[PATCH v2 0/3] perf more ELF machine usage

[PATCH v2 0/3] perf more ELF machine usage

[Ian Rogers]

Switch 3 more cases where the perf_env arch was being used to
determine the architecture type and switch to using the ELF
machine. This includes the addition of minimal instruction parsing in
perf script for architectures other than x86, or x86 when
cross-platform.

v2: Resolve conflicts with:
    https://lore.kernel.org/lkml/20260123222209.1181249-1-irogers@google.com/

v1: https://lore.kernel.org/lkml/20260124001611.1332019-1-irogers@google.com/

Ian Rogers (3):
  perf script: Fix script_fetch_insn for more than just x86
  perf callchain: Switch callchain_param_setup from an arch to an
    e_machine
  perf thread-stack: Switch thread_stack__init to use e_machine

 tools/perf/arch/x86/util/Build                |  1 -
 tools/perf/arch/x86/util/archinsn.c           | 27 -------
 tools/perf/builtin-report.c                   |  6 +-
 tools/perf/builtin-script.c                   | 20 +-----
 .../scripts/python/Perf-Trace-Util/Context.c  |  2 +-
 tools/perf/tests/dlfilter-test.c              |  1 -
 tools/perf/util/archinsn.h                    | 12 ----
 tools/perf/util/callchain.c                   |  4 +-
 tools/perf/util/callchain.h                   |  2 +-
 tools/perf/util/dlfilter.c                    |  3 +-
 tools/perf/util/sample.c                      | 71 +++++++++++++++++++
 tools/perf/util/sample.h                      |  7 ++
 tools/perf/util/thread-stack.c                |  4 +-
 tools/perf/util/trace-event-scripting.c       | 16 -----
 tools/perf/util/trace-event.h                 |  3 -
 15 files changed, 90 insertions(+), 89 deletions(-)
 delete mode 100644 tools/perf/arch/x86/util/archinsn.c
 delete mode 100644 tools/perf/util/archinsn.h

-- 
2.52.0.457.g6b5491de43-goog

[PATCH v2 1/3] perf script: Fix script_fetch_insn for more than just x86

[Ian Rogers]

The script_fetch_insn code was only supported on natively running
x86. Implement a crude elf_machine_max_instruction_length function and
use to give an instruction length on more than just x86. Use the ELF
machine to determine the length to use to support cross-architecture
development.

Signed-off-by: Ian Rogers <irogers@google.com>
---
 tools/perf/arch/x86/util/Build                |  1 -
 tools/perf/arch/x86/util/archinsn.c           | 27 -------
 tools/perf/builtin-script.c                   | 16 +----
 .../scripts/python/Perf-Trace-Util/Context.c  |  2 +-
 tools/perf/tests/dlfilter-test.c              |  1 -
 tools/perf/util/archinsn.h                    | 12 ----
 tools/perf/util/dlfilter.c                    |  3 +-
 tools/perf/util/sample.c                      | 71 +++++++++++++++++++
 tools/perf/util/sample.h                      |  7 ++
 tools/perf/util/trace-event-scripting.c       | 16 -----
 tools/perf/util/trace-event.h                 |  3 -
 11 files changed, 81 insertions(+), 78 deletions(-)
 delete mode 100644 tools/perf/arch/x86/util/archinsn.c
 delete mode 100644 tools/perf/util/archinsn.h

diff --git a/tools/perf/arch/x86/util/Build b/tools/perf/arch/x86/util/Build
index fad256252bb9..76127eefde8b 100644
--- a/tools/perf/arch/x86/util/Build
+++ b/tools/perf/arch/x86/util/Build
@@ -14,6 +14,5 @@ perf-util-y += iostat.o
 perf-util-$(CONFIG_LOCAL_LIBUNWIND)    += unwind-libunwind.o
 
