Skip to contentSkip to conversation

drm/hyperv: Replace simple-KMS with regular atomic helpers

PATCH Superseded Ryosuke Yasuoka 2 participants 4 messages Apr 25, 2025 – Apr 27, 2025

Subsystems (5)

v2 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 drm/hyperv: Replace simple-KMS with regular atomic helpers

Drop simple-KMS in favor of regular atomic helpers to make the code more modular. The simple-KMS helper mix up plane and CRTC state, so it is obsolete and should go away [1]. Since it just split the simple-pipe functions into per-plane and per-CRTC, no functional changes is expected.

[1] https://lore.kernel.org/lkml/[email protected]/

---
v2:
- Remove hyperv_crtc_helper_mode_valid
- Remove hyperv_format_mod_supported
- Call helper_add after {plane,crtc}_init
- Move drm_plane_enable_fb_damage_clips to a place close to plane init
- Move hyperv_conn_init() into hyperv_pipe_init
- Remove hyperv_blit_to_vram_fullscreen
- Remove format check
- Replace hyperv_crtc_helper_atomic_check to drm_crtc_helper_atomic_check

v1:
https://lore.kernel.org/all/[email protected]/

 drivers/gpu/drm/hyperv/hyperv_drm.h         |   4 +-
 drivers/gpu/drm/hyperv/hyperv_drm_modeset.c | 159 +++++++++++++-------
 2 files changed, 107 insertions(+), 56 deletions(-)
drivers/gpu/drm/hyperv/hyperv_drm.h +3 -1
index d2d8582b36df..9e776112c03e 100644
@@ -11,7 +11,9 @@
 struct hyperv_drm_device {
 	/* drm */
 	struct drm_device dev;
-	struct drm_simple_display_pipe pipe;
+	struct drm_plane plane;
+	struct drm_crtc crtc;
+	struct drm_encoder encoder;
 	struct drm_connector connector;
 
 	/* mode */
drivers/gpu/drm/hyperv/hyperv_drm_modeset.c +104 -55

230 lines not on this page · raw message

base-commit: b60301774a8fe6c30b14a95104ec099290a2e904

Signature · 1 line
2.49.0
Acked-by on v2 1/1
#1 lore raw

Ryosuke Yasuoka <[email protected]> writes:

Hello Ryosuke,

Quoted 15 lines

Drop simple-KMS in favor of regular atomic helpers to make the code more modular. The simple-KMS helper mix up plane and CRTC state, so it is obsolete and should go away [1]. Since it just split the simple-pipe functions into per-plane and per-CRTC, no functional changes is expected.

[1] https://lore.kernel.org/lkml/[email protected]/

drivers/gpu/drm/hyperv/hyperv_drm_modeset.c +0 -2
-static void hyperv_pipe_enable(struct drm_simple_display_pipe *pipe,
-			       struct drm_crtc_state *crtc_state,

drivers/gpu/drm/hyperv/hyperv_drm_modeset.c static int hyperv_check_size(struct hyperv_drm_device *hv, int w, int h, · line 99

Quoted 10 lines
drivers/gpu/drm/hyperv/hyperv_drm_modeset.c +9 -1
-			       struct drm_plane_state *plane_state)
+static const uint32_t hyperv_formats[] = {
+	DRM_FORMAT_XRGB8888,
+};
+
+static const uint64_t hyperv_modifiers[] = {
+	DRM_FORMAT_MOD_LINEAR,
+	DRM_FORMAT_MOD_INVALID
+};
+

I think the kernel u32 and u64 types are preferred ?

