[PATCH v8 1/2] perf: Reveal PMU type in fdinfo

[PATCH v8 1/2] perf: Reveal PMU type in fdinfo

[Chun-Tse Shao]

It gives useful info on knowing which PMUs are reserved by this process.
Also add config which would be useful.
Testing cycles:

  $ ./perf stat -e cycles &
  $ cat /proc/`pidof perf`/fdinfo/3
  pos:    0
  flags:  02000002
  mnt_id: 16
  ino:    3081
  perf_event_attr.type:   0
  perf_event_attr.config: 0x0
  perf_event_attr.config1:        0x0
  perf_event_attr.config2:        0x0
  perf_event_attr.config3:        0x0
  perf_event_attr.config4:        0x0

Testing L1-dcache-load-misses:

  $ ./perf stat -e L1-dcache-load-misses &
  $ cat /proc/`pidof perf`/fdinfo/3
  pos:    0
  flags:  02000002
  mnt_id: 16
  ino:    1072
  perf_event_attr.type:   3
  perf_event_attr.config: 0x10000
  perf_event_attr.config1:        0x0
  perf_event_attr.config2:        0x0
  perf_event_attr.config3:        0x0
  perf_event_attr.config4:        0x0

Reviewed-by: Ian Rogers <irogers@google.com>
Signed-off-by: Chun-Tse Shao <ctshao@google.com>
Assisted-by: Gemini:gemini-3.1-pro-preview
---
v8:
  - Fix a race condition on 32-bit architectures by acquiring parent or
    child child_mutex inside perf_show_fdinfo() to prevent concurrent
    mutations via perf_event_modify_breakpoint() from causing torn reads
    of breakpoint fields (which alias config1 and config2).
  - Expose missing config4 field in the fdinfo output.
---
 kernel/events/core.c | 29 +++++++++++++++++++++++++++++
 1 file changed, 29 insertions(+)

diff --git a/kernel/events/core.c b/kernel/events/core.c
index 7935d5663944..95d806bba654 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -58,6 +58,7 @@
 #include <linux/percpu-rwsem.h>
 #include <linux/unwind_deferred.h>
 #include <linux/kvm_types.h>
+#include <linux/seq_file.h>

 #include "internal.h"

@@ -7546,6 +7547,33 @@ static int perf_fasync(int fd, struct file *filp, int on)
 	return 0;
 }

+static void perf_show_fdinfo(struct seq_file *m, struct file *f)
+{
+	struct perf_event *event = f->private_data;
+	struct perf_event_context *ctx;
+	struct mutex *child_mutex;
+
+	ctx = perf_event_ctx_lock(event);
+	child_mutex = event->parent ? &event->parent->child_mutex : &event->child_mutex;
+	mutex_lock(child_mutex);
+
+	seq_printf(m, "perf_event_attr.type:\t%u\n", event->orig_type);
+	if (event->pmu)
+		seq_printf(m, "pmu_type:\t%u\n", event->pmu->type);
+	seq_printf(m, "perf_event_attr.config:\t0x%llx\n", (unsigned long long)event->attr.config);
+	seq_printf(m, "perf_event_attr.config1:\t0x%llx\n",
+		   (unsigned long long)event->attr.config1);
+	seq_printf(m, "perf_event_attr.config2:\t0x%llx\n",
+		   (unsigned long long)event->attr.config2);
+	seq_printf(m, "perf_event_attr.config3:\t0x%llx\n",
+		   (unsigned long long)event->attr.config3);
+	seq_printf(m, "perf_event_attr.config4:\t0x%llx\n",
+		   (unsigned long long)event->attr.config4);
+
+	mutex_unlock(child_mutex);
+	perf_event_ctx_unlock(event, ctx);
+}
+
 static const struct file_operations perf_fops = {
 	.release		= perf_release,
 	.read			= perf_read,
@@ -7554,6 +7582,7 @@ static const struct file_operations perf_fops = {
 	.compat_ioctl		= perf_compat_ioctl,
 	.mmap			= perf_mmap,
 	.fasync			= perf_fasync,
+	.show_fdinfo		= perf_show_fdinfo,
 };

 /*
--
2.54.0.1013.g208068f2d8-goog

[PATCH v8 2/2] perf evsel: Find process with busy PMUs for EBUSY

[Chun-Tse Shao]

It parses fdinfo with PMU type, comparing with the event which failed to
open, and report the processes causing EBUSY error.

Testing cycles and intel_pt//

  $ ./perf stat -e cycles &
  [1] 55569
  $ ./perf stat -e intel_pt// &
  [2] 55683
  $ ./perf stat -e intel_pt//
  Error:
  The PMU intel_pt counters are busy and in use by another process.
  Possible processes:
  55683 ./perf stat -e intel_pt//

Only perf with intel_pt was reported.

