There is a TLB U bit which when set prevents cache move in. This should be set for pages which map to the I/O space. I think it should be checked that it is correctly being set for these cards. On 2016-04-03, at 11:56 PM, Simone Mannori wrote: > Dear Ilia, > > this page summarize all the possible options of the "nouveau" kernel module: > > https://nouveau.freedesktop.org/wiki/KernelModuleParameters/ > > I will test the options that you have suggested as soon as possible. > > Try a PCI video card is a very good advice because the > AGP/FX1(chipset) is on the top of the "usual suspects" list; > unfortunately the c8000 has 3.3 PCI slots and all the (very old) PCI > video cards in my hands have 5V PCI connector > > Many thanks for your suggestions and advice. > > Simone Mannori - Italy > > > On 4 April 2016 at 04:18, Ilia Mirkin <imirkin@xxxxxxxxxxxx> wrote: >> Not sure about the radeon issue, but >> >> "DRM: GPU lockup - switching to software fbcon" >> >> basically means "the CPU isn't able to submit commands, or the GPU >> isn't executing the commands". Basically the GPU can only have so many >> commands outstanding [well, command buffers], and we hit that limit. >> From what I understand, PA-RISC has a "funny" architecture that is >> very unlike x86 in terms of memory coherency, and is thus more likely >> to hit issues that don't exist on other architectures. Perhaps you can >> summarize the key points of oddness? Perhaps there's some debug >> "disable caches" mode or something? Also note that nouveau won't work >> well on an architecture with a non-4K page size. Not sure if that's >> the situation for you. >> >> I see that it's using AGP - there's a high chance that something >> AGP-related is broken - try booting with nouveau.config=NvAGP=0 (or >> nouveau.agpmode=0 for pre-4.3 kernels). >> >> Perhaps it has issues DMA'ing the command buffer from system memory, >> we can try forcing the pushbuf to be in vram - boot with >> nouveau.vram_pushbuf=1 . >> >> If there are PCI slots and you have a PCI video card handy, I'd >> definitely try that too. >> >> Good luck, >> >> -ilia >> >> On Sun, Apr 3, 2016 at 3:51 AM, Simone Mannori <simone.mannori@xxxxxxxxx> wrote: >>> Dear "nouveau" developers, >>> >>> I know that many very competent guys have already spent a lot of time >>> and efforts on this issue without success. I have started to play with >>> "hppa" two weeks ago and, with the support of the linux-parisc mailing >>> list people, now I have a - almost fully - working workstation (hp >>> c8000). >>> >>> Everythings work perfectly BUT the video card. No matter the model, >>> type, driver, etc. the results are always the same: a very slow frame >>> buffer mode only. >>> >>> I'm forwarding the results of my investigations with the hope that >>> someone of you will help me to find the right path to fix this issue. >>> >>> Thanks in advance for your help. >>> >>> Simone Mannori - Italy >>> >>> //**------------------------------------------------------------------------------------------------------------ >>> >>> ---------- Forwarded message ---------- >>> From: Simone Mannori <simone.mannori@xxxxxxxxx> >>> Date: 3 April 2016 at 09:34 >>> Subject: Video cards init failure loading the device driver kernel module >>> To: John David Anglin <dave.anglin@xxxxxxxx> >>> Cc: Helge Deller <deller@xxxxxx>, Graham Gower >>> <graham.gower@xxxxxxxxx>, linux-parisc <linux-parisc@xxxxxxxxxxxxxxx> >>> >>> >>> Dear All, >>> >>> following you suggestions, now I have an - almost - perfectly working >>> hp c8000 workstation. >>> >>> Mission accomplished? Not exactly: despite a lot of