[igt-dev] [PATCH i-g-t 2/2] lib/xe: Add hang library support

Kumar, Janga Rahul janga.rahul.kumar at intel.com
Mon Aug 7 04:06:33 UTC 2023



> -----Original Message-----
> From: Karas, Anna <anna.karas at intel.com>
> Sent: 27 July 2023 02:15
> To: igt-dev at lists.freedesktop.org
> Cc: Kamil Konieczny <kamil.konieczny at linux.intel.com>; Kumar, Janga Rahul
> <janga.rahul.kumar at intel.com>; Gandi, Ramadevi
> <ramadevi.gandi at intel.com>; Ch, Sai Gowtham <sai.gowtham.ch at intel.com>;
> Joshi, Kunal1 <kunal1.joshi at intel.com>
> Subject: Re: [igt-dev] [PATCH i-g-t 2/2] lib/xe: Add hang library support
> 
> Hi all,
> 
> On 25.07.2023 20:04, Kamil Konieczny wrote:
> > Hi Janga,
> >
> > On 2023-07-25 at 12:46:22 +0530, janga.rahul.kumar at intel.com wrote:
> >> From: Janga Rahul Kumar <janga.rahul.kumar at intel.com>
> >>
> >> Add hang library support using spinners.
> >>
> >> Cc: Sai Gowtham Ch <sai.gowtham.ch at intel.com>
> >> Cc: Kunal Joshi <kunal1.joshi at intel.com>
> >> Signed-off-by: Janga Rahul Kumar <janga.rahul.kumar at intel.com>
> >> ---
> >>   lib/igt_gt.c      | 21 +++++++++++++++++++++
> > ------ ^^^^^^
> > Move this change into separate patch.
> >
> >>   lib/xe/xe_ioctl.c | 20 ++++++++++++++++++++
> >>   lib/xe/xe_ioctl.h |  6 ++++++
> >>   3 files changed, 47 insertions(+)
> >>
> >> diff --git a/lib/igt_gt.c b/lib/igt_gt.c index 2ce464ba6..6821179ca
> >> 100644
> >> --- a/lib/igt_gt.c
> >> +++ b/lib/igt_gt.c
> >> @@ -170,6 +170,15 @@ static void context_set_ban(int fd, unsigned
> >> ctx, unsigned ban)
> >>
> >>   igt_hang_t igt_allow_hang(int fd, unsigned ctx, unsigned flags)
> >>   {
> >> +	if (is_xe_device(fd)) {
> >> +		if (!igt_check_boolean_env_var("IGT_HANG", true))
> >> +			igt_skip("hang injection disabled by user
> [IGT_HANG=0]\n");
> >> +
> >> +		igt_require(has_gpu_reset(fd));
> >> +
> >> +		return (struct igt_hang){ 0, ctx, 0, flags };
> >> +       }
> >> +
> >
> > Move code after declarations of vars.
> >
> >>   	struct drm_i915_gem_context_param param = {
> >>   		.ctx_id = ctx,
> >>   	};
> >> @@ -219,6 +228,9 @@ igt_hang_t igt_allow_hang(int fd, unsigned ctx,
> >> unsigned flags)
> >>
> >>   void igt_disallow_hang(int fd, igt_hang_t arg)
> >>   {
> >> +	if (is_xe_device(fd))
> >> +		return;
> >> +
> >>   	context_set_ban(fd, arg.ctx, arg.ban);
> >>
> >>   	if ((arg.flags & HANG_ALLOW_CAPTURE) == 0) { @@ -290,6 +302,9 @@
> >> static bool has_ctx_exec(int fd, unsigned ring, uint32_t ctx)
> >>   static igt_hang_t __igt_hang_ctx(int fd, uint64_t ahnd, uint32_t ctx, int ring,
> >>   				 unsigned flags)
> >>   {
> >> +	if (is_xe_device(fd))
> >> +		return xe_hang_engine(fd, ahnd, ctx, ring, flags);
> >> +
> >
> > Move this after var declatarions.
> One more thing: with your change the doc of __igt_hang_ctx is no longer valid.
> Please delete i915 reference in the doc, as now fd can be both
> i915 and xe.
> You can also fix already existing typo: @ctx: the contxt -> context
Done.
> 
> >
> >>   	struct drm_i915_gem_context_param param;
> >>   	igt_spin_t *spin;
> >>   	unsigned ban;
> >> @@ -372,6 +387,12 @@ void igt_post_hang_ring(int fd, igt_hang_t arg)
> >>   	if (!arg.spin)
> >>   		return;
> >>
> >> +	if (is_xe_device(fd)) {
> >> +		igt_spin_free(fd, arg.spin);
> >> +		xe_post_hang_ring(fd, arg);
> >
> > It looks like spinner is not used here?
> Please update the doc also for igt_post_hang_ring - @fd.
> 
> >
> >> +		return;
> >> +	}
> >> +
> >>   	gem_sync(fd, arg.spin->handle); /* Wait until it hangs */
> >>   	igt_spin_free(fd, arg.spin);
> >>
> >> diff --git a/lib/xe/xe_ioctl.c b/lib/xe/xe_ioctl.c index
> >> 488aa218e..5ec85239a 100644
> >> --- a/lib/xe/xe_ioctl.c
> >> +++ b/lib/xe/xe_ioctl.c
> >> @@ -520,6 +520,26 @@ void xe_force_gt_reset(int fd, int gt)
> >>   	system(reset_string);
> >>   }
> >>
> >> +
> >> +igt_hang_t xe_hang_engine(int fd, uint64_t ahnd, uint32_t ctx, int ring,
> >> +				unsigned flags)
> >> +{
> >> +	uint32_t vm;
> >> +	unsigned int engine;
> >> +	igt_spin_t *spin_t;
> >
> > Add newline.
> >
> >> +	vm = xe_vm_create(fd, 0, 0);
> >> +	engine = xe_engine_create_class(fd, vm,
> DRM_XE_ENGINE_CLASS_COPY);
> > ----------------------------------------------- ^ Only one engine
> > class here?
Updated it in v2
> >
> >> +
> >> +	spin_t = igt_spin_new(fd, .ahnd = ahnd, .engine = engine, .vm = vm,
> >> +.flags = IGT_SPIN_NO_PREEMPTION);
> >
> > Add newline.
> >
> >> +	return (igt_hang_t){ spin_t, engine, 0, flags };
> >
> > Unused ring var?
> Ctx also seems to be unused.
In Xe, we need to create engine(exec_queue) hence not used here.

