Re: [PATCH RFC] more progress with radeon on C8000

[Helge Deller <deller@gmx.de>]

On 5/17/23 21:59, John David Anglin wrote:
> On 2023-05-16 4:48 p.m., Helge Deller wrote:
>> * Helge Deller<deller@gmx.de>:
>>> There have been various approaches to fix the radeon graphic cards
>>> on C8000 workstations, e.g. this is the last thread:
>>> https://marc.info/?l=linux-parisc&m=156971832128700&w=2
>>>
>>> With the patch below the radeon ring and ib tests don't fail any longer.
>>> It uses (relatively) lightweight pdc/fdc instructions.
>>>
>>> The patch below
>>> - uses the pdc instruction (purge data cache) before reading back the
>>>    ring test result from memory.
>>> - modifies the parisc-agp code, based on Thomas patch
>>>
>>> The patch is not intended to by applied as-is.
>> attached is a cleaned up version of the patch.
> I tried this change on c8000 with kernel 6.3.3:
>
> [   89.531274] drm_kms_helper alternatives: applied 0 out of 27 patches
> [   89.630885] ipmi_si alternatives: applied 0 out of 8 patches
> [   89.731041] ipmi_si: IPMI System Interface driver
> [   89.821872] drm_display_helper alternatives: applied 0 out of 6 patches
> [   89.833885] ipmi_si: Adding device-tree-specified kcs state machine
> [   89.971367] ipmi_si: Trying device-tree-specified kcs state machine at mem address 0xfffffff0f05b0000, slave address 0x0, irq 0
> [   90.271358] ttm alternatives: applied 0 out of 13 patches
> [   90.421083] drm_ttm_helper alternatives: applied 0 out of 2 patches
> [   90.881506] radeon alternatives: applied 0 out of 4803 patches
> [   90.980838] [drm] radeon kernel modesetting enabled.
> [   91.041955] [drm] initializing kernel modesetting (RV350 0x1002:0x4154 0x1002:0x0002 0x80).
> [   91.140773] [drm] Forcing AGP to PCI mode
> [   91.290915] [drm] GPU not posted. posting now...
> [   91.551801] [drm] Generation 2 PCI interface, using max accessible memory
> [   91.620801] radeon 0000:80:00.0: VRAM: 128M 0xFFFFFFFFC0000000 - 0xFFFFFFFFC7FFFFFF (128M used)
> [   91.671141] ipmi 16: IPMI message handler: Found new BMC (man_id: 0x00000b, prod_id: 0x8201, dev_id: 0x32)
> [   91.710747] radeon 0000:80:00.0: GTT: 512M 0xFFFFFFFFA0000000 - 0xFFFFFFFFBFFFFFFF
> [   91.814064] ipmi 16: IPMI kcs interface initialized
> [   91.880842] [drm] Detected VRAM RAM=128M, BAR=128M
> [   91.970765] [drm] RAM width 128bits DDR
> [   91.971192] [drm] radeon: 128M of VRAM memory ready
> [   92.050830] [drm] radeon: 512M of GTT memory ready.
> [   92.110946] [drm] GART: num cpu pages 131072, num gpu pages 131072
> [   92.222903] [drm] radeon: 1 quad pipes, 1 Z pipes initialized
> [   92.290747] [drm] PCI GART of 512M enabled (table at 0x0000000047040000).
> [   92.370907] radeon 0000:80:00.0: WB enabled
> [   92.431273] radeon 0000:80:00.0: fence driver on ring 0 use gpu addr 0xffffffffa0000000
> [   92.531273] [drm] radeon: irq initialized.
> [   92.560906] [drm] Loading R300 Microcode
> [   92.631141] radeon 0000:80:00.0: Direct firmware load for radeon/R300_cp.bin failed with error -2
> [   92.743052] radeon_cp: Failed to load firmware "radeon/R300_cp.bin"
> [   92.819159] [drm:r100_cp_init [radeon]] *ERROR* Failed to load firmware!
> [   92.890738] radeon 0000:80:00.0: failed initializing CP (-2).
> [   92.960765] radeon 0000:80:00.0: Disabling GPU acceleration
> [   93.030744] [drm] radeon: cp finalized
> [   93.142149] [drm] Radeon Display Connectors
> [   93.170762] [drm] Connector 0:
> [   93.170771] [drm]   VGA-1
> [   93.170775] [drm]   DDC: 0x60 0x60 0x60 0x60 0x60 0x60 0x60 0x60
> [   93.170786] [drm]   Encoders:
> [   93.290760] [drm]     CRT1: INTERNAL_DAC1
> [   93.340770] [drm] Connector 1:
> [   93.380749] [drm]   DVI-I-1
> [   93.410745] [drm]   HPD1
> [   93.440846] [drm]   DDC: 0x64 0x64 0x64 0x64 0x64 0x64 0x64 0x64
> [   93.520737] [drm]   Encoders:
> [   93.559678] [drm]     CRT2: INTERNAL_DAC2
> [   93.590733] [drm]     DFP1: INTERNAL_TMDS1
> [   93.853779] [drm] fb mappable at 0xFFFFFFFFC0040000
> [   93.900758] [drm] vram apper at 0xFFFFFFFFC0000000
> [   93.940732] [drm] size 5242880
> [   93.960731] [drm] fb depth is 24
> [   93.981757] [drm]    pitch is 5120
> [   94.080657] Console: switching to colour frame buffer device 160x64
> [   94.290810] radeon 0000:80:00.0: [drm] fb0: radeondrmfb frame buffer device
> [   94.420892] [drm] Initialized radeon 2.50.0 20080528 for 0000:80:00.0 on minor 0
>
> 80:00.0 VGA compatible controller: Advanced Micro Devices, Inc. [AMD/ATI] RV350 GL [FireGL T2] (rev 80) (prog-if 00 [VGA controller])
>          Subsystem: Advanced Micro Devices, Inc. [AMD/ATI] RV350 GL [FireGL T2]
>          Control: I/O+ Mem+ BusMaster+ SpecCycle- MemWINV- VGASnoop- ParErr- Stepping- SERR+ FastB2B- DisINTx-
>          Status: Cap+ 66MHz+ UDF- FastB2B+ ParErr- DEVSEL=medium >TAbort- <TAbort- <MAbort- >SERR- <PERR- INTx-
>          Latency: 192 (2000ns min), Cache Line Size: 128 bytes
>          Interrupt: pin A routed to IRQ 75
>          Region 0: Memory at ffffffffc0000000 (32-bit, prefetchable) [size=128M]
>          Region 1: I/O ports at 3c000 [size=256]
>          Region 2: Memory at ffffffffc8020000 (32-bit, non-prefetchable) [size=64K]
>          Expansion ROM at ffffffffc8000000 [disabled] [size=128K]
>          Capabilities: <access denied>
>          Kernel driver in use: radeon
>          Kernel modules: radeon
>
> Firmware still fails to load

