Thread (4 messages) flat view 4 messages, 2 authors, 22h ago
HOTtoday

Revision v4 of 4 in this series.

Revisions (4)
  1. rfc [diff vs current]
  2. v2 [diff vs current]
  3. v3 [diff vs current]
  4. v4 current

[PATCH v4 0/2] drm: Add DRM driver for GlandaGPU (VHDL soft-IP GPU)

From: Leander Kieweg <hidden>
Date: 2026-09-13 20:50:14
Also in: dri-devel

This is v4 of the GlandaGPU DRM driver series [1][2][3]. It addresses
review feedback from Thomas Zimmermann and the automated review bot
on v3.

GlandaGPU is a small VHDL soft-IP 2D display controller, currently
targeting a Terasic DE10-Standard (Cyclone V SoC). This series has
been tested against a QEMU digital twin and on real hardware.

Hardware/VHDL:   https://github.com/stiangglanda/GlandaGPU
QEMU fork:       https://github.com/stiangglanda/qemu-glandagpu
Userspace tests: https://github.com/stiangglanda/GlandaGPU-userspace-tests

Changes since v3:

dt-bindings:
- Fix alphabetical ordering of the '^kieweg,.*' vendor prefix in
  vendor-prefixes.yaml.

driver core:
- Use devm_ioremap_wc() instead of devm_ioremap() for the VRAM
  mapping in both the platform and PCI probe paths, avoiding a severe
  performance regression from uncached writes (Sashiko bot).
- Only enable the hardware VSYNC interrupt when an IRQ handler is
  actually registered. enable_vblank() now returns -EINVAL when
  falling back to polling mode, instead of risking an unhandled
  interrupt storm (Sashiko bot).
- Stop unconditionally enabling the VSYNC interrupt during probe.
  Let the DRM core enable/disable it through enable_vblank()/
  disable_vblank() as needed (Thomas Zimmermann, Sashiko bot).
- Guard the PCI probe/remove code and pci_driver structure with
  #ifdef CONFIG_PCI so the driver builds with
  CONFIG_COMPILE_TEST=y && CONFIG_PCI=n (Sashiko bot).
- Use container_of_const() instead of container_of() (Thomas
  Zimmermann).
- Add glanda_plane_atomic_disable(), which blanks VRAM when the
  plane is disabled, instead of silently returning on a NULL fb
  (Thomas Zimmermann).
- Wrap direct access to the shadow-plane buffer object in
  drm_gem_fb_begin_cpu_access()/drm_gem_fb_end_cpu_access() to
  synchronize against imported buffers (Thomas Zimmermann).
- Switch atomic_update() to damage-clipped blitting via
  drm_atomic_helper_damage_iter instead of copying the whole frame
  on every update (Thomas Zimmermann).
- Always call drm_atomic_helper_check_plane_state() in
  atomic_check(), even when the plane has no CRTC yet (Thomas
  Zimmermann).
- Support panning within a larger, system-allocated framebuffer by
  calculating the correct source offset when reading pixel data 
  (Thomas Zimmermann).
- Remove glanda_connector_detect(). The default "connected" status
  is sufficient (Thomas Zimmermann).
- Wrap hardware register access in enable_vblank()/disable_vblank()
  with drm_dev_enter()/drm_dev_exit() (Thomas Zimmermann).
- Use drm_crtc_vblank_atomic_enable()/drm_crtc_vblank_atomic_disable()
  instead of custom wrapper functions (Thomas Zimmermann).
- Use drmm_mode_config_init() so the mode-config pipeline is cleaned
  up automatically (Thomas Zimmermann).
- Raise mode_config.max_width/max_height to
  DRM_SHADOW_PLANE_MAX_WIDTH/DRM_SHADOW_PLANE_MAX_HEIGHT instead of
  the fixed 640x480, so userspace can allocate larger framebuffers
  (Thomas Zimmermann).
- Call drm_plane_enable_fb_damage_clips() to enable damage clipping
  (Thomas Zimmermann).
- Move drm_vblank_init() to right before drm_mode_config_reset()
  (Thomas Zimmermann).
- Remove the manual drm_helper_probe_single_connector_modes() call
  during init. The DRM core probes modes on demand (Thomas
  Zimmermann).
- Simplify glanda_drm_fini() to just drm_dev_unplug(). The DRM core
  handles vblank/IRQ teardown after unplug (Thomas Zimmermann).
- Drop "Hardware Accelerated" from the driver description (Thomas
  Zimmermann).

Regarding the panning support: Since the physical VRAM is fixed to 
640x480, the display output itself cannot be panned. Instead, the 
panning is handled on the source side. If userspace allocates a 
larger framebuffer, the driver now calculates the correct src_x and 
src_y offsets from the plane state and copies only the requested 
sub-region into VRAM. Please let me know if this implementation 
matches what you had in mind with the sysfb reference.

[1] v1: https://lore.kernel.org/dri-devel/20260714101146.200416-1-kieweg.leander@gmail.com/T/#t (local)
[2] v2: https://lore.kernel.org/dri-devel/20260730173643.256052-1-kieweg.leander@gmail.com/T/#t (local)
[3] v3: https://lore.kernel.org/dri-devel/20260824195418.17707-1-kieweg.leander@gmail.com/T/#t (local)

Leander Kieweg (2):
  dt-bindings: display: Add GlandaGPU binding
  drm/glanda: Add initial DRM driver for GlandaGPU

 .../bindings/display/kieweg,gpu.yaml          |  55 ++
 .../devicetree/bindings/vendor-prefixes.yaml  |   2 +
 MAINTAINERS                                   |   6 +
 drivers/gpu/drm/tiny/Kconfig                  |  11 +
 drivers/gpu/drm/tiny/Makefile                 |   1 +
 drivers/gpu/drm/tiny/glandagpu.c              | 642 ++++++++++++++++++
 6 files changed, 717 insertions(+)
 create mode 100644 Documentation/devicetree/bindings/display/kieweg,gpu.yaml
 create mode 100644 drivers/gpu/drm/tiny/glandagpu.c

-- 
2.43.0
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help