On 06/09/16 16:33, Goel, Akash wrote:
On 9/6/2016 6:47 PM, Tvrtko Ursulin wrote:
Hi,
On 06/09/16 11:43, akash.goel@xxxxxxxxx wrote:
From: Akash Goel <akash.goel@xxxxxxxxx>
This patch provides a test utility which helps capture GuC firmware
logs and
then dump them to file.
The logs are pulled from a debugfs file
'/sys/kernel/debug/dri/guc_log' and
stored into a file '/tmp/guc_log_dump.dat', the name of the output
file can
be changed through a command line argument.
The utility goes into an infinite loop where it waits for the arrival
of new
logs and as soon as new set of logs are produced it captures them in
its local
buffer which is then flushed out to the file on disk.
Any time when logging needs to be ended, User can stop this utility
(CTRL+C).
Before entering into a loop, it first discards whatever logs are
present in
the debugfs file.
This way User can first launch this utility and then start a
workload/activity
for which GuC firmware logs are to be actually captured and keep
running the
utility for as long as its needed, like once the workload is over this
utility
can be forcefully stopped.
If the logging wasn't enabled on GuC side by the Driver at boot time,
utility
will first enable the logging and later on when it is stopped (CTRL+C)
it will
also pause the logging on GuC side.
Signed-off-by: Akash Goel <akash.goel@xxxxxxxxx>
---
tools/Makefile.sources | 1 +
tools/intel_guc_logger.c | 441
+++++++++++++++++++++++++++++++++++++++++++++++
2 files changed, 442 insertions(+)
create mode 100644 tools/intel_guc_logger.c
diff --git a/tools/Makefile.sources b/tools/Makefile.sources
index 2bb6c8e..be58871 100644
--- a/tools/Makefile.sources
+++ b/tools/Makefile.sources
@@ -19,6 +19,7 @@ tools_prog_lists = \
intel_gpu_time \
intel_gpu_top \
intel_gtt \
+ intel_guc_logger \
intel_infoframes \
intel_l3_parity \
intel_lid \
diff --git a/tools/intel_guc_logger.c b/tools/intel_guc_logger.c
new file mode 100644
index 0000000..92172fa
--- /dev/null
+++ b/tools/intel_guc_logger.c
@@ -0,0 +1,441 @@
+
+#define _GNU_SOURCE /* For using O_DIRECT */
+#include <inttypes.h>
+#include <stdio.h>
+#include <fcntl.h>
+#include <sched.h>
+#include <sys/ioctl.h>
+#include <string.h>
+#include <errno.h>
+#include <signal.h>
+#include <stdlib.h>
+#include <unistd.h>
+#include <sys/poll.h>
+#include <sys/wait.h>
+#include <sys/types.h>
+#include <sys/syscall.h>
+#include <sys/mman.h>
+#include <assert.h>
+#include <pthread.h>
+
+#include "igt.h"
+
+#define MB(x) ((uint64_t)(x) * 1024 * 1024)
+#ifndef PAGE_SIZE
+ #define PAGE_SIZE 4096
+#endif
+#define SUBBUF_SIZE (19*PAGE_SIZE)
+/* Need large buffering from logger side to hide the DISK IO latency,
Driver
+ * can only store 8 snapshots of GuC log buffer in relay.
+ */
+#define NUM_SUBBUFS 100
+
+#define RELAY_FILE_NAME "guc_log"
+#define CONTROL_FILE_NAME "i915_guc_log_control"
+
+char *read_buffer;
+char *out_filename;
+int poll_timeout = 2; /* by default 2ms timeout */
+pthread_mutex_t mutex;
+pthread_t flush_thread;
+int verbosity_level = 3; /* by default capture logs at max verbosity */
+uint32_t produced, consumed;
+uint64_t total_bytes_written;
+int num_buffers = NUM_SUBBUFS;
+int relay_fd, outfile_fd = -1;
+bool stop_logging, discard_oldlogs;
+uint32_t test_duration, max_filesize;
+pthread_cond_t underflow_cond, overflow_cond;
+
+static void guc_log_control(bool enable_logging)
+{
+ int control_fd;
+ char data[19];
+ uint64_t val;
+ int ret;
+
+ control_fd = igt_debugfs_open(CONTROL_FILE_NAME, O_WRONLY);
+ if (control_fd < 0)
+ igt_assert_f(0, "Couldn't open the guc log control file");
+
+ val = enable_logging ? ((verbosity_level << 4) | 0x1) : 0;
+
+ snprintf(data, sizeof(data), "0x%" PRIx64, val);
+ ret = write(control_fd, data, strlen(data) + 1);
Minor: It looks safe like it is but something like below would maybe be
more robust?
ret = snprintf(data, sizeof(data), "0x%" PRIx64, val);
igt_assert(ret > 2 && ret < sizeof(data));
ok will add, but possibility of failure will be really remote here.
but igt_assert(ret > 0) should suffice.