 perf-util-y += auxtrace.o
-perf-util-y += archinsn.o
 perf-util-y += intel-pt.o
 perf-util-y += intel-bts.o
diff --git a/tools/perf/arch/x86/util/archinsn.c b/tools/perf/arch/x86/util/archinsn.c
deleted file mode 100644
index 546feda08428..000000000000
--- a/tools/perf/arch/x86/util/archinsn.c
+++ /dev/null
@@ -1,27 +0,0 @@
-// SPDX-License-Identifier: GPL-2.0
-#include "archinsn.h"
-#include "event.h"
-#include "machine.h"
-#include "thread.h"
-#include "symbol.h"
-#include "../../../../arch/x86/include/asm/insn.h"
-
-void arch_fetch_insn(struct perf_sample *sample,
-		     struct thread *thread,
-		     struct machine *machine)
-{
-	struct insn insn;
-	int len, ret;
-	bool is64bit = false;
-
-	if (!sample->ip)
-		return;
-	len = thread__memcpy(thread, machine, sample->insn, sample->ip, sizeof(sample->insn), &is64bit);
-	if (len <= 0)
-		return;
-
-	ret = insn_decode(&insn, sample->insn, len,
-			  is64bit ? INSN_MODE_64 : INSN_MODE_32);
-	if (ret >= 0 && insn.length <= len)
-		sample->insn_len = insn.length;
-}
diff --git a/tools/perf/builtin-script.c b/tools/perf/builtin-script.c
index 6ec225c697a4..69af25780fc5 100644
--- a/tools/perf/builtin-script.c
+++ b/tools/perf/builtin-script.c
@@ -37,7 +37,6 @@
 #include "ui/ui.h"
 #include "print_binary.h"
 #include "print_insn.h"
-#include "archinsn.h"
 #include <linux/bitmap.h>
 #include <linux/compiler.h>
 #include <linux/kernel.h>
@@ -90,7 +89,6 @@ static bool			print_flags;
 static const char		*cpu_list;
 static DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS);
 static int			max_blocks;
-static bool			native_arch;
 static struct dlfilter		*dlfilter;
 static int			dlargc;
 static char			**dlargv;
@@ -1627,7 +1625,7 @@ static int perf_sample__fprintf_insn(struct perf_sample *sample,
 {
 	int printed = 0;
 
-	script_fetch_insn(sample, thread, machine, native_arch);
+	perf_sample__fetch_insn(sample, thread, machine);
 
 	if (PRINT_FIELD(INSNLEN))
 		printed += fprintf(fp, " ilen: %d", sample->insn_len);
@@ -4034,7 +4032,6 @@ int cmd_script(int argc, const char **argv)
 		.set = false,
 		.default_no_sample = true,
 	};
-	struct utsname uts;
 	char *script_path = NULL;
 	const char *dlfilter_file = NULL;
 	const char **__argv;
@@ -4456,17 +4453,6 @@ int cmd_script(int argc, const char **argv)
 	if (symbol__init(env) < 0)
 		goto out_delete;
 
-	uname(&uts);
-	if (data.is_pipe) { /* Assume pipe_mode indicates native_arch */
-		native_arch = true;
-	} else if (env->arch) {
-		if (!strcmp(uts.machine, env->arch))
-			native_arch = true;
-		else if (!strcmp(uts.machine, "x86_64") &&
-			 !strcmp(env->arch, "i386"))
-			native_arch = true;
-	}
-
 	script.session = session;
 	script__setup_sample_type(&script);
 
diff --git a/tools/perf/scripts/python/Perf-Trace-Util/Context.c b/tools/perf/scripts/python/Perf-Trace-Util/Context.c
index 60dcfe56d4d9..c19f44610983 100644
--- a/tools/perf/scripts/python/Perf-Trace-Util/Context.c
+++ b/tools/perf/scripts/python/Perf-Trace-Util/Context.c
@@ -93,7 +93,7 @@ static PyObject *perf_sample_insn(PyObject *obj, PyObject *args)
 	if (c->sample->ip && !c->sample->insn_len && thread__maps(c->al->thread)) {
 		struct machine *machine =  maps__machine(thread__maps(c->al->thread));
 
