[igt-dev] [PATCH i-g-t v4 1/1] tests/xe/xe_uevent: add new test for uevent gt reset failure

Ghimiray, Himal Prasad himal.prasad.ghimiray at intel.com
Wed Jul 26 18:10:28 UTC 2023


Hi Kamil,

On 25-07-2023 23:00, Kamil Konieczny wrote:
> Hi Himal,
>
> On 2023-07-25 at 17:48:52 +0530, Himal Prasad Ghimiray wrote:
>> This test is to cause the fake reset failure and capture the uevent
>> sent in case of gt reset failure. We are relying on debugfs to
>> cause fake reset failure because there is no hardware mechanism using
>> which we can force gt to fail its reset.
>>
>> v2:
>> - Make fake reset for all GT's in platform.
>> - Add test names in order (Rahul)
>> - Add headers in order.
>> - Instead of system call use igt helper for reading debugfs.
>> - Use igt_debug instead of prints.
>> - Use break instead of unnecessary goto.
>> - Change test name.
>> - Define values as constants. (Kamil)
>>
>> v3:
>> - Use fault injection exposed debugfs.
>>
>> v4:
>> - Use pci subsystem for uevent listening.
>>
>> Cc: Kamil Konieczny <kamil.konieczny at linux.intel.com>
>> Cc: Janga Rahul Kumar <janga.rahul.kumar at intel.com>
>> Cc: Francois Dugast <francois.dugast at intel.com>
>> Cc: Rodrigo Vivi <rodrigo.vivi at intel.com>
>> Cc: Lucas De Marchi <lucas.demarchi at intel.com>
>>
>> Reviewed-by: Janga Rahul Kumar <janga.rahul.kumar at intel.com>
>> Signed-off-by: Himal Prasad Ghimiray <himal.prasad.ghimiray at intel.com>
>> ---
>> The sending of Uevent and fault injection is handled by:
>> https://patchwork.freedesktop.org/series/120919/
>>
>>   tests/meson.build    |   1 +
>>   tests/xe/xe_uevent.c | 130 +++++++++++++++++++++++++++++++++++++++++++
>>   2 files changed, 131 insertions(+)
>>   create mode 100644 tests/xe/xe_uevent.c
>>
>> diff --git a/tests/meson.build b/tests/meson.build
>> index 944a0941f..b9f0a9c15 100644
>> --- a/tests/meson.build
>> +++ b/tests/meson.build
>> @@ -291,6 +291,7 @@ xe_progs = [
>>   	'xe_prime_self_import',
>>   	'xe_query',
>>   	'xe_sysfs_tile',
>> +	'xe_uevent',
>>   	'xe_vm',
>>   	'xe_waitfence',
>>   	'xe_spin_batch',
>> diff --git a/tests/xe/xe_uevent.c b/tests/xe/xe_uevent.c
>> new file mode 100644
>> index 000000000..aabeb89b8
>> --- /dev/null
>> +++ b/tests/xe/xe_uevent.c
>> @@ -0,0 +1,130 @@
>> +// SPDX-License-Identifier: MIT
>> +/*
>> + * Copyright © 2023 Intel Corporation
>> + */
>> +
>> +/**
>> + * TEST: cause fake gt reset failure and listen uevent from KMD
>> + * SUBTEST:fake_reset_uevent_listener
>> + * Description:
>> + *		Test creates uevent listener and causes fake reset failure for gt0
>> + *		and returns success if uevent is sent by driver and listened by listener.
>> + */
>> +
>> +#include <libudev.h>
>> +#include <string.h>
>> +#include <sys/stat.h>
>> +
>> +#include "igt.h"
>> +
>> +#include "xe_drm.h"
>> +#include "xe/xe_ioctl.h"
>> +#include "xe/xe_query.h"
>> +
>> +static bool xe_fail_gt_reset(int fd, int gt)
>> +{
>> +	igt_debugfs_write(fd, "fail_gt_reset/probability", "100");
>> +	igt_debugfs_write(fd, "fail_gt_reset/times", "2");
>> +
>> +	xe_force_gt_reset(fd, gt);
>> +
>> +	return true;
> -------------- ^^^^
> Better make this function void with no return.

Sure will address in next patch.