Reviewed-by: Ian Rogers <irogers@google.com>
Signed-off-by: Chun-Tse Shao <ctshao@google.com>
Assisted-by: Gemini:gemini-3.1-pro-preview
---
v8:
  - Fix a boundary mismatch bug in strncmp when matching anon_inode
    symlinks.
  - Switch to thread-safe strtok_r to prevent use-after-free or static
    state corruption.
  - Expand the stack-allocated fdinfo parsing buffer from 256 bytes to
    1024 bytes to completely prevent truncation before the PMU type is
    parsed.
  - Open proc files (fdinfo and cmdline) with O_NONBLOCK to prevent
    malicious or slow mounts from hanging the perf session indefinitely.
  - Match EBUSY conflicting processes using the physical PMU type
    (pmu_type) instead of the requested event type, ensuring legacy/raw
    events are correctly resolved.
---
 tools/perf/util/evsel.c | 99 ++++++++++++++++++++++++++++++++---------
 1 file changed, 77 insertions(+), 22 deletions(-)

diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
index 34c03f47a913..a6b1005db5e8 100644
--- a/tools/perf/util/evsel.c
+++ b/tools/perf/util/evsel.c
@@ -4104,8 +4104,11 @@ static bool find_process(const char *name)
 	return ret ? false : true;
 }

-static int dump_perf_event_processes(char *msg, size_t size)
+static int dump_perf_event_processes(const struct evsel *evsel,
+				     char *msg, size_t size)
 {
+	const struct perf_event_attr *failed_attr = &evsel->core.attr;
+	u32 target_pmu_type = evsel->pmu ? evsel->pmu->type : UINT32_MAX;
 	DIR *proc_dir;
 	struct dirent *proc_entry;
 	int printed = 0;
@@ -4136,6 +4139,8 @@ static int dump_perf_event_processes(char *msg, size_t size)
 			continue;
 		}
 		while ((fd_entry = readdir(fd_dir)) != NULL) {
+			const char *target_lnk = "anon_inode:[perf_event]";
+			size_t target_lnk_len = sizeof("anon_inode:[perf_event]") - 1;
 			ssize_t link_size;

 			if (fd_entry->d_type != DT_LNK)
@@ -4144,30 +4149,78 @@ static int dump_perf_event_processes(char *msg, size_t size)
 			if (link_size < 0)
 				continue;
 			/* Take care as readlink doesn't null terminate the string. */
-			if (!strncmp(buf, "anon_inode:[perf_event]", link_size)) {
-				int cmdline_fd;
-				ssize_t cmdline_size;
-
-				scnprintf(buf, sizeof(buf), "%s/cmdline", proc_entry->d_name);
-				cmdline_fd = openat(dirfd(proc_dir), buf, O_RDONLY);
-				if (cmdline_fd == -1)
+			if (link_size == (ssize_t)target_lnk_len &&
+			    !strncmp(buf, target_lnk, target_lnk_len)) {
+				char fdinfo_buf[1024];
+				int fdinfo_fd;
+				ssize_t fdinfo_size;
+				char *line;
+				char *saveptr;
+				u32 perf_event_type = UINT32_MAX;
+				u32 pmu_type = UINT32_MAX;
+
+				/* Let's check the PMU type reserved by this process */
+				scnprintf(buf, sizeof(buf), "%s/fdinfo/%s",
+					  proc_entry->d_name, fd_entry->d_name);
+				fdinfo_fd = openat(dirfd(proc_dir), buf, O_RDONLY);
+				if (fdinfo_fd == -1)
 					continue;
-				cmdline_size = read(cmdline_fd, buf, sizeof(buf) - 1);
-				close(cmdline_fd);
-				if (cmdline_size < 0)
+				fdinfo_size = read(fdinfo_fd, fdinfo_buf, sizeof(fdinfo_buf) - 1);
+				close(fdinfo_fd);
+				if (fdinfo_size < 0)
 					continue;
-				buf[cmdline_size] = '\0';
-				for (ssize_t i = 0; i < cmdline_size; i++) {
-					if (buf[i] == '\0')
-						buf[i] = ' ';
+				fdinfo_buf[fdinfo_size] = '\0';
+
+				line = strtok_r(fdinfo_buf, "\n", &saveptr);
+				while (line) {
+					if (sscanf(line,
+						   "perf_event_attr.type:\t%u",
+						   &perf_event_type) == 1) {
+						/* continue parsing */
+					} else if (sscanf(line,
+							  "pmu_type:\t%u",
+							  &pmu_type) == 1) {
+						/* continue parsing */
+					}
+					line = strtok_r(NULL, "\n", &saveptr);
 				}

