diff mbox

[I-g-t,V2,2/2] tests/gem_dummy_reloc_loop: Add one subtest based on multi drm_fd to test CPU<->GPU sync under multi BSD rings

Message ID 1397529528-7114-3-git-send-email-yakui.zhao@intel.com (mailing list archive)
State New, archived
Headers show

Commit Message

Zhao, Yakui April 15, 2014, 2:38 a.m. UTC
The Broadwell GT3 machine has two independent BSD rings in kernel driver while
it is transparent to the user-space driver. In such case it needs to check
the CPU<->GPU sync for the second BSD ring.

V1->V2: Follow Daniel's comment to add one subtext instead of one individual
test case, which is used to test the CPU<->GPU sync under multi BSD rings

Signed-off-by: Zhao Yakui <yakui.zhao@intel.com>
---
 tests/gem_dummy_reloc_loop.c |  102 +++++++++++++++++++++++++++++++++++++++++-
 1 file changed, 101 insertions(+), 1 deletion(-)

Comments

Imre Deak April 22, 2014, 12:05 p.m. UTC | #1
On Tue, 2014-04-15 at 10:38 +0800, Zhao Yakui wrote:
> The Broadwell GT3 machine has two independent BSD rings in kernel driver while
> it is transparent to the user-space driver. In such case it needs to check
> the CPU<->GPU sync for the second BSD ring.
> 
> V1->V2: Follow Daniel's comment to add one subtext instead of one individual
> test case, which is used to test the CPU<->GPU sync under multi BSD rings
> 
> Signed-off-by: Zhao Yakui <yakui.zhao@intel.com>
> ---
>  tests/gem_dummy_reloc_loop.c |  102 +++++++++++++++++++++++++++++++++++++++++-
>  1 file changed, 101 insertions(+), 1 deletion(-)
> 
> diff --git a/tests/gem_dummy_reloc_loop.c b/tests/gem_dummy_reloc_loop.c
> index a61b59b..660d8e1 100644
> --- a/tests/gem_dummy_reloc_loop.c
> +++ b/tests/gem_dummy_reloc_loop.c
> @@ -48,6 +48,13 @@ static drm_intel_bufmgr *bufmgr;
>  struct intel_batchbuffer *batch;
>  static drm_intel_bo *target_buffer;
>  
> +#define NUM_FD	50
> +
> +static int mfd[NUM_FD];
> +static drm_intel_bufmgr *mbufmgr[NUM_FD];
> +static struct intel_batchbuffer *mbatch[NUM_FD];
> +static drm_intel_bo *mbuffer[NUM_FD];
> +
>  /*
>   * Testcase: Basic check of ring<->cpu sync using a dummy reloc
>   *
> @@ -124,6 +131,50 @@ dummy_reloc_loop_random_ring(int num_rings)
>  	}
>  }
>  
> +static void
> +dummy_reloc_loop_random_ring_multi_fd(int num_rings)
> +{
> +	int i;
> +	struct intel_batchbuffer *saved_batch;
> +
> +	saved_batch = batch;
> +
> +	srandom(0xdeadbeef);
> +
> +	for (i = 0; i < 0x100000; i++) {
> +		int mindex;
> +		int ring = random() % num_rings + 1;
> +
> +		mindex = random() % NUM_FD;
> +		batch = mbatch[mindex];
> +
> +		if (ring == I915_EXEC_RENDER) {
> +			BEGIN_BATCH(4);
> +			OUT_BATCH(MI_COND_BATCH_BUFFER_END | MI_DO_COMPARE);
> +			OUT_BATCH(0xffffffff); /* compare dword */
> +			OUT_RELOC(mbuffer[mindex], I915_GEM_DOMAIN_RENDER,
> +					I915_GEM_DOMAIN_RENDER, 0);
> +			OUT_BATCH(MI_NOOP);
> +			ADVANCE_BATCH();
> +		} else {
> +			BEGIN_BATCH(4);
> +			OUT_BATCH(MI_FLUSH_DW | 1);
> +			OUT_BATCH(0); /* reserved */
> +			OUT_RELOC(mbuffer[mindex], I915_GEM_DOMAIN_RENDER,
> +					I915_GEM_DOMAIN_RENDER, 0);
> +			OUT_BATCH(MI_NOOP | (1<<22) | (0xf));
> +			ADVANCE_BATCH();
> +		}
> +		intel_batchbuffer_flush_on_ring(batch, ring);
> +
> +		drm_intel_bo_map(target_buffer, 0);
> +		// map to force waiting on rendering
> +		drm_intel_bo_unmap(target_buffer);
> +	}
> +
> +	batch = saved_batch;
> +}
> +
>  int fd;
>  int devid;
>  int num_rings;
> @@ -133,6 +184,7 @@ igt_main
>  	igt_skip_on_simulation();
>  
>  	igt_fixture {
> +		int i;
>  		fd = drm_open_any();
>  		devid = intel_get_drm_devid(fd);
>  		num_rings = gem_get_num_rings(fd);
> @@ -148,6 +200,35 @@ igt_main
>  
>  		target_buffer = drm_intel_bo_alloc(bufmgr, "target bo", 4096, 4096);
>  		igt_assert(target_buffer);
> +
> +		/* Create multi drm_fd and map one gem object to multi gem_contexts */
> +		{
> +			unsigned int target_flink;
> +			char buffer_name[32];
> +			if (dri_bo_flink(target_buffer, &target_flink)) {
> +				printf("fail to get flink for target buffer\n");
> +				igt_assert(0);