-		script_fetch_insn(c->sample, c->al->thread, machine, /*native_arch=*/true);
+		perf_sample__fetch_insn(c->sample, c->al->thread, machine);
 	}
 	if (!c->sample->insn_len)
 		Py_RETURN_NONE; /* N.B. This is a return statement */
diff --git a/tools/perf/tests/dlfilter-test.c b/tools/perf/tests/dlfilter-test.c
index 80a1c941138d..e63790c61d53 100644
--- a/tools/perf/tests/dlfilter-test.c
+++ b/tools/perf/tests/dlfilter-test.c
@@ -30,7 +30,6 @@
 #include "symbol.h"
 #include "synthetic-events.h"
 #include "util.h"
-#include "archinsn.h"
 #include "dlfilter.h"
 #include "tests.h"
 #include "util/sample.h"
diff --git a/tools/perf/util/archinsn.h b/tools/perf/util/archinsn.h
deleted file mode 100644
index 448cbb6b8d7e..000000000000
--- a/tools/perf/util/archinsn.h
+++ /dev/null
@@ -1,12 +0,0 @@
-#ifndef INSN_H
-#define INSN_H 1
-
-struct perf_sample;
-struct machine;
-struct thread;
-
-void arch_fetch_insn(struct perf_sample *sample,
-		     struct thread *thread,
-		     struct machine *machine);
-
-#endif
diff --git a/tools/perf/util/dlfilter.c b/tools/perf/util/dlfilter.c
index c0afcbd954f8..dc31b5e7149e 100644
--- a/tools/perf/util/dlfilter.c
+++ b/tools/perf/util/dlfilter.c
@@ -234,8 +234,7 @@ static const __u8 *dlfilter__insn(void *ctx, __u32 *len)
 			struct machine *machine = maps__machine(thread__maps(al->thread));
 
 			if (machine)
-				script_fetch_insn(d->sample, al->thread, machine,
-						  /*native_arch=*/true);
+				perf_sample__fetch_insn(d->sample, al->thread, machine);
 		}
 	}
 
diff --git a/tools/perf/util/sample.c b/tools/perf/util/sample.c
index 605fee971f55..d885c02a9861 100644
--- a/tools/perf/util/sample.c
+++ b/tools/perf/util/sample.c
@@ -1,9 +1,12 @@
 /* SPDX-License-Identifier: GPL-2.0 */
 #include "sample.h"
 #include "debug.h"
+#include "thread.h"
+#include <elf.h>
 #include <linux/zalloc.h>
 #include <stdlib.h>
 #include <string.h>
+#include "../../arch/x86/include/asm/insn.h"
 
 void perf_sample__init(struct perf_sample *sample, bool all)
 {
@@ -41,3 +44,71 @@ struct regs_dump *perf_sample__intr_regs(struct perf_sample *sample)
 	}
 	return sample->intr_regs;
 }
+
+static int elf_machine_max_instruction_length(uint16_t e_machine)
+{
+	switch (e_machine) {
+	/* Fixed 4-byte (32-bit) architectures */
+	case EM_AARCH64:
+	case EM_PPC:
+	case EM_PPC64:
+	case EM_MIPS:
+	case EM_SPARC:
+	case EM_SPARCV9:
+	case EM_ALPHA:
+	case EM_LOONGARCH:
+	case EM_PARISC:
+	case EM_SH:
+		return 4;
+
+	/* Variable length or mixed-mode architectures */
+	case EM_ARM:    /* Variable due to Thumb/Thumb-2 */
+	case EM_RISCV:  /* Variable due to Compressed (C) extension */
+	case EM_CSKY:   /* Variable (16 or 32 bit) */
+	case EM_ARC:    /* Variable (ARCompact) */
+		return 4;
+	case EM_S390:   /* Variable (2, 4, or 6 bytes) */
+		return 6;
+	case EM_68K:
+		return 10;
+	case EM_386:
+	case EM_X86_64:
+		return 15;
+	case EM_XTENSA: /* Variable (FLIX) */
+		return 16;
+	default:
+		return MAX_INSN;
+	}
+}
+
+void perf_sample__fetch_insn(struct perf_sample *sample,
+			     struct thread *thread,
+			     struct machine *machine)
+{
+	int ret, len;
+	bool is64bit = false;
+	uint16_t e_machine;
+
+	if (!sample->ip || sample->insn_len != 0)
+		return;
+
+	e_machine = thread__e_machine(thread, machine);
+	len = elf_machine_max_instruction_length(e_machine);
+	len = thread__memcpy(thread, machine, sample->insn,
+			     sample->ip, len,
+			     &is64bit);
+	if (len <= 0)
+		return;
+
+	sample->insn_len = len;
+
+	if (e_machine == EM_386 || e_machine == EM_X86_64) {
+		/* Refine the x86 instruction length with the decoder. */
+		struct insn insn;
+
+		ret = insn_decode(&insn, sample->insn, len,
+				  is64bit ? INSN_MODE_64 : INSN_MODE_32);
+		if (ret >= 0 && insn.length <= len)
+			sample->insn_len = insn.length;
+	}
+}
diff --git a/tools/perf/util/sample.h b/tools/perf/util/sample.h
index a8307b20a9ea..3cce8dd202aa 100644
--- a/tools/perf/util/sample.h
+++ b/tools/perf/util/sample.h
@@ -5,6 +5,9 @@
 #include <linux/perf_event.h>
 #include <linux/types.h>
 
