Radeon GPU card failing while barfing some "*ERROR* failed to bind X pages at 0x00000000"

shimmy00 · 10-09-2022, 11:38 PM

Quote:

Originally Posted by business_kid

There's also the little matter whether it's trying to load at 0x00h in GPU memory or in normal memory.

The latter was a favourite ploy by hackers. Engineer a buffer overrun, and have an Assembler instruction like "Goto 0" above the buffer. But every step was cut off by software. Buffers have 'canaries' to prevent overrun. The kernel won't let you near low memory.

Hmm. The reason though that I wonder if it's not more fundamental is because as I mentioned there was this thread:

https://www.spinics.net/lists/amd-gfx/msg63400.html
https://www.spinics.net/lists/amd-gfx/msg75949.html

which talked of a very similar error on a different, but related, setup and makes it sound like the Rockchip SoC itself has a quirky PCIe controller implementation (not perfectly standard compliant, in particular it cannot snoop the CPU cache), which while it may be possible to work around with clever software, would be a highly nontrivial programming task and would likely require comment from the Rockchip team themselves. However, that was RK3566, so I don't know if it is the same problem on the RK3588; unfortunately, getting responses out of message boards about this board&chip often proves elusive (this has been one of the most fecund threads of all I've had about it). And RasPi is not of a lot of help here because while still ARM, it has a different SoC by a different manufacturer (Broadcom) - and yes its PCIe is also dodgy, but in a different way ("BAR is too small", or at least that's what I hear).

Hence if, if this thing is this broken, I don't want to waste money on trying to shotgun graphics cards if I don't have to (as I suspect this feature is necessary on all GPU cards - it makes sense because directly accessing CPU cache [as opposed to, say, accessing RAM as an intermediate] sounds pretty essential to the "acceleration" function of a GPU). I just wanna really make sure this is a wise move before I start hosing away money.

mrmazda · 10-10-2022, 12:25 AM

I put another old PC online a few days after my last post in this thread, with a Radeon closer to OP's, and a 5.10 kernel:

Code:

# pinxi -Gaz
Graphics:
  Device-1: AMD RV730 PRO [Radeon HD 4650] vendor: Micro-Star MSI
    driver: radeon v: kernel arch: TeraScale process: TSMC 55-65nm
    built: 2005-13 pcie: gen: 1 speed: 2.5 GT/s lanes: 16 link-max: gen: 2
    speed: 5 GT/s ports: active: DVI-I-1,HDMI-A-1 empty: VGA-1
    bus-ID: 01:00.0 chip-ID: 1002:9498 class-ID: 0300 temp: 41.0 C
  Display: x11 server: X.Org v: 1.20.11 driver: X: loaded: modesetting
    alternate: fbdev,vesa dri: r600 gpu: radeon display-ID: :0 screens: 1
  Screen-1: 0 s-res: 3600x1200 s-dpi: 120 s-size: 762x254mm (30.00x10.00")
    s-diag: 803mm (31.62")
  Monitor-1: DVI-I-1 pos: primary,left model: NEC EA243WM serial: <filter>
    built: 2011 res: 1920x1200 hz: 60 dpi: 94 gamma: 1.2
    size: 519x324mm (20.43x12.76") diag: 612mm (24.1") ratio: 16:10 modes:
    max: 1920x1200 min: 640x480
  Monitor-2: HDMI-A-1 mapped: HDMI-1 pos: right model: Dell P2213
    serial: <filter> built: 2012 res: 1680x1050 hz: 60 dpi: 90 gamma: 1.2
    size: 473x296mm (18.62x11.65") diag: 558mm (22") ratio: 16:10 modes:
    max: 1680x1050 min: 720x400
  OpenGL: renderer: AMD RV730 (DRM 2.50.0 / 5.10.0-18-amd64 LLVM 11.0.1)
    v: 3.3 Mesa 20.3.5 compat-v: 3.0 direct render: Yes
# uname -a
Linux hp750 5.10.0-18-amd64 #1 SMP Debian 5.10.140-1 (2022-09-02) x86_64 GNU/Linux

Not that I expect it to solve any Rockchip SoC PCI problems.

business_kid · 10-10-2022, 04:11 AM

What surprises me is use of the r600 driver for a 2005 GPU, when I was firmly instructed to use the r300 driver for a later one. I tried the r600 driver anyhow, but it tanked.

Perhaps the fact that yours is external might affect things. Mine was in a Northbridge (Remember them?) so I suppose it counted as integrated graphics. It certainly was a tad faster than a 33/66Mhz PCI slot.

mrmazda · 10-10-2022, 07:56 AM

Quote:

Originally Posted by business_kid

What surprises me is use of the r600 driver for a 2005 GPU, when I was firmly instructed to use the r300 driver for a later one. I tried the r600 driver anyhow, but it tanked.

Both AMD and NVidia introduce new fabs as additions, while keeping the old, and in many cases adding new names for older fabs. https://en.wikipedia.org/wiki/List_o...ocessing_units charts include model introduction dates that show this.

business_kid · 10-10-2022, 12:07 PM

AMD/ATI used a bunch of names for the same device: device name; gpu design series; part number, etc. Perhaps thaty explains it. Mine was made from 2004-2008, and I got one of the last ones, so that explains it.

Here's the specs on my RS690M
https://www.techpowerup.com/gpu-specs/ati-rs690.g183

73mm² for the GPU

. It's not worth giving them away free in breakfast cereal packets

.

Anyhow, my rant doesn't solve anyone's problem.