Re: usb raw gadget impossible to sent buffer bigger than 3MB

[Peter Chen <peter.chen@nxp.com>]

On 20-05-28 08:19:46, Belisko Marek wrote:
> Hi,
> 
> we're using USB RAW gadget for communicating with PC application. We
> have created loopback test which send file (any size) from PC to
> device and then same data are sent back to PC to verify roundtrip time
> (using bulk data transfer). Everything works fine up to 3MB file size.
> If we sent bigger file like e.g. 5M we can receive file on device but
> when we want to write to output endpoint we got:
> 
> WARNING: CPU: 0 PID: 12299 at /kernel-source//mm/page_alloc.c:3725
> __alloc_pages_nodemask+0x1b0/0xde4
> [<c010fa08>] (unwind_backtrace) from [<c010c45c>] (show_stack+0x20/0x24)
> [<c010c45c>] (show_stack) from [<c04cc0d4>] (dump_stack+0x20/0x28)
> [<c04cc0d4>] (dump_stack) from [<c01324ac>] (__warn+0xec/0x108)
> [<c01324ac>] (__warn) from [<c0132598>] (warn_slowpath_null+0x30/0x38)
> [<c0132598>] (warn_slowpath_null) from [<c021ccb8>]
> (__alloc_pages_nodemask+0x1b0/0xde4)
> [<c021ccb8>] (__alloc_pages_nodemask) from [<c023b888>]
> (kmalloc_order+0x2c/0x48)
> [<c023b888>] (kmalloc_order) from [<c023b8d0>] (kmalloc_order_trace+0x2c/0xd4)
> [<c023b8d0>] (kmalloc_order_trace) from [<c0261c24>] (__kmalloc+0x40/0x264)
> [<c0261c24>] (__kmalloc) from [<bf039bc4>] (ffs_epfile_io+0x13c/0x570
> [usb_f_fs])
> [<bf039bc4>] (ffs_epfile_io [usb_f_fs]) from [<bf03a0c0>]
> (ffs_epfile_write_iter+0xc8/0x120 [usb_f_fs])
> [<bf03a0c0>] (ffs_epfile_write_iter [usb_f_fs]) from [<c02734bc>]
> (new_sync_write+0xc8/0xec)
> [<c02734bc>] (new_sync_write) from [<c027351c>] (__vfs_write+0x3c/0x48)
> [<c027351c>] (__vfs_write) from [<c02749a0>] (vfs_write+0xcc/0x158)
> [<c02749a0>] (vfs_write) from [<c02756ac>] (SyS_write+0x50/0x88)
> [<c02756ac>] (SyS_write) from [<c0107a20>] (ret_fast_syscall+0x0/0x54)
> ---[ end trace fe5f79fe415b9881 ]---
> 
> and write ends up with: write /run/ffs/ep1: cannot allocate memory
> 
> When checked free command there should be plenty of available memory.
> Is there some limitation when writing to endpoint? We tried to split
> buffer to e.g. 3M and sent it and this works but looks like there is
> penalty when sending bigger files (100MB file received in 5secs whicle
> sending back it took 2minutes). Thanks for ideas and hints.
> 

Although you see there are plenty of available memory, but the block number
for the size(eg 64KB) you request may be zero. Since you use old
kernel, the f_fs doesn't support scatter list. To fix it, you may use
small buffer at use space, eg: 16KB per transfer.

-- 

Thanks,
Peter Chen