-				if (printed == 0)
-					printed += scnprintf(msg, size, "Possible processes:\n");
-
-				printed += scnprintf(msg + printed, size - printed,
-						"%s %s\n", proc_entry->d_name, buf);
-				break;
+				/* Report the process which reserves the conflicted PMU. */
+				/* If fdinfo does not contain PMU type, report it too. */
+				if (perf_event_type == failed_attr->type ||
+				    pmu_type == failed_attr->type ||
+				    (target_pmu_type != UINT32_MAX &&
+				     pmu_type == target_pmu_type) ||
+				    (perf_event_type == UINT32_MAX &&
+				     pmu_type == UINT32_MAX)) {
+					int cmdline_fd;
+					ssize_t cmdline_size;
+
+					scnprintf(buf, sizeof(buf),
+						  "%s/cmdline",
+						  proc_entry->d_name);
+					cmdline_fd = openat(dirfd(proc_dir), buf, O_RDONLY);
+					if (cmdline_fd == -1)
+						continue;
+					cmdline_size = read(cmdline_fd, buf, sizeof(buf) - 1);
+					close(cmdline_fd);
+					if (cmdline_size < 0)
+						continue;
+					buf[cmdline_size] = '\0';
+					for (ssize_t i = 0; i < cmdline_size; i++) {
+						if (buf[i] == '\0')
+							buf[i] = ' ';
+						else if (!isprint((unsigned char)buf[i]))
+							buf[i] = '.';
+					}
+					if (printed == 0)
+						printed += scnprintf(msg, size,
+								     "Possible processes:\n");
+
+					printed += scnprintf(msg + printed, size - printed,
+							"%s %s\n", proc_entry->d_name, buf);
+					break;
+				}
 			}
 		}
 		closedir(fd_dir);
@@ -4285,7 +4338,9 @@ int evsel__open_strerror(struct evsel *evsel, struct target *target,
 			msg, size,
 			"The PMU %s counters are busy and in use by another process.\n",
 			evsel->pmu ? evsel->pmu->name : "");
-		return printed + dump_perf_event_processes(msg + printed, size - printed);
+		return printed + dump_perf_event_processes(evsel,
+							   msg + printed,
+							   size - printed);
 		break;
 	case EINVAL:
 		if (evsel->core.attr.sample_type & PERF_SAMPLE_CODE_PAGE_SIZE && perf_missing_features.code_page_size)
--
2.54.0.1013.g208068f2d8-goog

Re: [PATCH v8 1/2] perf: Reveal PMU type in fdinfo

[Peter Zijlstra]

On Tue, Jun 02, 2026 at 11:13:48AM -0700, Chun-Tse Shao wrote:
> It gives useful info on knowing which PMUs are reserved by this process.
> Also add config which would be useful.
> Testing cycles:
> 
>   $ ./perf stat -e cycles &
>   $ cat /proc/`pidof perf`/fdinfo/3
>   pos:    0
>   flags:  02000002
>   mnt_id: 16
>   ino:    3081
>   perf_event_attr.type:   0
>   perf_event_attr.config: 0x0
>   perf_event_attr.config1:        0x0
>   perf_event_attr.config2:        0x0
>   perf_event_attr.config3:        0x0
>   perf_event_attr.config4:        0x0
> 
> Testing L1-dcache-load-misses:
> 
>   $ ./perf stat -e L1-dcache-load-misses &
>   $ cat /proc/`pidof perf`/fdinfo/3
>   pos:    0
>   flags:  02000002
>   mnt_id: 16
>   ino:    1072
>   perf_event_attr.type:   3
>   perf_event_attr.config: 0x10000
>   perf_event_attr.config1:        0x0
>   perf_event_attr.config2:        0x0
>   perf_event_attr.config3:        0x0
>   perf_event_attr.config4:        0x0
> 
> Reviewed-by: Ian Rogers <irogers@google.com>
> Signed-off-by: Chun-Tse Shao <ctshao@google.com>
> Assisted-by: Gemini:gemini-3.1-pro-preview
> ---

Yeah, I suppose I'll take this. But I'm not sure about the wording here.
PMUs aren't really reserved as such.

