-tarxz-src-pkg - Build perf-5.18.0-rc7.tar.xz source tarball perf-tarzst-src-pkg - Build perf-5.18.0-rc7.tar.zst source tarball
⬢[acme@toolbox perf-urgent]$

That will pick just what is in tools/perf/MANIFEST and that doesn´t include the kernel headers, just what is in tools/ then building with this patchkit would fail.

To recap:

⬢[acme@toolbox perf-urgent]$ make perf-tarxz-src-pkg
TAR
PERF_VERSION = 5.18.rc7.gc645054ef517
⬢[acme@toolbox perf-urgent]$ ls -la perf-5.18.0-rc7.tar.xz -rw-r--r--. 1 acme acme 2364832 May 20 09:32 perf-5.18.0-rc7.tar.xz ⬢[acme@toolbox perf-urgent]$ tar tvf perf-5.18.0-rc7.tar.xz | tail

(quoted diff) +0 -4
-rw-rw-r-- root/root      7364 2022-05-20 09:32 perf-5.18.0-rc7/tools/perf/util/vdso.c
-rw-rw-r-- root/root       639 2022-05-20 09:32 perf-5.18.0-rc7/tools/perf/util/vdso.h
-rw-rw-r-- root/root      1793 2022-05-20 09:32 perf-5.18.0-rc7/tools/perf/util/zlib.c
-rw-rw-r-- root/root      2672 2022-05-20 09:32 perf-5.18.0-rc7/tools/perf/util/zstd.c

drwxrwxr-x root/root 0 2022-05-20 09:32 perf-5.18.0-rc7/tools/scripts/

(quoted diff) +0 -5
-rw-rw-r-- root/root      1091 2022-05-20 09:32 perf-5.18.0-rc7/tools/scripts/Makefile.arch
-rw-rw-r-- root/root      5905 2022-05-20 09:32 perf-5.18.0-rc7/tools/scripts/Makefile.include
-rw-rw-r-- root/root      5288 2022-05-20 09:32 perf-5.18.0-rc7/tools/scripts/utilities.mak
-rw-r--r-- acme/acme        41 2022-05-20 09:32 perf-5.18.0-rc7/HEAD
-rw-r--r-- acme/acme        46 2022-05-20 09:32 perf-5.18.0-rc7/PERF-VERSION-FILE

⬢[acme@toolbox perf-urgent]$

- Arnaldo

Quoted 13 lines
tools/perf/util/libunwind/arm64.c +1 -1
index 15f60fd09424..014d82159656 100644
@@ -24,7 +24,7 @@
  #include "unwind.h"
  #include "libunwind-aarch64.h"
  #define perf_event_arm_regs perf_event_arm64_regs
-#include <../../../../arch/arm64/include/uapi/asm/perf_regs.h>
+#include <../../../arch/arm64/include/uapi/asm/perf_regs.h>
  #undef perf_event_arm_regs
  #include "../../arch/arm64/util/unwind-libunwind.c"