ASRock Rack Ampere bundle - #40 by Civiloid - starting from here and down.
Yes, in my case, it was recognised as a PCI device out of the box; just the driver was OOPSing.
2 Likes
Also the open source NVIDIA driver: NVIDIA Releases Open-Source GPU Kernel Modules | NVIDIA Technical Blog
a quick look at the mmap syscall and I see the mmio size argument is 0:
mmap(NULL, 0, PROT_READ|PROT_WRITE, MAP_SHARED, 3, 0) = -1 EINVAL (Invalid argument)
and using a dirty fix:
flashrom -dirty on Linux 6.1.0-21-arm64 (aarch64)
flashrom is free software, get the source code at https://flashrom.org
Using clock_gettime for delay loops (clk_id: 1, resolution: 1ns).
Detected 1002:67ff@02:00.0 "Advanced Micro Devices, Inc. [AMD/ATI] Baffin [Radeon RX 550 640SP / RX 560/560X]"
===
This PCI device is UNTESTED. Please report the 'flashrom -p xxxx' output
to flashrom@flashrom.org if it works for you. Please add the name of your
PCI device to the subject. Thank you for your help!
===
Found GigaDevice flash chip "GD25Q20(B)" (256 kB, SPI) on ati_spi.
Reading flash... done.```
1 Like
From what I read, it appears that this NVIDIA open source driver is just a driver that gives access to the GPU for computing purposes. It is not a display driver. So you still need to choose between reverse engineered nouveau drivers, or binary-only drivers.
2 Likes
I managed to port the Erratum PCIE-65 patches:
and get it running with a WX 3100 on v6.19.9:
however, I failed to get an RX 6400 and an RX 9600 XT to work:
(I tried it on 6.1.155, 6.12.50, 6.15.4 and 6.16.9)
[ 6.171966] amdgpu 0000:03:00.0: amdgpu: [drm] Display Core failed to initialize with v3.2.334!
[ 6.171980] [drm:amdgpu_show_reset_mask [amdgpu]] *ERROR* hw_init of IP block <dm> failed -22
[ 6.172412] amdgpu 0000:03:00.0: amdgpu: amdgpu_device_ip_init failed
[ 6.172416] amdgpu 0000:03:00.0: amdgpu: Fatal error during GPU init
[ 6.172421] amdgpu 0000:03:00.0: amdgpu: amdgpu: finishing device.
[ 6.172934] WARNING: CPU: 0 PID: 495 at drivers/gpu/drm/amd/amdgpu/amdgpu_irq.c:635 amdgpu_irq_put+0xac/0xf4 [amdgpu]
[ 6.173271] Modules linked in: amdgpu(+) snd_hda_codec_hdmi drm_panel_backlight_quirks
amdxcp snd_hda_intel drm_suballoc_helper mt7921e btusb drm_ttm_helper snd_intel_dspcfg
snd_hda_codec ttm btrtl mt7921_common video snd_hda_core btmtk mt792x_lib gpu_sched
btbcm snd_hwdep snd_pcm drm_exec btintel cdc_ether mt76_connac_lib usbnet drm_buddy
mt76 snd_timer joydev onboard_usb_dev mii bluetooth ast drm_display_helper snd
drm_client_lib mac80211 soundcore polyval_ce cec drm_shmem_helper acpi_ipmi libarc4
drm_kms_helper igb cfg80211 ipmi_ssif(+) ghash_ce ipmi_devintf ixgbe(+) sbsa_gwdt
arm_spe_pmu i2c_algo_bit rfkill mdio ipmi_msghandler xgene_hwmon acpiphp_ampere_altra
arm_cmn arm_dmc620_pmu cppc_cpufreq arm_dsu_pmu acpi_tad
[ 6.173383] pc : amdgpu_irq_put+0xac/0xf4 [amdgpu]
[ 6.173710] lr : amdgpu_irq_put+0x30/0xf4 [amdgpu]
[ 6.174090] amdgpu_irq_put+0xac/0xf4 [amdgpu] (P)
[ 6.174413] amdgpu_fence_driver_hw_fini+0x5c/0x164 [amdgpu]
[ 6.174733] amdgpu_device_fini_hw+0x108/0x664 [amdgpu]
[ 6.175053] amdgpu_driver_load_kms+0xac/0xf4 [amdgpu]
[ 6.175372] amdgpu_file_to_fpriv+0x348/0x1268 [amdgpu]
Has anyone gone further down this rabbit hole and can help?
Turns out DRM_AMD_DC_FP has been disabled on clang based builds due to a clang bug, which has been resolved now. Therefore, GPUs using DCN weren’t working, while the one using DCE was working. I removed the safeguard:
It’s now running with the RX 6400 and RX 9060 XT. Will report on stability and eventually gaming performance once I managed to further setup things.
I used “those patches” daily on my machines. Between 6.13 and 6.17 they only need refreshing to apply.
RX6700XT works for me.
Too bad that those patches are still needed ;(
I’ve not gotten a kernel newer than 6.14 to boot
Reporting in for my regular every-few-months testing.
As of 6.18 this patch is now failing to correct alignment on my 9060 xt:
DOOM 2016 would open, but then freeze in the menus.
I found this alternative patch, which appears to do the same thing and is much more robust:
It needed a very small rebase to work on 6.19 (just 3 lines):
Which allowed my 9060 xt to behave – tested DOOM 2016 (vulkan native) and Path of Exile 1 in all 3 modes (dx11, dx12, vulkan native)
This one of course was still needed:
0002-ampere-arm64-Work-around-Ampere-Altra-erratum-82288-.patch
both kernel options for the patches:
CONFIG_ALTRA_ERRATUM_82288=y
CONFIG_ARM64_ALIGNMENT_FIXUPS=y
and kernel boot options:
rd.driver.blacklist=ast modprobe.blacklist=ast initcall_blacklist=sysfb_init pcie_aspm=off amdgpu.runpm=0 amdgpu.aspm=0
Much of this goes over my head, but I am sort of curious why a fork, .iso, or image hasn’t been made with the amd gpu patches baked in. I may be just ignorant to a larger issue at hand reguarding the problem but it’s something I’ve wondered.
A fork, iso or image of what?
Some people use Debian, some Fedora or RHEL. Or SUSE, Alpine or whatever else.
What should be done instead is getting those patches upstream into mainline kernel.
3 Likes
Hi all, during my regular rebase for my AVA, I updated the alignment patches for kernel 7.0. Also a fix for amdgpu for my RX 7600 to boot. Branch includes also some patches for virtio-gpu you might not need. Posting this here in case it’s useful for anyone else. GitHub - epilys/linux at 7.0-ampere-pci-fixes · GitHub
1 Like
I wasn’t really thinking of a distro in particular. Just something with the patched baked it is all. I would also love for it to be in the kernel from the start. Maybe one day.