USB config SX lock deadlock

[Hans Petter Selasky]

On 10/10/13 09:38, Daniel O'Connor wrote:
>
> On 10/10/2013, at 16:59, Hans Petter Selasky <hps at bitfrost.no> wrote:
>>> It might under some circumstances but not directly.
>>>
>>> It has 3 sub interfaces, if there is an error on one the the DAQ program will call abort() which would result in FDs being closed.
>>>
>>> However that sub interface does not use read/poll only ioctl.
>>>
>> How do these IOCTLs work?
>
> They get turned into UT_READ_VENDOR_DEVICE requests in the driver.
>
>> Do you wakeup any sleepers at "surprise" detach?

Hi,

If you use synchronous USB control requests, then those should always 
error out.

> The usb_fifo_* code handles all wakeups so I am not sure.

The usb_fifo code only will only do refcounting. If you do USB control 
requests, you should use the f_ioctl_post, callback, because the f_ioctl 
callback does not protect against attach and detach or the enumeration 
thread running.

         err = (f->methods->f_ioctl) (f, cmd, addr, fflags);

         DPRINTFN(2, "f_ioctl cmd 0x%lx = %d\n", cmd, err);

         if (err != ENOIOCTL)
                 goto done;

         if (usb_usb_ref_device(cpd, &refs)) {
                 err = ENXIO;
                 goto done;
         }

         err = (f->methods->f_ioctl_post) (f, cmd, addr, fflags);

         DPRINTFN(2, "f_ioctl_post cmd 0x%lx = %d\n", cmd, err);

--HPS