Yes there is no possibility for failure as it stands, just more robust
implementation should someone change something in the future. That's why
I said you could also decide to keep it as is. My version also avoided
the strlen since snprintf already tells you that.
ret = write(control_fd, data, ret);
igt_assert(ret > 0); // assuming short writes can't happen
Up to you.
+ if (ret < 0)
+ igt_assert_f(0, "Couldn't write to the log control file");
+
+ close(control_fd);
+}
+
+static void int_sig_handler(int sig)
+{
+ igt_info("Received signal %d\n", sig);
+
+ stop_logging = true;
+}
+
+static void pull_leftover_data(void)
+{
+ unsigned int bytes_read = 0;
+ int ret;
+
+ while (1) {
+ /* Read the logs from relay buffer */
+ ret = read(relay_fd, read_buffer, SUBBUF_SIZE);
+ if (!ret)
+ break;
+ else if (ret < 0)
+ igt_assert_f(0, "Failed to read from the guc log file");
+ else if (ret < SUBBUF_SIZE)
+ igt_assert_f(0, "invalid read from relay file");
+
+ bytes_read += ret;
+
+ if (outfile_fd > 0) {
= 0 I think. Or is it even needed since open_output_file asserts if it
fails to open?
Actually pull_leftover_data() will be called twice, once before opening
the outfile. This is for purging the old logs, if enabled.
Oh right, OK then.
+ ret = write(outfile_fd, read_buffer, SUBBUF_SIZE);
+ if (ret < 0)
+ igt_assert_f(0, "Couldn't dump the logs in a file");
Do you need to check for short writes as well here? Man page for write
says it can happen with out of disk space or signals.
To keep things simple can just check 'ret < SUBBUF_SIZE' considering
out of disk space as a serious failure or add a loop here which will run
until SUBBUF_SIZE bytes have been written.
Just assert, no point in complicating things.
+ total_bytes_written += ret;
+ }
+ };
+
+ igt_debug("%u bytes flushed\n", bytes_read);
+}
+
+static int num_filled_bufs(void)
+{
+ return (produced - consumed);
+}
+
+static void pull_data(void)
+{
+ char *ptr;
+ int ret;
+
+ pthread_mutex_lock(&mutex);
+ while (num_filled_bufs() >= num_buffers) {
+ igt_debug("overflow, will wait, produced %u, consumed %u\n",
produced, consumed);
+ /* Stall the main thread in case of overflow, as there are no
+ * buffers available to store the new logs, otherwise there
+ * could be corruption if both threads work on the same buffer.
+ */
+ pthread_cond_wait(&overflow_cond, &mutex);
+ };
+ pthread_mutex_unlock(&mutex);
+
+ ptr = read_buffer + (produced % num_buffers) * SUBBUF_SIZE;
+
+ /* Read the logs from relay buffer */
+ ret = read(relay_fd, ptr, SUBBUF_SIZE);
+ if (ret < 0)
+ igt_assert_f(0, "Failed to read from the guc log file");
+ else if (ret && (ret < SUBBUF_SIZE))
+ igt_assert_f(0, "invalid read from relay file");
+
+ pthread_mutex_lock(&mutex);
+ if (ret) {
+ produced++;
+ pthread_cond_signal(&underflow_cond);
+ }
+ pthread_mutex_unlock(&mutex);
Very minor: can move the mutex under the if.
Fine will move under the 'if'.
+}
+
+static void *flusher(void *arg)
+{
+ char *ptr;
+ int ret;
+
+ igt_debug("Execution started of flusher thread\n");
+
+ do {
+ pthread_mutex_lock(&mutex);
+ while (!num_filled_bufs()) {
+ /* Exit only after completing the flush of all the filled
+ * buffers as User would expect that all logs captured up
+ * till the point of interruption/exit are written out to
+ * the disk file.
