[RFC] Patch X server 1.19.3 to fix VGA arbitration overflow

[biergaizi]

In QubesOS/qubes-issues#7076, I reported a known VGA arbitration bug in xorg-server 1.19.3 causes the QubesOS X server to crash randomly when both an Intel & an extra GPU are installed on the system, regardless of whether the Intel GPU is used.

Because this bug crashes my desktop 30 times a day, I'm now trying to fix the issue by myself. and after spending a day on fighting with and abusing the QubesOS build system, I came up with the following hack.

It basically,

1. Add a new list called PACKAGES_FROM_FC25_NEED_PATCH, within this variable is a list of packages which we need to apply for a custom patch. Now only xorg-x11-server-1.19.3-1.fc25.src.rpm is listed.
2. But we can't use the existing file name, otherwise it would be added to ALL_FILES and triggers a RPM auto build, without a chance of applying the patch. Thus, I did a hack by renaming all packages in PACKAGES_FROM_FC25_NEED_PATCH from *.fc25.src.rpm to .need-patch.fc25.src.rpm
3. Because the .need-patch target doesn't exist, I can now trigger make to build to target using my rules. Thus, I then subsequently wrote a build rule for xorg-x11-server-1.19.3-1.need-patch.fc25.src.rpm for applying the patch. These build rules need to be written on a case-by-case basis, manually.
4. The rule for xorg-x11-server-1.19.3-1.need-patch.fc25.src.rpm unpacks xorg-x11-server-1.19.3-1.need-patch.fc25.src.rpm to the directory xorg-x11-server-1.19.3-1.need-patch.fc25.src.rpm, then apply a patch xorg-x11-server-1.19.3-1-fix-vga-arbitration-overflow.patch, which was created manually.
5. Finally, we add the RPM spec location to RPM_SPEC_FILES, and with the appropriate RPM_SOURCE_DIR := $(DIST_SRC_ROOT)/$(COMPONENT)/xorg-x11-server-1.19.3-1.need-patch.fc25.src.rpm/, and let the QubesOS build system to generate a patched RPM package.

As you see, I don't know much about Makefile or the QubesOS build system, and this fix uses a lot of hacks, especially, the files and paths in step 5 are hardcoded. Thus I don't think it's upstreamable in its current condition, nevertheless I think it's still a good idea to publish the patch in its half-finished form.

I think what we need is a general solution for applying a custom patch to an arbitrary RPM package, once finished, the maintainer should be able to apply patches just by modifying a variable or two.

I'd like to hear your suggestion on the proper way to extend the build script.

[marmarek]

PipelineRetry