From: Adrian Hunter <adrian.hunter@intel.com>
To: Athira Rajeev <atrajeev@linux.ibm.com>, <acme@kernel.org>,
<jolsa@kernel.org>, <maddy@linux.ibm.com>, <irogers@google.com>,
<namhyung@kernel.org>, <linux-perf-users@vger.kernel.org>
Cc: <linuxppc-dev@lists.ozlabs.org>, <aboorvad@linux.ibm.com>,
<sshegde@linux.ibm.com>, <hbathini@linux.vnet.ibm.com>,
<Aditya.Bodkhe1@ibm.com>, <venkat88@linux.ibm.com>
Subject: Re: [PATCH V2 4/6] tools/perf: Allocate and setup aux buffer queue to help co-relate with other events across CPU's
Date: Mon, 15 Sep 2025 18:08:22 +0300 [thread overview]
Message-ID: <3a294b79-165e-4922-b4fc-8d85d65c8004@intel.com> (raw)
In-Reply-To: <20250915072754.99850-5-atrajeev@linux.ibm.com>
On 15/09/2025 10:27, Athira Rajeev wrote:
> When the Dispatch Trace Log data is collected along with other events
> like sched tracepoint events, it needs to be correlated and present
> interleaved along with these events. Perf events can be collected
> parallely across the CPUs. Hence it needs to be ensured events/dtl
> entries are processed in timestamp order.
>
> An auxtrace_queue is created for each CPU. Data within each queue is in
> increasing order of timestamp. Each auxtrace queue has a array/list of
> auxtrace buffers. When processing the auxtrace buffer, the data is
> mmapp'ed. All auxtrace queues is maintained in auxtrace heap. Each queue
> has a queue number and a timestamp. The queues are sorted/added to head
> based on the time stamp. So always the lowest timestamp (entries to be
> processed first) is on top of the heap.
>
> The auxtrace queue needs to be allocated and heap needs to be populated
> in the sorted order of timestamp. The queue needs to be filled with data
> only once via powerpc_vpadtl__update_queues() function.
> powerpc_vpadtl__setup_queues() iterates through all the entries to
> allocate and setup the auxtrace queue. To add to auxtrace heap, it is
> required to fetch the timebase of first entry for each of the queue.
>
> The first entry in the queue for VPA DTL PMU has the boot timebase,
> frequency details which are needed to get timestamp which is required to
> correlate with other events. The very next entry is the actual trace data
> that provides timestamp for occurrence of DTL event. Formula used to get
> the timestamp from dtl entry is:
>
> ((timbase from DTL entry - boot time) / frequency) * 1000000000
>
> powerpc_vpadtl_decode() adds the boot time and frequency as part of
> powerpc_vpadtl_queue structure so that it can be reused. Each of the
> dtl_entry is of 48 bytes size. Sometimes it could happen that one buffer
> is only partially processed (if the timestamp of occurrence of another
> event is more than currently processed element in queue, it will move on
> to next event). In order to keep track of position of buffer, additional
> fields is added to powerpc_vpadtl_queue structure.