+ */
+ if (stop_logging) {
+ igt_debug("flusher to exit now\n");
+ pthread_mutex_unlock(&mutex);
+ return NULL;
+ }
+ pthread_cond_wait(&underflow_cond, &mutex);
+ };
+ pthread_mutex_unlock(&mutex);
+
+ ptr = read_buffer + (consumed % num_buffers) * SUBBUF_SIZE;
+
+ ret = write(outfile_fd, ptr, SUBBUF_SIZE);
+ if (ret < 0)
+ igt_assert_f(0, "Couldn't dump the logs in a file");
Question about short writes applies here as well.
Fine will do the same here whatever is decided for above.
+
+ total_bytes_written += ret;
+ if (max_filesize && (total_bytes_written > MB(max_filesize))) {
+ igt_debug("Reached the target of %" PRIu64 " bytes\n",
MB(max_filesize));
+ stop_logging = true;
+ }
+
+ pthread_mutex_lock(&mutex);
+ consumed++;
+ pthread_cond_signal(&overflow_cond);
+ pthread_mutex_unlock(&mutex);
+ } while(1);
Minor again: You use both "while(1) {}" and "do { } while(1)" in here.
Maybe stick with one? :D (I think "for (;;)" is preferred, but maybe I
am misremembering things.)
Sorry for the inconsistency, will clean this up.
+
+ return NULL;
+}
+
+static void init_flusher_thread(void)
+{
+ struct sched_param thread_sched;
+ pthread_attr_t p_attr;
+
+ pthread_cond_init(&underflow_cond, NULL);
+ pthread_cond_init(&overflow_cond, NULL);
+ pthread_mutex_init(&mutex, NULL);
+
+ if (pthread_attr_init(&p_attr))
+ igt_assert_f(0, "Error obtaining default thread attributes");
+
+ if (pthread_attr_setinheritsched(&p_attr, PTHREAD_EXPLICIT_SCHED))
+ igt_assert_f(0, "Couldn't set inheritsched");
+
+ if (pthread_attr_setschedpolicy(&p_attr, SCHED_RR))
+ igt_assert_f(0, "Couldn't set thread scheduling policy");
+
+ /* Keep the flusher task also at rt priority, so that it doesn't
get
+ * too late in flushing the collected logs in local buffers to
the disk,
+ * and so main thread always have spare buffers to collect the
logs.
+ */
+ thread_sched.sched_priority = 5;
+ if (pthread_attr_setschedparam(&p_attr, &thread_sched))
+ igt_assert_f(0, "Couldn't set thread priority");
+
+ if (pthread_create(&flush_thread, &p_attr, flusher, NULL))
+ igt_assert_f(0, "thread creation failed");
More nitpicks: You mix starting error messages with upper case and lower
case.
Will use the lower case for all error messages. fine ?
Fine with me.
+
+ if (pthread_attr_destroy(&p_attr))
+ igt_assert_f(0, "Error destroying thread attributes");
+}
+
+static void open_relay_file(void)
+{
+ relay_fd = igt_debugfs_open(RELAY_FILE_NAME, O_RDONLY);
+ if (relay_fd < 0)
+ igt_assert_f(0, "Couldn't open the guc log file");
+
+ /* Purge the old/boot-time logs from the relay buffer.
+ * This is more for Val team's requirement, where they have to
first
+ * purge the existing logs before starting the tests for which
the logs
+ * are actually needed. After this logger will enter into a loop
and
+ * wait for the new data, at that point benchmark can be launched
from
+ * a different shell.
+ */
+ if (discard_oldlogs)
+ pull_leftover_data();
+}
+
+static void open_output_file(void)
+{
+ char *filename;
+
+ if (asprintf(&filename, "/tmp/%s", out_filename ? :
"guc_log_dump.dat") < 0)
+ igt_assert_f(0, "Couldn't allocate the filename");
I think you need to take the out_filename when specified by the user
as-is and not stick it in /tmp. It is perfectly reasonable that someone
would want to put it somewhere else.
Ok so User should provide the full path.
It can be relative, absolute, or just a filename. My suggestion was to
just use what the user provided for the open(2) call.
Also better to move the default name of "guc_log_dump.dat" to where
defaults like RELAY_FILE_NAME and such are defined.
fine.
And I would not default to /tmp but the current directory.
Chris had suggested to use /tmp. Fine will use the current directory.
+
+ /* Use Direct IO mode for the output file, as the data written is
not
+ * supposed to be accessed again, this saves a copy of data from
App's
+ * buffer to kernel buffer (Page cache). Due to no buffering on
kernel
+ * side, data is flushed out to disk faster and more buffering
can be
+ * done on the logger side to hide the disk IO latency.
