[igt-dev] [ PATCH i-g-t v1 ] tests/i915/i915_pm_dc: Fix DC9 test

[Gupta, Anshuman]

> -----Original Message-----
> From: Deak, Imre <imre.deak at intel.com>
> Sent: Thursday, September 9, 2021 12:25 AM
> To: Bhatt, Jigar <jigar.bhatt at intel.com>
> Cc: igt-dev at lists.freedesktop.org; Shankar, Uma <uma.shankar at intel.com>;
> Varide, Nischal <nischal.varide at intel.com>; Gupta, Anshuman
> <anshuman.gupta at intel.com>
> Subject: Re: [ PATCH i-g-t v1 ] tests/i915/i915_pm_dc: Fix DC9 test
> 
> On Wed, Sep 08, 2021 at 04:45:43PM +0530, Jigar Bhatt wrote:
> > Currently, check_dc9 is being called with reference
> > (previous) counter being read after dpms_off call.
> > At this time, already the counter is 0.
> > We need to read the counter before dpms_off is called so that it holds
> > the previous value which had incremented while testing shallow states
> > (DC5 or DC6).
> >
> > Signed-off-by: Jigar Bhatt <jigar.bhatt at intel.com>
> > ---
> >  tests/i915/i915_pm_dc.c | 11 ++++++-----
> >  1 file changed, 6 insertions(+), 5 deletions(-)
> >
> > diff --git a/tests/i915/i915_pm_dc.c b/tests/i915/i915_pm_dc.c index
> > 9d0a15d8..d1925e5a 100644
> > --- a/tests/i915/i915_pm_dc.c
> > +++ b/tests/i915/i915_pm_dc.c
> > @@ -400,8 +400,8 @@ static bool check_dc9(uint32_t debugfs_fd, int
> prev_dc, bool dc6_supported, int
> >  	 * so we rely on dc5/dc6 counter reset to check if display engine was in
> DC9.
> >  	 */
> >  	return igt_wait(dc6_supported ? read_dc_counter(debugfs_fd,
> CHECK_DC6) <
> > -			prev_dc : read_dc_counter(debugfs_fd, CHECK_DC5) <
> > -			prev_dc, seconds, 100);
> > +			prev_dc : read_dc_counter(debugfs_fd, CHECK_DC5) <
> prev_dc
> > +			, seconds, 100);
> >  }
> >
> >  static void setup_dc9_dpms(data_t *data, int prev_dc, bool
> > dc6_supported) @@ -418,15 +418,16 @@ static void setup_dc9_dpms(data_t
> > *data, int prev_dc, bool dc6_supported)  static void
> > test_dc9_dpms(data_t *data)
> >
> >  	bool dc6_supported;
> 
> The fix looks good, but while at it could you also make things a bit cleaner by
> 
> 	int dc_target = support_dc6() ? CHECK_DC6 : CHECK_DC5;
> 
> > +	int prev_dc;
> >
> >  	require_dc_counter(data->debugfs_fd, CHECK_DC5);
> >  	dc6_supported = support_dc6(data->debugfs_fd);
> >  	setup_dc9_dpms(data, dc6_supported ? read_dc_counter(data-
> >debugfs_fd, CHECK_DC6) :
> >  			read_dc_counter(data->debugfs_fd, CHECK_DC5),
> dc6_supported);
> 
> 	setup_dc9_dpms(data, dc_target);
> 
> and read out the counter within setup_dc9_dpms().
> 
> > +	prev_dc = dc6_supported ? read_dc_counter(data->debugfs_fd,
> CHECK_DC6) :
> > +		read_dc_counter(data->debugfs_fd, CHECK_DC5);
> 
> 	prev_dc = read_dc_counter(dc_target);
> 
> >  	dpms_off(data);
> > -	igt_assert_f(check_dc9(data->debugfs_fd, dc6_supported ?
> > -				read_dc_counter(data->debugfs_fd,
> CHECK_DC6) :
> > -				read_dc_counter(data->debugfs_fd,
> CHECK_DC5),
> > +	igt_assert_f(check_dc9(data->debugfs_fd, prev_dc,
> >  				dc6_supported, 3000), "Not in DC9\n");
This test can also fail due to any wakeup which lead to runtime resume (kms connector polling is one of example), which will disable
DC9 in DC_STATE_EN register. To debug such failure it would be better to dump the i915_pm_runtime_status debugfs  file in case of test failure.
Like below example:
igt_assert_f(dc_state_wait_entry(data->debugfs_fd, dc_flag, prev_dc_count),
                     "%s st\n%s:\n%s\n", tmp, PWR_DOMAIN_INFO,
                     data->pwr_dmn_info = igt_sysfs_get(data->debugfs_fd, PWR_DOMAIN_INFO));

Br,
Anshuman Gupta.
> 
> 		     check_dc9(data->debugfs_fd, target_dc, 3000, ...
> 
> and simplify check_dc9() accordingly?
> 
> >  	dpms_on(data);
> >  }
> > --
> > 2.25.1
> >