Re: [PATCH v6 3/5] fbdev: Disable sysfb device registration when removing conflicting FBs

[Javier Martinez Canillas]

On 6/16/22 23:03, Zack Rusin wrote:
> On Thu, 2022-06-16 at 21:55 +0200, Javier Martinez Canillas wrote:
>> Hello Zack,
>>
>> On 6/16/22 21:29, Zack Rusin wrote:
>>> On Tue, 2022-06-07 at 20:23 +0200, Javier Martinez Canillas wrote:
>>>> The platform devices registered by sysfb match with firmware-based DRM or
>>>> fbdev drivers, that are used to have early graphics using a framebuffer
>>>> provided by the system firmware.
>>>>
>>
>> [snip]
>>
>>>
>>> Hi, Javier.
>>>
>>> This change broke arm64 with vmwgfx. We get a kernel oops at boot (let me know if
>>> you'd like .config or just have us test something directly for you):
>>>
>>
>> Yes please share your .config and I'll try to reproduce on an arm64 machine.
> 
> Attached. It might be a little hard to reproduce unless you have an arm64 machine
> with a dedicated gpu. You'll need a system that actually transitions from a generic
> fb driver (e.g. efifb) to the dedicated one.
>

Yes, all my testing for this was done with a rpi4 so I should be able to reproduce
that case. I'm confused though because I tested efifb -> vc4, simplefb -> vc4 and
simpledrm -> vc4.
 
>>>
>>>  Unable to handle kernel NULL pointer dereference at virtual address
>>> 0000000000000008
>>>  Mem abort info:
>>>    ESR = 0x96000004
>>>    EC = 0x25: DABT (current EL), IL = 32 bits
>>>    SET = 0, FnV = 0
>>>    EA = 0, S1PTW = 0
>>>    FSC = 0x04: level 0 translation fault
>>>  Data abort info:
>>>    ISV = 0, ISS = 0x00000004
>>>    CM = 0, WnR = 0
>>>  user pgtable: 4k pages, 48-bit VAs, pgdp=00000001787ee000
>>>  [0000000000000008] pgd=0000000000000000, p4d=0000000000000000
>>>  Internal error: Oops: 96000004 [#1] SMP
>>>  Modules linked in: vmwgfx(+) e1000e(+) nvme ahci(+) xhci_pci drm_ttm_helper ttm
>>> sha256_arm64 sha1_ce nvme_core xhci_pci_renesas aes_neon_bs aes_neon_blk aes>
>>>  CPU: 3 PID: 215 Comm: systemd-udevd Tainted: G     U            5.18.0-rc5-vmwgfx
>>> #12
>>
>> I'm confused, your kernel version seems to be 5.18.0-rc5 but this patch
>> is only in drm-misc-next now and will land in 5.20...
>>
>> Did you backport it? Can you please try to reproduce with latest drm-tip ?
> 
> No, this is drm-misc-next as of yesterday. drm-misc-next was still on 5.18.0-rc5
> yesterday.
> 

Right! I looked at the base for drm-tip but forgot that drm-misc was still on 5.18.

I'll look at this tomorrow but in the meantime, could you please look if the following
commits on top of drm-misc-next help ?

d258d00fb9c7 fbdev: efifb: Cleanup fb_info in .fb_destroy rather than .remove
1b5853dfab7f fbdev: efifb: Fix a use-after-free due early fb_info cleanup

-- 
Best regards,

Javier Martinez Canillas
Linux Engineering
Red Hat