[tip: perf/core] perf: Reveal PMU type in fdinfo

[tip-bot2 for Chun-Tse Shao]

The following commit has been merged into the perf/core branch of tip:

Commit-ID:     2369ce0f16b89e3d85c9683c06eb104545999378
Gitweb:        https://git.kernel.org/tip/2369ce0f16b89e3d85c9683c06eb104545999378
Author:        Chun-Tse Shao <ctshao@google.com>
AuthorDate:    Tue, 02 Jun 2026 11:13:48 -07:00
Committer:     Peter Zijlstra <peterz@infradead.org>
CommitterDate: Thu, 04 Jun 2026 11:38:38 +02:00

perf: Reveal PMU type in fdinfo

It gives useful info on knowing which PMUs are reserved by this process.
Also add config which would be useful.
Testing cycles:

  $ ./perf stat -e cycles &
  $ cat /proc/`pidof perf`/fdinfo/3
  pos:    0
  flags:  02000002
  mnt_id: 16
  ino:    3081
  perf_event_attr.type:   0
  perf_event_attr.config: 0x0
  perf_event_attr.config1:        0x0
  perf_event_attr.config2:        0x0
  perf_event_attr.config3:        0x0
  perf_event_attr.config4:        0x0

Testing L1-dcache-load-misses:

  $ ./perf stat -e L1-dcache-load-misses &
  $ cat /proc/`pidof perf`/fdinfo/3
  pos:    0
  flags:  02000002
  mnt_id: 16
  ino:    1072
  perf_event_attr.type:   3
  perf_event_attr.config: 0x10000
  perf_event_attr.config1:        0x0
  perf_event_attr.config2:        0x0
  perf_event_attr.config3:        0x0
  perf_event_attr.config4:        0x0

Signed-off-by: Chun-Tse Shao <ctshao@google.com>
Assisted-by: Gemini:gemini-3.1-pro-preview
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Ian Rogers <irogers@google.com>
Link: https://patch.msgid.link/20260602181349.3969429-1-ctshao@google.com
---
 kernel/events/core.c | 29 +++++++++++++++++++++++++++++
 1 file changed, 29 insertions(+)

diff --git a/kernel/events/core.c b/kernel/events/core.c
index 7935d56..95d806b 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -58,6 +58,7 @@
 #include <linux/percpu-rwsem.h>
 #include <linux/unwind_deferred.h>
 #include <linux/kvm_types.h>
+#include <linux/seq_file.h>
 
 #include "internal.h"
 
@@ -7546,6 +7547,33 @@ static int perf_fasync(int fd, struct file *filp, int on)
 	return 0;
 }
 
+static void perf_show_fdinfo(struct seq_file *m, struct file *f)
+{
+	struct perf_event *event = f->private_data;
+	struct perf_event_context *ctx;
+	struct mutex *child_mutex;
+
+	ctx = perf_event_ctx_lock(event);
+	child_mutex = event->parent ? &event->parent->child_mutex : &event->child_mutex;
+	mutex_lock(child_mutex);
+
+	seq_printf(m, "perf_event_attr.type:\t%u\n", event->orig_type);
+	if (event->pmu)
+		seq_printf(m, "pmu_type:\t%u\n", event->pmu->type);
+	seq_printf(m, "perf_event_attr.config:\t0x%llx\n", (unsigned long long)event->attr.config);
+	seq_printf(m, "perf_event_attr.config1:\t0x%llx\n",
+		   (unsigned long long)event->attr.config1);
+	seq_printf(m, "perf_event_attr.config2:\t0x%llx\n",
+		   (unsigned long long)event->attr.config2);
+	seq_printf(m, "perf_event_attr.config3:\t0x%llx\n",
+		   (unsigned long long)event->attr.config3);
+	seq_printf(m, "perf_event_attr.config4:\t0x%llx\n",
+		   (unsigned long long)event->attr.config4);
+
+	mutex_unlock(child_mutex);
+	perf_event_ctx_unlock(event, ctx);
+}
+
 static const struct file_operations perf_fops = {
 	.release		= perf_release,
 	.read			= perf_read,
@@ -7554,6 +7582,7 @@ static const struct file_operations perf_fops = {
 	.compat_ioctl		= perf_compat_ioctl,
 	.mmap			= perf_mmap,
 	.fasync			= perf_fasync,
+	.show_fdinfo		= perf_show_fdinfo,
 };
 
 /*

Re: [PATCH v8 2/2] perf evsel: Find process with busy PMUs for EBUSY

[Ian Rogers]

On Tue, Jun 2, 2026 at 11:13 AM Chun-Tse Shao <ctshao@google.com> wrote:
>
> It parses fdinfo with PMU type, comparing with the event which failed to
> open, and report the processes causing EBUSY error.
>
> Testing cycles and intel_pt//
>
>   $ ./perf stat -e cycles &
>   [1] 55569
>   $ ./perf stat -e intel_pt// &
>   [2] 55683
>   $ ./perf stat -e intel_pt//
>   Error:
>   The PMU intel_pt counters are busy and in use by another process.
>   Possible processes:
>   55683 ./perf stat -e intel_pt//
>
> Only perf with intel_pt was reported.
>
> Reviewed-by: Ian Rogers <irogers@google.com>
> Signed-off-by: Chun-Tse Shao <ctshao@google.com>
> Assisted-by: Gemini:gemini-3.1-pro-preview

Arnaldo, as the kernel piece of this landed (thanks Peter!) could we
land the tool part too?

