On Sep 13, 2013, at 6:45 AM, Hans de Goede <hdego...@redhat.com> wrote:
> Hi,
>
> On 09/13/2013 02:35 PM, Nathan Hjelm wrote:
>>
>> On Sep 13, 2013, at 3:18 AM, Hans de Goede <hdego...@redhat.com> wrote:
>
> <snip>
>
>>> 2) The alloc_streams backend op should not fail when the user requests more
>>> streams
>>> then available, instead it should simply allocate the number of
>>> available streams
>>> available and return that (as mentioned in the doxygen docs)
>>>
>>> So this is wrong:
>>>
>>> + (*(cInterface->interface))->SupportsStreams (cInterface->interface,
>>> pipeRef, &supportsStreams);
>>> + if (num_streams > supportsStreams) {
>>> + rc = LIBUSB_ERROR_INVALID_PARAM;
>>> + break;
>>> + }
>>>
>>> And should be:
>>>
>>> + (*(cInterface->interface))->SupportsStreams (cInterface->interface,
>>> pipeRef, &supportsStreams);
>>> + if (num_streams > supportsStreams)
>>> + num_streams = supportsStreams
>>>
>>> Note the check for how many streams are supported is done inside the kernel
>>> with Linux, and it will continue with the maximum number of streams
>>> supported
>>> if the user request more streams, so it is hard to implement any other logic
>>> under Linux, while it seems easy to mimick the Linux behavior with darwin.
>>
>> Hmm, what exactly are the semantics under Linux if the endpoints support a
>> different number of streams?
>
> The xhci driver (which implements this) first iterates over all endpoints to
> find the minimum support by
> all of them. Then if the minimum <= 1 it bails with an error (no streams
> supported on 1 or more eps), if
> the minimum > 1, it allocates the minimum found on all eps.
Makes sense. I updated the doxygen of alloc_streams to make those semantics
clear.
>> To mimic I will either need to take the minimum first and allocate that for
>> each endpoint (returning an error if any endpoint returns 0 supported
>> streams) or just allocate the smaller of num_streams or the maximum
>> supported streams for each endpoint and return the mimimum.
>
> Right, you need to "calculate" the minimum first.
I made the relevant changes and force-pushed an update to my tree. Let me know
if there are any other issues.
So far, I think the semantics are good. Will need some feedback from the
Windows and BSD developers on how well they will work for them.
Pete, I know the Microsoft compiler is a sub-par C compiler so I want to make
sure uint32_t is available before chaning the type of num_streams. Does it
support standard C99 integer types (stdint.h)?
-Nathan
------------------------------------------------------------------------------
How ServiceNow helps IT people transform IT departments:
1. Consolidate legacy IT systems to a single system of record for IT
2. Standardize and globalize service processes across IT
3. Implement zero-touch automation to replace manual, redundant tasks
http://pubads.g.doubleclick.net/gampad/clk?id=51271111&iu=/4140/ostg.clktrk
_______________________________________________
libusbx-devel mailing list
libusbx-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/libusbx-devel