For the future: could be just igt_assert_f().

> +			}
> +			for (i = 0; i < NUM_FD; i++) {
> +				mfd[i] = 0;
> +				mbufmgr[i] = NULL;
> +				mbuffer[i] = NULL;
> +			}

Nitpick: the above are all statics, so no need to init them.

Other than the above this looks good:
Reviewed-by: Imre Deak <imre.deak@intel.com>

> +			for (i = 0; i < NUM_FD; i++) {
> +				sprintf(buffer_name, "Target buffer %d\n", i);
> +				mfd[i] = drm_open_any();
> +				mbufmgr[i] = drm_intel_bufmgr_gem_init(mfd[i], 4096);
> +				igt_assert(mbufmgr[i]);
> +				drm_intel_bufmgr_gem_enable_reuse(mbufmgr[i]);
> +				mbatch[i] = intel_batchbuffer_alloc(mbufmgr[i], devid);
> +				igt_assert(mbufmgr[i]);
> +				mbuffer[i] = intel_bo_gem_create_from_name(
> +								mbufmgr[i],
> +								buffer_name,
> +								target_flink);
> +				igt_assert(mbuffer[i]);
> +			}
> +		}
>  	}
>  
>  	igt_subtest("render") {
> @@ -190,8 +271,27 @@ igt_main
>  			printf("dummy loop run on random rings completed\n");
>  		}
>  	}
> -
> +	igt_subtest("mixed_multi_fd") {
> +		if (num_rings > 1) {
> +			sleep(2);
> +			printf("running dummy loop on random rings based on "
> +					"multi drm_fd\n");
> +			dummy_reloc_loop_random_ring_multi_fd(num_rings);
> +			printf("dummy loop run on random rings based on "
> +					"multi drm_fd completed\n");
> +		}
> +	}
>  	igt_fixture {
> +		int i;
> +		/* Free the buffer/batchbuffer/buffer mgr for multi-fd */
> +		{
> +			for (i = 0; i < NUM_FD; i++) {
> +				dri_bo_unreference(mbuffer[i]);
> +				intel_batchbuffer_free(mbatch[i]);
> +				drm_intel_bufmgr_destroy(mbufmgr[i]);
> +				close(mfd[i]);
> +			}
> +		}
>  		drm_intel_bo_unreference(target_buffer);
>  		intel_batchbuffer_free(batch);
>  		drm_intel_bufmgr_destroy(bufmgr);
Daniel Vetter April 22, 2014, 7:48 p.m. UTC | #2
On Tue, Apr 22, 2014 at 03:05:03PM +0300, Imre Deak wrote:
> On Tue, 2014-04-15 at 10:38 +0800, Zhao Yakui wrote:
> > The Broadwell GT3 machine has two independent BSD rings in kernel driver while
> > it is transparent to the user-space driver. In such case it needs to check
> > the CPU<->GPU sync for the second BSD ring.
> > 
> > V1->V2: Follow Daniel's comment to add one subtext instead of one individual
> > test case, which is used to test the CPU<->GPU sync under multi BSD rings
> > 
> > Signed-off-by: Zhao Yakui <yakui.zhao@intel.com>
> > ---
> >  tests/gem_dummy_reloc_loop.c |  102 +++++++++++++++++++++++++++++++++++++++++-
> >  1 file changed, 101 insertions(+), 1 deletion(-)
> > 
> > diff --git a/tests/gem_dummy_reloc_loop.c b/tests/gem_dummy_reloc_loop.c
> > index a61b59b..660d8e1 100644