Quoted 15 lines
drivers/gpu/drm/hyperv/hyperv_drm_modeset.c +6 -2
+static void hyperv_crtc_helper_atomic_enable(struct drm_crtc *crtc,
+					     struct drm_atomic_state *state)
 {
-	struct hyperv_drm_device *hv = to_hv(pipe->crtc.dev);
-	struct drm_shadow_plane_state *shadow_plane_state = to_drm_shadow_plane_state(plane_state);
+	struct hyperv_drm_device *hv = to_hv(crtc->dev);
+	struct drm_plane *plane = &hv->plane;
+	struct drm_plane_state *plane_state = plane->state;
+	struct drm_crtc_state *crtc_state = crtc->state;
 
 	hyperv_hide_hw_ptr(hv->hdev);
 	hyperv_update_situation(hv->hdev, 1,  hv->screen_depth,
 				crtc_state->mode.hdisplay,
 				crtc_state->mode.vdisplay,
 				plane_state->fb->pitches[0]);

drivers/gpu/drm/hyperv/hyperv_drm_modeset.c static int hyperv_check_size(struct hyperv_drm_device *hv, int w, int h, · line 118

Quoted 10 lines
drivers/gpu/drm/hyperv/hyperv_drm_modeset.c +4 -4
-	hyperv_blit_to_vram_fullscreen(plane_state->fb, &shadow_plane_state->data[0]);
 }
 
-static int hyperv_pipe_check(struct drm_simple_display_pipe *pipe,
-			     struct drm_plane_state *plane_state,
-			     struct drm_crtc_state *crtc_state)
+static void hyperv_crtc_helper_atomic_disable(struct drm_crtc *crtc,
+					      struct drm_atomic_state *state)
+{ }
+

Why do you need an empty CRTC atomic disable callback? Can you just not set it instead?

drivers/gpu/drm/hyperv/hyperv_drm_modeset.c static int hyperv_pipe_check(struct drm_simple_display_pipe *pipe, · line 171

Quoted 11 lines
drivers/gpu/drm/hyperv/hyperv_drm_modeset.c +5 -4
-static void hyperv_pipe_update(struct drm_simple_display_pipe *pipe,
-			       struct drm_plane_state *old_state)
+static void hyperv_plane_atomic_update(struct drm_plane *plane,
+						      struct drm_atomic_state *old_state)
 {
-	struct hyperv_drm_device *hv = to_hv(pipe->crtc.dev);
-	struct drm_plane_state *state = pipe->plane.state;
+	struct drm_plane_state *old_pstate = drm_atomic_get_old_plane_state(old_state, plane);
+	struct hyperv_drm_device *hv = to_hv(plane->dev);
+	struct drm_plane_state *state = plane->state;

You should never access the plane->state directly, instead the helper drm_atomic_get_new_plane_state() should be used. You can also rename the old_state paramete to just state, since it will be used to lookup both the old and new atomic states.

More info is in the following email from Ville:

https://lore.kernel.org/dri-devel/[email protected]/

drivers/gpu/drm/hyperv/hyperv_drm_modeset.c static int hyperv_pipe_check(struct drm_simple_display_pipe *pipe, · line 175

Quoted 5 lines
drivers/gpu/drm/hyperv/hyperv_drm_modeset.c +1 -1
 	struct drm_shadow_plane_state *shadow_plane_state = to_drm_shadow_plane_state(state);
 	struct drm_rect rect;
 
-	if (drm_atomic_helper_damage_merged(old_state, state, &rect)) {
+	if (drm_atomic_helper_damage_merged(old_pstate, state, &rect)) {

I know that most of the simple-KMS drivers do this but since this driver enables FB damage clips support, it is better to iterate over the damage areas. For example:

	struct drm_atomic_helper_damage_iter iter;
        struct drm_rect dst_clip;
	struct drm_rect damage;

	drm_atomic_helper_damage_iter_init(&iter, old_pstate, state);
	drm_atomic_for_each_plane_damage(&iter, &damage) {
		dst_clip = state->dst;

		if (!drm_rect_intersect(&dst_clip, &damage))
			continue;

                hyperv_blit_to_vram_rect(state->fb, &shadow_plane_state->data[0], &damage);
                hyperv_update_dirt(hv->hdev, &damage);
        }

Other than these small comments, the patch looks good to me. So if you take into account my suggestions, feel free to add:

Signature · 5 lines
Best regards,

Javier Martinez Canillas
Core Platforms
Red Hat
#2 lore raw

Hi Javier,

Quoted 18 lines · Javier Martinez Canillas

Ryosuke Yasuoka <[email protected]> writes:

Hello Ryosuke,

Quoted 10 lines

Drop simple-KMS in favor of regular atomic helpers to make the code more modular. The simple-KMS helper mix up plane and CRTC state, so it is obsolete and should go away [1]. Since it just split the simple-pipe functions into per-plane and per-CRTC, no functional changes is expected.