>
> Signed-off-by: Athira Rajeev <atrajeev@linux.ibm.com>
> ---
> Changelog:
> Addressed review comments from Adrian
> - Moved time calculation to separate function
>
> tools/perf/util/powerpc-vpadtl.c | 226 ++++++++++++++++++++++++++++++-
> 1 file changed, 223 insertions(+), 3 deletions(-)
>
> diff --git a/tools/perf/util/powerpc-vpadtl.c b/tools/perf/util/powerpc-vpadtl.c
> index 9098cbe00bfd..637abde60f44 100644
> --- a/tools/perf/util/powerpc-vpadtl.c
> +++ b/tools/perf/util/powerpc-vpadtl.c
> @@ -10,6 +10,8 @@
> #include "session.h"
> #include "debug.h"
> #include "powerpc-vpadtl.h"
> +#include "sample.h"
> +#include "tool.h"
>
> /*
> * Structure to save the auxtrace queue
> @@ -38,6 +40,14 @@ struct powerpc_vpadtl_queue {
> struct auxtrace_buffer *buffer;
> struct thread *thread;
> bool on_heap;
> + struct powerpc_vpadtl_entry *dtl;
> + u64 timestamp;
> + unsigned long pkt_len;
> + unsigned long buf_len;
> + u64 boot_tb;
> + u64 tb_freq;
> + unsigned int tb_buffer;
> + unsigned int size;
> bool done;
> pid_t pid;
> pid_t tid;
> @@ -112,6 +122,33 @@ static void powerpc_vpadtl_dump(struct powerpc_vpadtl *vpa __maybe_unused,
> }
> }
>
> +static unsigned long long powerpc_vpadtl_timestamp(struct powerpc_vpadtl_queue *vpaq)
> +{
> + struct powerpc_vpadtl_entry *record = vpaq->dtl;
> + double result, div;
> + double boot_freq;
> + unsigned long long boot_tb;
> + unsigned long long diff;
> + unsigned long long timestamp = 0;
Prettier in descending line length e.g.
struct powerpc_vpadtl_entry *record = vpaq->dtl;
unsigned long long timestamp = 0;
unsigned long long boot_tb;
unsigned long long diff;
double result, div;
double boot_freq;
> +
> + /*
> + * Formula used to get timestamp that can be co-related with
> + * other perf events:
> + * ((timbase from DTL entry - boot time) / frequency) * 1000000000
> + */
> + if (record->timebase) {
> + boot_tb = vpaq->boot_tb;
> + boot_freq = vpaq->tb_freq;
> + diff = be64_to_cpu(record->timebase) - boot_tb;
> + div = diff / boot_freq;
> + result = div;
> + result = result * 1000000000;
> + timestamp = result;
> + }
> +
> + return timestamp;
> +}
> +
> static struct powerpc_vpadtl *session_to_vpa(struct perf_session *session)
> {
> return container_of(session->auxtrace, struct powerpc_vpadtl, auxtrace);
> @@ -124,12 +161,195 @@ static void powerpc_vpadtl_dump_event(struct powerpc_vpadtl *vpa, unsigned char
> powerpc_vpadtl_dump(vpa, buf, len);
> }
>
> +static int powerpc_vpadtl_get_buffer(struct powerpc_vpadtl_queue *vpaq)
> +{
> + struct auxtrace_buffer *buffer = vpaq->buffer;
> + struct auxtrace_queues *queues = &vpaq->vpa->queues;
> + struct auxtrace_queue *queue;
> +
> + queue = &queues->queue_array[vpaq->queue_nr];
> + buffer = auxtrace_buffer__next(queue, buffer);
> +
> + if (!buffer)
> + return 0;
> +
> + vpaq->buffer = buffer;
> + vpaq->size = buffer->size;
> +
> + /* If the aux_buffer doesn't have data associated, try to load it */
> + if (!buffer->data) {
> + /* get the file desc associated with the perf data file */
> + int fd = perf_data__fd(vpaq->vpa->session->data);
> +
> + buffer->data = auxtrace_buffer__get_data(buffer, fd);
> + if (!buffer->data)
> + return -ENOMEM;
> + }
> +
> + vpaq->buf_len = buffer->size;
> +
> + if (buffer->size % dtl_entry_size)
> + vpaq->buf_len = buffer->size - (buffer->size % dtl_entry_size);
> +
> + if (vpaq->tb_buffer != buffer->buffer_nr) {
> + vpaq->pkt_len = 0;
> + vpaq->tb_buffer = 0;
> + }
> +
> + return 1;
> +}
> +
> +/*
> + * The first entry in the queue for VPA DTL PMU has the boot timebase,
> + * frequency details which are needed to get timestamp which is required to
> + * correlate with other events. Save the boot_tb and tb_freq as part of
> + * powerpc_vpadtl_queue. The very next entry is the actual trace data to
> + * be returned.
