[Intel-gfx] [PATCH 1/2] lib: add igt_get_stop_rings and igt_set_stop_rings

[Daniel Vetter]

On Tue, Mar 25, 2014 at 07:02:21PM +0200, Mika Kuoppala wrote:
> Multiple tests are introducing hangs by fidding with i915_ring_stop
> debugfs entry.
> 
> Signed-off-by: Mika Kuoppala <mika.kuoppala at intel.com>

Both patches look good to me. One thing we could do is add an exit handler
to reset stop_rings to 0, just in case the test died and somehow the
stop_rings value leaked.

With or w/o that change both patches lgtm, so feel free to push.
-Daniel

> ---
>  lib/igt_debugfs.c |  107 +++++++++++++++++++++++++++++++++++++++++++++++++++++
>  lib/igt_debugfs.h |   28 ++++++++++++++
>  2 files changed, 135 insertions(+)
> 
> diff --git a/lib/igt_debugfs.c b/lib/igt_debugfs.c
> index e04f8c5..c4cf50d 100644
> --- a/lib/igt_debugfs.c
> +++ b/lib/igt_debugfs.c
> @@ -28,9 +28,11 @@
>  #include <errno.h>
>  #include <stdio.h>
>  #include <stdlib.h>
> +#include <limits.h>
>  #include <string.h>
>  #include <fcntl.h>
>  #include <unistd.h>
> +#include <i915_drm.h>
>  
>  #include "drmtest.h"
>  #include "igt_display.h"
> @@ -615,3 +617,108 @@ int igt_open_forcewake_handle(void)
>  {
>  	return igt_debugfs_open("i915_forcewake_user", O_WRONLY);
>  }
> +
> +/**
> + * igt_to_stop_ring_flag:
> + * @ring: ring to get flag for, in drm I915_EXEC_* namespace
> + *
> + * This converts ring specified (drm namespace) to a ring flag
> + * to be used with igt_get_stop_rings() and igt_set_stop_rings().
> + *
> + * Returns:
> + * Ring flag for the give ring
> + */
> +enum stop_ring_flags igt_to_stop_ring_flag(int ring) {
> +	if (ring == I915_EXEC_DEFAULT)
> +		return STOP_RING_RENDER;
> +
> +	igt_assert(ring && ((ring & ~I915_EXEC_RING_MASK) == 0));
> +	return 1 << (ring - 1);
> +}
> +
> +static void stop_rings_write(uint32_t mask)
> +{
> +	int fd;
> +	char buf[80];
> +
> +	igt_assert(snprintf(buf, sizeof(buf), "0x%08x", mask) == 10);
> +	fd = igt_debugfs_open("i915_ring_stop", O_WRONLY);
> +	igt_assert(fd >= 0);
> +
> +	igt_assert(write(fd, buf, strlen(buf)) == strlen(buf));
> +	close(fd);
> +}
> +
> +/**
> + * igt_get_stop_rings:
> + *
> + * Read current ring flags from 'i915_ring_stop' debugfs entry.
> + *
> + * Returns:
> + * Current ring flags
> + */
> +enum stop_ring_flags igt_get_stop_rings(void)
> +{
> +	int fd;
> +	char buf[80];
> +	int l;
> +	unsigned long long ring_mask;
> +
> +	fd = igt_debugfs_open("i915_ring_stop", O_RDONLY);
> +	igt_assert(fd >= 0);
> +	l =  read(fd, buf, sizeof(buf));
> +	igt_assert(l > 0);
> +	igt_assert(l < sizeof(buf));
> +
> +	buf[l] = '\0';
> +
> +	close(fd);
> +
> +	errno = 0;
> +	ring_mask = strtoull(buf, NULL, 0);
> +	igt_assert(errno == 0);
> +	return ring_mask;
> +}
> +
> +/**
> + * igt_set_stop_rings:
> + * @flags: Ring flags to write
> + *
> + * This writes @flags to 'i915_ring_stop' debugfs entry. Driver will
> + * prevent cpu from writing tail pointer for ring @flags specify. Note
> + * that the ring is not stopped right away. Instead any further command
> + * emissions won't be executed after the flag is set.
> + *
> + * This is the least invasive way to make the gpu stuck. Hence you must
> + * set this after a batch submission with it's own invalid or endless
> + * looping instructions. In this case it is merely to give a log
> + * notification that this was simulated hang, as the batch would have
> + * caused hang in any case. Or if you use valid or noop batch and want
> + * to hang the ring (gpu), you must set this flag before submitting the
> + * batch.
> + *
> + * Driver checks periodically if ring is making any progress, and if
> + * not, it will declare the ring to be hung and will reset the GPU.
> + * After reset, the driver will clear flags in 'i915_ring_stop'
> + *
> + * Note: Always when hanging the gpu, use igt_set_stop_rings() to
> + * notify the driver. Driver tunes down gpu hang log messaging
> + * based on these flags and thus prevents false positives on logs.
> + *
> + */
> +void igt_set_stop_rings(enum stop_ring_flags flags)
> +{
> +	enum stop_ring_flags current;
> +
> +	igt_assert((flags & ~STOP_RING_ALL) == 0);
> +
> +	current = igt_get_stop_rings();
> +	igt_assert_f(current == 0,
> +		     "previous i915_ring_stop is still 0x%x\n", current);
> +
> +	stop_rings_write(flags);
> +	current = igt_get_stop_rings();
> +	if (current != flags)
> +		igt_warn("i915_ring_stop readback mismatch 0x%x vs 0x%x\n",
> +			 flags, current);
> +}
> diff --git a/lib/igt_debugfs.h b/lib/igt_debugfs.h
> index 3312a8b..72f4999 100644
> --- a/lib/igt_debugfs.h
> +++ b/lib/igt_debugfs.h
> @@ -143,4 +143,32 @@ void igt_enable_prefault(void);
>  
>  int igt_open_forcewake_handle(void);
>  
> +/**
> + * stop_ring_flags:
> + *
> + * @STOP_RING_NONE: Can be used to clear the pending stop (warning: hang might
> + * be declared already). Returned by igt_get_stop_rings() if there is
> + * no currently stopped rings.
> + * @STOP_RING_RENDER: Render ring
> + * @STOP_RING_BSD: Video encoding/decoding ring
> + * @STOP_RING_BLT: Blitter ring
> + * @STOP_RING_VEBOX: Video enchanment ring
> + * @STOP_RING_ALL: All rings
> + *
> + * Enumeration of all supported flags for igt_set_stop_rings().
> + *
> + */
> +enum stop_ring_flags {
> +	STOP_RING_NONE = 0,
> +	STOP_RING_RENDER = (1 << 0),
> +	STOP_RING_BSD = (1 << 1),
> +	STOP_RING_BLT = (1 << 2),
> +	STOP_RING_VEBOX = (1 << 3),
> +	STOP_RING_ALL = 0xff,
> +};
> +
> +enum stop_ring_flags igt_to_stop_ring_flag(int ring);
> +void igt_set_stop_rings(enum stop_ring_flags flags);
> +enum stop_ring_flags igt_get_stop_rings(void);
> +
>  #endif /* __IGT_DEBUGFS_H__ */
> -- 
> 1.7.9.5
> 
> _______________________________________________
> Intel-gfx mailing list
> Intel-gfx at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/intel-gfx

-- 
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch