Panic booting qemu-system-sparc64 with bochs_drm

[Mark Cave-Ayland]

Hi all,

I've been receiving reports that newer sparc64 kernels have started to panic on boot
under qemu-system-sparc64 with bochs_drm enabled which I was able to confirm locally
building git master:


[    9.007161] [drm] Found bochs VGA, ID 0xb0c5.
[    9.007840] [drm] Framebuffer size 16384 kB @ 0x1ff22000000, mmio @ 0x1ff23000000.
[    9.012567] [TTM] Zone  kernel: Available graphics memory: 51496 KiB
[    9.013551] [TTM] Initializing pool allocator
[    9.032757] [drm] Found EDID data blob.
[    9.061904] [drm] Initialized bochs-drm 1.0.0 20130925 for 0000:01:02.0 on minor 0
[    9.336819] Unable to handle kernel paging request at virtual address 000001ff221d0000
[    9.337177] tsk->{mm,active_mm}->context = 0000000000000000
[    9.337283] tsk->{mm,active_mm}->pgd = fffff80000402000
[    9.337372]               \|/ ____ \|/
[    9.337372]               "@'/ .. \`@"
[    9.337372]               /_| \__/ |_\
[    9.337372]                  \__U_/
[    9.337468] kworker/0:0(5): Oops [#1]
[    9.339359] CPU: 0 PID: 5 Comm: kworker/0:0 Not tainted 5.8.0-rc3+ #55
[    9.341360] Workqueue: events drm_fb_helper_dirty_work
[    9.341775] TSTATE: 0000000080001605 TPC: 000000000077441c TNPC: 0000000000774420
Y: 00000000    Not tainted
[    9.341894] TPC: <memcpy+0x121c/0x13c0>
[    9.342015] g0: 0000000000000000 g1: 0000000000000000 g2: 0000000000000000 g3:
fffff800043d2c00
[    9.342094] g4: fffff8000410eac0 g5: fffff800064cc000 g6: fffff80004124000 g7:
0000000000000010
[    9.342173] o0: 000001ff221d0000 o1: 0000000100220000 o2: 0000000000000000 o3:
000001fe21fb0000
[    9.342254] o4: 000001ff221d0000 o5: 0000000000000000 sp: fffff800041273d1 ret_pc:
0000000000805b18
[    9.342325] RPC: <drm_fb_helper_dirty_work+0xf8/0x180>
[    9.342591] l0: fffff80007819cc0 l1: fffff800043df8cc l2: 0000000001356200 l3:
fffff800064cc000
[    9.342670] l4: fffff80004004200 l5: 0000000000000000 l6: 0000000000000025 l7:
fffff80004002500
[    9.342750] i0: fffff800043df8d0 i1: fffff800040106b0 i2: 0000000000000020 i3:
fffff800043e5500
[    9.342829] i4: 00000000000001d1 i5: 0000000100220000 i6: fffff80004127491 i7:
0000000000481fec
[    9.342960] I7: <process_one_work+0x18c/0x540>
[    9.343308] Call Trace:
[    9.344077] [<0000000000481fec>] process_one_work+0x18c/0x540
[    9.344267] [<00000000004824c4>] worker_thread+0x124/0x580
[    9.344310] [<0000000000489758>] kthread+0xf8/0x120
[    9.344357] [<00000000004060a4>] ret_from_fork+0x1c/0x2c
[    9.344714] [<0000000000000000>] 0x0


The error "Unable to handle kernel paging request at virtual address
000001ff221d0000" is caused by trying to access the framebuffer using a virtual
address, rather than using IO accessors which access the framebuffer correctly using
SPARC ASI_PHYS (physical) loads and stores. In some ways this is similar to the bug I
reported a couple of years back at
https://lists.freedesktop.org/archives/dri-devel/2017-June/145793.html which was
fixed with https://lists.freedesktop.org/archives/dri-devel/2017-July/145935.html.

According to git bisect the regression is introduced by the following commit:

$ git bisect bad
7a0483ac4ffca4998945c159b28afdde8353cc84 is the first bad commit
commit 7a0483ac4ffca4998945c159b28afdde8353cc84
Author: Gerd Hoffmann <kraxel at redhat.com>
Date:   Fri Jan 11 06:37:50 2019 +0100

    drm/bochs: switch to generic drm fbdev emulation

    Signed-off-by: Gerd Hoffmann <kraxel at redhat.com>
    Acked-by: Daniel Vetter <daniel.vetter at ffwll.ch>
    Link:
http://patchwork.freedesktop.org/patch/msgid/20190111053752.4004-15-kraxel@redhat.com

:040000 040000 1917943277034f620af03ac1a2fa5db48b7b224c
6d7a3c316a68efbffd398d6c2b7eebefb47bc92d M      drivers


The commit following this one at
https://patchwork.freedesktop.org/patch/276488/?series=54269&rev=4 removes
bochsfb_ops and the cfb helpers which was the original fix introduced by my second
patch above, so I'm unsure how to approach fixing this with the switch to
drm_fbdev_generic_setup().

Can anyone point me in the right direction?


Many thanks,

Mark.