> + */
> +static int powerpc_vpadtl_decode(struct powerpc_vpadtl_queue *vpaq)
> +{
> + int ret;
> + char *buf;
> + struct boottb_freq *boottb;
> +
> + ret = powerpc_vpadtl_get_buffer(vpaq);
> + if (ret <= 0)
> + return ret;
> +
> + boottb = (struct boottb_freq *)vpaq->buffer->data;
> + if (boottb->timebase == 0) {
> + vpaq->boot_tb = boottb->boot_tb;
> + vpaq->tb_freq = boottb->tb_freq;
> + vpaq->pkt_len += dtl_entry_size;
> + }
> +
> + buf = vpaq->buffer->data;
> + buf += vpaq->pkt_len;
> + vpaq->dtl = (struct powerpc_vpadtl_entry *)buf;
> +
> + vpaq->tb_buffer = vpaq->buffer->buffer_nr;
> + vpaq->buffer = NULL;
> + vpaq->buf_len = 0;
> +
> + return 1;
> +}
> +
> +static struct powerpc_vpadtl_queue *powerpc_vpadtl__alloc_queue(struct powerpc_vpadtl *vpa,
> + unsigned int queue_nr)
> +{
> + struct powerpc_vpadtl_queue *vpaq;
> +
> + vpaq = zalloc(sizeof(*vpaq));
> + if (!vpaq)
> + return NULL;
> +
> + vpaq->vpa = vpa;
> + vpaq->queue_nr = queue_nr;
> +
> + return vpaq;
> +}
> +
> +/*
> + * When the Dispatch Trace Log data is collected along with other events
> + * like sched tracepoint events, it needs to be correlated and present
> + * interleaved along with these events. Perf events can be collected
> + * parallely across the CPUs.
> + *
> + * An auxtrace_queue is created for each CPU. Data within each queue is in
> + * increasing order of timestamp. Allocate and setup auxtrace queues here.
> + * All auxtrace queues is maintained in auxtrace heap in the increasing order
> + * of timestamp. So always the lowest timestamp (entries to be processed first)
> + * is on top of the heap.
> + *
> + * To add to auxtrace heap, fetch the timestamp from first DTL entry
> + * for each of the queue.
> + */
> +static int powerpc_vpadtl__setup_queue(struct powerpc_vpadtl *vpa,
> + struct auxtrace_queue *queue,
> + unsigned int queue_nr)
> +{
> + struct powerpc_vpadtl_queue *vpaq = queue->priv;
> +
> + if (list_empty(&queue->head) || vpaq)
> + return 0;
> +
> + vpaq = powerpc_vpadtl__alloc_queue(vpa, queue_nr);
> + if (!vpaq)
> + return -ENOMEM;
> +
> + queue->priv = vpaq;
> +
> + if (queue->cpu != -1)
> + vpaq->cpu = queue->cpu;
> +
> + if (!vpaq->on_heap) {
> + int ret;
> +retry:
> + ret = powerpc_vpadtl_decode(vpaq);
> + if (!ret)
> + return 0;
> +
> + if (ret < 0)
> + goto retry;
> +
> + vpaq->timestamp = powerpc_vpadtl_timestamp(vpaq);
> +
> + ret = auxtrace_heap__add(&vpa->heap, queue_nr, vpaq->timestamp);
> + if (ret)
> + return ret;
> + vpaq->on_heap = true;
> + }
> +
> + return 0;
> +}
> +
> +static int powerpc_vpadtl__setup_queues(struct powerpc_vpadtl *vpa)
> +{
> + unsigned int i;
> + int ret;
> +
> + for (i = 0; i < vpa->queues.nr_queues; i++) {
> + ret = powerpc_vpadtl__setup_queue(vpa, &vpa->queues.queue_array[i], i);
> + if (ret)
> + return ret;
> + }
> +
> + return 0;
> +}
> +