Thanks,
Ian

> ---
> v8:
>   - Fix a boundary mismatch bug in strncmp when matching anon_inode
>     symlinks.
>   - Switch to thread-safe strtok_r to prevent use-after-free or static
>     state corruption.
>   - Expand the stack-allocated fdinfo parsing buffer from 256 bytes to
>     1024 bytes to completely prevent truncation before the PMU type is
>     parsed.
>   - Open proc files (fdinfo and cmdline) with O_NONBLOCK to prevent
>     malicious or slow mounts from hanging the perf session indefinitely.
>   - Match EBUSY conflicting processes using the physical PMU type
>     (pmu_type) instead of the requested event type, ensuring legacy/raw
>     events are correctly resolved.
> ---
>  tools/perf/util/evsel.c | 99 ++++++++++++++++++++++++++++++++---------
>  1 file changed, 77 insertions(+), 22 deletions(-)
>
> diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
> index 34c03f47a913..a6b1005db5e8 100644
> --- a/tools/perf/util/evsel.c
> +++ b/tools/perf/util/evsel.c
> @@ -4104,8 +4104,11 @@ static bool find_process(const char *name)
>         return ret ? false : true;
>  }
>
> -static int dump_perf_event_processes(char *msg, size_t size)
> +static int dump_perf_event_processes(const struct evsel *evsel,
> +                                    char *msg, size_t size)
>  {
> +       const struct perf_event_attr *failed_attr = &evsel->core.attr;
> +       u32 target_pmu_type = evsel->pmu ? evsel->pmu->type : UINT32_MAX;
>         DIR *proc_dir;
>         struct dirent *proc_entry;
>         int printed = 0;
> @@ -4136,6 +4139,8 @@ static int dump_perf_event_processes(char *msg, size_t size)
>                         continue;
>                 }
>                 while ((fd_entry = readdir(fd_dir)) != NULL) {
> +                       const char *target_lnk = "anon_inode:[perf_event]";
> +                       size_t target_lnk_len = sizeof("anon_inode:[perf_event]") - 1;
>                         ssize_t link_size;
>
>                         if (fd_entry->d_type != DT_LNK)
> @@ -4144,30 +4149,78 @@ static int dump_perf_event_processes(char *msg, size_t size)
>                         if (link_size < 0)
>                                 continue;
>                         /* Take care as readlink doesn't null terminate the string. */
> -                       if (!strncmp(buf, "anon_inode:[perf_event]", link_size)) {
> -                               int cmdline_fd;
> -                               ssize_t cmdline_size;
> -
> -                               scnprintf(buf, sizeof(buf), "%s/cmdline", proc_entry->d_name);
> -                               cmdline_fd = openat(dirfd(proc_dir), buf, O_RDONLY);
> -                               if (cmdline_fd == -1)
> +                       if (link_size == (ssize_t)target_lnk_len &&
> +                           !strncmp(buf, target_lnk, target_lnk_len)) {
> +                               char fdinfo_buf[1024];
> +                               int fdinfo_fd;
> +                               ssize_t fdinfo_size;
> +                               char *line;
> +                               char *saveptr;
> +                               u32 perf_event_type = UINT32_MAX;
> +                               u32 pmu_type = UINT32_MAX;
> +
> +                               /* Let's check the PMU type reserved by this process */
> +                               scnprintf(buf, sizeof(buf), "%s/fdinfo/%s",
> +                                         proc_entry->d_name, fd_entry->d_name);
> +                               fdinfo_fd = openat(dirfd(proc_dir), buf, O_RDONLY);
> +                               if (fdinfo_fd == -1)
>                                         continue;
> -                               cmdline_size = read(cmdline_fd, buf, sizeof(buf) - 1);
> -                               close(cmdline_fd);
> -                               if (cmdline_size < 0)
> +                               fdinfo_size = read(fdinfo_fd, fdinfo_buf, sizeof(fdinfo_buf) - 1);
> +                               close(fdinfo_fd);
> +                               if (fdinfo_size < 0)
>                                         continue;
> -                               buf[cmdline_size] = '\0';
> -                               for (ssize_t i = 0; i < cmdline_size; i++) {
> -                                       if (buf[i] == '\0')
> -                                               buf[i] = ' ';
> +                               fdinfo_buf[fdinfo_size] = '\0';
> +
> +                               line = strtok_r(fdinfo_buf, "\n", &saveptr);
> +                               while (line) {
> +                                       if (sscanf(line,
> +                                                  "perf_event_attr.type:\t%u",
> +                                                  &perf_event_type) == 1) {
> +                                               /* continue parsing */
> +                                       } else if (sscanf(line,
> +                                                         "pmu_type:\t%u",
> +                                                         &pmu_type) == 1) {
> +                                               /* continue parsing */
> +                                       }
> +                                       line = strtok_r(NULL, "\n", &saveptr);
>                                 }
>
> -                               if (printed == 0)
> -                                       printed += scnprintf(msg, size, "Possible processes:\n");
> -
> -                               printed += scnprintf(msg + printed, size - printed,
> -                                               "%s %s\n", proc_entry->d_name, buf);
> -                               break;
> +                               /* Report the process which reserves the conflicted PMU. */
> +                               /* If fdinfo does not contain PMU type, report it too. */
> +                               if (perf_event_type == failed_attr->type ||
> +                                   pmu_type == failed_attr->type ||
> +                                   (target_pmu_type != UINT32_MAX &&
> +                                    pmu_type == target_pmu_type) ||
> +                                   (perf_event_type == UINT32_MAX &&
> +                                    pmu_type == UINT32_MAX)) {
> +                                       int cmdline_fd;
> +                                       ssize_t cmdline_size;
> +
> +                                       scnprintf(buf, sizeof(buf),
> +                                                 "%s/cmdline",
> +                                                 proc_entry->d_name);
> +                                       cmdline_fd = openat(dirfd(proc_dir), buf, O_RDONLY);
> +                                       if (cmdline_fd == -1)
> +                                               continue;
> +                                       cmdline_size = read(cmdline_fd, buf, sizeof(buf) - 1);
> +                                       close(cmdline_fd);
> +                                       if (cmdline_size < 0)
> +                                               continue;
> +                                       buf[cmdline_size] = '\0';
> +                                       for (ssize_t i = 0; i < cmdline_size; i++) {
> +                                               if (buf[i] == '\0')
> +                                                       buf[i] = ' ';
> +                                               else if (!isprint((unsigned char)buf[i]))
> +                                                       buf[i] = '.';
> +                                       }
> +                                       if (printed == 0)
> +                                               printed += scnprintf(msg, size,
> +                                                                    "Possible processes:\n");
> +
> +                                       printed += scnprintf(msg + printed, size - printed,
> +                                                       "%s %s\n", proc_entry->d_name, buf);
> +                                       break;
> +                               }
>                         }
>                 }
>                 closedir(fd_dir);
> @@ -4285,7 +4338,9 @@ int evsel__open_strerror(struct evsel *evsel, struct target *target,
>                         msg, size,
>                         "The PMU %s counters are busy and in use by another process.\n",
>                         evsel->pmu ? evsel->pmu->name : "");
> -               return printed + dump_perf_event_processes(msg + printed, size - printed);
> +               return printed + dump_perf_event_processes(evsel,
> +                                                          msg + printed,
> +                                                          size - printed);
>                 break;
>         case EINVAL:
>                 if (evsel->core.attr.sample_type & PERF_SAMPLE_CODE_PAGE_SIZE && perf_missing_features.code_page_size)
> --
> 2.54.0.1013.g208068f2d8-goog
>

