[v2] drm/xe/fbdev: Limit the usage of stolen for LNL+

[Lucas De Marchi]

On Mon, Jul 15, 2024 at 02:26:59AM GMT, Uma Shankar wrote:
>As per recommendation in the workarounds:
>WA_22019338487
>
>There is an issue with accessing Stolen memory pages due a
>hardware limitation. Limit the usage of stolen memory for
>fbdev for LNL+. Don't use BIOS FB from stolen on LNL+ and
>assign the same from system memory.
>
>v2: Corrected the WA Number, limited WA to LNL and
>    Adopted XE_WA framework as suggested by Lucas and Matt.
>
>Signed-off-by: Uma Shankar <uma.shankar at intel.com>
>---
> drivers/gpu/drm/xe/display/intel_fbdev_fb.c   | 20 ++++++++++++++++++-
> drivers/gpu/drm/xe/display/xe_plane_initial.c | 12 +++++++++++
> drivers/gpu/drm/xe/xe_wa_oob.rules            |  1 +
> 3 files changed, 32 insertions(+), 1 deletion(-)
>
>diff --git a/drivers/gpu/drm/xe/display/intel_fbdev_fb.c b/drivers/gpu/drm/xe/display/intel_fbdev_fb.c
>index 816ad13821a8..9c70c9158108 100644
>--- a/drivers/gpu/drm/xe/display/intel_fbdev_fb.c
>+++ b/drivers/gpu/drm/xe/display/intel_fbdev_fb.c
>@@ -10,6 +10,8 @@
> #include "xe_bo.h"
> #include "xe_gt.h"
> #include "xe_ttm_stolen_mgr.h"
>+#include "xe_wa.h"

missing newline

>+#include <generated/xe_wa_oob.h>
>
> struct intel_framebuffer *intel_fbdev_fb_alloc(struct drm_fb_helper *helper,
> 					       struct drm_fb_helper_surface_size *sizes)
>@@ -20,6 +22,9 @@ struct intel_framebuffer *intel_fbdev_fb_alloc(struct drm_fb_helper *helper,
> 	struct drm_mode_fb_cmd2 mode_cmd = {};
> 	struct drm_i915_gem_object *obj;
> 	int size;
>+	bool wa_22019338487 = false;
>+	struct xe_gt *gt;
>+	u8 id;
>
> 	/* we don't do packed 24bpp */
> 	if (sizes->surface_bpp == 24)
>@@ -37,7 +42,19 @@ struct intel_framebuffer *intel_fbdev_fb_alloc(struct drm_fb_helper *helper,
> 	size = PAGE_ALIGN(size);
> 	obj = ERR_PTR(-ENODEV);
>
>-	if (!IS_DGFX(xe)) {
>+	/*
>+	 * WA_22019338487:
>+	 * There is an issue with accessing Stolen memory pages
>+	 * due a hardware limitation. Limit the usage of stolen
>+	 * memory for fbdev for LNL+. Don't use BIOS FB from
>+	 * stolen on LNL+ and assign the same from system memory
>+	 */
>+	for_each_gt(gt, xe, id) {

why do you loop here, but in the other path you use main_gt of tile0?

I think at this point it's pretty safe to just do:

	if (XE_WA(xe_root_mmio_gt(xe), 22019338487))

Also, no need for the comment above, the commit message and WA
documentation is sufficient.

>+		if (XE_WA(gt, 22019338487))
>+			wa_22019338487 = true;
>+	}
>+
>+	if (!IS_DGFX(xe) && !wa_22019338487) {
> 		obj = xe_bo_create_pin_map(xe, xe_device_get_root_tile(xe),
> 					   NULL, size,
> 					   ttm_bo_type_kernel, XE_BO_FLAG_SCANOUT |
>@@ -48,6 +65,7 @@ struct intel_framebuffer *intel_fbdev_fb_alloc(struct drm_fb_helper *helper,
> 		else
> 			drm_info(&xe->drm, "Allocated fbdev into stolen failed: %li\n", PTR_ERR(obj));
> 	}
>+
> 	if (IS_ERR(obj)) {
> 		obj = xe_bo_create_pin_map(xe, xe_device_get_root_tile(xe), NULL, size,
> 					   ttm_bo_type_kernel, XE_BO_FLAG_SCANOUT |
>diff --git a/drivers/gpu/drm/xe/display/xe_plane_initial.c b/drivers/gpu/drm/xe/display/xe_plane_initial.c
>index 5eccd6abb3ef..7e93ddad6df8 100644
>--- a/drivers/gpu/drm/xe/display/xe_plane_initial.c
>+++ b/drivers/gpu/drm/xe/display/xe_plane_initial.c
>@@ -18,6 +18,8 @@
> #include "intel_frontbuffer.h"
> #include "intel_plane_initial.h"
> #include "xe_bo.h"
>+#include "xe_wa.h"
>+#include <generated/xe_wa_oob.h>
>
> static bool
> intel_reuse_initial_plane_obj(struct intel_crtc *this,
>@@ -104,6 +106,16 @@ initial_plane_bo(struct xe_device *xe,
> 		phys_base = base;
> 		flags |= XE_BO_FLAG_STOLEN;
>
>+		/*
>+		 * WA_22019338487:
>+		 * There is an issue with accessing Stolen memory pages
>+		 * due a hardware limitation. Limit the usage of stolen
>+		 * memory for fbdev for LNL+. Don't use BIOS FB from
>+		 * stolen on LNL+ and assign the same from system memory
>+		 */
>+		if (XE_WA(tile0->primary_gt, 22019338487))

just use the same xe_root_mmio_gt() as suggested above.

>+			return NULL;
>+
> 		/*
> 		 * If the FB is too big, just don't use it since fbdev is not very
> 		 * important and we should probably use that space with FBC or other
>diff --git a/drivers/gpu/drm/xe/xe_wa_oob.rules b/drivers/gpu/drm/xe/xe_wa_oob.rules
>index 08f7336881e3..9b08fedbf85c 100644
>--- a/drivers/gpu/drm/xe/xe_wa_oob.rules
>+++ b/drivers/gpu/drm/xe/xe_wa_oob.rules
>@@ -29,4 +29,5 @@
> 13011645652	GRAPHICS_VERSION(2004)
> 22019338487	MEDIA_VERSION(2000)
> 		GRAPHICS_VERSION(2001)
>+		GRAPHICS_VERSION(2004)

This will end up matching the graphics GT in LNL for other paths it was
not previously taking. Looking at the code, main change will be:

drivers/gpu/drm/xe/xe_guc_pc.c:pc_max_freq_cap()
drivers/gpu/drm/xe/xe_gt.c:xe_gt_sanitize_freq()

about the freq handling for the GT. And the change will be wrong

I think we could just make this a new entry with:

	22019338487_display GRAPHICS_VERSION(2024)
or
	22019338487_display PLATFORM(LUNARLAKE)

I like the second more as then it doesn't matter what gt you use in the
code. Matt Roper, thoughts?

Lucas De Marchi

> 16023588340	GRAPHICS_VERSION(2001)
>-- 
>2.42.0
>