Created attachment 50104 [details] kernel log with drm.debug=15 I must currently use 'noaccel=1' to use nouveau on my NVA5 (GT 220, 1GB vram). While debugging, it appeared the extra created FIFOs simply do not run at all, and trigger PFIFO_DMA_PUSHER errors, with IB_EMPTY status. The first of these errors happen in nouveau_dma_init() function, which is the first one to be run after creation of a new FIFO. I must add 'vram_pushbuf=1' to kernel parameters to not have these errors at start, however, adding 'vram_notify=1' appears to have not effect. X crashes at start if I do not use 'noaccel=1', with or without 'vram_pushbuf=1', triggering PFIFO_DMA_PUSHER errors, with INVALID_CMD status.
Created attachment 50105 [details] kernel log, when trying to start X
I've noticed you mention this issue on IRC previously. Am I misremembering that you also mentioned that some other board doesn't work in that machine either? And that the NVIDIA binary driver does not work? If the binary driver works, please post a kmmio trace of it.
The blob works, and also no problem in Windows either. I'll do an mmio trace, and also update/retest everything with the nv46.
mmio trace uploaded to mmio.dumps@gmail.com
Created attachment 50120 [details] [review] try harder at forcing vram pushbufs Thanks for the trace, I shall attempt to look at it tomorrow (bed time now!). From what you said, I presume console accel works okay with vram_pushbuf=1? If that's the case, does X work too with this patch?
Yes, console acceleration works with vram_pushbuf=1. With your patch, X starts now, albeit showing full garbage.
Created attachment 50132 [details] X screenshot with patch
Created attachment 50147 [details] [review] reset pci bus mastering at init time Hmmm.. Yes, it looks like your card is completely failing to access system memory for some reason. I can't believe it's a general issue with us setting up page tables etc, we'd surely have seen this report a *lot* more otherwise.. The binary driver resets PCI bus mastering really early on during init, lets give that a go for starters..
Created attachment 50161 [details] X screenshot with pci bus mastering patch It's a bit better with this second patch. KDE Menu, and many other stuff, are readable now, when they were partly full of garbage previously. I also think it's something specific to my system, as you say. My motherboard has Via PT880 chipset, and can support an AGP or a PCI-E video card.
My motherboard BIOS had option "PCIE VC1 Request Queue" set to [Auto]. When putting this setting as [Disabled], there is no problem anymore, and everything works correctly. Will check that is also fixes the issue with my NV46 (bug #32494 ).
*** Bug 32494 has been marked as a duplicate of this bug. ***
Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.