igt/gem_partial_pwrite_pread: Add set-cache subtest to validate JIRA VIZ-3721

Commit Message

Reese, Armin C June 12, 2014, 9:18 p.m. UTC

From: Armin Reese <[e-mail]>

This subtest forces the driver down the code path in i915_gem_object_set_cache_level()
which unbinds VMAs and triggers the kernel panic described in VIZ-3721.  This
test will pass if the bug fix is in place.

Bugzilla:  https://bugs.fredesktop.org/show_bug.cgi?id=76384
Signed-off-by:  Armin Reese <[e-mail]>
---
 tests/gem_partial_pwrite_pread.c | 11 +++++++++++
 1 file changed, 11 insertions(+)

Comments

Chris Wilson June 13, 2014, 6:18 a.m. UTC | #1

On Thu, Jun 12, 2014 at 02:18:37PM -0700, [e-mail] wrote:
> From: Armin Reese <[e-mail]>
> 
> This subtest forces the driver down the code path in i915_gem_object_set_cache_level()
> which unbinds VMAs and triggers the kernel panic described in VIZ-3721.  This
> test will pass if the bug fix is in place.
> 
> Bugzilla:  https://bugs.fredesktop.org/show_bug.cgi?id=76384
Where's that?
-Chris

Daniel Vetter June 13, 2014, 7:13 a.m. UTC | #2

On Thu, Jun 12, 2014 at 02:18:37PM -0700, [e-mail] wrote:
> From: Armin Reese <[e-mail]>
> 
> This subtest forces the driver down the code path in i915_gem_object_set_cache_level()
> which unbinds VMAs and triggers the kernel panic described in VIZ-3721.  This
> test will pass if the bug fix is in place.
> 
> Bugzilla:  https://bugs.fredesktop.org/show_bug.cgi?id=76384
> Signed-off-by:  Armin Reese <[e-mail]>
> ---
>  tests/gem_partial_pwrite_pread.c | 11 +++++++++++
>  1 file changed, 11 insertions(+)
> 
> diff --git a/tests/gem_partial_pwrite_pread.c b/tests/gem_partial_pwrite_pread.c
> index 7945ba7..3e3ecc9 100644
> --- a/tests/gem_partial_pwrite_pread.c
> +++ b/tests/gem_partial_pwrite_pread.c
> @@ -131,6 +131,14 @@ static void test_partial_reads(void)
>  
>  }
>  
> +static void test_set_cache(void)
> +{
> +	igt_info("checking set-cache\n");
> +
> +	blt_bo_fill(staging_bo, scratch_bo, 0);
> +
> +}

Ok, my apologies for the unclear JIRA description. My idea was to have a
tailor-made testcase with a minimal reproduction scenario (i.e. map into 2
ppgtt or ggtt + ppgtt, then do a set_caching ioctl).

This here blows up because we do the set_caching as part of the fixture,
so has a high risk that when someone decides to refactor the testcase
we'll no longer test this bug. Hence the task to create a new, specific
testcase.
-Daniel

> +
>  static void test_partial_writes(void)
>  {
>  	int i, j;
> @@ -244,6 +252,9 @@ static void do_tests(int cache_level, const char *suffix)
>  			gem_set_caching(fd, scratch_bo->handle, cache_level);
>  	}
>  
> +	igt_subtest_f("set-cache%s", suffix)
> +		test_set_cache();
> +
>  	igt_subtest_f("reads%s", suffix)
>  		test_partial_reads();
>  
> -- 
> 1.9.1
> 
> _______________________________________________
> Intel-gfx mailing list
> [e-mail]
> http://lists.freedesktop.org/mailman/listinfo/intel-gfx

Reese, Armin C July 11, 2014, 5:55 p.m. UTC | #3

Which of the I-G-Ts would be the best to house the set-cache subtest?  Is gem_partial_pwrite_pread a good candidate?
- Armin

-----Original Message-----
From: Daniel Vetter [mailto:[e-mail]] On Behalf Of Daniel Vetter
Sent: Friday, June 13, 2014 12:14 AM
To: Reese, Armin C
Cc: [e-mail]
Subject: Re: [Intel-gfx] [PATCH] igt/gem_partial_pwrite_pread: Add set-cache subtest to validate JIRA VIZ-3721

On Thu, Jun 12, 2014 at 02:18:37PM -0700, [e-mail] wrote:
> From: Armin Reese <[e-mail]>
> 
> This subtest forces the driver down the code path in 
> i915_gem_object_set_cache_level() which unbinds VMAs and triggers the 
> kernel panic described in VIZ-3721.  This test will pass if the bug fix is in place.
> 
> Bugzilla:  https://bugs.fredesktop.org/show_bug.cgi?id=76384
> Signed-off-by:  Armin Reese <[e-mail]>
> ---
>  tests/gem_partial_pwrite_pread.c | 11 +++++++++++
>  1 file changed, 11 insertions(+)
> 
> diff --git a/tests/gem_partial_pwrite_pread.c 
> b/tests/gem_partial_pwrite_pread.c
> index 7945ba7..3e3ecc9 100644
> --- a/tests/gem_partial_pwrite_pread.c
> +++ b/tests/gem_partial_pwrite_pread.c
> @@ -131,6 +131,14 @@ static void test_partial_reads(void)
>  
>  }
>  
> +static void test_set_cache(void)
> +{
> +	igt_info("checking set-cache\n");
> +
> +	blt_bo_fill(staging_bo, scratch_bo, 0);
> +
> +}

Ok, my apologies for the unclear JIRA description. My idea was to have a tailor-made testcase with a minimal reproduction scenario (i.e. map into 2 ppgtt or ggtt + ppgtt, then do a set_caching ioctl).