> > --- a/tests/gem_dummy_reloc_loop.c
> > +++ b/tests/gem_dummy_reloc_loop.c
> > @@ -48,6 +48,13 @@ static drm_intel_bufmgr *bufmgr;
> >  struct intel_batchbuffer *batch;
> >  static drm_intel_bo *target_buffer;
> >  
> > +#define NUM_FD	50
> > +
> > +static int mfd[NUM_FD];
> > +static drm_intel_bufmgr *mbufmgr[NUM_FD];
> > +static struct intel_batchbuffer *mbatch[NUM_FD];
> > +static drm_intel_bo *mbuffer[NUM_FD];
> > +
> >  /*
> >   * Testcase: Basic check of ring<->cpu sync using a dummy reloc
> >   *
> > @@ -124,6 +131,50 @@ dummy_reloc_loop_random_ring(int num_rings)
> >  	}
> >  }
> >  
> > +static void
> > +dummy_reloc_loop_random_ring_multi_fd(int num_rings)
> > +{
> > +	int i;
> > +	struct intel_batchbuffer *saved_batch;
> > +
> > +	saved_batch = batch;
> > +
> > +	srandom(0xdeadbeef);
> > +
> > +	for (i = 0; i < 0x100000; i++) {
> > +		int mindex;
> > +		int ring = random() % num_rings + 1;
> > +
> > +		mindex = random() % NUM_FD;
> > +		batch = mbatch[mindex];
> > +
> > +		if (ring == I915_EXEC_RENDER) {
> > +			BEGIN_BATCH(4);
> > +			OUT_BATCH(MI_COND_BATCH_BUFFER_END | MI_DO_COMPARE);
> > +			OUT_BATCH(0xffffffff); /* compare dword */
> > +			OUT_RELOC(mbuffer[mindex], I915_GEM_DOMAIN_RENDER,
> > +					I915_GEM_DOMAIN_RENDER, 0);
> > +			OUT_BATCH(MI_NOOP);
> > +			ADVANCE_BATCH();
> > +		} else {
> > +			BEGIN_BATCH(4);
> > +			OUT_BATCH(MI_FLUSH_DW | 1);
> > +			OUT_BATCH(0); /* reserved */
> > +			OUT_RELOC(mbuffer[mindex], I915_GEM_DOMAIN_RENDER,
> > +					I915_GEM_DOMAIN_RENDER, 0);
> > +			OUT_BATCH(MI_NOOP | (1<<22) | (0xf));
> > +			ADVANCE_BATCH();
> > +		}
> > +		intel_batchbuffer_flush_on_ring(batch, ring);
> > +
> > +		drm_intel_bo_map(target_buffer, 0);
> > +		// map to force waiting on rendering
> > +		drm_intel_bo_unmap(target_buffer);
> > +	}
> > +
> > +	batch = saved_batch;
> > +}
> > +
> >  int fd;
> >  int devid;
> >  int num_rings;
> > @@ -133,6 +184,7 @@ igt_main
> >  	igt_skip_on_simulation();
> >  
> >  	igt_fixture {
> > +		int i;
> >  		fd = drm_open_any();
> >  		devid = intel_get_drm_devid(fd);
> >  		num_rings = gem_get_num_rings(fd);
> > @@ -148,6 +200,35 @@ igt_main
> >  
> >  		target_buffer = drm_intel_bo_alloc(bufmgr, "target bo", 4096, 4096);
> >  		igt_assert(target_buffer);
> > +
> > +		/* Create multi drm_fd and map one gem object to multi gem_contexts */
> > +		{
> > +			unsigned int target_flink;
> > +			char buffer_name[32];
> > +			if (dri_bo_flink(target_buffer, &target_flink)) {
> > +				printf("fail to get flink for target buffer\n");
> > +				igt_assert(0);
> 
> For the future: could be just igt_assert_f().

