RE: [PATCH] usb: gadget: f_fs: data_len used before properly set

[Yang, Fei]

>> The following line of code in function ffs_epfile_io is trying to set 
>> flag io_data->use_sg in case buffer required is larger than one page.
>>
>>     io_data->use_sg = gadget->sg_supported && data_len > PAGE_SIZE;
>>
>> However at this point of time the variable data_len has not been set 
>> to the proper buffer size yet. The consequence is that io_data->use_sg 
>> is always set regardless what buffer size really is, because the 
>> condition (data_len > PAGE_SIZE) is effectively an unsigned comparison 
>> between -EINVAL and PAGE_SIZE which would always result in TRUE.
>>
>> Fixes: 772a7a724f69 ("usb: gadget: f_fs: Allow scatter-gather 
>> buffers")
>> Signed-off-by: Fei Yang <fei.yang@xxxxxxxxx>
>> Cc: stable <stable@xxxxxxxxxxxxxxx>
>
> Hey Fei! Thanks so much for sending this out! I was excited that this might resolve
> the ffs stalls I've been seeing on dwc3/dwc2 hardware, but when I gave it a shot, it
> doesn't seem to help. In fact, rather then a stall, I end up seeing the following panic:
> 
> [  383.415362] Unable to handle kernel read from unreadable memory at virtual address 0000000000000018
> [  383.431935] Mem abort info:
> [  383.431937]   ESR = 0x96000005
> [  383.431940]   Exception class = DABT (current EL), IL = 32 bits
> [  383.431941]   SET = 0, FnV = 0
> [  383.431942]   EA = 0, S1PTW = 0
> [  383.431943] Data abort info:
> [  383.431945]   ISV = 0, ISS = 0x00000005
> [  383.431946]   CM = 0, WnR = 0
> [  383.431951] user pgtable: 4k pages, 39-bit VAs, pgdp=00000000aae1f000
> [  383.431953] [0000000000000018] pgd=000000009f064003, pud=000000009f064003, pmd=0000000000000000
> [  383.482560] Internal error: Oops: 96000005 [#1] PREEMPT SMP
> [  383.488128] Modules linked in:
> [  383.491181] CPU: 0 PID: 399 Comm: irq/69-dwc3 Tainted: G S
>       5.2.0-rc4-00092-gf5f12f5d3fdd #296 [  383.501002] Hardware name: HiKey960 (DT)
> [  383.504918] pstate: 20400085 (nzCv daIf +PAN -UAO)
> [  383.509714] pc : dma_direct_unmap_sg+0x38/0x80
> [  383.514151] lr : dma_direct_unmap_sg+0x5c/0x80
> [  383.518586] sp : ffffff8011fcbc40
> [  383.521893] x29: ffffff8011fcbc40 x28: ffffffc0bad9c180
> [  383.527199] x27: ffffffc0bae05300 x26: 0000000000000002
> [  383.532504] x25: ffffffc0b9a9fd00 x24: 0000000000000000
> [  383.537809] x23: 0000000000000001 x22: ffffffc0bad9fc10
> [  383.543114] x21: 0000000000000002 x20: 0000000000000001
> [  383.548420] x19: 0000000000000000 x18: 0000000000000000
> [  383.553726] x17: 0000000000000000 x16: 0000000000000000
> [  383.559032] x15: 0000000000000000 x14: ffffff8010eb6ad0
> [  383.564338] x13: 0000000000000000 x12: 0000000000000000
> [  383.569643] x11: 0000000000000000 x10: 00000000000009d0
> [  383.574949] x9 : ffffff8011fcbd20 x8 : ffffffc0b63c3a30
> [  383.580254] x7 : ffffffc0bc515c00 x6 : 0000000000000007
> [  383.585560] x5 : 0000000000000001 x4 : 0000000000000004
> [  383.590865] x3 : 0000000000000001 x2 : 0000000000000001
> [  383.596169] x1 : 000000000006bf42 x0 : 0000000000000000
> [  383.601477] Call trace:
> [  383.603916]  dma_direct_unmap_sg+0x38/0x80
> [  383.608013]  usb_gadget_unmap_request_by_dev+0xb0/0xc8
> [  383.613148]  dwc3_gadget_del_and_unmap_request.isra.13+0x78/0x150
> [  383.619235]  dwc3_gadget_giveback+0x30/0x68
> [  383.623412]  dwc3_thread_interrupt+0x694/0x14e0
> [  383.627938]  irq_thread_fn+0x28/0x78
> [  383.631506]  irq_thread+0x124/0x1c0
> [  383.634991]  kthread+0x128/0x130
> [  383.638214]  ret_from_fork+0x10/0x1c
> [  383.641786] Code: 2a0303f7 aa0403f8 52800014 d503201f (b9401a62)
> [  383.647874] ---[ end trace f48053c2040c5658 ]---
> 
> From the looks of it though, I suspect your fix is a good one, and maybe its just helping
> expose some related underlying issues in the dwc3 driver?

It doesn't fix the ffs stall issue for me either, but I have not seen this panic though.

It's interesting to see this panic because dma unmapping is what I'm looking at right now.
When I see the ffs stalls, the problem seems to be that a read request of 512 bytes returning
right away with a buffer filled with all zeros. And that is happening after a bunch of requests
of 16384 bytes. I'm suspecting the dma as well, because the completion interrupt for this
request of 512 bytes seems to be fired mistakenly.

I don't always have time to work on this issue though, might not update for a while. Please
keep me posted if you find anything.

Thanks,
-Fei