On 07/01/2014 03:31 PM, Stephen Warren wrote:
On 07/01/2014 03:25 PM, Jörg Krause wrote:
On 07/01/2014 07:41 PM, Stephen Warren wrote:
From: Stephen Warren swarren@nvidia.com
This is a series of small fixes and cleanups either required by those fixes, or enabled now that the fixes are made.
I hope that either patch 1 or 4 might fix the issues Jörg is seeing, but I'm not sure that will happen. The other patches shouldn't change any behaviour.
Stephen Warren (6): usb: ci_udc: fix ci_flush_{qh,qtd} calls in ci_udc_probe() usb: ci_udc: don't assume QTDs are adjacent when transmitting ZLPs usb: ci_udc: lift ilist size calculations to global scope usb: ci_udc: fix items array size/stride calculation usb: ci_udc: remove controller.items array usb: ci_udc: don't memalign() struct ci_req allocations
drivers/usb/gadget/ci_udc.c | 62 ++++++++++++++++++++++----------------------- drivers/usb/gadget/ci_udc.h | 1 - 2 files changed, 30 insertions(+), 33 deletions(-)
Good news! The last patch usb: ci_udc: don't memalign() struct ci_req allocations removes the timeout error after starting the fourth run of tftp in a row.
This is how I tested. Checked out u-boot-usb/master branch. Applied the necessary patches to support our board. Applied the patches step after step. After applying a patch reset the board and run tftp from console until an error occured, which is always the fourth run.
How many times did you boot the board for each patch? I'm more interested in how often the first TFTP after a reset/power-on passes or fails, so it would be nice to boot the board many times to see what happens to the first TFTP invocation too. This is especially true since you'd indicated before that the problem was (at least sometimes) visible on the first TFTP invocation, and this "it fails the fourth time" symptom is something completely new.
This is the case until applying patch usb: ci_udc: don't memalign() struct ci_req allocations, which throws no timeout error within running tftp about 60 times in a row.
That's quite odd. That patch definitely should not affect behaviour (and indeed I only sent it as an after-thought). If it does, then the only explanation I can think of is that the malloc heap's alignment changed, which just happens to hide the bug you're seeing. No doubt, there is still some lingering cache-flushing bug or similar...
BTW, did you fix the U-Boot header files in your board support patches, so that everything correctly knows that the cache line size is 32? I think it's mandatory to fix that issue before testing the USB code.
Can you please try one more thing:
Go back to u-boot-usb/master plus just your board support patches. Apply the following and test that:
diff --git a/drivers/usb/gadget/ci_udc.c b/drivers/usb/gadget/ci_udc.c index a6433e8d2d8d..13be1b73d3d1 100644 --- a/drivers/usb/gadget/ci_udc.c +++ b/drivers/usb/gadget/ci_udc.c @@ -209,9 +209,9 @@ ci_ep_alloc_request(struct usb_ep *ep, unsigned int gfp_flags) ci_req = memalign(ARCH_DMA_MINALIGN, sizeof(*ci_req)); if (!ci_req) return NULL;
memset(ci_req, 0, sizeof(*ci_req)); INIT_LIST_HEAD(&ci_req->queue);
ci_req->b_buf = 0; if (num == 0) controller.ep0_req = ci_req;
Thanks.