Bug 64808

Created attachment 79772 [details]
backtrace and variables with gdb

Confirmed, trying to start QEMU (in SDL) full-screen triggers a segfault. Backtrace attached.

Problematic source:
 956 AdjustFrame(ADJUST_FRAME_ARGS_DECL)
 957 {
 958     SCRN_INFO_PTR(arg);
 959     xf86CrtcConfigPtr config = XF86_CRTC_CONFIG_PTR(pScrn);
 960     xf86OutputPtr output = config->output[config->compat_output];
 961     xf86CrtcPtr crtc = output->crtc;
 962 
 963     if (crtc && crtc->enabled) {
 964         crtc->funcs->mode_set(crtc, pScrn->currentMode,
                          ^^^^^^^^
pScrn->currentMode, x,
 965                               y);
 966         crtc->x = output->initial_x + x;
 967         crtc->y = output->initial_y + y;
 968     }


mode_set is not set in drmmode_display.c, static const xf86CrtcFuncsRec drmmode_crtc_funcs. If I disable this function within gdb (break AdjustFrame; return; continue), there is no crash. (but the picture is shown outside the screen boundaries, probably unrelated?)

By the way, that was Xorg 1.14.1 on Arch Linux 64-bit. Using gma500_gfx as well.

Created attachment 79781 [details] [review]
Prevent AdjustFrame from calling uninitialized mode_set

In my tests with qemu, output->initial_x, x and crtc->x (and p;s/x/y/g) are always 0. That, and since I saw other drivers that also do nothing for AdjustFrame, I decided to strip this whole function.

The code that is stripped by this patch originates from a different driver in 2008.

After applying this patch, X no longer segfaults (obviously, mode_set is a NULL pointer...). In order to get the picture correctly displayed on-screen, I had to apply another quirk though. There was a ghost output device which I had to disable before starting this application. It is documented on
https://wiki.archlinux.org/index.php/Poulsbo#SDL_fullscreen_viewport_is_too_large.2Fsmall

I have also tried the below patch, with the same results:
--- src/xf86-video-modesetting-0.7.0/src/driver.c	2013-05-25 08:35:26.372159926 +0000
+++ -	2013-05-25 08:35:31.281344995 +0000
@@ -961,10 +961,8 @@
     xf86CrtcPtr crtc = output->crtc;
 
     if (crtc && crtc->enabled) {
-	crtc->funcs->mode_set(crtc, pScrn->currentMode, pScrn->currentMode, x,
-			      y);
-	crtc->x = output->initial_x + x;
-	crtc->y = output->initial_y + y;
+        crtc->funcs->set_mode_major(crtc, pScrn->currentMode, crtc->rotation,
+                output->initial_x + x, output->initial_y + y);
     }
 }

This problem actual for x86 too.

should be fixed in master now.

I'll do a release in a day or two.