[PATCH] test/intel/xe_pmt: Add testing for BMG crashlog

Ruhl, Michael J michael.j.ruhl at intel.com
Mon Jun 30 15:42:29 UTC 2025


>-----Original Message-----
>From: Kamil Konieczny <kamil.konieczny at linux.intel.com>
>Sent: Monday, June 30, 2025 6:13 AM
>To: Ruhl, Michael J <michael.j.ruhl at intel.com>
>Cc: igt-dev at lists.freedesktop.org; De Marchi, Lucas
><lucas.demarchi at intel.com>; Vivi, Rodrigo <rodrigo.vivi at intel.com>;
>david.e.box at linux.intel.com
>Subject: Re: [PATCH] test/intel/xe_pmt: Add testing for BMG crashlog
>
>Hi Michael,
>On 2025-06-27 at 16:48:42 -0400, Michael J. Ruhl wrote:
>> The BMG devices has the PMT crashlog feature. If the devices present
>> is a BMG, test PMT api.
>>
>> NOTE: the testing order is not flexible and must be done in
>> the currently specified order.
>>
>> Signed-off-by: Michael J. Ruhl <michael.j.ruhl at intel.com>
>> ---
>>  tests/intel/xe_pmt.c | 561
>+++++++++++++++++++++++++++++++++++++++++++
>>  tests/meson.build    |   1 +
>>  2 files changed, 562 insertions(+)
>>  create mode 100644 tests/intel/xe_pmt.c
>>
>> diff --git a/tests/intel/xe_pmt.c b/tests/intel/xe_pmt.c
>> new file mode 100644
>> index 000000000..21a2e74e7
>> --- /dev/null
>> +++ b/tests/intel/xe_pmt.c
>> @@ -0,0 +1,561 @@
>> +// SPDX-License-Identifier: MIT
>> +/*
>> + * Copyright © 2025 Intel Corporation
>> + */
>> +
>> +/**
>> + * TEST: Verify BMG PMT files operations
>
>Please remove 'BMG' from test description here, as this could be
>used for future GPUs. Also write full name for 'PMT', my guess
>it is "Power Menagement Telemetry" but I could be wrong here.

Will do.

>Second note - this is for crash logs, what about PM telemetry
>for GPU? Could it also be tested?

A majority of this is for the crashlog functionality.  The telemetry part is just verifying that the
correct directories/files are in place.

>> + * Category: Core
>> + * Mega feature: General Core features
>> + * Sub-category: uapi
>> + * Functionality: sysfs
>> + * Description: Verify BMG PMT files are created and are accessable
>
>Same here, remove 'BMG'
>> + */
>> +
>> +#include <dirent.h>
>> +#include <limits.h>
>> +#include <string.h>
>> +#include <fcntl.h>
>
>Move 'fcntl.h' to proper place.
>
>> +#include <unistd.h>
>
>You could place it as first one (this is an exception), up to
>you.
>
>> +
>> +#include "igt.h"
>> +#include "igt_sysfs.h"
>> +#include "linux_scaffold.h"
>> +#include "xe_drm.h"
>> +#include "xe/xe_ioctl.h"
>> +#include "xe/xe_query.h"
>> +
>> +/* base directory names */
>> +#define VSEC_CRASHLOG_DIR "intel_vsec.crashlog."
>> +#define VSEC_TELEMETRY_DIR "intel_vsec.telemetry."
>> +#define CRASHLOG_DIR "crashlog"
>> +#define TELEMETRY_DIR "telem"
>> +#define BMG_CRASHLOG_CNT 2
>> +#define BMG_TELEMETRY_CNT 2
>
>Why 'BMG_' prefix here?

This define was replace by the enum below.  Should have been removed.

>> +
>> +enum bmg_crashlog_instances {
>> +	bmg_crashlog_punit = 0,
>> +	bmg_crashlog_oobmsm,
>> +	bmg_crashlog_max
>> +};
>
>What is "oobmsm" shortcut?

PMT allows you to define multiple crashlog and telemetry instances in a device.  For BMG, there are two
instances each of crashlog and telemetry in the device.  PUNIT and OOBMSM.  I thought this would be
more descriptive than index = 0 and index = 1...