Re: [PATCH v8 2/2] perf evsel: Find process with busy PMUs for EBUSY

[Arnaldo Carvalho de Melo]

On Tue, Jun 09, 2026 at 08:58:03AM -0700, Ian Rogers wrote:
> On Tue, Jun 2, 2026 at 11:13 AM Chun-Tse Shao <ctshao@google.com> wrote:
> >
> > It parses fdinfo with PMU type, comparing with the event which failed to
> > open, and report the processes causing EBUSY error.
> >
> > Testing cycles and intel_pt//
> >
> >   $ ./perf stat -e cycles &
> >   [1] 55569
> >   $ ./perf stat -e intel_pt// &
> >   [2] 55683
> >   $ ./perf stat -e intel_pt//
> >   Error:
> >   The PMU intel_pt counters are busy and in use by another process.
> >   Possible processes:
> >   55683 ./perf stat -e intel_pt//
> >
> > Only perf with intel_pt was reported.
> >
> > Reviewed-by: Ian Rogers <irogers@google.com>
> > Signed-off-by: Chun-Tse Shao <ctshao@google.com>
> > Assisted-by: Gemini:gemini-3.1-pro-preview
> 
> Arnaldo, as the kernel piece of this landed (thanks Peter!) could we
> land the tool part too?

Thanks for pointing this out to me, I'll do some housekeeping and
process this.

