Hi Harald

I made a bunch of changes that significantly improved my test scenario:

Before the changes, the LTE SIM card being traced on a USB modem would fairly consistently "restart" the simtrace board (I later found out that the WDT was kicking in). And the same sim card on a Galaxy SII phone would get corrupted data (ie., wireshark would no longer be able to follow the conversation)

The changes are as follow:

1) The req_ctx are loaded into a double-linked list for consistency (FIFO manner). Previously, in the old code when multiple req_ctx are ready for transmit to USB, the lowered numbered (which may be the most recently filled req_ctx) would always be transmitted first. Additionally, to simplify the queue, there is only a single sized req_ctx now, 1 kb. There are 19 of them allocated.

2) The debug ring buffer is now transmitted using Peripheral DMA. Receive is left alone. Also, the append/flush and AT..._PrintFrame is written to allow for re-entrance (via Interrupt handler). Additionally, debugp will always prefix a line number to help / assist in debugging.

3) The watch dog (wdt.c) is modified so I can see when the watch dog is activated (which it no longer does in my current test, so you can comment out the #define WDT_DEBUG line...)

4) Both the usb interrupt and the uart interrupt handlers first enable core interrupt so that the AIC can do interrupt nesting (I observed at least 1 uart overrun per DEBUGPCR prior to these changes)

5) Minor additional debugpcr in req_ctx_init and in udp_refill_ep to indicate where the req_ctx are and containing what data to assist debug

Please review and comment

Issues I will work on further:

1) The first APDU after the ATR is almost always broken up (I don't think it's due to my changes). And occasionally, the length byte (p3?) sometimes is missing (ONLY on this LTE sim card / MODEM).

2) I need the simtrace to transmit a complete APDU as soon as possible. It seems not consistent on when it sends out received data to the USB. Sometimes, I would see GSM AUTHENTICATE apdu without a response for quiet a while...

----

I would like to have an additional field (2 byte short) in the simtrace header (the beginning of the req_ctx to indicate the position of the first byte of an APDU, so that in the event the synchronization is lost, the pc side can recover, because it appears the firmware code tracks the state of the APDU so it should be able to indicate it.

Thanks

Best Regards

On Sat, Aug 24, 2013 at 4:05 AM, Harald Welte <laforge@gnumonks.org> wrote:

Hi Min Xu,

On Fri, Aug 09, 2013 at 10:54:28AM -1000, Min Xu wrote:

> Thanks for getting back to me. After analyzing the issue, I believe the
> issue is the overrun error the USART reports. The issue appear to stem
> from the fact that each byte received by the USART is handled in the
> interrupt handler, but the USB sending code locks out the interrupt for the
> entirety of its call.

Thanks again for analyzing the problem. I was pondering to apply your
patch, but think a different approach might be better: Use the USART
DMA mode. Thiss way we don't get one interrupt per character, but one
interrupt every 'dma buffer size' characters. Using double buffering we
should be able to always process one particular dma buffer, while the
other buffer is being filled by the hardware without interrupts.

Processing of the DMA buffer could happen either in an interrupt handler
of the PDC (dma controller) or even in the main loop.

Unfortunately the SAM7S USB controller does not support DMA, so it is
impossible to improve on that side.

I will give this some more thought. If I'm unable to implement this
within short time, I might merge your patch anyway as a interim fix.

Regards,
Harald
--
- Harald Welte <laforge@gnumonks.org> http://laforge.gnumonks.org/
============================================================================
"Privacy in residential applications is a desirable marketing option."
(ETSI EN 300 175-7 Ch. A6)