>> +
>> +enum bmg_telemety_instances {
>> +	bmg_telemetry_punit = 0,
>> +	bmg_telemetry_oobmsm,
>> +	bmg_telemetry_max
>> +};
>> +
>> +static char dev_path[PATH_MAX];
>> +static char work_path[PATH_MAX * 2];
>> +
>> +/*
>> + * In most case there should be a single instance of crashlog an telemetry
>directories.
>
>'an'? Did you mean 'in'?

And...

If the DVSEC entries for the PMT instance are consecutive, there is one sysfs directory
that contains both entries.

<pci device>/intel_vsec.crashlog.<x>

Where <x> is the PMT increment.  In this directory you should see:

<pci device>/intel_vsec.crashlog.<x>/intel_pmt/crashlog1
<pci device>/intel_vsec.crashlog.<x>/intel_pmt/crashlog2

(two instances of crashlog).

The current definition for the DVSEC entry in the Xe driver has two separate entries so you get this:

<pci device>/intel_vsec.crashlog.<x>/intel_pmt/crashlog1
<pci device>/intel_vsec.crashlog.<y>/intel_pmt/crashlog1

(I have posted a fix for the Xe driver to "fix" this).

https://patchwork.freedesktop.org/patch/661321/?series=149120&rev=5

So the test here is to verify that the expected directory structure is in place.

>
>> + * If DVSEC entries are separate the structure will be different.
>> + */
>> +static char crashlog_vsec_dir[32];
>> +static char telemetry_vsec_dir[32];
>> +
>> +/* This needs to be specific for each supported device */
>> +static char crashlog_dir[bmg_crashlog_max][32];
>> +static char telemetry_dir[bmg_telemetry_max][32];
>> +
>> +/* telemetry file names */
>> +static const char *telem = "telem";
>> +
>> +/* crashlog filenames and descriptors */
>> +static const char *clear = "clear";
>> +static const char *consumed = "consumed";
>> +static const char *crashlog = "crashlog";
>> +static const char *enable = "enable";
>> +static const char *error = "error";
>> +static const char *dev_guid = "guid";
>> +static const char *rearm = "rearm";
>> +static const char *trigger = "trigger";
>> +
>> +struct crashlog_v2_info {
>> +	int clear_fd;
>> +	int consumed_fd;
>> +	int crashlog_fd;
>> +	int enable_fd;
>> +	int error_fd;
>> +	int guid_fd;
>> +	int rearm_fd;
>> +	int trigger_fd;
>> +	u_int32_t guid;
>> +} bmg_info[bmg_crashlog_max];
>> +
>> +#define DEV_PATH_LEN 80
>> +
>> +/**
>> + * device_sysfs_path:
>> + * @fd: opened device file descriptor
>> + * @path: buffer to store sysfs path to device directory
>> + *
>> + * Returns:
>> + * On successfull path resolution sysfs path to device directory,
>> + * NULL otherwise
>> + */
>> +static char *device_sysfs_path(int fd, char *path)
>> +{
>> +        char sysfs[DEV_PATH_LEN];
>> +
>> +        if (!igt_sysfs_path(fd, sysfs, sizeof(sysfs)))
>> +                return NULL;
>> +
>> +        if (DEV_PATH_LEN <= (strlen(sysfs) + strlen("/device")))
>> +                return NULL;
>> +
>> +        strcat(sysfs, "/device");
>> +
>> +        return realpath(sysfs, path);
>> +}
>> +
>> +/**
>> + * SUBTEST: test_pmt_directories
>> + * BMG PMT directory structure:
>> + * device/intel_vsec.crashlog.x/intel_pmt/crashlog<a,b>
>> + * device/intel_vsec.telemetry.x/intel_pmt/telemetry<c,d>
>> + *
>> + * If this is done for a different platform, this could be
>> + * different.
>> + *
>> + */
>> +static void test_pmt_directories(int dev_fd)
>> +{
>> +	struct dirent *ent;
>> +	int index;
>> +	DIR *dir;
>> +
>> +        igt_assert(device_sysfs_path(dev_fd, dev_path));
>> +
>> +	/* verify top level PMT directories */
>> +	dir = opendir(dev_path);
>> +	igt_assert_f(dir, "no directories found\n");
>
>Why not igt_require_f()?

Will update.