efforts, the video >>> card section is still not fully operative. Let me resume the >>> situation. >>> >>> I'm using a c8000 with two PA-8800 and Debian 8.0, kernel: >>> >>> Linux version 4.4.0-1-parisc64-smp (debian-kernel@xxxxxxxxxxxxxxxx) >>> (gcc version 4.9.3 (GCC) ) #1 SMP Debian 4.4.6-1 (2016-03-17) >>> >>> I have two very different AGP video cards (ATI FireGL T2 and NVIDIA >>> QUADRO MXR) that have the same issue, despite the two complety >>> different drivers ("radeon" and "nouveau"). The full logs are in >>> bottom of this post. Just to make a long history short: >>> >>> ATI FIRE GL T2 ("radeon") >>> [ 47.836000] [drm:r100_ring_test [radeon]] *ERROR* radeon: ring test >>> failed (scratch(0x15E4)=0xCAFEDEAD) >>> [ 47.948000] [drm:r100_cp_init [radeon]] *ERROR* radeon: cp isn't >>> working (-22). >>> [ 48.036000] radeon 0000:80:00.0: failed initializing CP (-22). >>> [ 48.036000] radeon 0000:80:00.0: Disabling GPU acceleration >>> >>> NVIDIA QUADRO MXR ("nouveau") >>> [ 54.388000] nouveau 0000:80:00.0: DRM: GPU lockup - switching to >>> software fbcon >>> [ 54.684000] nouveau 0000:80:00.0: fb0: nouveaufb frame buffer device >>> [ 54.784000] [drm] Initialized nouveau 1.3.1 20120801 for >>> 0000:80:00.0 on minor 0 >>> >>> Looking inside the posts on many developers mailing lists, I see that >>> most of the efforts have been focused on the video device drivers init >>> code. I have tried to force the video drivers using the "modeset" >>> parameter without success. I have tried also to boot without video >>> driver, then load it manually using "modprobe". No way: the results >>> are the same. >>> >>> "lsmod" tell me that "radeon"/"nouveau" are used by the "ttm" module >>> and they cannot be removed (despite the "- r --force" modprobe >>> parameters). >>> >>> Let suppose for a moment that the issue is NOT inside the video driver >>> but in the "ttm" module: something is not fully correct in the >>> initialization of the ZX1 chipset, therefore the video driver code go >>> crazy because cannot match the video card with the chipset. >>> >>> The only way that I know to investigate this issue is to recompile the >>> kernel with debug support and execute "single step"/breakpoint the >>> code that initialize the video card and the chipset near the zone of >>> the error message. >>> >>> Question: "What are the deb packages that I need to install to >>> recompile the 4.4.0-1-parisc64-smp kernel (or another kernel version >>> suitable for this investigation) ? >>> >>> Thanks in advance for you help >>> >>> Simone Mannori - Italy >>> >>> //**----------------------------------------------------------------------------------------------------------------- >>> >>> ATI FIRE GL T2 >>> [ 45.460000] [drm] radeon kernel modesetting enabled. >>> [ 45.576000] radeon 0000:80:00.0: enabling SERR and PARITY (0107 -> 0147) >>> [ 45.660000] [drm] initializing kernel modesetting (RV350 >>> 0x1002:0x4154 0x1002:0x0002). >>> [ 45.812000] [drm] register mmio base: 0xC8020000 >>> [ 45.868000] [drm] register mmio size: 65536 >>> [ 46.084000] radeon 0000:80:00.0: putting AGP V3 device into 8x mode >>> [ 46.160000] radeon 0000:80:00.0: GTT: 512M 0x60000000 - 0x7FFFFFFF >>> [ 46.236000] [drm] Generation 2 PCI interface, using max accessible memory >>> [ 46.316000] radeon 0000:80:00.0: VRAM: 128M 0xFFFFFFFFC0000000 - >>> 0xFFFFFFFFC7FFFFFF (128M used) >>> [ 46.420000] [drm] Detected VRAM RAM=128M, BAR=128M >>> [ 46.476000] [drm] RAM width 128bits DDR >>> [ 46.524000] [TTM] Zone kernel: Available graphics memory: 3077188 kiB >>> [ 46.604000] [TTM] Zone dma32: Available graphics memory: 2097152 kiB >>> [ 46.680000] [TTM] Initializing pool allocator >>> [ 46.732000] [drm] radeon: 128M of VRAM memory ready >>> [ 46.792000] [drm] radeon: 512M of GTT memory ready. >>> [ 46.852000] [drm] radeon: 1 quad pipes, 1 Z pipes initialized. >>> [ 46.944000] radeon 0000:80:00.0: WB disabled >>> [ 46.996000] radeon 0000:80:00.0: fence driver on ring 0 use gpu >>> addr 0x0000000060000000 and cpu addr 0x0000000011279000 >>> [ 47.124000] [drm] Supports vblank timestamp caching Rev 2 (21.10.2013). >>> [ 47.204000] [drm] Driver supports precise vblank timestamp query. >>> [ 47.276000] [drm] radeon: irq initialized. >>> [ 47.328000] [drm] Loading R300 Microcode >>> [ 47.468000] radeon 0000:80:00.0: firmware: direct-loading firmware >>> radeon/R300_cp.bin >>> [ 47.584000] [drm] radeon: ring at 0x0000000060001000 >>> [ 47.836000] [drm:r100_ring_test [radeon]] *ERROR* radeon: ring test >>> failed (scratch(0x15E4)=0xCAFEDEAD) >>> [ 47.948000] [drm:r100_cp_init [radeon]] *ERROR* radeon: cp isn't >>> working (-22). >>> [ 48.036000] radeon 0000:80:00.0: failed initializing CP (-22). >>> [ 48.036000] radeon 0000:80:00.0: Disabling GPU acceleration >>> [ OK ] Found device ST373207LW 2. >>> [ 48.236000] [drm:r100_cp_fini [radeon]] *ERROR* Wait for CP idle >>> timeout, shutting down CP. >>> [ 48.384000] [drm] radeon: cp finalized >>> [ 48.448000] [drm] radeon: cp finalized >>> [ 48.496000] [TTM] Finalizing pool allocator >>> [ 48.544000] [TTM] Zone kernel: Used memory at exit: 0 kiB >>> [ 48.612000] [TTM] Zone dma32: Used memory at exit: 0 kiB >>> [ 48.676000] [drm] radeon: ttm finalized >>> [ 48.724000] [drm] Forcing AGP to PCI mode >>> Starting File System Check on /d[ 48.872000] [drm] >>> Generation 2 PCI interface, using max accessible memory >>> ev/...2-638b-4a0[ 48.968000] radeon 0000:80:00.0: VRAM: 128M >>> 0xFFFFFFFFC0000000 - 0xFFFFFFFFC7FFFFFF (128M used) >>> c-be55-7444e126d[ 49.088000] radeon 0000:80:00.0: GTT: 512M >>> 0xFFFFFFFFA0000000 - 0xFFFFFFFFBFFFFFFF >>> 283... >>> [ 49.196000] [drm] Detected VRAM RAM=128M, BAR=128M >>> [ 49.264000] [drm] RAM width 128bits DDR >>> [ 49.328000] [TTM] Zone kernel: Available graphics memory: 3077188 kiB >>> [ 49412000] [TTM] Zone dma32: Available graphics memory: 2097152 kiB >>> [ 49.492000] [TTM] Initializing pool allocator >>> [ 49.492000] [drm] radeon: 128M of VRAM memory ready >>> [ 49.492000] [drm] radeon: 512M of GTT memory ready. >>> [ 49.492000] [drm] GART: num cpu pages 131072, num gpu pages 131072 >>> [ 49.636000] [drm] radeon: 1 quad pipes, 1 Z pipes initialized. >>> [ 49.636000] [drm] PCI GART of 512M enabled (table at 0x0000000043300000). >>> [ 49.636000] radeon 0000:80:00.0: WB enabled >>> [ 49.636000] radeon 0000:80:00.0: fence driver on ring 0 use gpu >>> addr 0xffffffffa0000000 and cpu addr 0x000000007d239000 >>> [ 49.636000] [drm] Supports vblank timestamp caching Rev 2 (21.10.2013). >>> [ 49.636000] [drm] Driver supports precise vblank timestamp query. >>> [ 49.636000] [drm] radeon: irq initialized. >>> [ 49.780000] [drm] radeon: ring at 0xFFFFFFFFA0001000 >>> [ 49.972000] [drm:r100_ring_test [radeon]] *ERROR* radeon: ring test >>> failed (scratch(0x15E4)=0xCAFEDEAD) >>> [ 49.976000] [drm:r100_cp_init [radeon]] *ERROR* radeon: cp isn't >>> working (-22). >>> [ 49.976000] radeon 0000:80:00.0: failed initializing CP (-22). >>> [ 49.976000] radeon 0000:80:00.0: Disabling GPU acceleration >>> [ 50.168000] [drm:r100_cp_fini [radeon]] *ERROR* Wait for CP idle >>> timeout, shutting down CP. >>> [ 50.280000] [drm] radeon: cp finalized >>> [ 50.592000] [drm] Radeon Display Connectors >>> [ 50.592000] [drm] Connector 0: >>> [ 