On Wed, Sep 21, 2016 at 03:16:57PM +0200, Krzysztof Hałasa wrote: > Hans Verkuil <hverkuil@xxxxxxxxx> writes: > > > That was probably the reason for the pci_read_config_word in the reg_write > > code. Try putting that back (and just that). > > Yes. I guess a single pci_read_config_word() would suffice. > > Though it would obviously be much better to identify the place in the > driver which needs to have the write buffers flushed, and add a read() > just there. > > The interrupt handler maybe (e.g. just before the return IRQ_HANDLED)? > > OTOH this may be some sort of timing problem, I mean the faster code may > put too much stress on the SOLO chip. > > Doesn't happen here so I can't test the cure. It happens in solo_disp_init at uploading default motion thresholds array. I've got a prints trace with solo6010-fix-lockup branch https://github.com/bluecherrydvr/linux/tree/solo6010-fix-lockup/drivers/media/pci/solo6x10 the trace itself in jpg: https://decent.im:5281/upload/3793f393-e285-4514-83dd-bf08d1c8b4a2/e7ad898b-515b-4522-86a9-553daaeb0860.jpg Indeed, targeted fixing would be more reasonable than making register r/w routines follow blocking fashion. But the driver is already complete and was known to be working, and I seems all places in code assume the blocking fashion of reg r/w, and changing that assumption may lead to covert bugs anywhere else, not just at probing, which may be hard to nail down. For now, I'll try setting pci_read_config_word() back instead of full revert. Does it need to be just in reg_write? No need for it in reg_read, right? -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