>> +
>> +	while ((ent = readdir(dir)) != NULL) {
>> +		if (strncmp(VSEC_CRASHLOG_DIR, ent->d_name,
>sizeof(VSEC_CRASHLOG_DIR) - 1) == 0)
>> +			strcpy(crashlog_vsec_dir, ent->d_name);
>> +		if (strncmp(VSEC_TELEMETRY_DIR, ent->d_name,
>sizeof(VSEC_TELEMETRY_DIR) - 1) == 0)
>> +			strcpy(telemetry_vsec_dir, ent->d_name);
>> +	}
>> +
>> +	closedir(dir);
>> +
>> +	igt_assert_f(strlen(crashlog_vsec_dir), "missing crashlog directory\n");
>> +	igt_assert_f(strlen(telemetry_vsec_dir), "missing telemetry
>directory\n");
>> +
>> +	/* verify crashlog directory structure */
>> +	sprintf(work_path, "%s/%s/%s", dev_path, crashlog_vsec_dir,
>"intel_pmt");
>> +
>> +	dir = opendir(work_path);
>> +	igt_assert_f(dir, "no intel_pmt directories found\n");
>> +
>> +	index = 0;
>> +	/* find the crashlog<x> directory instances */
>> +	while ((ent = readdir(dir)) != NULL) {
>> +		if (strncmp(CRASHLOG_DIR, ent->d_name,
>sizeof(CRASHLOG_DIR) - 1) == 0) {
>> +			if (index < bmg_crashlog_max)
>> +				strcpy(crashlog_dir[index], ent->d_name);
>> +			index++;
>> +		}
>> +	}
>> +
>> +	closedir(dir);
>> +
>> +	igt_assert_f(index == bmg_crashlog_max, "too many crashlog entries
>%d\n", index);
>> +	for (int i = 0; i < ARRAY_SIZE(crashlog_dir); i++)
>> +		igt_assert_f(strlen(crashlog_dir[i]), "missing crashlog[%d]
>directory\n", i);
>> +
>> +	/* verify telemetry directory structure */
>> +	sprintf(work_path, "%s/%s/%s", dev_path, telemetry_vsec_dir,
>"intel_pmt");
>> +
>> +	dir = opendir(work_path);
>> +	igt_assert_f(dir, "no telemetry intel_pmt directories found\n");
>> +
>> +	index = 0;
>> +	while ((ent = readdir(dir)) != NULL) {
>> +		if (strncmp(TELEMETRY_DIR, ent->d_name,
>sizeof(TELEMETRY_DIR) - 1) == 0) {
>> +			if (index < BMG_TELEMETRY_CNT)
>> +				strcpy(telemetry_dir[index], ent->d_name);
>> +			index++;
>> +		}
>> +	}
>> +
>> +	closedir(dir);
>> +
>> +	igt_assert_f(index == bmg_telemetry_max, "too many telemetry entries
>%d\n", index);
>> +	for (int i = 0; i < ARRAY_SIZE(telemetry_dir); i++)
>> +		igt_assert_f(strlen(telemetry_dir[i]), "missing telemetry[%d]
>directory\n", i);
>> +
>> +}
>> +
>> +static void find_pmt_file(const char *path, const char *file)
>> +{
>> +	struct dirent *ent;
>> +	bool found;
>> +	DIR *dir;
>> +
>> +	dir = opendir(path);
>> +	igt_assert_f(dir, "no intel_pmt directories found\n");
>> +
>> +	found = false;
>> +	while ((ent = readdir(dir)) != NULL)
>> +		if (strcmp(file, ent->d_name) == 0)
>> +			found = true;
>> +	closedir(dir);
>> +
>> +	igt_assert_f(found, "missing %s from %s\n", file, path);
>> +}
>> +
>> +static void open_pmt_file(const char *path, const char *file, int *fd, int flags)
>> +{
>> +	char file_path[PATH_MAX];
>> +
>> +	sprintf(file_path, "%s/%s", path, file);
>> +
>> +	*fd = open(file_path, flags);
>> +	igt_assert_f(*fd > -1, "failed to open %s\n", file_path);
>> +
>> +	/* TODO: match flags to file attributes */
>> +}
>> +
>> +/**
>> + * SUBTEST: test_pmt_telemetry_files
>> + * Description: validate the expected telemetry file(s)
>> + * Test category: functionality test
>> + *
>> + */
>> +static void test_pmt_telemetry_files(int dev_fd)
>> +{
>> +	int i;
>> +
>> +	for (i = 0; i < BMG_TELEMETRY_CNT; i++) {
>> +		sprintf(work_path, "%s/%s/%s/%s", dev_path,
>telemetry_vsec_dir,
>> +			"intel_pmt", telemetry_dir[i]);
>> +		find_pmt_file(work_path, telem);
>> +	}
>> +}
>> +
>> +/**
>> + * SUBTEST: test_pmt_crashlog_files
