Re: [RFC PATCH 1/1] drm/meson: Support drm_panic

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On 02/10/2024 09:59, Neil Armstrong wrote:
Hi !

On 01/10/2024 23:04, Yao Zi wrote:
This patch implements drm_plane_helper_funcs.get_scanout_buffer for
primary plane, enabling meson-drm to work with drm_panic.

This implementation tries to use current framebuffer as scanout buffer.
In case of AFBC enabled, we disable the decoder path and adjust OSD1
parameters in get_scanout_buffer to make the buffer linear.

Tested on TTY and Wayland session (Sway).

Thanks for enabling this!


Signed-off-by: Yao Zi <ziyao@xxxxxxxxxxx>
---
  drivers/gpu/drm/meson/meson_plane.c | 47 +++++++++++++++++++++++++++--
  1 file changed, 44 insertions(+), 3 deletions(-)

diff --git a/drivers/gpu/drm/meson/meson_plane.c b/drivers/gpu/drm/ meson/meson_plane.c
index b43ac61201f3..b2def784c00d 100644
--- a/drivers/gpu/drm/meson/meson_plane.c
+++ b/drivers/gpu/drm/meson/meson_plane.c
@@ -20,6 +20,8 @@
  #include <drm/drm_framebuffer.h>
  #include <drm/drm_gem_atomic_helper.h>
  #include <drm/drm_gem_dma_helper.h>
+#include <drm/drm_gem_framebuffer_helper.h>
+#include <drm/drm_panic.h>
  #include "meson_plane.h"
  #include "meson_registers.h"
@@ -419,10 +421,49 @@ static void meson_plane_atomic_disable(struct drm_plane *plane,
      priv->viu.osd1_enabled = false;
  }
+static int meson_plane_get_scanout_buffer(struct drm_plane *plane,
+                      struct drm_scanout_buffer *sb)
+{
+    struct meson_plane *meson_plane = to_meson_plane(plane);
+    struct meson_drm *priv = meson_plane->priv;
+    struct drm_framebuffer *fb;
+
+    if (!meson_plane->enabled)
+        return -ENODEV;
+
+    if (priv->viu.osd1_afbcd) {
+        if (meson_vpu_is_compatible(priv, VPU_COMPATIBLE_GXM)) {

This should be meson_vpu_is_compatible(priv, VPU_COMPATIBLE_G12A)

You should call:

             if (priv->afbcd.ops) {
                 priv->afbcd.ops->reset(priv);
                 priv->afbcd.ops->disable(priv);
             }

+            writel_relaxed(0, priv->io_base +
+                      _REG(VIU_OSD1_BLK1_CFG_W4));
+            writel_relaxed(0, priv->io_base +
+                      _REG(VIU_OSD1_BLK2_CFG_W4));
+            writel_bits_relaxed(OSD_ENDIANNESS_LE, OSD_ENDIANNESS_LE,
+                        priv->io_base +
+                        _REG(VIU_OSD1_BLK0_CFG_W0));

This won't work, drop it, the canvas isn't correctly configured, you should instead call:
             meson_canvas_config(priv->canvas, priv->canvas_id_osd1,
                         priv->viu.osd1_addr,
                         priv->viu.osd1_stride,
                         priv->viu.osd1_height,
                         MESON_CANVAS_WRAP_NONE,
                         MESON_CANVAS_BLKMODE_LINEAR, 0);

The problem is meson_canvas_config() takes a spin_lock, which can block forever in a panic handler.
So you can rewrite a version without lock.
Considering the addr, stride, and height don't change in a panic, it can probably be simplified for this case.

Also this will be the last image to be drawn, and the last driver code to be executed before a reboot, so it's ok if it's a bit dirty, and if the software state is not up-to-date with the hardware state.


+            meson_viu_g12a_disable_osd1_afbc(priv);
+        } else if (meson_vpu_is_compatible(priv, VPU_COMPATIBLE_G12A)) {

And here meson_vpu_is_compatible(priv, VPU_COMPATIBLE_GXM)

+            writel_bits_relaxed(OSD_DPATH_MALI_AFBCD, 0,
+                        priv->io_base +
+                        _REG(VIU_OSD1_CTRL_STAT2));

Ok, you should also call meson_canvas_config()

You should call:

             if (priv->afbcd.ops) {
                 priv->afbcd.ops->reset(priv);
                 priv->afbcd.ops->disable(priv);
             }

+            meson_viu_gxm_disable_osd1_afbc(priv);
+        }
+    }

I thing the code should look like:

if (priv->viu.osd1_afbcd) {
     meson_canvas_config(priv->canvas, priv->canvas_id_osd1,
                 priv->viu.osd1_addr,
                 priv->viu.osd1_stride,
                 priv->viu.osd1_height,
                 MESON_CANVAS_WRAP_NONE,
                 MESON_CANVAS_BLKMODE_LINEAR, 0);

     if (priv->afbcd.ops) {
         priv->afbcd.ops->reset(priv);
         priv->afbcd.ops->disable(priv);
     }

     if (meson_vpu_is_compatible(priv, VPU_COMPATIBLE_G12A)) {
         writel_bits_relaxed(OSD_ENDIANNESS_LE, OSD_ENDIANNESS_LE,
                     priv->io_base +
                     _REG(VIU_OSD1_BLK0_CFG_W0));
         meson_viu_g12a_disable_osd1_afbc(priv);
     } else if (meson_vpu_is_compatible(priv, VPU_COMPATIBLE_GXM)) {
         writel_bits_relaxed(OSD_DPATH_MALI_AFBCD, 0,
                     priv->io_base +
                     _REG(VIU_OSD1_CTRL_STAT2));
         meson_viu_gxm_disable_osd1_afbc(priv);
     }
}

AFBC is quite hard to test since it requires DRM_FORMAT_XBGR8888, but
I think sway should perhaps support it, Mesa should also support AFBC.

At some point I made some memory dumps of AFBC buffers, perhaps they could
be useful here.

Another way would be to simply ignore the AFBC case, and bail out since
it would be a very rare case.

panic handler is a best effort mode, so if it's too complex to support AFBC, and it's not very widespread, then it's ok to bail out in this case.

+
+    fb = plane->state->fb;
+    sb->format    = fb->format;
+    sb->width    = fb->width;
+    sb->height    = fb->height;
+    sb->pitch[0]    = fb->pitches[0];
+    drm_gem_fb_vmap(fb, sb->map, NULL);
+
+    return 0;
+}
+
  static const struct drm_plane_helper_funcs meson_plane_helper_funcs = {
-    .atomic_check    = meson_plane_atomic_check,
-    .atomic_disable    = meson_plane_atomic_disable,
-    .atomic_update    = meson_plane_atomic_update,
+    .atomic_check        = meson_plane_atomic_check,
+    .atomic_disable        = meson_plane_atomic_disable,
+    .atomic_update        = meson_plane_atomic_update,
+    .get_scanout_buffer    = meson_plane_get_scanout_buffer,
  };
  static bool meson_plane_format_mod_supported(struct drm_plane *plane,

Thanks,
Neil


Best regards,

--

Jocelyn




[Index of Archives]     [Linux DRI Users]     [Linux Intel Graphics]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]     [XFree86]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Linux Kernel]     [Linux SCSI]     [XFree86]
  Powered by Linux