| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <CAM9d7cgPpeCyjd2Z3Ld6yya7BCE3KxVg0P4xeF2gDZ0Y5mw2HQ@mail.gmail.com>
Date: Wed, 27 Jan 2021 16:37:51 +0900
From: Namhyung Kim <namhyung@...nel.org>
To: Bixuan Cui <cuibixuan@...wei.com>
Cc: Peter Zijlstra <peterz@...radead.org>,
Ingo Molnar <mingo@...hat.com>,
Arnaldo Carvalho de Melo <acme@...nel.org>,
Mark Rutland <mark.rutland@....com>,
Alexander Shishkin <alexander.shishkin@...ux.intel.com>,
Jiri Olsa <jolsa@...hat.com>,
linux-kernel <linux-kernel@...r.kernel.org>,
john.wanghui@...wei.com, Alexei Budankov <abudankov@...wei.com>,
yz.yuzhou@...wei.com
Subject: Re: [PATCH v3 1/2] perf tools: add 'perf irq' to measure the hardware interrupts
Hello,
On Sat, Jan 16, 2021 at 10:20 AM Bixuan Cui <cuibixuan@...wei.com> wrote:
>
> Add 'perf irq' to trace/measure the hardware interrupts.
>
> Now three functions are provided:
> 1. 'perf irq record <command>' to record the irq handler events.
> 2. 'perf irq script' to see a detailed trace of the workload that
> was recorded.
> 3. 'perf irq report' to calculate the time consumed by each
> hardware interrupt processing function.
>
> Signed-off-by: Bixuan Cui <cuibixuan@...wei.com>
> ---
> tools/perf/Build | 1 +
> tools/perf/builtin-irq.c | 283 +++++++++++++++++++++++++++++++++++++++
> tools/perf/builtin.h | 1 +
> tools/perf/perf.c | 1 +
> 4 files changed, 286 insertions(+)
> create mode 100644 tools/perf/builtin-irq.c
>
> diff --git a/tools/perf/Build b/tools/perf/Build
> index 5f392dbb88fc..d52a1e1d6d8a 100644
> --- a/tools/perf/Build
> +++ b/tools/perf/Build
> @@ -24,6 +24,7 @@ perf-y += builtin-mem.o
> perf-y += builtin-data.o
> perf-y += builtin-version.o
> perf-y += builtin-c2c.o
> +perf-y += builtin-irq.o
>
> perf-$(CONFIG_TRACE) += builtin-trace.o
> perf-$(CONFIG_LIBELF) += builtin-probe.o
> diff --git a/tools/perf/builtin-irq.c b/tools/perf/builtin-irq.c
> new file mode 100644
> index 000000000000..25ba0669a875
> --- /dev/null
> +++ b/tools/perf/builtin-irq.c
[SNIP]
> +
> +#define IRQ_NAME_LEN 20
> +#define MAX_CPUS 4096
> +
> +static const char *cpu_list;
> +static DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS);
> +
> +struct perf_irq;
Seems unnecessary.
> +
> +struct perf_irq {
> + struct perf_tool tool;
> + bool force;
> +
> + u32 irq_entry_irq;
> + char irq_name[IRQ_NAME_LEN];
> + u32 cpu;
> + u64 irq_entry_time;
> + u32 irq_entry_pid;
> + u32 irq_exit_irq;
> + u64 irq_exit_time;
> + u32 irq_exit_pid;
> +};
> +
> +typedef int (*irq_handler)(struct perf_tool *tool,
> + union perf_event *event,
> + struct evsel *evsel,
> + struct perf_sample *sample,
> + struct machine *machine);
You don't need to pass all the arguments if unused.
> +
> +static int perf_report_process_sample(struct perf_tool *tool,
> + union perf_event *event,
> + struct perf_sample *sample,
> + struct evsel *evsel,
> + struct machine *machine)
> +{
> + int err = 0;
> +
> + if (evsel->handler != NULL) {
> + irq_handler f = evsel->handler;
> + err = f(tool, event, evsel, sample, machine);
> + }
> +
> + return err;
> +}
> +
> +static void output_report(struct perf_irq *irq)
> +{
> + int ret, i;
> + char irq_entry_time[30], irq_exit_time[30], irq_diff[30];
> +
> + /* The entry and exit of the hardware irq function
> + * exist at the same time. Check it by irq and pid.
> + */
> + if (irq->irq_entry_pid != irq->irq_exit_pid ||
> + irq->irq_entry_irq != irq->irq_exit_irq)
> + return;
Is there only a single instance of the perf_irq here?
Then I don't think this is correct and you should keep
pairs of irq entry/exit per cpu. Otherwise overlapped
irqs from different cpus will be discarded (wrongly).
> +
> + timestamp__scnprintf_usec(irq->irq_entry_time,
> + irq_entry_time, sizeof(irq_entry_time));
> + timestamp__scnprintf_usec(irq->irq_exit_time,
> + irq_exit_time, sizeof(irq_exit_time));
> + timestamp__scnprintf_usec(irq->irq_exit_time - irq->irq_entry_time,
> + irq_diff, sizeof(irq_diff));
> +
> + ret = printf(" %s ", irq->irq_name);
> + for (i = 0; i < IRQ_NAME_LEN - ret; i++)
> + printf(" ");
> +
> + printf("| [%04d] | %13s s | %16s s | %16s s\n",
> + irq->cpu, irq_diff, irq_entry_time, irq_exit_time);
> +}
> +
> +static int report_irq_handler_entry_event(struct perf_tool *tool,
> + union perf_event *event __maybe_unused,
> + struct evsel *evsel,
> + struct perf_sample *sample,
> + struct machine *machine __maybe_unused)
> +{
> + int err = 0;
> + struct perf_irq *irq = container_of(tool, struct perf_irq, tool);
> +
> + const char *name = evsel__strval(evsel, sample, "name");
> +
> + irq->irq_entry_pid = evsel__intval(evsel, sample, "pid");
> + irq->irq_entry_irq = evsel__intval(evsel, sample, "irq");
> + irq->irq_entry_time = sample->time;
> + strncpy(irq->irq_name, name, IRQ_NAME_LEN);
Note that strncpy doesn't guarantee the NUL-termination.
You'd better do it by yourself just in case.
Thanks,
Namhyung
> +
> + return err;
> +}
> +
> +static int report_irq_handler_exit_event(struct perf_tool *tool,
> + union perf_event *event __maybe_unused,
> + struct evsel *evsel,
> + struct perf_sample *sample,
> + struct machine *machine __maybe_unused)
> +{
> + int err = 0;
> + struct perf_irq *irq = container_of(tool, struct perf_irq, tool);
> +
> + irq->irq_exit_pid = evsel__intval(evsel, sample, "pid");
> + irq->irq_exit_irq = evsel__intval(evsel, sample, "irq");
> + irq->irq_exit_time = sample->time;
> + irq->cpu = sample->cpu;
> +
> + if (cpu_list && !test_bit(sample->cpu, cpu_bitmap))
> + return err;
> +
> + output_report(irq);
> +
> + return err;
> +}
> +
Powered by blists - more mailing lists