[1] https://lore.kernel.org/lkml/[email protected]/

Quoted 14 lines

drivers/gpu/drm/hyperv/hyperv_drm_modeset.c static int hyperv_check_size(struct hyperv_drm_device *hv, int w, int h, · line 99

Quoted 12 lines
drivers/gpu/drm/hyperv/hyperv_drm_modeset.c +9 -3
-static void hyperv_pipe_enable(struct drm_simple_display_pipe *pipe,
-                            struct drm_crtc_state *crtc_state,
-                            struct drm_plane_state *plane_state)
+static const uint32_t hyperv_formats[] = {
+     DRM_FORMAT_XRGB8888,
+};
+
+static const uint64_t hyperv_modifiers[] = {
+     DRM_FORMAT_MOD_LINEAR,
+     DRM_FORMAT_MOD_INVALID
+};
+

I think the kernel u32 and u64 types are preferred ?

I'm not sure if I should fix this in this patch because I did not add these variables. IMO, we need to split the commit if we fix them.

Quoted 18 lines
Quoted 18 lines
drivers/gpu/drm/hyperv/hyperv_drm_modeset.c +6 -3
+static void hyperv_crtc_helper_atomic_enable(struct drm_crtc *crtc,
+                                          struct drm_atomic_state *state)
 {
-     struct hyperv_drm_device *hv = to_hv(pipe->crtc.dev);
-     struct drm_shadow_plane_state *shadow_plane_state = to_drm_shadow_plane_state(plane_state);
+     struct hyperv_drm_device *hv = to_hv(crtc->dev);
+     struct drm_plane *plane = &hv->plane;
+     struct drm_plane_state *plane_state = plane->state;
+     struct drm_crtc_state *crtc_state = crtc->state;
 
      hyperv_hide_hw_ptr(hv->hdev);
      hyperv_update_situation(hv->hdev, 1,  hv->screen_depth,
                              crtc_state->mode.hdisplay,
                              crtc_state->mode.vdisplay,
                              plane_state->fb->pitches[0]);
-     hyperv_blit_to_vram_fullscreen(plane_state->fb, &shadow_plane_state->data[0]);
 }
Quoted 10 lines

drivers/gpu/drm/hyperv/hyperv_drm_modeset.c static int hyperv_check_size(struct hyperv_drm_device *hv, int w, int h, · line 118

Quoted 7 lines
drivers/gpu/drm/hyperv/hyperv_drm_modeset.c +4 -3
-static int hyperv_pipe_check(struct drm_simple_display_pipe *pipe,
-                          struct drm_plane_state *plane_state,
-                          struct drm_crtc_state *crtc_state)
+static void hyperv_crtc_helper_atomic_disable(struct drm_crtc *crtc,
+                                           struct drm_atomic_state *state)
+{ }
+

Why do you need an empty CRTC atomic disable callback? Can you just not set it instead?

OK. I'll fix it in my next patch.

Quoted 17 lines

drivers/gpu/drm/hyperv/hyperv_drm_modeset.c static int hyperv_pipe_check(struct drm_simple_display_pipe *pipe, · line 171

Quoted 11 lines
drivers/gpu/drm/hyperv/hyperv_drm_modeset.c +5 -4
-static void hyperv_pipe_update(struct drm_simple_display_pipe *pipe,
-                            struct drm_plane_state *old_state)
+static void hyperv_plane_atomic_update(struct drm_plane *plane,
+                                                   struct drm_atomic_state *old_state)
 {
-     struct hyperv_drm_device *hv = to_hv(pipe->crtc.dev);
-     struct drm_plane_state *state = pipe->plane.state;
+     struct drm_plane_state *old_pstate = drm_atomic_get_old_plane_state(old_state, plane);
+     struct hyperv_drm_device *hv = to_hv(plane->dev);
+     struct drm_plane_state *state = plane->state;

You should never access the plane->state directly, instead the helper drm_atomic_get_new_plane_state() should be used. You can also rename the old_state paramete to just state, since it will be used to lookup both the old and new atomic states.