- Arnaldo

Re: [PATCH v8 2/2] perf evsel: Find process with busy PMUs for EBUSY

[Arnaldo Carvalho de Melo]

On Tue, Jun 09, 2026 at 02:20:43PM -0300, Arnaldo Carvalho de Melo wrote:
> On Tue, Jun 09, 2026 at 08:58:03AM -0700, Ian Rogers wrote:
> > On Tue, Jun 2, 2026 at 11:13 AM Chun-Tse Shao <ctshao@google.com> wrote:
> > >
> > > It parses fdinfo with PMU type, comparing with the event which failed to
> > > open, and report the processes causing EBUSY error.
> > >
> > > Testing cycles and intel_pt//
> > >
> > >   $ ./perf stat -e cycles &
> > >   [1] 55569
> > >   $ ./perf stat -e intel_pt// &
> > >   [2] 55683
> > >   $ ./perf stat -e intel_pt//
> > >   Error:
> > >   The PMU intel_pt counters are busy and in use by another process.
> > >   Possible processes:
> > >   55683 ./perf stat -e intel_pt//
> > >
> > > Only perf with intel_pt was reported.
> > >
> > > Reviewed-by: Ian Rogers <irogers@google.com>
> > > Signed-off-by: Chun-Tse Shao <ctshao@google.com>
> > > Assisted-by: Gemini:gemini-3.1-pro-preview
> > 
> > Arnaldo, as the kernel piece of this landed (thanks Peter!) could we
> > land the tool part too?
> 
> Thanks for pointing this out to me, I'll do some housekeeping and
> process this.

I think this hasn't landed, I tried to find the userland part of v8 but
didn't manage to, can you please point it to me?

- Arnaldo

Re: [PATCH v8 2/2] perf evsel: Find process with busy PMUs for EBUSY

[Chun-Tse Shao]

On Sun, Sep 13, 2026 at 6:44 PM Arnaldo Carvalho de Melo
<acme@kernel.org> wrote:
>
> On Tue, Jun 09, 2026 at 02:20:43PM -0300, Arnaldo Carvalho de Melo wrote:
> > On Tue, Jun 09, 2026 at 08:58:03AM -0700, Ian Rogers wrote:
> > > On Tue, Jun 2, 2026 at 11:13 AM Chun-Tse Shao <ctshao@google.com> wrote:
> > > >
> > > > It parses fdinfo with PMU type, comparing with the event which failed to
> > > > open, and report the processes causing EBUSY error.
> > > >
> > > > Testing cycles and intel_pt//
> > > >
> > > >   $ ./perf stat -e cycles &
> > > >   [1] 55569
> > > >   $ ./perf stat -e intel_pt// &
> > > >   [2] 55683
> > > >   $ ./perf stat -e intel_pt//
> > > >   Error:
> > > >   The PMU intel_pt counters are busy and in use by another process.
> > > >   Possible processes:
> > > >   55683 ./perf stat -e intel_pt//
> > > >
> > > > Only perf with intel_pt was reported.
> > > >
> > > > Reviewed-by: Ian Rogers <irogers@google.com>
> > > > Signed-off-by: Chun-Tse Shao <ctshao@google.com>
> > > > Assisted-by: Gemini:gemini-3.1-pro-preview
> > >
> > > Arnaldo, as the kernel piece of this landed (thanks Peter!) could we
> > > land the tool part too?
> >
> > Thanks for pointing this out to me, I'll do some housekeeping and
> > process this.
>
> I think this hasn't landed, I tried to find the userland part of v8 but
> didn't manage to, can you please point it to me?
>
> - Arnaldo

Hi Arnaldo,

Thanks for taking care this. The link to lkml for this userland part
is: https://lore.kernel.org/all/20260602181349.3969429-2-ctshao@google.com/

The kernel part
(https://lore.kernel.org/all/20260602181349.3969429-1-ctshao@google.com/)
was merged by Peter Zijlstra and I can see a tip-bot replied on it.

-CT

Re: [PATCH v8 2/2] perf evsel: Find process with busy PMUs for EBUSY

[Ian Rogers]

