There is a pull request with a fix for the PCIe driver in the 6.1.15 bianbu kernel:
PCIe PR
I built that kernel, copied the firmware files over to /lib/firmware/amdgpu and booted the system.
The driver starts to load but then throws this error:
[ 2.584592] [drm] Detected VRAM RAM=4096M, BAR=0M
[ 2.589346] [drm] RAM width 128bits GDDR5
[ 2.593507] [drm] amdgpu: 4096M of VRAM memory ready
[ 2.598535] [drm] amdgpu: 7938M of GTT memory ready.
[ 2.603637] [drm] GART: num cpu pages 65536, num gpu pages 65536
[ 2.609732] amdgpu 0002:01:00.0: amdgpu: (-22) kernel bo map failed
[ 2.616086] [drm:amdgpu_device_init] ERROR sw_init of IP block <gmc_v8_0> f
ailed -22
[ 2.624117] amdgpu 0002:01:00.0: amdgpu: amdgpu_device_ip_init failed
[ 2.630630] amdgpu 0002:01:00.0: amdgpu: Fatal error during GPU init
[ 2.637057] amdgpu 0002:01:00.0: amdgpu: amdgpu: finishing device.
[ 2.643385] amdgpu: probe of 0002:01:00.0 failed with error -22
[ 2.678480] [drm] init_lt8911exb()
I found some statements about that error meaning the firmware is not loaded but as I said, it exists.
Any hints or anybody found another way to get it to run?
Ok, from hints i got, itâs really that the PCIe interface is having 32bits only so every GPU beyond 4GB is not working.
I then used an old GeForce 7300 GT and it worked until the login screen.
But USB didnât work whenever the GPU was initialized by its driver.
Not with the 6.1 kernel and not with 6.6.
Now Iâm stuck againâŚany hints?
Thank you for that link, very insightful.
What I donât understand is, why you need a converter in the first place. I mean, as far as I know, a card with fewer lanes should work without it just slower!?
I mean PCIe 1x is the short connector in front and everything else is just optional for more speed, right?
Main problem for me is still, that PCIe seems to interfere with USB 2 and 3 ports.
And, as I said: the Geforce did work without an adapter it was just interfering with USB so I couldnât type anything for login.
dmesg and journalctl didnât show anything either (was logged in via serial console).
Any more insights on how to debug such stuff?
Thanks!
1 Like
There are IO problems with this SOC I donât really understand. I have an RX 590 working without using a converter using this kernel: Commits ¡ StackDoubleFlow/spacemit-k1-linux-6.1 ¡ GitHub
I never had that exact error, but I had a similar one which led to amdgpu: (-14) failed to allocate kernel bo
This was caused by swiotlb buffer is full earlier in the log, though it wasnât really full, it was just trying to allocate a block size greater than the max, and thatâs what led to me changing IO_TLB_SEGSIZE.
If you try again, let me know how that goes, Iâm curious to see if the RX 550 will run more smoothly, since I still have crashes with the RX 590 after ~20-30 minutes of use.
2 Likes
Sorry for the late answerâŚIâve been on vacation 
Thatâs a nice catchâŚIâll build a new kernel later and report back!
Cheers!
1 Like
IO_TLB_SEGSIZE
I canât find such a config option in the kernel you mentioned:
~/repositories/spacemit-k1-linux-6.1$ grep TLB .config
CONFIG_CGROUP_HUGETLB=y
CONFIG_ARCH_WANT_GENERAL_HUGETLB=y
# CONFIG_BT_HCIBTUSB_RTLBTUSB is not set
CONFIG_ARCH_SUPPORTS_HUGETLBFS=y
CONFIG_HUGETLBFS=y
CONFIG_HUGETLB_PAGE=y
CONFIG_SWIOTLB=y
.config is a copy of arch/riscv/configs/k1_extern_gpu_defconfig
1 Like
Itâs not a config option. Itâs this code patch in âinclude/linux/swiotlb.h. If youâre using my kernel, then itâs already there.
2 Likes
UnderstoodâŚI was wondering already why there is no CONFIG_ in front of that setting
Built the kernel yesterday and Iâll try it out today with my RX550
1 Like
HmmâŚmy kernel build from yesterday didnât produce an imageâŚmust have overseen an error. Today I rebuilt it after âmake distcleanâ and it stopped building when compiling the radeon module and after configuring radeon out, it did the same thing with the nouveau driver.
Did you build it on Bianbuz 1.0.15 stock or on a different system?