+struct machine;
+struct thread;
+
 /* number of register is bound by the number of bits in regs_dump::mask (64) */
 #define PERF_SAMPLE_REGS_CACHE_SIZE (8 * sizeof(u64))
 
@@ -127,6 +130,10 @@ void perf_sample__exit(struct perf_sample *sample);
 struct regs_dump *perf_sample__user_regs(struct perf_sample *sample);
 struct regs_dump *perf_sample__intr_regs(struct perf_sample *sample);
 
+void perf_sample__fetch_insn(struct perf_sample *sample,
+			     struct thread *thread,
+			     struct machine *machine);
+
 /*
  * raw_data is always 4 bytes from an 8-byte boundary, so subtract 4 to get
  * 8-byte alignment.
diff --git a/tools/perf/util/trace-event-scripting.c b/tools/perf/util/trace-event-scripting.c
index 72abb28b7b5a..fa850e44cb46 100644
--- a/tools/perf/util/trace-event-scripting.c
+++ b/tools/perf/util/trace-event-scripting.c
@@ -13,7 +13,6 @@
 #include <event-parse.h>
 #endif
 
-#include "archinsn.h"
 #include "debug.h"
 #include "event.h"
 #include "trace-event.h"
@@ -274,21 +273,6 @@ void setup_perl_scripting(void)
 #endif
 #endif
 
-#if !defined(__i386__) && !defined(__x86_64__)
-void arch_fetch_insn(struct perf_sample *sample __maybe_unused,
-		     struct thread *thread __maybe_unused,
-		     struct machine *machine __maybe_unused)
-{
-}
-#endif
-
-void script_fetch_insn(struct perf_sample *sample, struct thread *thread,
-		       struct machine *machine, bool native_arch)
-{
-	if (sample->insn_len == 0 && native_arch)
-		arch_fetch_insn(sample, thread, machine);
-}
-
 static const struct {
 	u32 flags;
 	const char *name;
diff --git a/tools/perf/util/trace-event.h b/tools/perf/util/trace-event.h
index 71e680bc3d4b..914d9b69ed62 100644
--- a/tools/perf/util/trace-event.h
+++ b/tools/perf/util/trace-event.h
@@ -116,9 +116,6 @@ extern unsigned int scripting_max_stack;
 struct scripting_ops *script_spec__lookup(const char *spec);
 int script_spec__for_each(int (*cb)(struct scripting_ops *ops, const char *spec));
 
-void script_fetch_insn(struct perf_sample *sample, struct thread *thread,
-		       struct machine *machine, bool native_arch);
-
 void setup_perl_scripting(void);
 void setup_python_scripting(void);
 
-- 
2.52.0.457.g6b5491de43-goog

[PATCH v2 2/3] perf callchain: Switch callchain_param_setup from an arch to an e_machine

[Ian Rogers]

Increase use of e_machine by replacing callchain_param_setup's arch
argument to be an e_machine typically read from the session.