>
>> +}
>> +
>> +static bool listen_reset_fail_uevent(struct udev_device *device, const char *source, int gt_id)
>> +{
>> +	struct udev_list_entry *list_entry;
>> +	bool dev_needs_reset = false;
>> +	bool reset_unit_is_gt = false;
>> +	bool gt_id_matches = false;
>> +	const char *name, *val;
>> +
>> +	udev_list_entry_foreach(list_entry, udev_device_get_properties_list_entry(device))
>> +	{
>> +		name = udev_list_entry_get_name(list_entry);
>> +		val = udev_list_entry_get_value(list_entry);
>> +
>> +		if (!strcmp(name, "DEVICE_STATUS") && !strcmp(val, "NEEDS_RESET")) {
>> +			igt_debug("%s = %s\n", name, val);
>> +			dev_needs_reset = true;
>> +			continue;
>> +		}
>> +
>> +		if (!strcmp(name, "RESET_FAILED") && !strcmp(val, "gt")) {
>> +			igt_debug("%s = %s\n", name, val);
>> +			reset_unit_is_gt = true;
>> +			continue;
>> +		}
>> +
>> +		if (!strcmp(name, "RESET_ID") && (atoi(val) == gt_id)) {
>> +			igt_debug("%s = %s\n", name, val);
>> +			gt_id_matches = true;
>> +			continue;
>> +		}
>> +	}
>> +
>> +	return (dev_needs_reset && reset_unit_is_gt && gt_id_matches);
>> +}
>> +
>> +static void fake_reset_uevent_listener(int fd, int gt_id)
>> +{
>> +	struct udev *udev;
>> +	struct udev_device *dev;
>> +	struct udev_monitor *mon;
>> +	bool event_received = false;
>> +	bool event_sent = false;
>> +	const u32 listener_timeout = 5;
>> +
>> +	/* create udev object */
>> +	udev = udev_new();
>> +	if (!udev)
>> +		igt_assert_f(false, "New udev object creation failed");
>> +
>> +	mon = udev_monitor_new_from_netlink(udev, "kernel");
>> +	udev_monitor_filter_add_match_subsystem_devtype(mon, "pci", NULL);
>> +	udev_monitor_enable_receiving(mon);
>> +	igt_until_timeout(listener_timeout) {
>> +		if (event_sent) {
>> +			dev = udev_monitor_receive_device(mon);
>> +			if (dev) {
>> +				event_received = listen_reset_fail_uevent(dev, "kernel", gt_id);
>> +				udev_device_unref(dev);
>> +			}
>> +		}
>> +
>> +		if (event_received)
>> +			break;
>> +
>> +		if (!event_sent)
>> +			event_sent = xe_fail_gt_reset(fd, gt_id);
> ----------------------- ^^^^^^^^^^^^
> This looks like the only purpose function return true is to set
> this var here. Loop looks convoluted, why not:
>
> 	igt_until_timeout(listener_timeout) {
> 		if (event_sent) {
> 			...your old code here...
> 		} else {
> 			event_sent = true;
> 			xe_fail_gt_reset(fd, gt_id);
> 		}
Makes sense. Will address in next patch.
>
>> +			}
>> +		}
>> +	}
> Add newline here.
>
>> +	udev_unref(udev);
> ------------ ^
> Why unref here?
>
>> +	igt_assert_f(event_received, "Event not received");
>> +}
>> +
>> +igt_main
>> +{
>> +	int fd;
>> +	int gt;
>> +	const u32 settle_xe_load_uevents = 50000;
>> +
>> +	igt_fixture {
>> +		fd = drm_open_driver(DRIVER_XE);
>> +		xe_device_get(fd);
> --------------- ^
> Not needed as drm_open_driver will do this.
>
>> +	}
>> +
>> +	/* Ensures uevents triggered in case of driver
>> +	 * load are settled down.
>> +	 */
>> +	usleep(settle_xe_load_uevents);
>> +
>> +	igt_subtest("fake_reset_uevent_listener")
>> +		xe_for_each_gt(fd, gt) {
>> +			fake_reset_uevent_listener(fd, gt);
>> +		}
>> +
>> +	igt_fixture {
>> +		xe_device_put(fd);
> --------------- ^
>> +		close(fd);
> --------------- ^
> Same here, just use drm_close_driver(fd);

Sure

BR

Himal

>
> Regards,
> Kamil
>
>> +	}
>> +}
>> -- 
>> 2.25.1
>>


More information about the igt-dev mailing list