> Date: Thu, 26 Nov 2020 14:19:01 -0300 > From: Martin Pieuchot <m...@openbsd.org> > > As reported by AIsha Tammy on bugs@, there's a current recursion > possible with the pageqlock: > > ddb> trace > db_enter() at db_enter+0x10 > panic(ffffffff81dcd47c) at panic+0x12a > mtx_enter(ffffffffB219ed00) at mtx_enter+0x81 > uvm_objfree(fffffd8015f2c9a0) at uvm_objfree+0x61 > buf_dealloc_mem(fffffd8015f2c8e0) at buf_dealloc_mem+0x7c > buf_put(fffffd8O15f2c8e0) at buf _put+Oxcd > brelse(fffffd8O15f2c8e0) at brelse+0x3df > sd_buf_done(fffffd8014a27028) at sd_buf_done+0xf9 > vioblk_vq_done (ffff8000000c 7050) at vioblk_vq_done+0x6f > virtio_check_vqs (ffff80000009ce00) at virtio_check_vqs+0xfe > intr_handler (ffff80000e6dea70, ffff800000047700) at intr_handler+0x38 > Xintr_ioapic_edge19_untramp() at Xintr_ioapic_edge19_untramp+0x18f > mtx_enter (fffffd8002699680) at mtx_enter+0xb > uvm_pagedeactivate(fffffd800Z2699680) at uvm_pagedeactivate+0x105 > uyvmpd_scan() at uvmpd_scan+0x259 > uvm_pageout (ffff8000ffff9718) at uvm_pageout +0x375 > end trace frame: 0x0, count: -16 > > Diff below should prevent that, ok?
Yes, that is necessary now that you're taking the lock in uvm_objfree(). ok kettenis@ > Index: uvm/uvm_page.c > =================================================================== > RCS file: /cvs/src/sys/uvm/uvm_page.c,v > retrieving revision 1.151 > diff -u -p -r1.151 uvm_page.c > --- uvm/uvm_page.c 24 Nov 2020 13:49:09 -0000 1.151 > +++ uvm/uvm_page.c 26 Nov 2020 17:17:55 -0000 > @@ -180,7 +180,7 @@ uvm_page_init(vaddr_t *kvm_startp, vaddr > TAILQ_INIT(&uvm.page_active); > TAILQ_INIT(&uvm.page_inactive_swp); > TAILQ_INIT(&uvm.page_inactive_obj); > - mtx_init(&uvm.pageqlock, IPL_NONE); > + mtx_init(&uvm.pageqlock, IPL_VM); > mtx_init(&uvm.fpageqlock, IPL_VM); > uvm_pmr_init(); > > >