Signed-off-by: Ian Rogers <irogers@google.com>
---
 tools/perf/builtin-report.c | 6 ++----
 tools/perf/builtin-script.c | 4 ++--
 tools/perf/util/callchain.c | 4 ++--
 tools/perf/util/callchain.h | 2 +-
 tools/perf/util/sample.c    | 2 +-
 5 files changed, 8 insertions(+), 10 deletions(-)

diff --git a/tools/perf/builtin-report.c b/tools/perf/builtin-report.c
index 2e936928e8c0..810ffd66b11c 100644
--- a/tools/perf/builtin-report.c
+++ b/tools/perf/builtin-report.c
@@ -448,7 +448,7 @@ static int report__setup_sample_type(struct report *rep)
 		}
 	}
 
-	callchain_param_setup(sample_type, perf_env__arch(perf_session__env(rep->session)));
+	callchain_param_setup(sample_type, perf_session__e_machine(session));
 
 	if (rep->stitch_lbr && (callchain_param.record_mode != CALLCHAIN_LBR)) {
 		ui__warning("Can't find LBR callchain. Switch off --stitch-lbr.\n"
@@ -1283,7 +1283,6 @@ static int process_attr(const struct perf_tool *tool __maybe_unused,
 			struct evlist **pevlist)
 {
 	struct perf_session *session;
-	struct perf_env *env;
 	u64 sample_type;
 	int err;
 
@@ -1297,8 +1296,7 @@ static int process_attr(const struct perf_tool *tool __maybe_unused,
 	 */
 	sample_type = evlist__combined_sample_type(*pevlist);
 	session = (*pevlist)->session;
-	env = perf_session__env(session);
-	callchain_param_setup(sample_type, perf_env__arch(env));
+	callchain_param_setup(sample_type, perf_session__e_machine(session));
 	return 0;
 }
 
diff --git a/tools/perf/builtin-script.c b/tools/perf/builtin-script.c
index 69af25780fc5..c7d5a325b5cb 100644
--- a/tools/perf/builtin-script.c
+++ b/tools/perf/builtin-script.c
@@ -2859,7 +2859,7 @@ static int process_attr(const struct perf_tool *tool, union perf_event *event,
 	 * on events sample_type.
 	 */
 	sample_type = evlist__combined_sample_type(evlist);
-	callchain_param_setup(sample_type, perf_env__arch(perf_session__env(scr->session)));
+	callchain_param_setup(sample_type, perf_session__e_machine(evsel__session(evsel)));
 
 	/* Enable fields for callchain entries */
 	if (symbol_conf.use_callchain &&
@@ -3834,7 +3834,7 @@ static void script__setup_sample_type(struct perf_script *script)
 	struct perf_session *session = script->session;
 	u64 sample_type = evlist__combined_sample_type(session->evlist);
 
-	callchain_param_setup(sample_type, perf_env__arch(session->machines.host.env));
+	callchain_param_setup(sample_type, perf_session__e_machine(session));
 
 	if (script->stitch_lbr && (callchain_param.record_mode != CALLCHAIN_LBR)) {
 		pr_warning("Can't find LBR callchain. Switch off --stitch-lbr.\n"
diff --git a/tools/perf/util/callchain.c b/tools/perf/util/callchain.c
index 515bb8b5da01..8ff0898799ee 100644
--- a/tools/perf/util/callchain.c
+++ b/tools/perf/util/callchain.c
@@ -1681,7 +1681,7 @@ void callchain_cursor_reset(struct callchain_cursor *cursor)
 		map_symbol__exit(&node->ms);
 }
 
