Bug 443357

I'm trying to do a PR for this, but me writing C++ is like the old gag about dancing bears, so it may take a minute. :D

Hum, well. Turning off atomic mode setting doesn't actually solve the problem, for some reason, on Plasma. It does on GNOME, but not Plasma. I thought my patch wasn't working so I tested with KWIN_DRM_NO_AMS=1 in /etc/environment and confirmed via `qdbus org.kde.KWin / KWin supportInformation` output that AMS is disabled:

Atomic Mode Setting on GPU 0: false

but the cursor offset is still present when running Plasma on Wayland. When running Plasma on X11, there is no offset.

jadahl - who dealt with the similar problems on the mutter (GNOME) side - passes this along:

"one possibility is that kde changed to do what mutter did for a while, which is to emulate atomic mode setting with legacy mode setting. what mutter did was to offset things manually, always setting a 0,0 hotspot. when the virtual machine viewer bugs showed up, I had to re-introduce the hotspot to the atomic mode setting layer even if it couldn't be used"

Aha. Looks like this was fixed on the development branch:

https://github.com/KDE/kwin/commit/998bbf4eba724a9b94a5bae62182456b85b11383#diff-034885748897f413c645e3efd125d7c0db9a8454d580f6093e40c885d974c818

unfortunately, that was after a big change that moved a ton of code around:

https://github.com/KDE/kwin/commit/586efe94d4f3b2c9759a26f27661edcaa54dce0b#diff-034885748897f413c645e3efd125d7c0db9a8454d580f6093e40c885d974c818

I'll have a go at backporting it to the 5.22 stable branch today, as that's what we need downstream for Fedora 35.

So, it looks like the DrmPipeline addition changed all the cursor setting code considerably, such that it's beyond me how the bug can be fixed on the 5.22 branch. CCing Andrey Butirsky, who wrote the fix on the development branch - Andrey, would you possibly have time to see if this can be fixed on the 5.22 code? Thanks!

Oh, hmm, think I may have managed it myself after all. Does anyone see anything wrong with this?

diff --git a/src/plugins/platforms/drm/drm_output.cpp b/src/plugins/platforms/drm/drm_output.cpp
index 917bb857d..7ee387219 100644
--- a/src/plugins/platforms/drm/drm_output.cpp
+++ b/src/plugins/platforms/drm/drm_output.cpp
@@ -102,10 +102,15 @@ bool DrmOutput::hideCursor()
     return drmModeSetCursor(m_gpu->fd(), m_crtc->id(), 0, 0, 0) == 0;
 }
 
-bool DrmOutput::showCursor(DrmDumbBuffer *c)
+bool DrmOutput::showCursor(DrmDumbBuffer *c, const QPoint &hotspot)
 {
     const QSize &s = c->size();
-    if (drmModeSetCursor(m_gpu->fd(), m_crtc->id(), c->handle(), s.width(), s.height()) == 0) {
+    int ret = drmModeSetCursor2(m_gpu->fd(), m_crtc->id(), c->handle(), s.width(), s.height(), hotspot.x(), hotspot.y());
+    if (ret == ENOTSUP) {
+        // for NVIDIA case that does not support drmModeSetCursor2
+        ret = drmModeSetCursor(m_gpu->fd(), m_crtc->id(), c->handle(), s.width(), s.height());
+    }
+    if (ret == 0) {
         if (RenderLoopPrivate::get(m_renderLoop)->presentMode == RenderLoopPrivate::SyncMode::Adaptive
             && isCursorVisible()) {
             m_renderLoop->scheduleRepaint();
@@ -122,7 +127,8 @@ bool DrmOutput::showCursor()
         return false;
     }
 
-    const bool ret = showCursor(m_cursor[m_cursorIndex].data());
+    const Cursor * const cursor = Cursors::self()->currentCursor();
+    const bool ret = showCursor(m_cursor[m_cursorIndex].data(), logicalToNativeMatrix(cursor->rect(), scale(), transform()).map(cursor->hotspot()));
     if (!ret) {
         qCDebug(KWIN_DRM) << "DrmOutput::showCursor(DrmDumbBuffer) failed";
         return ret;
diff --git a/src/plugins/platforms/drm/drm_output.h b/src/plugins/platforms/drm/drm_output.h
index 1f89f9064..af46c88a0 100644
--- a/src/plugins/platforms/drm/drm_output.h
+++ b/src/plugins/platforms/drm/drm_output.h
@@ -45,7 +45,7 @@ public:
     ///queues deleting the output after a page flip has completed.
     void teardown();
     void releaseGbm();
-    bool showCursor(DrmDumbBuffer *buffer);
+    bool showCursor(DrmDumbBuffer *buffer, const QPoint &hotspot);
     bool showCursor();
     bool hideCursor();
     bool updateCursor();
-- 
2.32.0

>Aha. Looks like this was fixed on the development branch:

Indeed.

That patch looks ok.

(In reply to Adam Williamson from comment #5)
> CCing Andrey Butirsky, who wrote the fix on the development
> branch - Andrey, would you possibly have time to see if this can be fixed on
> the 5.22 code? Thanks!

If it's still needed - your patch above looks OK

Thanks for the check, both of you. I've applied that to our package downstream. Sounds like no more upstream 5.22 releases are planned, so I won't bother submitting it for that branch.

*** This bug has been marked as a duplicate of bug 427060 ***