50.592000] [drm] VGA-1 >>> [ 50.592000] [drm] DDC: 0x60 0x60 0x60 0x60 0x60 0x60 0x60 0x60 >>> [ 50.592000] [drm] Encoders: >>> [ 50.592000] [drm] CRT1: INTERNAL_DAC1 >>> [ 50.592000] [drm] Connector 1: >>> [ 50.592000] [drm] DVI-I-1 >>> [ 50.592000] [drm] HPD1 >>> [ 50.592000] [drm] DDC: 0x64 0x64 0x64 0x64 0x64 0x64 0x64 0x64 >>> [ 50.592000] [drm] Encoders: >>> [ 50.592000] [drm] CRT2: INTERNAL_DAC2 >>> [ 50.592000] [drm] DFP1: INTERNAL_TMDS1 >>> [ 50.608000] radeon 0000:80:00.0: No connectors reported connected with modes >>> [ 50.608000] [drm] Cannot find any crtc or sizes - going 1024x768 >>> [ 50.828000] [drm] fb mappable at 0xFFFFFFFFC0040000 >>> [ 50.828000] [drm] vram apper at 0xFFFFFFFFC0000000 >>> [ 50.828000] [drm] size 3145728 >>> [ 50.828000] [drm] fb depth is 24 >>> [ 50.828000] [drm] pitch is 4096 >>> [ 51.848000] Console: switching to colour frame buffer device 128x48 >>> [ 52.080000] radeon 0000:80:00.0: fb0: radeondrmfb frame buffer device >>> [ 52172000] [drm] Initialized radeon 2.43.0 20080528 for >>> 0000:80:00.0 on minor 0 >>> >>> NVIDIA QUADRO MXR >>> [ 51.880000] nouveau 0000:80:00.0: enabling SERR and PARITY (0007 -> 0147) >>> [ 52.044000] nouveau 0000:80:00.0: NVIDIA NV11 (011300a1) >>> [ 52.364000] nouveau 0000:80:00.0: bios: version 03.11.01.24.00 >>> [ 52.488000] nouveau 0000:80:00.0: putting AGP V2 device into 4x mode >>> [ 52.592000] nouveau 0000:80:00.0: putting AGP V2 device into 4x mode >>> [ 52.668000] nouveau 0000:80:00.0: tmr: unknown input clock freq >>> [ 52.740000] nouveau 0000:80:00.0: fb: 32 MiB SDRAM >>> [ 52.844000] [TTM] Zone kernel: Available graphics memory: 3077188 kiB >>> [ 52.920000] [TTM] Zone dma32: Available graphics memory: 2097152 kiB >>> [ 52.920000] [TTM] Initializing pool allocator >>> [ 52.928000] nouveau 0000:80:00.0: DRM: VRAM: 31 MiB >>> [ 52.928000] nouveau 0000:80:00.0: DRM: GART: 512 MiB >>> [ 52.928000] nouveau 0000:80:00.0: DRM: BMP version 5.20 >>> [ 52.928000] nouveau 0000:80:00.0: DRM: No useful DCB data in VBIOS >>> [ 52.928000] nouveau 0000:80:00.0: DRM: Adaptor not initialised, >>> running VBIOS init tables. >>> [ 53.016000] nouveau 0000:80:00.0: DRM: Saving VGA fonts >>> [ 53.336000] nouveau 0000:80:00.0: DRM: No useful DCB data in VBIOS >>> [ 53.336000] [drm] Supports vblank timestamp caching Rev 2 (21.10.2013). >>> [ 53.336000] [drm] Driver supports precise vblank timestamp query. >>> [ 53.336000] nouveau 0000:80:00.0: DRM: 0xABA2: Parsing digital >>> output script table >>> [ 53.920000] nouveau 0000:80:00.0: DRM: MM: using M2MF for buffer copies >>> [ 54.144000] nouveau 0000:80:00.0: DRM: allocated 1920x1200 fb: >>> 0x8000, bo 000000007cee1000 >>> [ 54.276000] nouveau 0000:80:00.0: DRM: 0xABCE: Parsing digital >>> output script table >>> [ 54.312000] Console: switching to colour frame buffer device 240x75 >>> [ 54.388000] nouveau 0000:80:00.0: DRM: GPU lockup - switching to >>> software fbcon >>> [ 54.684000] nouveau 0000:80:00.0: fb0: nouveaufb frame buffer device >>> [ 54.784000] [drm] Initialized nouveau 1.3.1 20120801 for >>> 0000:80:00.0 on minor 0 >>> _______________________________________________ >>> Nouveau mailing list >>> Nouveau@xxxxxxxxxxxxxxxxxxxxx >>> https://lists.freedesktop.org/mailman/listinfo/nouveau > -- > To unsubscribe from this list: send the line "unsubscribe linux-parisc" in > the body of a message to majordomo@xxxxxxxxxxxxxxx > More majordomo info at http://vger.kernel.org/majordomo-info.html -- John David Anglin dave.anglin@xxxxxxxx -- To unsubscribe from this list: send the line "unsubscribe linux-parisc" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