You need to install the debian firware package:
apt install firmware-linux-nonfree firmware-linux-free

> and acceleration is disabled.

Yes, if firware isn't loaded the acceleration will be disabled too.
But even with firmware loaded the ring tests will fail on AGP cards.


> Maybe this is also a cache flush issue?

Probably.
Interestingly the radeon driver doesn't use the parisc-agp driver,
but it's internal GART. Since it works with radeon-PCI cards,
maybe it's not useable on parisc and we would need the parisc-agp
driver.

> FB console looks okay.
>
> This is a significant advance 😁

Not really :-(
The unaccelerated FB console comes up even without my patch.
The latest patch I sent also missed the hunk below.
With that ugly patch, my radeon *PCI* (not AGP) card seems to work.
Additionally it's unclear how to cope with user-mapped memory.

Helge

@@ -3689,6 +3690,16 @@ void r100_ring_ib_execute(struct radeon_device *rdev, struct radeon_ib *ib)
  {
  	struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];

+	u32 *ptr;
+	int len;
+	ptr = ib->ptr;
+	len = ib->length_dw;
+	while (len >= 0) {
+		asm_io_fdc(ptr);
+		ptr++;
+		len--;
+	}
+
  	if (ring->rptr_save_reg) {
  		u32 next_rptr = ring->wptr + 2 + 3;
  		radeon_ring_write(ring, PACKET0(ring->rptr_save_reg, 0));