Yeah I think for new testcases we should try to use the latest igt_*
macros and helpers as much as possible. Reducing control flow and
replacing it by the right igt_assert/require/... macro imo really helps
the readability of testcases.
-Daniel
> 
> > +			}
> > +			for (i = 0; i < NUM_FD; i++) {
> > +				mfd[i] = 0;
> > +				mbufmgr[i] = NULL;
> > +				mbuffer[i] = NULL;
> > +			}
> 
> Nitpick: the above are all statics, so no need to init them.
> 
> Other than the above this looks good:
> Reviewed-by: Imre Deak <imre.deak@intel.com>
> 
> > +			for (i = 0; i < NUM_FD; i++) {
> > +				sprintf(buffer_name, "Target buffer %d\n", i);
> > +				mfd[i] = drm_open_any();
> > +				mbufmgr[i] = drm_intel_bufmgr_gem_init(mfd[i], 4096);
> > +				igt_assert(mbufmgr[i]);
> > +				drm_intel_bufmgr_gem_enable_reuse(mbufmgr[i]);
> > +				mbatch[i] = intel_batchbuffer_alloc(mbufmgr[i], devid);
> > +				igt_assert(mbufmgr[i]);
> > +				mbuffer[i] = intel_bo_gem_create_from_name(
> > +								mbufmgr[i],
> > +								buffer_name,
> > +								target_flink);
> > +				igt_assert(mbuffer[i]);
> > +			}
> > +		}
> >  	}
> >  
> >  	igt_subtest("render") {
> > @@ -190,8 +271,27 @@ igt_main
> >  			printf("dummy loop run on random rings completed\n");
> >  		}
> >  	}
> > -
> > +	igt_subtest("mixed_multi_fd") {
> > +		if (num_rings > 1) {
> > +			sleep(2);
> > +			printf("running dummy loop on random rings based on "
> > +					"multi drm_fd\n");
> > +			dummy_reloc_loop_random_ring_multi_fd(num_rings);
> > +			printf("dummy loop run on random rings based on "
> > +					"multi drm_fd completed\n");
> > +		}
> > +	}
> >  	igt_fixture {
> > +		int i;
> > +		/* Free the buffer/batchbuffer/buffer mgr for multi-fd */
> > +		{
> > +			for (i = 0; i < NUM_FD; i++) {
> > +				dri_bo_unreference(mbuffer[i]);
> > +				intel_batchbuffer_free(mbatch[i]);
> > +				drm_intel_bufmgr_destroy(mbufmgr[i]);
> > +				close(mfd[i]);
> > +			}
> > +		}
> >  		drm_intel_bo_unreference(target_buffer);
> >  		intel_batchbuffer_free(batch);
> >  		drm_intel_bufmgr_destroy(bufmgr);
> 