This here blows up because we do the set_caching as part of the fixture, so has a high risk that when someone decides to refactor the testcase we'll no longer test this bug. Hence the task to create a new, specific testcase.
-Daniel

> +
>  static void test_partial_writes(void)  {
>  	int i, j;
> @@ -244,6 +252,9 @@ static void do_tests(int cache_level, const char *suffix)
>  			gem_set_caching(fd, scratch_bo->handle, cache_level);
>  	}
>  
> +	igt_subtest_f("set-cache%s", suffix)
> +		test_set_cache();
> +
>  	igt_subtest_f("reads%s", suffix)
>  		test_partial_reads();
>  
> --
> 1.9.1
> 
> _______________________________________________
> Intel-gfx mailing list
> [e-mail]
> http://lists.freedesktop.org/mailman/listinfo/intel-gfx

--
Daniel Vetter
Software Engineer, Intel Corporation
[phone] - http://blog.ffwll.ch

Daniel Vetter July 12, 2014, 10:11 a.m. UTC | #4

On Fri, Jul 11, 2014 at 05:55:41PM +0000, Reese, Armin C wrote:
> Which of the I-G-Ts would be the best to house the set-cache subtest?
> Is gem_partial_pwrite_pread a good candidate?

I guess you could just create a new test binary if there's nothing
suitable. Especially for tests I'm leaning towards "code reuse is
overrated". And with igt_simple_main there's really no boilerplate
(besides the obligatory license header) left.
-Daniel

> - Armin
> 
> -----Original Message-----
> From: Daniel Vetter [mailto:[e-mail]] On Behalf Of Daniel Vetter
> Sent: Friday, June 13, 2014 12:14 AM
> To: Reese, Armin C
> Cc: [e-mail]
> Subject: Re: [Intel-gfx] [PATCH] igt/gem_partial_pwrite_pread: Add set-cache subtest to validate JIRA VIZ-3721
> 
> On Thu, Jun 12, 2014 at 02:18:37PM -0700, [e-mail] wrote:
> > From: Armin Reese <[e-mail]>
> > 
> > This subtest forces the driver down the code path in 
> > i915_gem_object_set_cache_level() which unbinds VMAs and triggers the 
> > kernel panic described in VIZ-3721.  This test will pass if the bug fix is in place.
> > 
> > Bugzilla:  https://bugs.fredesktop.org/show_bug.cgi?id=76384
> > Signed-off-by:  Armin Reese <[e-mail]>
> > ---
> >  tests/gem_partial_pwrite_pread.c | 11 +++++++++++
> >  1 file changed, 11 insertions(+)
> > 
> > diff --git a/tests/gem_partial_pwrite_pread.c 
> > b/tests/gem_partial_pwrite_pread.c
> > index 7945ba7..3e3ecc9 100644
> > --- a/tests/gem_partial_pwrite_pread.c
> > +++ b/tests/gem_partial_pwrite_pread.c
> > @@ -131,6 +131,14 @@ static void test_partial_reads(void)
> >  
> >  }
> >  
> > +static void test_set_cache(void)
> > +{
> > +	igt_info("checking set-cache\n");
> > +
> > +	blt_bo_fill(staging_bo, scratch_bo, 0);
> > +
> > +}
> 
> Ok, my apologies for the unclear JIRA description. My idea was to have a tailor-made testcase with a minimal reproduction scenario (i.e. map into 2 ppgtt or ggtt + ppgtt, then do a set_caching ioctl).
> 
> This here blows up because we do the set_caching as part of the fixture, so has a high risk that when someone decides to refactor the testcase we'll no longer test this bug. Hence the task to create a new, specific testcase.
> -Daniel
> 
> > +
> >  static void test_partial_writes(void)  {
> >  	int i, j;
> > @@ -244,6 +252,9 @@ static void do_tests(int cache_level, const char *suffix)
> >  			gem_set_caching(fd, scratch_bo->handle, cache_level);
> >  	}
> >  
> > +	igt_subtest_f("set-cache%s", suffix)
> > +		test_set_cache();
> > +
> >  	igt_subtest_f("reads%s", suffix)
> >  		test_partial_reads();
> >  
> > --
> > 1.9.1
> > 
> > _______________________________________________
> > Intel-gfx mailing list
> > [e-mail]
> > http://lists.freedesktop.org/mailman/listinfo/intel-gfx
> 
> --
> Daniel Vetter
> Software Engineer, Intel Corporation
> [phone] - http://blog.ffwll.ch

Patch

diff --git a/tests/gem_partial_pwrite_pread.c b/tests/gem_partial_pwrite_pread.c
index 7945ba7..3e3ecc9 100644
--- a/tests/gem_partial_pwrite_pread.c
+++ b/tests/gem_partial_pwrite_pread.c
@@ -131,6 +131,14 @@  static void test_partial_reads(void)
 
 }
 
+static void test_set_cache(void)
+{
+	igt_info("checking set-cache\n");
+
+	blt_bo_fill(staging_bo, scratch_bo, 0);
+
+}
+
 static void test_partial_writes(void)
 {
 	int i, j;
@@ -244,6 +252,9 @@  static void do_tests(int cache_level, const char *suffix)
 			gem_set_caching(fd, scratch_bo->handle, cache_level);
 	}
 
+	igt_subtest_f("set-cache%s", suffix)
+		test_set_cache();
+
 	igt_subtest_f("reads%s", suffix)
 		test_partial_reads();