Date: Fri, 29 Dec 2017 15:45:49 -0800 From: Peter Grehan <grehan@freebsd.org> To: Andriy Gapon <avg@FreeBSD.org> Cc: freebsd-virtualization@freebsd.org Subject: Re: bhyve/amd: interrupt delivered when it shouldn't be? Message-ID: <450137ba-52dd-8b4c-63d2-3c3ce1909d69@freebsd.org> In-Reply-To: <42c22179-ae42-e4bb-e77d-a1d49fe634ed@FreeBSD.org>
index | next in thread | previous in thread | raw e-mail
Hi Andriy, > The hardware is AMD. Ryzen ? > But what I see suggests that at this point a Local APIC timer interrupt gets > delivered to the thread. And that causes all the mess as the thread holding the > spinlock gets preempted. > > Does this ring a bell to anyone? I have seen something similar to this after about ~20 mins when doing a current -j 16 buildworld in a guest, with the symptom being a spinlock timeout, with one vCPU spinning in smp_targeted_tlb_shootdown() at smp_targeted_tlb_shootdown+0x352/frame 0xfffffe02c80098d0 smp_masked_invlpg() at smp_masked_invlpg+0x4c/frame 0xfffffe02c8009900 pmap_invalidate_page() at pmap_invalidate_page+0x191/frame 0xfffffe02c8009950 pmap_ts_referenced() at pmap_ts_referenced+0x7b3/frame 0xfffffe02c8009a00 vm_pageout() at vm_pageout+0xe04/frame 0xfffffe02c8009a70 ... and all the others eventually spinning on that held lock. However, NMIs are still able to get through (the post-panic ddb NMI IPI) so the VM isn't completely locked up - either an interrupt is missed, or a write isn't seen by the vCPU issuing the tlb shootdown. > Is there any suspect code? Not sure yet, but the interrupt-injection path could do with a close inspection. > It seems that we set v_intr_masking bit, so the rFLAGS / eFLAGS should be > completely virtualized. So, maybe a hardware issue? Hard to say. Running with all vCPUs pinned makes the problem go away, but that could just mean the issue is isolated to when vCPUs migrate. later, Peter.home | help
Want to link to this message? Use this
URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?450137ba-52dd-8b4c-63d2-3c3ce1909d69>