1 Like
Did a clean build again:
make distclean
echo ARCH=riscv
make k1_extern_gpu_defconfig
make menuconfig (disabled radeon and nouveau)
make -j8
build error:
drivers/gpu/drm/amd/amdgpu/../display/dc/core/dc_stream.c: In function âdc_stream_remove_writebackâ:
drivers/gpu/drm/amd/amdgpu/../display/dc/core/dc_stream.c:531:55: warning: array subscript -1 is below array bounds of âstruct dc_writeback_info[1]â [-Warray-bounds=]
531 | stream->writeback_info[j] = stream->writeback_info[i];
| ~~~~~~~~~~~~~~~~~~~~~~^~~
In file included from ./drivers/gpu/drm/amd/amdgpu/../display/dc/dc.h:1272,
from ./drivers/gpu/drm/amd/amdgpu/../display/dc/inc/core_types.h:29,
from ./drivers/gpu/drm/amd/amdgpu/../display/dc/basics/dc_common.h:29,
from drivers/gpu/drm/amd/amdgpu/../display/dc/core/dc_stream.c:27:
./drivers/gpu/drm/amd/amdgpu/../display/dc/dc_stream.h:241:34: note: while referencing âwriteback_infoâ
241 | struct dc_writeback_info writeback_info[MAX_DWB_PIPES];
| ^~~~~~~~~~~~~~
CC drivers/gpu/drm/amd/amdgpu/../display/dc/core/dc_vm_helper.o
CC drivers/gpu/drm/amd/amdgpu/../display/dc/dc_helper.o
CC drivers/gpu/drm/amd/amdgpu/../display/dc/dc_dmub_srv.o
CC drivers/gpu/drm/amd/amdgpu/../display/dc/dc_edid_parser.o
CC drivers/gpu/drm/amd/amdgpu/../display/modules/freesync/freesync.o
CC drivers/gpu/drm/amd/amdgpu/../display/modules/color/color_gamma.o
CC drivers/gpu/drm/amd/amdgpu/../display/modules/color/color_table.o
CC drivers/gpu/drm/amd/amdgpu/../display/modules/info_packet/info_packet.o
CC drivers/gpu/drm/amd/amdgpu/../display/modules/power/power_helpers.o
CC drivers/gpu/drm/amd/amdgpu/../display/dmub/src/dmub_srv.o
CC drivers/gpu/drm/amd/amdgpu/../display/dmub/src/dmub_srv_stat.o
CC drivers/gpu/drm/amd/amdgpu/../display/dmub/src/dmub_reg.o
CC drivers/gpu/drm/amd/amdgpu/../display/dmub/src/dmub_dcn20.o
CC drivers/gpu/drm/amd/amdgpu/../display/dmub/src/dmub_dcn21.o
CC drivers/gpu/drm/amd/amdgpu/../display/dmub/src/dmub_dcn30.o
CC drivers/gpu/drm/amd/amdgpu/../display/dmub/src/dmub_dcn301.o
CC drivers/gpu/drm/amd/amdgpu/../display/dmub/src/dmub_dcn302.o
CC drivers/gpu/drm/amd/amdgpu/../display/dmub/src/dmub_dcn303.o
CC drivers/gpu/drm/amd/amdgpu/../display/dmub/src/dmub_dcn31.o
CC drivers/gpu/drm/amd/amdgpu/../display/dmub/src/dmub_dcn315.o
CC drivers/gpu/drm/amd/amdgpu/../display/dmub/src/dmub_dcn316.o
CC drivers/gpu/drm/amd/amdgpu/../display/dmub/src/dmub_dcn32.o
AR drivers/gpu/drm/amd/amdgpu/built-in.a
AR drivers/gpu/drm/built-in.a
AR drivers/gpu/built-in.a
make[1]: *** [scripts/Makefile.build:500: drivers] Fehler 2
make: *** [Makefile:2007: .] Fehler 2
1 Like
Is echo ARCH=riscv a command you ran? That does not set or display any actual environment variable.
The way I compile it (cross-compile from x86_64 arch linux) is like this
make ARCH=riscv CROSS_COMPILE=riscv64-linux-gnu- k1_extern_gpu_defconfig
make ARCH=riscv CROSS_COMPILE=riscv64-linux-gnu- -j 32
The cross compile toolchain is from the riscv64-linux-gnu-gcc package on arch.
2 Likes
No, I just copied the wrong line. I just checked whether the variable is set before building and copied that line instead of the correspondent âexportâ line.
So, the difference seems to be, that I build that kernel on the Jupiter itself and Bianbu created their own build chain.
Iâll try building it under Archlinux on one of the next days.
1 Like
To check whether the variable âARCHâ is set, you would need to type âecho $ARCHâ. In your reply you forgot the dollar sign. And to set the variable ARCH to the value âriscvâ, you would type âexport ARCH=riscvâ
2 Likes
Yes, I knowâŚas I said, copied the wrong line but this is not the issue in this thread. Itâs about building on bare metal which isnât working.
1 Like