>> + * Description: validate the expected crashlog files
>> + * Test category: functionality test
>> + *
>> + */
>> +static void test_pmt_crashlog_files(int dev_fd)
>> +{
>> +	char buf[64] = {};
>> +	int ret;
>> +	int i;
>> +
>> +	for (i = 0; i < ARRAY_SIZE(bmg_info); i++) {
>> +		sprintf(work_path, "%s/%s/%s/%s", dev_path,
>crashlog_vsec_dir, "intel_pmt",
>> +			crashlog_dir[i]);
>> +
>> +		open_pmt_file(work_path, clear, &bmg_info[i].clear_fd,
>O_RDONLY);
>> +		open_pmt_file(work_path, consumed,
>&bmg_info[i].consumed_fd, O_RDWR);
>> +		open_pmt_file(work_path, crashlog, &bmg_info[i].crashlog_fd,
>O_RDONLY);
>> +		open_pmt_file(work_path, enable, &bmg_info[i].enable_fd,
>O_RDWR);
>> +		open_pmt_file(work_path, error, &bmg_info[i].error_fd,
>O_RDONLY);
>> +		open_pmt_file(work_path, dev_guid, &bmg_info[i].guid_fd,
>O_RDONLY);
>> +		open_pmt_file(work_path, rearm, &bmg_info[i].rearm_fd,
>O_RDWR);
>> +		open_pmt_file(work_path, trigger, &bmg_info[i].trigger_fd,
>O_RDWR);
>> +
>> +		ret = pread(bmg_info[i].guid_fd, buf, sizeof(buf), 0);
>> +		igt_assert_f(ret > 0, "failed to read guid for device %d\n", i);
>> +		bmg_info[i].guid = strtol(buf, NULL, 16);
>> +		igt_assert_f(bmg_info[i].guid > 0, "failed to set guid for device
>%d\n", i);
>> +	}
>> +}
>> +
>> +#define ENABLE_MSG "1\n"
>> +#define DISABLE_MSG "0\n"
>> +
>> +static bool send_msg(int fd, const char *msg, const char *file) {
>> +	size_t len = strlen(msg);
>> +	int ret;
>> +
>> +	errno = 0;
>> +	ret = pwrite(fd, msg, len, 0);
>> +	if (ret != len)
>> +		igt_info("%s failed: len: %ld vs %d  errno: %d\n", file, len, ret,
>> +			 errno);
>> +
>> +	return ret == len;
>> +}
>> +
>> +static bool verify_msg(int fd, const char *msg, const char *file) {
>> +	size_t len = strlen(msg);
>> +	char buf[32] = {};
>> +	int ret;
>> +
>> +	errno = 0;
>> +	ret = pread(fd, buf, sizeof(buf), 0);
>> +	if (ret != len)
>> +		igt_info("%s failed: len: %ld vs %d  errno: %d\n", file, len, ret,
>errno);
>> +
>> +	return ret == len && strcmp(buf, msg) == 0;
>> +}
>> +
>> +/**
>> + * SUBTEST: test_pmt_crashlog_enable
>> + * Description: Set enable enable/disable bit and verify usage
>> + * Test category: functionality test
>> + *
>> + */
>> +static void test_pmt_crashlog_enable(int dev_fd)
>> +{
>> +	u_int32_t guid;
>> +	int fd;
>> +	int i;
>> +
>> +	for (i = 0; i < ARRAY_SIZE(bmg_info); i++) {
>> +		fd = bmg_info[i].enable_fd;
>> +		guid = bmg_info[i].guid;
>> +
>> +		/* force enable so we are in a known state */
>> +		igt_assert_f(send_msg(fd, ENABLE_MSG, enable), "0x%x: send
>enable\n", guid);
>> +		igt_assert_f(verify_msg(fd, ENABLE_MSG, enable), "0x%x:
>verify enable\n", guid);
>> +
>> +		/* disable */
>> +		igt_assert_f(send_msg(fd, DISABLE_MSG, enable), "0x%x: send
>disable\n", guid);
>> +		igt_assert_f(verify_msg(fd, DISABLE_MSG, enable), "0x%x:
>verify disable\n", guid);
>> +
>> +		/* re-enable so we can do more testing */
>> +		igt_assert_f(send_msg(fd, ENABLE_MSG, enable), "0x%x: re-
>enable\n", guid);
>> +		igt_assert_f(verify_msg(fd, ENABLE_MSG, enable), "0x%x:
>verify re-enable\n", guid);
>> +	}
>> +
>> +}
>> +
>> +/**
>> + * SUBTEST: test_pmt_crashlog_clear
>> + * Description:
>> + *   Test the clear crashlog bit. After setting the crashlog data buffer should be
>> + *   set to 0xdeadbeef.
>> + *   "0" (DISABLE_MSG) is written to the trigger file to set the clear bit.  BMG
>does
>> + *   writing to the clear file, but once the bit is set it cannot be cleared with a