> _______________________________________________
> Intel-gfx mailing list
> Intel-gfx@lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Zhao, Yakui April 23, 2014, 12:26 a.m. UTC | #3
On Tue, 2014-04-22 at 13:48 -0600, Daniel Vetter wrote:
> On Tue, Apr 22, 2014 at 03:05:03PM +0300, Imre Deak wrote:
> > On Tue, 2014-04-15 at 10:38 +0800, Zhao Yakui wrote:
> > > The Broadwell GT3 machine has two independent BSD rings in kernel driver while
> > > it is transparent to the user-space driver. In such case it needs to check
> > > the CPU<->GPU sync for the second BSD ring.
> > > 
> > > V1->V2: Follow Daniel's comment to add one subtext instead of one individual
> > > test case, which is used to test the CPU<->GPU sync under multi BSD rings
> > > 
> > > Signed-off-by: Zhao Yakui <yakui.zhao@intel.com>
> > > ---
> > >  tests/gem_dummy_reloc_loop.c |  102 +++++++++++++++++++++++++++++++++++++++++-
> > >  1 file changed, 101 insertions(+), 1 deletion(-)
> > > 
> > > diff --git a/tests/gem_dummy_reloc_loop.c b/tests/gem_dummy_reloc_loop.c
> > > index a61b59b..660d8e1 100644
> > > --- a/tests/gem_dummy_reloc_loop.c
> > > +++ b/tests/gem_dummy_reloc_loop.c
> > > @@ -48,6 +48,13 @@ static drm_intel_bufmgr *bufmgr;
> > >  struct intel_batchbuffer *batch;
> > >  static drm_intel_bo *target_buffer;
> > >  
> > > +#define NUM_FD	50
> > > +
> > > +static int mfd[NUM_FD];
> > > +static drm_intel_bufmgr *mbufmgr[NUM_FD];
> > > +static struct intel_batchbuffer *mbatch[NUM_FD];
> > > +static drm_intel_bo *mbuffer[NUM_FD];
> > > +
> > >  /*
> > >   * Testcase: Basic check of ring<->cpu sync using a dummy reloc
> > >   *
> > > @@ -124,6 +131,50 @@ dummy_reloc_loop_random_ring(int num_rings)
> > >  	}
> > >  }
> > >  
> > > +static void
> > > +dummy_reloc_loop_random_ring_multi_fd(int num_rings)
> > > +{
> > > +	int i;
> > > +	struct intel_batchbuffer *saved_batch;
> > > +
> > > +	saved_batch = batch;
> > > +
> > > +	srandom(0xdeadbeef);
> > > +
> > > +	for (i = 0; i < 0x100000; i++) {
> > > +		int mindex;
> > > +		int ring = random() % num_rings + 1;
> > > +
> > > +		mindex = random() % NUM_FD;
> > > +		batch = mbatch[mindex];
> > > +
> > > +		if (ring == I915_EXEC_RENDER) {
> > > +			BEGIN_BATCH(4);
> > > +			OUT_BATCH(MI_COND_BATCH_BUFFER_END | MI_DO_COMPARE);
> > > +			OUT_BATCH(0xffffffff); /* compare dword */
> > > +			OUT_RELOC(mbuffer[mindex], I915_GEM_DOMAIN_RENDER,
> > > +					I915_GEM_DOMAIN_RENDER, 0);
> > > +			OUT_BATCH(MI_NOOP);
> > > +			ADVANCE_BATCH();
> > > +		} else {
> > > +			BEGIN_BATCH(4);
> > > +			OUT_BATCH(MI_FLUSH_DW | 1);
> > > +			OUT_BATCH(0); /* reserved */
> > > +			OUT_RELOC(mbuffer[mindex], I915_GEM_DOMAIN_RENDER,
> > > +					I915_GEM_DOMAIN_RENDER, 0);
> > > +			OUT_BATCH(MI_NOOP | (1<<22) | (0xf));
> > > +			ADVANCE_BATCH();
> > > +		}
> > > +		intel_batchbuffer_flush_on_ring(batch, ring);
> > > +
> > > +		drm_intel_bo_map(target_buffer, 0);
> > > +		// map to force waiting on rendering
> > > +		drm_intel_bo_unmap(target_buffer);
> > > +	}
> > > +
> > > +	batch = saved_batch;
> > > +}
> > > +
> > >  int fd;
> > >  int devid;
> > >  int num_rings;
> > > @@ -133,6 +184,7 @@ igt_main
> > >  	igt_skip_on_simulation();
> > >  
> > >  	igt_fixture {
> > > +		int i;
> > >  		fd = drm_open_any();
> > >  		devid = intel_get_drm_devid(fd);
> > >  		num_rings = gem_get_num_rings(fd);
> > > @@ -148,6 +200,35 @@ igt_main
> > >  
> > >  		target_buffer = drm_intel_bo_alloc(bufmgr, "target bo", 4096, 4096);
> > >  		igt_assert(target_buffer);
> > > +
> > > +		/* Create multi drm_fd and map one gem object to multi gem_contexts */
> > > +		{
> > > +			unsigned int target_flink;
> > > +			char buffer_name[32];
> > > +			if (dri_bo_flink(target_buffer, &target_flink)) {
> > > +				printf("fail to get flink for target buffer\n");
> > > +				igt_assert(0);
> > 
> > For the future: could be just igt_assert_f().
> 
> Yeah I think for new testcases we should try to use the latest igt_*
> macros and helpers as much as possible. Reducing control flow and
> replacing it by the right igt_assert/require/... macro imo really helps
> the readability of testcases.

Hi, Daniel/Imre

    Thanks for your comments and advice.
    I will update it.

