RE: [PATCH] usb: gadget: udc: fix the kernel NULL pointer in composite_setup

[Felipe Balbi]

Hi,

(please don't top-post and also break your lines at 80-columns ;-)

"He, Bo" <bo.he@xxxxxxxxx> writes:
> 	1. the issue reproduced very rarely, we run reboot test
> 	reproduce the issue, it reproduced two times on two board after
> 	more than 1500 cycles reboot.

That's fine, we, somehow, got a use-after-free on the tracepoints. I'm
interested in fixing that without touching udc-core since that's a
dwc3-only bug.

> 	2. the kernel version is 4.4, the test case is cold reboot, I think it's not android patches cause it, it's the interrupt thread run after the udc->driver->unbind.

Yeah, I need you to try v4.13-rc1. v4.4 is *really* old. I can't accept
your patch unless I'm certain the bug still exists.

> 	3. I check more drivers, like amd5536_udc_stop, at91_stop,
> 	atmel_usba_stop, bcm63xx_udc_stop, s3c_hsudc_stop, all the
> 	interrupt disable will be in the udc_stop(), so we need
> 	guarantee to stop the interrupt then release the resource.

Right, we also disable the interrupt on ->udc_stop(). See below:

static void __dwc3_gadget_stop(struct dwc3 *dwc)
{
	dwc3_gadget_disable_irq(dwc);
	__dwc3_gadget_ep_disable(dwc->eps[0]);
	__dwc3_gadget_ep_disable(dwc->eps[1]);
}

static int dwc3_gadget_stop(struct usb_gadget *g)
{
	struct dwc3		*dwc = gadget_to_dwc(g);
	unsigned long		flags;
	int			epnum;

	spin_lock_irqsave(&dwc->lock, flags);

	if (pm_runtime_suspended(dwc->dev))
		goto out;

	__dwc3_gadget_stop(dwc);

	for (epnum = 2; epnum < DWC3_ENDPOINTS_NUM; epnum++) {
		struct dwc3_ep  *dep = dwc->eps[epnum];

		if (!dep)
			continue;

		if (!(dep->flags & DWC3_EP_END_TRANSFER_PENDING))
			continue;

		wait_event_lock_irq(dep->wait_end_transfer,
				    !(dep->flags & DWC3_EP_END_TRANSFER_PENDING),
				    dwc->lock);
	}

out:
	dwc->gadget_driver	= NULL;
	spin_unlock_irqrestore(&dwc->lock, flags);

	free_irq(dwc->irq_gadget, dwc->ev_buf);

	return 0;
}

-- 
balbi

Attachment: signature.asc
Description: PGP signature