Thanks,
Rahul
> 
> >
> >> +}
> >> +
> >> +void xe_post_hang_ring(int fd, igt_hang_t arg) {
> >> +	xe_engine_destroy(fd, arg.ctx);
> >> +	xe_vm_destroy(fd, arg.spin->vm);
> >> +}
> >> +
> >>   void xe_vm_madvise(int fd, uint32_t vm, uint64_t addr, uint64_t size,
> >>   		   uint32_t property, uint32_t value)
> >>   {
> >> diff --git a/lib/xe/xe_ioctl.h b/lib/xe/xe_ioctl.h index
> >> 5a528b345..0e9708194 100644
> >> --- a/lib/xe/xe_ioctl.h
> >> +++ b/lib/xe/xe_ioctl.h
> >> @@ -15,6 +15,9 @@
> >>   #include <stdint.h>
> >>   #include <xe_drm.h>
> >>
> >> +#include "lib/igt_gt.h"
> >> +#include "xe_spin.h"
> >> +
> >
> > Are you using new ioctl? If not maybe add xe_hang and xe_post_hang to
> > other library? Or create new one?
> >
> > Regards,
> > Kamil
> >
> >>   uint32_t xe_cs_prefetch_size(int fd);
> >>   uint32_t xe_vm_create(int fd, uint32_t flags, uint64_t ext);
> >>   int  __xe_vm_bind(int fd, uint32_t vm, uint32_t engine, uint32_t
> >> bo, @@ -89,6 +92,9 @@ int64_t xe_wait_ufence_abstime(int fd, uint64_t
> *addr, uint64_t value,
> >>   void xe_force_gt_reset(int fd, int gt);
> >>   void xe_force_gt_reset_all(int fd);
> >>   bool has_xe_gt_reset(int fd);
> >> +igt_hang_t xe_hang_engine(int fd, uint64_t ahnd, uint32_t ctx, int ring,
> >> +				unsigned flags);
> >> +void xe_post_hang_ring(int fd, igt_hang_t arg);
> >>   void xe_vm_madvise(int fd, uint32_t vm, uint64_t addr, uint64_t size,
> >>   		   uint32_t property, uint32_t value);
> >>
> >> --
> >> 2.25.1
> >>
> 
> As with previous patch, please remember to check it with checkpatch.pl
> before sending v2.
> 
> Ania



More information about the igt-dev mailing list