On Fri, Feb 20, 2015 at 09:55:09AM -0200, Paulo Zanoni wrote: > 2015-02-20 3:33 GMT-02:00 Ben Widawsky <[email protected]>: > > Register writes are supposed to be posted, and therefore "fast." In order > > to try > > to preserve this behavior, the patch uses percpu variables to make the > > overhead > > minimal. The slow data collection is done at error time, and that only > > occurs if > > the user specified mmio_debug=1 module parameter. > > > > I wrote this patch to try to track down annoying and sporadic errors on my > > Broadwell laptop. Sometimes it's obvious how you got to this point from the > > backtrace, but other times it isn't. > > > > Register reads can also trigger the unclaimed register error. Don't > you want to also record the last one we read on each CPU too? > > Also, just to remind you since you seem to be debugging something: > user space programs (such as intel_reg_dumper or anything else) and > other Kernel modules (like vgacon and others) can also trigger these > errors, and we can't easily catch these cases. > > Pinging Chris since he always has some ideas about this code.
In this case it is probably better to look at the automatic oneshot mmio_debug patch. -Chris -- Chris Wilson, Intel Open Source Technology Centre _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