+ */
+ outfile_fd = open(filename, O_CREAT | O_WRONLY | O_TRUNC |
O_DIRECT, 0444);
0400 or 0440 would be better I think.
fine, this might make the file inaccessible to Root User in case the
logger wasn't running as a Root ?
I assumed logger will run as root and it would be preferable not to have
log files accessible by anyone. Either way root can read anything
regardless of permissions, no?
+ if (outfile_fd < 0)
+ igt_assert_f(0, "Couldn't open the output file");
+
+ free(out_filename);
+ free(filename);
+}
+
+static void init_main_thread(void)
+{
+ struct sched_param thread_sched;
+ /* Run the main thread at highest priority to ensure that it always
+ * gets woken-up at earliest on arrival of new data and so is
always
+ * ready to pull the logs, otherwise there could be loss logs if
+ * GuC firmware is generating logs at a very high rate.
+ */
+ thread_sched.sched_priority = 1;
+ if (sched_setscheduler(getpid(), SCHED_FIFO, &thread_sched))
+ igt_assert_f(0, "Couldn't set the priority");
+
+ if (signal(SIGINT, int_sig_handler) == SIG_ERR)
+ igt_assert_f(0, "SIGINT handler registration failed");
+
+ /* Need an aligned pointer for direct IO */
+ if (posix_memalign((void **)&read_buffer, PAGE_SIZE, num_buffers
* SUBBUF_SIZE))
+ igt_assert_f(0, "Couldn't allocate the read buffer");
+
+ /* Keep the pages locked in RAM, avoid page fault overhead */
+ if (mlock(read_buffer, num_buffers * SUBBUF_SIZE))
+ igt_assert_f(0, "Failed to lock memory");
+
+ /* Enable the logging, it may not have been enabled from boot
and so
+ * the relay file also wouldn't have been created.
+ */
+ guc_log_control(true);
+
+ open_relay_file();
+ open_output_file();
+}
+
+static int parse_options(int opt, int opt_index, void *data)
+{
+ igt_debug("opt %c optarg %s\n", opt, optarg);
+
+ switch(opt) {
+ case 'v':
+ verbosity_level = atoi(optarg);
+ if (verbosity_level < 0)
+ verbosity_level = 0;
+ else if (verbosity_level > 3)
+ verbosity_level = 3;
Notify user if invalid value was provided?
Just notify but don't abort and continue by using a valid value ?
No strong opinion. Either is much better than silently using a different
value or ignoring the fact user has typed in garbage on the command line.
+
+ igt_debug("verbosity level to be used is %d\n",
verbosity_level);
+ break;
+ case 'o':
+ if (asprintf(&out_filename, "%s", optarg) < 0)
strdup?
ok will try to use strdup.
+ igt_assert_f(0, "Couldn't allocate the o/p filename");
+ igt_debug("logs to be stored in file /tmp/%s\n", out_filename);
+ break;
+ case 'b':
+ num_buffers = atoi(optarg);
+ if (num_buffers == 0)
+ num_buffers = NUM_SUBBUFS;
Again notify if invalid number/value.
same doubt as above.
+ igt_debug("number of buffers to be used is %d\n", num_buffers);
+ break;
+ case 't':
+ test_duration = atoi(optarg);
+ if (test_duration > 0)
If switch was specified it should have a valid optarg or error reported.
+ igt_debug("logger to run for %d seconds\n", test_duration);
+ break;
+ case 'p':
+ poll_timeout = atoi(optarg);
+ if (poll_timeout == 0)
+ poll_timeout = 2; /* at least use 2ms poll timeout */
Hm I would prefer if you allowed no poll period.
by no poll period, do you mean indefinite wait ?
That can be done by specifying -1 as a timeout value in the command line,
"-p --polltimeout=ms polling timeout in ms, -1 == indefinite wait for
the new data\n"
My bad, I missed that. All good then.
I even think it should be the default if we think relayfs will be
definitely fixed to not
require it.
Can keep the indefinite wait as a default.
In any case, commend is wrong since user can specify 1ms and get away
with it. :)
+ if (poll_timeout > 0)
+ igt_debug("polling to be done with %d milli seconds
timeout\n", poll_timeout);
milliseconds
+ break;
+ case 's':
+ max_filesize = atoi(optarg);
+ if (max_filesize > 0)
Same as for -t.
