Imre Deak <imre.deak@xxxxxxxxx> writes: > The poweroff handlers undo the actions of the thaw handlers. As the > original commit stated saving the registers is not needed there, but > it's also not a big overhead and there should be no problem doing it. We > are planning to optimize the hibernation sequence by for example not > shutting down the display between freeze and thaw, and also getting rid > of unnecessary steps at the power off phase. But before that we want to > actually unify things rather than having special cases, as maintaining > the special code paths caused already quite a lot of problems for us so > far. That sounds like a worthy goal. I don't understand what you hope to achieve by having a poweroff_late hook, since there aren't really anything useful left to do at the point it is called, but if you want a dummy callback there for code structure reasons then fine. But you cannot just run around breaking stuff while slowly moving towards this goal over multiple releases... v3.19 is currently broken and that seems totally unnecessary. In any case: You should have noticed this problem while testing your patches. The breakage is 100% reproducible. Unfortunately I had to do a bisect to realize what you had done to the i915 driver, something I unfortunately didn't find time to do before v3.19 was released. But I do find it unnecessary to release with such bugs. Any attempt to exercise the code path you modified would have revealed the bug. > Reverting the commit may hide some other issue, so before doing that > could you try the following patch: > http://lists.freedesktop.org/archives/intel-gfx/2015-February/060529.html Makes no difference. I assume that patch fixes an unrelated bug? The age and reported symptoms indicates so. Note that I am reporting a regression introduced after v3.18, while that seems to fix a bug introduced in v3.17. Both v3.17 and v3.18 (including v3.18.6), as well as earlier releases, work fine for me. > If with that you still get the hang could you try on top of that the > patch below, first having only pci_set_power_state uncommented, then > both pci_set_power_state and pci_disable_device uncommented? That patch fixes the problem, with only pci_set_power_state commented out. Do you still want me to try with pci_disable_device() commented out as well? Bjørn _______________________________________________ Intel-gfx mailing list Intel-gfx@xxxxxxxxxxxxxxxxxxxxx http://lists.freedesktop.org/mailman/listinfo/intel-gfx
