On 12/11/2018 10:40 PM, Lukas F. Hartmann wrote: > Hi, Hi,
> I tested the patch on i.MX6QP with: > Linux reform 4.20.0-rc2-00133-g1ce80e0fe98e-dirty #10 SMP Mon Nov 26 > 02:02:42 CET 2018 armv7l GNU/Linux > > Running a recent Xorg built from source with modesetting driver and > etnaviv. > > I was getting segfaults after a few seconds of usage and tracked them > down to a corrupt pointer (probably use-after-free): > > http://dump.mntmn.com/marexpatch-trace1.txt.html > > After looking at your patch for a while I came up with this addition > (last line) in etna_resource.c, to remove the resource from the > used_resources set when it is destroyed. > > @@ -464,6 +469,9 @@ etna_resource_destroy(struct pipe_screen *pscreen, > struct pipe_resource *prsc) { > struct etna_resource *rsc = etna_resource(prsc); > > + _mesa_set_destroy(rsc->pending_ctx, NULL); > + _mesa_set_remove_key(etna_screen(pscreen)->used_resources, rsc); > > I've been running with this version for a few days and it seems stable. > The patch seems to do its job (tested qupzilla and qutebrowser, no > more corrupted tiles). Just sometimes an occassional black tile shows > up. So something is still up, but the patch is a perceived improvement. Nice, thanks. Lemme try that here too. -- Best regards, Marek Vasut _______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/mesa-dev