> +static int powerpc_vpadtl__update_queues(struct powerpc_vpadtl *vpa)
> +{
> + if (vpa->queues.new_data) {
> + vpa->queues.new_data = false;
> + return powerpc_vpadtl__setup_queues(vpa);
> + }
> +
> + return 0;
> +}
> +
> static int powerpc_vpadtl_process_event(struct perf_session *session __maybe_unused,
session is not __maybe_unused
> union perf_event *event __maybe_unused,
> - struct perf_sample *sample __maybe_unused,
> - const struct perf_tool *tool __maybe_unused)
> + struct perf_sample *sample,
> + const struct perf_tool *tool)
> {
> - return 0;
> + struct powerpc_vpadtl *vpa = session_to_vpa(session);
> + int err = 0;
> +
> + if (dump_trace)
> + return 0;
> +
> + if (!tool->ordered_events) {
> + pr_err("VPA requires ordered events\n");
> + return -EINVAL;
> + }
> +
> + if (sample->time) {
> + err = powerpc_vpadtl__update_queues(vpa);
> + if (err)
> + return err;
> + }
> +
> + return err;
> }
>
> /*
next prev parent reply other threads:[~2025-09-15 15:08 UTC|newest]
Thread overview: 18+ messages / expand[flat|nested] mbox.gz Atom feed top
2025-09-15 7:27 [PATCH V2 0/6] perf/tools: Add interface to expose vpa dtl counters via Athira Rajeev
2025-09-15 7:27 ` [PATCH V2 1/6] tools/perf: Add basic CONFIG_AUXTRACE support for VPA pmu on powerpc Athira Rajeev
2025-09-15 15:07 ` Adrian Hunter
2025-09-16 4:37 ` Athira Rajeev
2025-09-15 7:27 ` [PATCH V2 2/6] tools/perf: process auxtrace events and display in perf report -D Athira Rajeev
2025-09-15 15:08 ` Adrian Hunter
2025-09-16 4:37 ` Athira Rajeev
2025-09-15 7:27 ` [PATCH V2 3/6] tools/perf: Add event name as vpa-dtl of PERF_TYPE_SYNTH type to present DTL samples Athira Rajeev
2025-09-15 7:27 ` [PATCH V2 4/6] tools/perf: Allocate and setup aux buffer queue to help co-relate with other events across CPU's Athira Rajeev
2025-09-15 15:08 ` Adrian Hunter [this message]
2025-09-16 4:37 ` Athira Rajeev
2025-09-15 7:27 ` [PATCH V2 5/6] tools/perf: Process the DTL entries in queue and deliver samples Athira Rajeev
2025-09-15 15:08 ` Adrian Hunter
2025-09-15 7:27 ` [PATCH V2 6/6] tools/perf: Enable perf script to present the DTL entries Athira Rajeev
2025-09-15 15:09 ` Adrian Hunter
2025-09-16 4:38 ` Athira Rajeev
2025-09-15 16:07 ` [PATCH V2 0/6] perf/tools: Add interface to expose vpa dtl counters via Venkat
2025-09-18 6:31 ` Athira Rajeev
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=3a294b79-165e-4922-b4fc-8d85d65c8004@intel.com \
--to=adrian.hunter@intel.com \
--cc=Aditya.Bodkhe1@ibm.com \
--cc=aboorvad@linux.ibm.com \
--cc=acme@kernel.org \
--cc=atrajeev@linux.ibm.com \
--cc=hbathini@linux.vnet.ibm.com \
--cc=irogers@google.com \
--cc=jolsa@kernel.org \
--cc=linux-perf-users@vger.kernel.org \
--cc=linuxppc-dev@lists.ozlabs.org \
--cc=maddy@linux.ibm.com \
--cc=namhyung@kernel.org \
--cc=sshegde@linux.ibm.com \
--cc=venkat88@linux.ibm.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).