+ igt_debug("max allowed size of the output file is %d
MBs\n", max_filesize);
+ break;
+ case 'd':
+ discard_oldlogs = true;
+ igt_debug("old/boot-time logs will be discarded\n");
+ break;
+ }
+
+ return 0;
+}
+
+static void process_command_line(int argc, char **argv)
+{
+ static struct option long_options[] = {
+ {"verbosity", required_argument, 0, 'v'},
+ {"outputfilename", required_argument, 0, 'o'},
+ {"buffers", required_argument, 0, 'b'},
+ {"testduration", required_argument, 0, 't'},
+ {"polltimeout", required_argument, 0, 'p'},
+ {"size", required_argument, 0, 's'},
+ {"discard", no_argument, 0, 'd'},
+ { 0, 0, 0, 0 }
+ };
+
+ const char *help =
+ " -v --verbosity=level verbosity level of GuC logging
(0-3)\n"
+ " -o --outfile=name name of the output file in /tmp
directory, where logs will be stored\n"
+ " -b --buffers=num number of buffers to be maintained
on logger side for storing logs\n"
+ " -t --testduration=sec max duration in seconds for which
the logger should run, 0 == forever\n"
+ " -p --polltimeout=ms polling timeout in ms, -1 ==
indefinite wait for the new data\n"
+ " -s --size=MB max size of output file in MBs
after which logging will be stopped, 0 == no limit\n"
+ " -d --discard discard the old/boot-time logs
before entering into the capture loop\n";
+
+ igt_simple_init_parse_opts(&argc, argv, "v:o:b:t:p:s:d",
long_options,
+ help, parse_options, NULL);
+}
+
+int main(int argc, char **argv)
+{
+ struct pollfd relay_poll_fd;
+ struct timespec start={};
+ int nfds;
+ int ret;
+
+ process_command_line(argc, argv);
+
+ init_main_thread();
+
+ /* Use a separate thread for flushing the logs to a file on disk.
+ * Main thread will buffer the data from relay file in its pool of
+ * buffers and other thread will flush the data to disk in
background.
+ * This is needed, albeit by default data is written out to disk in
+ * async mode, as when there are too many dirty pages in the RAM,
+ * (/proc/sys/vm/dirty_ratio), kernel starts blocking the processes
+ * doing the file writes.
+ */
+ init_flusher_thread();
+
+ relay_poll_fd.fd = relay_fd;
+ relay_poll_fd.events = POLLIN;
+ relay_poll_fd.revents = 0;
+
+ nfds = 1; /* only one fd to poll */
+
+ while (!stop_logging)
+ {
+ if (test_duration && (igt_seconds_elapsed(&start) >
test_duration)) {
If you agree to allow no poll period the this would not work right? In
that case you would need to use alarm(2) or something.
Can calculate the timeout value for poll call as,
if (poll_timeout < 0) {
timeout = test_duration - igt_seconds_elapsed(&start))
}
My point was that with indefinite poll loop will not run if there is not
log data so timeout will not work implemented like this.
+ igt_debug("Ran for stipulated %d seconds, exit now\n",
test_duration);
+ stop_logging = true;
+ break;
+ }
+
+ /* Wait/poll for the new data to be available, relay doesn't
+ * provide a blocking read.
+ * Need to do polling with a timeout instead of indefinite
wait,
+ * to avoid relying on relay for the wakeup, as relay does
wakeup
+ * in a deferred manner on jiffies granularity by scheduling a
+ * timer and moreover that timer is re-scheduled on every newly
+ * produced buffer so timer keeps getting pushed out if there
+ * are multiple flush interrupts in a very quick succession
(less
+ * than a jiffy gap between 2 flush interrupts) and relay runs
+ * out of sub buffers to store the new logs.
+ */
+ ret = poll(&relay_poll_fd, nfds, poll_timeout);
+ if (ret < 0) {
+ if (errno == EINTR)
+ break;
+ else
+ igt_assert_f(0, "poll call failed");
+ }
+
+ /* No data available yet, poll again, hopefully new data is
round the corner */
+ if (!relay_poll_fd.revents)
+ continue;
+
+ pull_data();
+ }
+
+ /* Pause logging on the GuC side*/
+ guc_log_control(false);
+ pthread_cond_signal(&underflow_cond);
+ pthread_join(flush_thread, NULL);
+ pull_leftover_data();
+ igt_info("total_bytes_written %" PRIu64 "\n", total_bytes_written);
+
+ free(read_buffer);
+ close(relay_fd);
+ close(outfile_fd);
+ igt_exit();
+}
So only minor comments. The rest looks reasonable to me and I trust you
tested it well and it works fine.
Thanks much for the thorough review.
Np.
Regards,
Tvrtko
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