Quoted 3 lines

More info is in the following email from Ville:

https://lore.kernel.org/dri-devel/[email protected]/

OK. I'll fix it in my next patch. Thank you for sharing the url.

Quoted 10 lines

drivers/gpu/drm/hyperv/hyperv_drm_modeset.c static int hyperv_pipe_check(struct drm_simple_display_pipe *pipe, · line 175

Quoted 5 lines
      struct drm_shadow_plane_state *shadow_plane_state = to_drm_shadow_plane_state(state);
      struct drm_rect rect;
drivers/gpu/drm/hyperv/hyperv_drm_modeset.c +1 -1
-     if (drm_atomic_helper_damage_merged(old_state, state, &rect)) {
+     if (drm_atomic_helper_damage_merged(old_pstate, state, &rect)) {

I know that most of the simple-KMS drivers do this but since this driver enables FB damage clips support, it is better to iterate over the damage areas. For example:

Quoted 15 lines
        struct drm_atomic_helper_damage_iter iter;
        struct drm_rect dst_clip;
        struct drm_rect damage;

        drm_atomic_helper_damage_iter_init(&iter, old_pstate, state);
        drm_atomic_for_each_plane_damage(&iter, &damage) {
                dst_clip = state->dst;

                if (!drm_rect_intersect(&dst_clip, &damage))
                        continue;

                hyperv_blit_to_vram_rect(state->fb, &shadow_plane_state->data[0], &damage);
                hyperv_update_dirt(hv->hdev, &damage);
        }

OK. As you said, other drivers like mgag200 implement like this. I'll fix them in my next patch.

Quoted 4 lines

Other than these small comments, the patch looks good to me. So if you take into account my suggestions, feel free to add:

Thank you for your review and comment. I'll fix them and add your ack.

Best regards,
Ryosuke

Quoted 7 lines

--
Best regards,

Javier Martinez Canillas
Core Platforms
Red Hat

#3 lore raw

Ryosuke Yasuoka <[email protected]> writes:

Hello Ryosuke,

Quoted 37 lines

Hi Javier,

Quoted 18 lines · Javier Martinez Canillas

Ryosuke Yasuoka <[email protected]> writes:

Hello Ryosuke,

Quoted 10 lines

Drop simple-KMS in favor of regular atomic helpers to make the code more modular. The simple-KMS helper mix up plane and CRTC state, so it is obsolete and should go away [1]. Since it just split the simple-pipe functions into per-plane and per-CRTC, no functional changes is expected.

[1] https://lore.kernel.org/lkml/[email protected]/

Quoted 14 lines

drivers/gpu/drm/hyperv/hyperv_drm_modeset.c static int hyperv_check_size(struct hyperv_drm_device *hv, int w, int h, · line 99

Quoted 12 lines
drivers/gpu/drm/hyperv/hyperv_drm_modeset.c +9 -3
-static void hyperv_pipe_enable(struct drm_simple_display_pipe *pipe,
-                            struct drm_crtc_state *crtc_state,
-                            struct drm_plane_state *plane_state)
+static const uint32_t hyperv_formats[] = {
+     DRM_FORMAT_XRGB8888,
+};
+
+static const uint64_t hyperv_modifiers[] = {
+     DRM_FORMAT_MOD_LINEAR,
+     DRM_FORMAT_MOD_INVALID
+};
+

I think the kernel u32 and u64 types are preferred ?

Quoted 3 lines

I'm not sure if I should fix this in this patch because I did not add these variables. IMO, we need to split the commit if we fix them.

Right, I got confused for how the diff showed the changes. But I agree with you that should be a separate patch since the variables already exist.

[...]

Quoted 5 lines
Quoted 2 lines

Thank you for your review and comment. I'll fix them and add your ack.

Thanks!

Quoted 3 lines

Best regards,
Ryosuke

Signature · 5 lines
Best regards,

Javier Martinez Canillas
Core Platforms
Red Hat