-void callchain_param_setup(u64 sample_type, const char *arch)
+void callchain_param_setup(u64 sample_type, uint16_t e_machine)
 {
 	if (symbol_conf.use_callchain || symbol_conf.cumulate_callchain) {
 		if ((sample_type & PERF_SAMPLE_REGS_USER) &&
@@ -1703,7 +1703,7 @@ void callchain_param_setup(u64 sample_type, const char *arch)
 	 * erroneous entries. Always skipping the LR and starting from the FP
 	 * can result in missing entries.
 	 */
-	if (callchain_param.record_mode == CALLCHAIN_FP && !strcmp(arch, "arm64"))
+	if (callchain_param.record_mode == CALLCHAIN_FP && e_machine == EM_AARCH64)
 		dwarf_callchain_users = true;
 }
 
diff --git a/tools/perf/util/callchain.h b/tools/perf/util/callchain.h
index 2a52af8c80ac..df54ddb8c0cb 100644
--- a/tools/perf/util/callchain.h
+++ b/tools/perf/util/callchain.h
@@ -303,7 +303,7 @@ int callchain_branch_counts(struct callchain_root *root,
 			    u64 *branch_count, u64 *predicted_count,
 			    u64 *abort_count, u64 *cycles_count);
 
-void callchain_param_setup(u64 sample_type, const char *arch);
+void callchain_param_setup(u64 sample_type, uint16_t e_machine);
 
 bool callchain_cnode_matched(struct callchain_node *base_cnode,
 			     struct callchain_node *pair_cnode);
diff --git a/tools/perf/util/sample.c b/tools/perf/util/sample.c
index d885c02a9861..b9089878ace5 100644
--- a/tools/perf/util/sample.c
+++ b/tools/perf/util/sample.c
@@ -92,7 +92,7 @@ void perf_sample__fetch_insn(struct perf_sample *sample,
 	if (!sample->ip || sample->insn_len != 0)
 		return;
 
-	e_machine = thread__e_machine(thread, machine);
+	e_machine = thread__e_machine(thread, machine, /*e_flags=*/NULL);
 	len = elf_machine_max_instruction_length(e_machine);
 	len = thread__memcpy(thread, machine, sample->insn,
 			     sample->ip, len,
-- 
2.52.0.457.g6b5491de43-goog

[PATCH v2 3/3] perf thread-stack: Switch thread_stack__init to use e_machine

[Ian Rogers]

The architecture type is used to set the retpoline state. Rather than
use the arch string switch to using the ELF machine that's readily
available within the thread.

Signed-off-by: Ian Rogers <irogers@google.com>
---
 tools/perf/util/thread-stack.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/tools/perf/util/thread-stack.c b/tools/perf/util/thread-stack.c
index c6a0a27b12c2..c5ce741b0744 100644
--- a/tools/perf/util/thread-stack.c
+++ b/tools/perf/util/thread-stack.c
@@ -157,10 +157,10 @@ static int thread_stack__init(struct thread_stack *ts, struct thread *thread,
 
 	if (thread__maps(thread) && maps__machine(thread__maps(thread))) {
 		struct machine *machine = maps__machine(thread__maps(thread));
-		const char *arch = perf_env__arch(machine->env);
+		uint16_t e_machine = thread__e_machine(thread, machine, /*e_flags=*/NULL);
 
 		ts->kernel_start = machine__kernel_start(machine);
-		if (!strcmp(arch, "x86"))
+		if (e_machine == EM_X86_64 || e_machine == EM_386)
 			ts->rstate = X86_RETPOLINE_POSSIBLE;
 	} else {
 		ts->kernel_start = 1ULL << 63;
-- 
2.52.0.457.g6b5491de43-goog

Re: [PATCH v2 0/3] perf more ELF machine usage

[Arnaldo Carvalho de Melo]

On Mon, Jan 26, 2026 at 02:05:47PM -0800, Ian Rogers wrote:
> Switch 3 more cases where the perf_env arch was being used to
> determine the architecture type and switch to using the ELF
> machine. This includes the addition of minimal instruction parsing in
> perf script for architectures other than x86, or x86 when
> cross-platform.
> 
> v2: Resolve conflicts with:
>     https://lore.kernel.org/lkml/20260123222209.1181249-1-irogers@google.com/
> 
> v1: https://lore.kernel.org/lkml/20260124001611.1332019-1-irogers@google.com/
> 
> Ian Rogers (3):
>   perf script: Fix script_fetch_insn for more than just x86
>   perf callchain: Switch callchain_param_setup from an arch to an
>     e_machine
>   perf thread-stack: Switch thread_stack__init to use e_machine

Thanks, applied to perf-tools-next,

- Arnaldo