<html> <head> <base href="https://bugs.freedesktop.org/"> </head> <body> <div> <a class="bz_bug_link bz_status_NEW " title="NEW - [SNB] igt@drv_module_reload@basic-reload-final fails and hard hangs the system" href="https://bugs.freedesktop.org/show_bug.cgi?id=100858#c2">Comment # 2</a> on <a class="bz_bug_link bz_status_NEW " title="NEW - [SNB] igt@drv_module_reload@basic-reload-final fails and hard hangs the system" href="https://bugs.freedesktop.org/show_bug.cgi?id=100858">bug 100858</a> from <a class="email" href="mailto:krisman@collabora.co.uk" title="krisman@collabora.co.uk">krisman@collabora.co.uk</a> <pre>(In reply to Chris Wilson from <a href="show_bug.cgi?id=100858#c1">comment #1</a>) > The answer lies in if the GPU didn't write the dword to where we wanted, > where did it write it? And the same worrying question will be for lots of > different ops. The HWS is intact otherwise it would have complained about a > gpu hang... Actually no, that test isn't checking so that's a more > reasonable explanation that the HWS write also went astray, i.e. it is > likely that nothing is right in the way the GPU is addressing memory. > > In load/unload we do a GPU reset, so everything should be sane... > > Remove the store dword check and see it it still hard hangs. If so, we can > start removing chunks from the load/unload sequence and see at what point it > works. It doesn't hard hang without the store_dword sequence. The hang happens on gem_write when executing the blt engine. One interesting aspect is that the issue is only reproducible with intel_iommu=on. If that parameter is disabled, the test succeeds and the system never hangs.</pre> </div> <hr> You are receiving this mail because: <ul> <li>You are on the CC list for the bug.</li> <li>You are the QA Contact for the bug.</li> </ul> </body> </html>