>> + *   reboot.  "0" to trigger is the "standard" usage, so test it.
>> + *
>> + * Test category: functionality test
>> + *
>> + */
>> +static void test_pmt_crashlog_clear(int dev_fd)
>> +{
>> +	char buf[64] = {};
>> +	u_int32_t guid;
>> +	int crashlog_fd;
>> +	int trigger_fd;
>> +	int clear_fd;
>> +	int *val;
>> +	int len;
>> +	int i;
>> +
>> +	for (i = 0; i < ARRAY_SIZE(bmg_info); i++) {
>> +		clear_fd = bmg_info[i].clear_fd;
>> +		crashlog_fd = bmg_info[i].crashlog_fd;
>> +		trigger_fd = bmg_info[i].trigger_fd;
>> +		guid = bmg_info[i].guid;
>> +
>> +		/* make sure the bit is clear */
>> +		igt_assert_f(verify_msg(clear_fd, DISABLE_MSG, clear), "0x%x:
>verify clear\n", guid);
>> +
>> +		/* set the clear bit (0 -> trigger)*/
>> +		igt_assert_f(send_msg(trigger_fd, DISABLE_MSG, trigger),
>"0x%x: send enable\n", guid);
>> +
>> +		/* make sure the bit is set.  sleep() to allow HW to set the bit */
>> +		sleep(1);
>> +		igt_assert_f(verify_msg(clear_fd, ENABLE_MSG, clear), "0x%x:
>clear set\n", guid);
>> +
>> +		len = read(crashlog_fd, buf, sizeof(buf));
>> +		igt_assert_f(len == sizeof(buf), "0x%x: failed to read crashlog
>data\n", guid);
>> +
>> +		/* wa punit issue for first crashlog (NOTE: this is fixed)*/
>> +		if (i == 0)
>> +			val = (int *) &buf[32];
>> +		else
>> +			val = (int *)buf;
>> +
>> +		igt_assert_f(*val == 0xdeadbeef, "0x%x: invalid clear data value:
>: 0x%x", guid, *val);
>> +	}
>> +
>> +}
>> +
>> +/**
>> + * SUBTEST: test_pmt_crashlog_consumed
>> + * Description:
>> + *   After a crashlog has been "consumed" (read), setting this bit can be done.
>> + *   Verify that it is set correctly.
>> + * Test category: functionality test
>> + *
>> + */
>> +static void test_pmt_crashlog_consumed(int dev_fd)
>> +{
>> +	uint32_t guid;
>> +	int fd;
>> +	int i;
>> +
>> +	for (i = 0; i < ARRAY_SIZE(bmg_info); i++) {
>> +		fd = bmg_info[i].consumed_fd;
>> +		guid = bmg_info[i].guid;
>> +
>> +		/* check, set, verify */
>> +		igt_assert_f(verify_msg(fd, DISABLE_MSG, consumed), "0x%x:
>consumed clear\n", guid);
>> +		igt_assert_f(send_msg(fd, ENABLE_MSG, consumed), "0x%x:
>set consumed\n", guid);
>> +		/* sleep(1) to allow HW to set the bit */
>> +		sleep(1);
>> +		igt_assert_f(verify_msg(fd, ENABLE_MSG, consumed), "0x%x:
>verify consumed\n", guid);
>> +	}
>> +}
>> +
>> +/**
>> + * SUBTEST: test_pmt_crashlog_error
>> + * Description:
>> + *    The error bit is set when a crashlog fails in HW.  It is read only so only
>> + *    need to verify that it is "0".
>> + * Test category: functionality test
>> + *
>> + */
>> +static void test_pmt_crashlog_error(int dev_fd)
>> +{
>> +	uint32_t guid;
>> +	int i;
>> +
>> +	for (i = 0; i < ARRAY_SIZE(bmg_info); i++) {
>> +		guid = bmg_info[i].guid;
>> +		igt_assert_f(verify_msg(bmg_info[i].error_fd, DISABLE_MSG,
>error), "0x%x: error clear\n", guid);
>> +	}
>> +}
>> +
>> +/**
>> + * SUBTEST: test_pmt_crashlog_rearm
>> + * Description:
>> + *    The rearm bit is set at cold boot.  It cannot be reset unless are real
>crashlog
>> + *    occurs (i.e. setting trigger will not change its value).  Verify that it is "1".
>> + * Test category: functionality test
>> + *
>> + */
>> +static void test_pmt_crashlog_rearm(int dev_fd)
>> +{
>> +	uint32_t guid;
>> +	int i;
>> +
>> +	for (i = 0; i < ARRAY_SIZE(bmg_info); i++) {
>> +		guid = bmg_info[i].guid;
>> +		igt_assert_f(verify_msg(bmg_info[i].rearm_fd, ENABLE_MSG,
