[Intel-xe] [PATCH v2 1/2] drm/xe: add skip_pcode flag

[Rodrigo Vivi]

On Tue, Oct 24, 2023 at 12:15:44PM -0500, Lucas De Marchi wrote:
> On Tue, Oct 24, 2023 at 01:40:11PM +0300, Jani Nikula wrote:
> > On Sun, 22 Oct 2023, Koby Elbaz <kelbaz at habana.ai> wrote:
> > > Per device, set this flag to enable access to the PCODE uC or to skip it.
> > > 
> > > Signed-off-by: Koby Elbaz <kelbaz at habana.ai>
> > > ---
> > > Changes in v2:
> > > - Added the flag 'skip_pcode' instead of 'has_pcode'
> > 
> > I don't know why change was made from positive to negative flag.
> > 
> > Generally it's better to use positive flags, because they just read
> > better. Negative flags often lead to double negatives in source, such as
> > !disable.
> > 
> > Please consider using a positive flag, even if it's not "has_something".
> 
> https://lore.kernel.org/all/qvuhtciy6gasz4s5f5opebdepse7x2ubfttnb4alhjq4e5zz4b@fjs6d67fcmex/
> 
> The positive here is harder to work with IMO when thing "stop to exist"
> or when they are added to cope to certain environment limitations.
> That's because a) you have to go over all the past platform info to add
> the opposite, you can't rely on it being zero-initialized anymore.
> b) it's very easy to forget that for platforms that are wip as there are
> no git conflicts to serve as a warning.
> 
> So, I think adding it as "skip" or another better verb could be
> appropriate. At least while there is no single user of that flag and
> platforms being worked on @habana and @intel, IMO this should help us
> not shoot on each others' foot.

When I first read this thread I was also with the impression that the
positive has_pcode would be better. But then after chatting offline
with Lucas, Matt Roper and others, I changed my mind.

For the cases where it is rare to have to skip, the negative is better
to maintain, and then, if down the road, that becomes the most common
behavior we can flip from negative to positive.

I also liked a suggestion from Matt about using 'lacks_' instead of 'skip_',
however I don't have a strong preference and on the second patch in this
series, the 'skip' is better.

So, let's just unblock the progress here and move forward with what we
have here and we can always improve or flip things later as needed.

Reviewed-by: Rodrigo Vivi <rodrigo.vivi at intel.com>

> 
> Lucas De Marchi
> 
> > 
> > BR,
> > Jani.
> > 
> > 
> > > 
> > >  drivers/gpu/drm/xe/xe_device_types.h | 2 ++
> > >  drivers/gpu/drm/xe/xe_pci.c          | 2 ++
> > >  drivers/gpu/drm/xe/xe_pcode.c        | 9 +++++++++
> > >  3 files changed, 13 insertions(+)
> > > 
> > > diff --git a/drivers/gpu/drm/xe/xe_device_types.h b/drivers/gpu/drm/xe/xe_device_types.h
> > > index 44d622d4cc3a..29199671d9e8 100644
> > > --- a/drivers/gpu/drm/xe/xe_device_types.h
> > > +++ b/drivers/gpu/drm/xe/xe_device_types.h
> > > @@ -257,6 +257,8 @@ struct xe_device {
> > >  		u8 enable_display:1;
> > >  		/** @bypass_mtcfg: Bypass Multi-Tile configuration from MTCFG register */
> > >  		u8 bypass_mtcfg:1;
> > > +		/** @skip_pcode: skip access to PCODE uC */
> > > +		u8 skip_pcode:1;
> > >  		/** @supports_mmio_ext: supports MMIO extension/s */
> > >  		u8 supports_mmio_ext:1;
> > >  		/** @has_heci_gscfi: device has heci gscfi */
> > > diff --git a/drivers/gpu/drm/xe/xe_pci.c b/drivers/gpu/drm/xe/xe_pci.c
> > > index 2fae45b9d88e..05b0aa9f2ab7 100644
> > > --- a/drivers/gpu/drm/xe/xe_pci.c
> > > +++ b/drivers/gpu/drm/xe/xe_pci.c
> > > @@ -61,6 +61,7 @@ struct xe_device_desc {
> > > 
> > >  	u8 has_llc:1;
> > >  	u8 bypass_mtcfg:1;
> > > +	u8 skip_pcode:1;
> > >  	u8 supports_mmio_ext:1;
> > >  };
> > > 
> > > @@ -578,6 +579,7 @@ static int xe_info_init(struct xe_device *xe,
> > >  	xe->info.media_name = media_desc ? media_desc->name : "none";
> > >  	xe->info.has_llc = desc->has_llc;
> > >  	xe->info.bypass_mtcfg = desc->bypass_mtcfg;
> > > +	xe->info.skip_pcode = desc->skip_pcode;
> > >  	xe->info.supports_mmio_ext = desc->supports_mmio_ext;
> > >  	xe->info.tile_mmio_ext_size = graphics_desc->tile_mmio_ext_size;
> > > 
> > > diff --git a/drivers/gpu/drm/xe/xe_pcode.c b/drivers/gpu/drm/xe/xe_pcode.c
> > > index 4a240acf7625..b324dc2a5deb 100644
> > > --- a/drivers/gpu/drm/xe/xe_pcode.c
> > > +++ b/drivers/gpu/drm/xe/xe_pcode.c
> > > @@ -61,6 +61,9 @@ static int pcode_mailbox_rw(struct xe_gt *gt, u32 mbox, u32 *data0, u32 *data1,
> > >  {
> > >  	int err;
> > > 
> > > +	if (gt_to_xe(gt)->info.skip_pcode)
> > > +		return 0;
> > > +
> > >  	lockdep_assert_held(&gt->pcode.lock);
> > > 
> > >  	if ((xe_mmio_read32(gt, PCODE_MAILBOX) & PCODE_READY) != 0)
> > > @@ -249,6 +252,9 @@ int xe_pcode_init(struct xe_gt *gt)
> > >  	int timeout_us = 180000000; /* 3 min */
> > >  	int ret;
> > > 
> > > +	if (gt_to_xe(gt)->info.skip_pcode)
> > > +		return 0;
> > > +
> > >  	if (!IS_DGFX(gt_to_xe(gt)))
> > >  		return 0;
> > > 
> > > @@ -280,6 +286,9 @@ int xe_pcode_probe(struct xe_gt *gt)
> > >  {
> > >  	drmm_mutex_init(&gt_to_xe(gt)->drm, &gt->pcode.lock);
> > > 
> > > +	if (gt_to_xe(gt)->info.skip_pcode)
> > > +		return 0;
> > > +
> > >  	if (!IS_DGFX(gt_to_xe(gt)))
> > >  		return 0;
> > 
> > -- 
> > Jani Nikula, Intel