Re: [PATCH v3 0/8] alpha: enable generic entry infrastructure
Magnus Lindholm <[email protected]> Mon, 15 Jun 2026 21:55:26 +0200
| Newsgroups | org.kernel.vger.linux-alpha,org.kernel.vger.linux-kernel |
|---|---|
| Message-ID | <CA+=Fv5QgH19VsbbQcJFMWJ-8rnzEwqUv_HrYVZLGVMUnj51=Fg@mail.gmail.com> |
On Sun, Jun 14, 2026 at 3:39 AM Michael Cree <[email protected]> wrote: > > On Fri, Jun 12, 2026 at 10:26:48PM +0200, Magnus Lindholm wrote: > > This series moves Alpha closer to the common architecture infrastructure > > and enables GENERIC_ENTRY and GENERIC_IRQ_ENTRY. This is v3 of this patch > > series, please see below for changes since v1 and v2. > > I am seeing sporadic Xorg crashes with this patchset applied to a 7.0 > stable series kernel, usually within a minute of logging into the > desktop. The rest of system keeps running fine and there are no > messsages in the kernel log. I am testing on an XP1000 with Radeon > HD4350 graphics card. The same kernel without this patchset works > fine -- no Xorg crashes. Indeed, it has been playing the dulcet > tones of Pink Floyd through my sound system for the last hour without > a problem. > > Example Xorg backtrace follows. In this case it is a fp exception > but sometimes it is a segfault: > Michael, Thanks a lot for taking the time to test this series. I was close to sending a pull request with it, but your testing clearly shows that it still needs more work, so I will hold off. Given the symptoms, my first suspicion is that the final generic-entry patch has changed the Alpha return-to-user/signal path in a way that loses the old explicit FPU save before signal delivery. The old do_work_pending() path did a save_fpu() before calling do_signal(), while the new path goes through arch_do_signal_or_restart() and the generic exit-to-user code. If I missed preserving that Alpha-specific FPU handling, that could plausibly explain sporadic SIGFPE/SEGV failures in FP-heavy code such as Xorg/pixman. I have an XP1000 with a similar Radeon card, so this sounds like something I should be able to reproduce locally. Unfortunately I do not have access to that machine right now, so any proper fix will have to wait until the next merge window. Thanks again for catching this before I sent the pull request. Magnus