>rearm), "0x%x: rearm set\n", guid);
>> +	}
>> +}
>> +
>> +/**
>> + * SUBTEST: test_pmt_crashlog_rearm_after_disable
>> + * Description:
>> + *    After a disable/enable sequence REARM will be set for PUNINT instaces
>and
>> + *    clear for OOBMSM instances.
>> + *    Verify that the bits are set as expected
>> + *
>> + * Test category: functionality test
>> + *
>> + */
>> +static void test_pmt_crashlog_rearm_after_disable(int dev_fd)
>> +{
>> +	uint32_t guid;
>> +	int i;
>> +
>> +	i = bmg_crashlog_punit;
>> +	guid = bmg_info[i].guid;
>> +	igt_assert_f(verify_msg(bmg_info[i].rearm_fd, ENABLE_MSG, rearm),
>"0x%x: rearm set\n", guid);
>> +
>> +	i = bmg_crashlog_oobmsm;
>> +	guid = bmg_info[i].guid;
>> +	igt_assert_f(verify_msg(bmg_info[i].rearm_fd, DISABLE_MSG, rearm),
>"0x%x: rearm set\n", guid);
>> +}
>> +
>> +/**
>> + * SUBTEST: test_pmt_crashlog_trigger
>> + * Description:
>> + *    Set the manual trigger bit and make sure the data is not 0xdeadbeef
>> + * Test category: functionality test
>> + *
>> + */
>> +static void test_pmt_crashlog_trigger(int dev_fd)
>> +{
>> +	char buf[64] = {};
>> +	u_int32_t *val;
>> +	int crashlog_fd;
>> +	int trigger_fd;
>> +	u_int32_t guid;
>> +	int len;
>> +	int i;
>> +
>> +	for (i = 0; i < ARRAY_SIZE(bmg_info); i++) {
>> +		crashlog_fd = bmg_info[i].crashlog_fd;
>> +		trigger_fd = bmg_info[i].trigger_fd;
>> +		guid = bmg_info[i].guid;
>> +
>> +		/* make sure the bit is clear */
>> +		igt_assert_f(verify_msg(trigger_fd, DISABLE_MSG, trigger),
>"0x%x: trigger clear\n",
>> +			     guid);
>> +		/* set the trigger bit (1 -> trigger)*/
>> +		igt_assert_f(send_msg(trigger_fd, ENABLE_MSG, trigger),
>"0x%x: set trigger\n", guid);
>> +
>> +		/* sleep to let the HW do its thing */
>> +		sleep(1);
>> +
>> +		/* make sure the bit is set */
>> +		igt_assert_f(verify_msg(trigger_fd, ENABLE_MSG, trigger),
>"0x%x: trigger not set\n",
>> +			     guid);
>> +
>> +		len = read(crashlog_fd, buf, sizeof(buf));
>> +		igt_assert_f(len == sizeof(buf), "0x%x: failed to read crashlog
>data\n", guid);
>> +
>> +		val = (u_int32_t *)buf;
>> +
>> +		igt_assert_f(*val != 0xdeadbeef, "0x%x: invalid trigger value: :
>0x%x", guid, *val);
>> +	}
>> +}
>> +
>> +igt_main
>> +{
>> +	const struct {
>> +		const char *name;
>> +		void (*func)(int);
>> +	} funcs[] = {
>> +		/*
>> +		 * NOTE:
>> +		 *  o These tests are ordered.  Do not use them individualy
>unless you understand
>> +		 *    the underlying HW behavior
>> +		 *  o Testing MUST be done after a cold reset
>> +		 *  o Once crashlog is triggered the device needs a cold reset,
>and some of the
>> +		 *    tests cannot be done.
>> +		 *  Only change this order if you understand this feature.
>> +		 */
>> +		{ "pmt-directories", test_pmt_directories },
>> +		{ "pmt-telemetry-files", test_pmt_telemetry_files },
>> +		{ "pmt-crashlog-files", test_pmt_crashlog_files },
>> +		{ "pmt-crashlog-error", test_pmt_crashlog_error },
>> +		{ "pmt-crashlog-rearm", test_pmt_crashlog_rearm },
>> +		{ "pmt-crashlog-enable", test_pmt_crashlog_enable },
>> +		{ "pmt-crashlog-rearm_after_disable",
>test_pmt_crashlog_rearm_after_disable },
>> +		{ "pmt-crashlog-trigger", test_pmt_crashlog_trigger },
>> +		{ "pmt-crashlog-consumed", test_pmt_crashlog_consumed },
>> +		{ "pmt-crashlog-clear", test_pmt_crashlog_clear },
>> +		{ }
>
>Will it work out of order? For example:
>sudo ./xe_pmt --run pmt-crashlog-enable