On Mon, Sep 14, 2026 at 1:30 PM Chun-Tse Shao <ctshao@google.com> wrote:
>
> On Sun, Sep 13, 2026 at 6:44 PM Arnaldo Carvalho de Melo
> <acme@kernel.org> wrote:
> >
> > On Tue, Jun 09, 2026 at 02:20:43PM -0300, Arnaldo Carvalho de Melo wrote:
> > > On Tue, Jun 09, 2026 at 08:58:03AM -0700, Ian Rogers wrote:
> > > > On Tue, Jun 2, 2026 at 11:13 AM Chun-Tse Shao <ctshao@google.com> wrote:
> > > > >
> > > > > It parses fdinfo with PMU type, comparing with the event which failed to
> > > > > open, and report the processes causing EBUSY error.
> > > > >
> > > > > Testing cycles and intel_pt//
> > > > >
> > > > >   $ ./perf stat -e cycles &
> > > > >   [1] 55569
> > > > >   $ ./perf stat -e intel_pt// &
> > > > >   [2] 55683
> > > > >   $ ./perf stat -e intel_pt//
> > > > >   Error:
> > > > >   The PMU intel_pt counters are busy and in use by another process.
> > > > >   Possible processes:
> > > > >   55683 ./perf stat -e intel_pt//
> > > > >
> > > > > Only perf with intel_pt was reported.
> > > > >
> > > > > Reviewed-by: Ian Rogers <irogers@google.com>
> > > > > Signed-off-by: Chun-Tse Shao <ctshao@google.com>
> > > > > Assisted-by: Gemini:gemini-3.1-pro-preview
> > > >
> > > > Arnaldo, as the kernel piece of this landed (thanks Peter!) could we
> > > > land the tool part too?
> > >
> > > Thanks for pointing this out to me, I'll do some housekeeping and
> > > process this.
> >
> > I think this hasn't landed, I tried to find the userland part of v8 but
> > didn't manage to, can you please point it to me?
> >
> > - Arnaldo
>
> Hi Arnaldo,
>
> Thanks for taking care this. The link to lkml for this userland part
> is: https://lore.kernel.org/all/20260602181349.3969429-2-ctshao@google.com/

Hi CT,

Would it be possible to rebase and resend the tool part as a v9? This
will also trigger the Sashiko review, which would be nice to have.

Thanks,
Ian

> The kernel part
> (https://lore.kernel.org/all/20260602181349.3969429-1-ctshao@google.com/)
> was merged by Peter Zijlstra and I can see a tip-bot replied on it.
>
> -CT

Re: [PATCH v8 2/2] perf evsel: Find process with busy PMUs for EBUSY

[Chun-Tse Shao]

On Tue, Sep 15, 2026 at 11:54 AM Ian Rogers <irogers@google.com> wrote:
>
> On Mon, Sep 14, 2026 at 1:30 PM Chun-Tse Shao <ctshao@google.com> wrote:
> >
> > On Sun, Sep 13, 2026 at 6:44 PM Arnaldo Carvalho de Melo
> > <acme@kernel.org> wrote:
> > >
> > > On Tue, Jun 09, 2026 at 02:20:43PM -0300, Arnaldo Carvalho de Melo wrote:
> > > > On Tue, Jun 09, 2026 at 08:58:03AM -0700, Ian Rogers wrote:
> > > > > On Tue, Jun 2, 2026 at 11:13 AM Chun-Tse Shao <ctshao@google.com> wrote:
> > > > > >
> > > > > > It parses fdinfo with PMU type, comparing with the event which failed to
> > > > > > open, and report the processes causing EBUSY error.
> > > > > >
> > > > > > Testing cycles and intel_pt//
> > > > > >
> > > > > >   $ ./perf stat -e cycles &
> > > > > >   [1] 55569
> > > > > >   $ ./perf stat -e intel_pt// &
> > > > > >   [2] 55683
> > > > > >   $ ./perf stat -e intel_pt//
> > > > > >   Error:
> > > > > >   The PMU intel_pt counters are busy and in use by another process.
> > > > > >   Possible processes:
> > > > > >   55683 ./perf stat -e intel_pt//
> > > > > >
> > > > > > Only perf with intel_pt was reported.
> > > > > >
> > > > > > Reviewed-by: Ian Rogers <irogers@google.com>
> > > > > > Signed-off-by: Chun-Tse Shao <ctshao@google.com>
> > > > > > Assisted-by: Gemini:gemini-3.1-pro-preview
> > > > >
> > > > > Arnaldo, as the kernel piece of this landed (thanks Peter!) could we
> > > > > land the tool part too?
> > > >
> > > > Thanks for pointing this out to me, I'll do some housekeeping and
> > > > process this.
> > >
> > > I think this hasn't landed, I tried to find the userland part of v8 but
> > > didn't manage to, can you please point it to me?
> > >
> > > - Arnaldo
> >
> > Hi Arnaldo,
> >
> > Thanks for taking care this. The link to lkml for this userland part
> > is: https://lore.kernel.org/all/20260602181349.3969429-2-ctshao@google.com/
>
> Hi CT,
>
> Would it be possible to rebase and resend the tool part as a v9? This
> will also trigger the Sashiko review, which would be nice to have.
>
> Thanks,
> Ian

Thank you Ian, please check
lore.kernel.org/20260915204951.2935261-2-ctshao@google.com for the
rebased v9 patch.


-CT

>
> > The kernel part
> > (https://lore.kernel.org/all/20260602181349.3969429-1-ctshao@google.com/)
> > was merged by Peter Zijlstra and I can see a tip-bot replied on it.
> >
> > -CT