Thanks.
    Yakui

> -Daniel
> > 
> > > +			}
> > > +			for (i = 0; i < NUM_FD; i++) {
> > > +				mfd[i] = 0;
> > > +				mbufmgr[i] = NULL;
> > > +				mbuffer[i] = NULL;
> > > +			}
> > 
> > Nitpick: the above are all statics, so no need to init them.
> > 
> > Other than the above this looks good:
> > Reviewed-by: Imre Deak <imre.deak@intel.com>
> > 
> > > +			for (i = 0; i < NUM_FD; i++) {
> > > +				sprintf(buffer_name, "Target buffer %d\n", i);
> > > +				mfd[i] = drm_open_any();
> > > +				mbufmgr[i] = drm_intel_bufmgr_gem_init(mfd[i], 4096);
> > > +				igt_assert(mbufmgr[i]);
> > > +				drm_intel_bufmgr_gem_enable_reuse(mbufmgr[i]);
> > > +				mbatch[i] = intel_batchbuffer_alloc(mbufmgr[i], devid);
> > > +				igt_assert(mbufmgr[i]);
> > > +				mbuffer[i] = intel_bo_gem_create_from_name(
> > > +								mbufmgr[i],
> > > +								buffer_name,
> > > +								target_flink);
> > > +				igt_assert(mbuffer[i]);
> > > +			}
> > > +		}
> > >  	}
> > >  
> > >  	igt_subtest("render") {
> > > @@ -190,8 +271,27 @@ igt_main
> > >  			printf("dummy loop run on random rings completed\n");
> > >  		}
> > >  	}
> > > -
> > > +	igt_subtest("mixed_multi_fd") {
> > > +		if (num_rings > 1) {
> > > +			sleep(2);
> > > +			printf("running dummy loop on random rings based on "
> > > +					"multi drm_fd\n");
> > > +			dummy_reloc_loop_random_ring_multi_fd(num_rings);
> > > +			printf("dummy loop run on random rings based on "
> > > +					"multi drm_fd completed\n");
> > > +		}
> > > +	}
> > >  	igt_fixture {
> > > +		int i;
> > > +		/* Free the buffer/batchbuffer/buffer mgr for multi-fd */
> > > +		{
> > > +			for (i = 0; i < NUM_FD; i++) {
> > > +				dri_bo_unreference(mbuffer[i]);
> > > +				intel_batchbuffer_free(mbatch[i]);
> > > +				drm_intel_bufmgr_destroy(mbufmgr[i]);
> > > +				close(mfd[i]);
> > > +			}
> > > +		}
> > >  		drm_intel_bo_unreference(target_buffer);
> > >  		intel_batchbuffer_free(batch);
> > >  		drm_intel_bufmgr_destroy(bufmgr);
> > 
> 
> 
> 
> > _______________________________________________
> > Intel-gfx mailing list
> > Intel-gfx@lists.freedesktop.org
> > http://lists.freedesktop.org/mailman/listinfo/intel-gfx
> 
>
diff mbox

Patch

diff --git a/tests/gem_dummy_reloc_loop.c b/tests/gem_dummy_reloc_loop.c
index a61b59b..660d8e1 100644
--- a/tests/gem_dummy_reloc_loop.c
+++ b/tests/gem_dummy_reloc_loop.c
@@ -48,6 +48,13 @@  static drm_intel_bufmgr *bufmgr;
 struct intel_batchbuffer *batch;
 static drm_intel_bo *target_buffer;
 
+#define NUM_FD	50
+
+static int mfd[NUM_FD];
+static drm_intel_bufmgr *mbufmgr[NUM_FD];
+static struct intel_batchbuffer *mbatch[NUM_FD];
+static drm_intel_bo *mbuffer[NUM_FD];
+
 /*
  * Testcase: Basic check of ring<->cpu sync using a dummy reloc
  *
@@ -124,6 +131,50 @@  dummy_reloc_loop_random_ring(int num_rings)
 	}
 }
 