Some of the tests can be done out of order... however once the crashlog_trigger occurs
things become very limited, and the bit states are dependent on the crashlog state.

>Will it leave an enviroment clean for next non-pmt tests after
>such call?

The device can continue to be used (if we did a manual crashlog), but the crashlog feature
cannot be reused until a cold reboot occurs.

My understanding is that the cold reboot may not be necessary in future releases, but 
the cold reboot is the current limitation.

>> +	}, *f;
>> +	uint16_t dev_id;
>> +	int dev_fd;
>> +
>> +	igt_fixture
>> +		dev_fd = drm_open_driver(DRIVER_XE);
>> +
>> +	dev_id = intel_get_drm_devid(dev_fd);
>
>This should be in fixture.
>
>> +
>> +	if (IS_BATTLEMAGE(dev_id)) {
>
>This also should be in fixture, like:
>	igt_require_f(IS_BATTLEMAGE(dev_id), "PMT currenty supported only
>for BMG GPU\n");

I will update.

Thank you for comments!

Mike

>Regards,
>Kamil
>
>> +		for (f = funcs; f->name; f++) {
>> +			igt_subtest_f("%s", f->name)
>> +				f->func(dev_fd);
>> +		}
>> +	}
>> +
>> +	igt_fixture
>> +		drm_close_driver(dev_fd);
>> +}
>> diff --git a/tests/meson.build b/tests/meson.build
>> index 9b87a0d24..4276e6967 100644
>> --- a/tests/meson.build
>> +++ b/tests/meson.build
>> @@ -315,6 +315,7 @@ intel_xe_progs = [
>>  	'xe_peer2peer',
>>  	'xe_pm',
>>  	'xe_pm_residency',
>> +	'xe_pmt',
>>  	'xe_pmu',
>>  	'xe_prime_self_import',
>>  	'xe_pxp',
>> --
>> 2.49.0
>>


More information about the igt-dev mailing list