+static void
+dummy_reloc_loop_random_ring_multi_fd(int num_rings)
+{
+	int i;
+	struct intel_batchbuffer *saved_batch;
+
+	saved_batch = batch;
+
+	srandom(0xdeadbeef);
+
+	for (i = 0; i < 0x100000; i++) {
+		int mindex;
+		int ring = random() % num_rings + 1;
+
+		mindex = random() % NUM_FD;
+		batch = mbatch[mindex];
+
+		if (ring == I915_EXEC_RENDER) {
+			BEGIN_BATCH(4);
+			OUT_BATCH(MI_COND_BATCH_BUFFER_END | MI_DO_COMPARE);
+			OUT_BATCH(0xffffffff); /* compare dword */
+			OUT_RELOC(mbuffer[mindex], I915_GEM_DOMAIN_RENDER,
+					I915_GEM_DOMAIN_RENDER, 0);
+			OUT_BATCH(MI_NOOP);
+			ADVANCE_BATCH();
+		} else {
+			BEGIN_BATCH(4);
+			OUT_BATCH(MI_FLUSH_DW | 1);
+			OUT_BATCH(0); /* reserved */
+			OUT_RELOC(mbuffer[mindex], I915_GEM_DOMAIN_RENDER,
+					I915_GEM_DOMAIN_RENDER, 0);
+			OUT_BATCH(MI_NOOP | (1<<22) | (0xf));
+			ADVANCE_BATCH();
+		}
+		intel_batchbuffer_flush_on_ring(batch, ring);
+
+		drm_intel_bo_map(target_buffer, 0);
+		// map to force waiting on rendering
+		drm_intel_bo_unmap(target_buffer);
+	}
+
+	batch = saved_batch;
+}
+
 int fd;
 int devid;
 int num_rings;
@@ -133,6 +184,7 @@  igt_main
 	igt_skip_on_simulation();
 
 	igt_fixture {
+		int i;
 		fd = drm_open_any();
 		devid = intel_get_drm_devid(fd);
 		num_rings = gem_get_num_rings(fd);
@@ -148,6 +200,35 @@  igt_main
 
 		target_buffer = drm_intel_bo_alloc(bufmgr, "target bo", 4096, 4096);
 		igt_assert(target_buffer);
+
+		/* Create multi drm_fd and map one gem object to multi gem_contexts */
+		{
+			unsigned int target_flink;
+			char buffer_name[32];
+			if (dri_bo_flink(target_buffer, &target_flink)) {
+				printf("fail to get flink for target buffer\n");
+				igt_assert(0);
+			}
+			for (i = 0; i < NUM_FD; i++) {
+				mfd[i] = 0;
+				mbufmgr[i] = NULL;
+				mbuffer[i] = NULL;
+			}
+			for (i = 0; i < NUM_FD; i++) {
+				sprintf(buffer_name, "Target buffer %d\n", i);
+				mfd[i] = drm_open_any();
+				mbufmgr[i] = drm_intel_bufmgr_gem_init(mfd[i], 4096);
+				igt_assert(mbufmgr[i]);
+				drm_intel_bufmgr_gem_enable_reuse(mbufmgr[i]);
+				mbatch[i] = intel_batchbuffer_alloc(mbufmgr[i], devid);
+				igt_assert(mbufmgr[i]);
+				mbuffer[i] = intel_bo_gem_create_from_name(
+								mbufmgr[i],
+								buffer_name,
+								target_flink);
+				igt_assert(mbuffer[i]);
+			}
+		}
 	}
 
 	igt_subtest("render") {
@@ -190,8 +271,27 @@  igt_main
 			printf("dummy loop run on random rings completed\n");
 		}
 	}
-
+	igt_subtest("mixed_multi_fd") {
+		if (num_rings > 1) {
+			sleep(2);
+			printf("running dummy loop on random rings based on "
+					"multi drm_fd\n");
+			dummy_reloc_loop_random_ring_multi_fd(num_rings);
+			printf("dummy loop run on random rings based on "
+					"multi drm_fd completed\n");
+		}
+	}
 	igt_fixture {
+		int i;
+		/* Free the buffer/batchbuffer/buffer mgr for multi-fd */
+		{
+			for (i = 0; i < NUM_FD; i++) {
+				dri_bo_unreference(mbuffer[i]);
+				intel_batchbuffer_free(mbatch[i]);
+				drm_intel_bufmgr_destroy(mbufmgr[i]);
+				close(mfd[i]);
+			}
+		}
 		drm_intel_bo_unreference(target_buffer);
 		intel_batchbuffer_free(batch);
 		drm_intel_bufmgr_destroy(bufmgr);