All of lore.kernel.org
 help / color / mirror / Atom feed
* Questions on carl9170 architecture
@ 2010-10-08 16:31 Francesco Gringoli
  2010-10-08 18:42 ` Christian Lamparter
  0 siblings, 1 reply; 2+ messages in thread
From: Francesco Gringoli @ 2010-10-08 16:31 UTC (permalink / raw)
  To: linux-wireless, Christian Lamparter

Hello Christian,

I have tried carl9170 for a while, everything is ok: what a great work! I have not noticed any problem so far, it seems almost stable.

I'm digging into carl9170 source code to understand how it works and what one can build starting from it. There are however a few points that I didn't understand, would you mind helping me?

- how are handled acknowledgments? I don't see anything that schedules ack transmission when a packet that need ack is received. Is that accomplished internally by the hw? If yes, do you know if it can be disabled and replaced with new code so that it is the firmware to control ack transmission? For instance, a number of techniques have been recently proposed to estimate the channel condition when receiving corrupted frames: it could be interesting to send back a "special-ack" when a packet is received corrupted to help the transmitter understanding how to best retransmit the packet. Obviously this should be done within a SIFS from the reception of the packet;

- to which precision is it possible to drive packet transmission? e.g., is it possible to start sending a packet a chosen interval after the previous?

- is it possible to send packet independently of CCA, e.g., starting sending a packet even if the channel is busy?

- is it possible to build a packet internally and decide to send it independently of the host?

- is it possible to modify packets sent by the host before they leave the adapter?

Sorry to bother you with so many questions.

Kind regards,
-Francesco Gringoli


Informativa sulla privacy: http://help.ing.unibs.it/privacy.php



^ permalink raw reply	[flat|nested] 2+ messages in thread

* Re: Questions on carl9170 architecture
  2010-10-08 16:31 Questions on carl9170 architecture Francesco Gringoli
@ 2010-10-08 18:42 ` Christian Lamparter
  0 siblings, 0 replies; 2+ messages in thread
From: Christian Lamparter @ 2010-10-08 18:42 UTC (permalink / raw)
  To: Francesco Gringoli; +Cc: linux-wireless

On Friday 08 October 2010 18:31:42 Francesco Gringoli wrote:
> Hello Christian,
> 
> I have tried carl9170 for a while, everything is ok: what a great work!
> I have not noticed any problem so far, it seems almost stable.
"almost". That's the biggest issue, the driver is "new" and all
the bugs are hidden from sight.

> - how are handled acknowledgments? I don't see anything that schedules ack
> transmission when a packet that need ack is received. Is that accomplished
> internally by the hw?
Yes, Atheros calls it the "Response Controller" (and it's pretty much
hardwired into the MAC HW)

> If yes, do you know if it can be disabled and replaced with new code so
> that it is the firmware to control ack transmission?
Maybe. The normal ACKs can be disabled by zeroing 0x1c36f8.

> For instance, a number of techniques have been recently proposed to estimate
> the channel condition when receiving corrupted frames: it could be
> interesting to send back a "special-ack" when a packet is received corrupted
> to help the transmitter understanding how to best retransmit the packet.
Hmm, by the time you've "generated" your special response, the peer
already knows that the attempt has failed.
anyway, that's OT ;)

> Obviously this should be done within a SIFS from the reception of the packet;
That's going to be though, especially under load.
Apparently, the HW doesn't provide real interrupts.
Therefore everything needs to be "polled"
(yes, that includes the USB DMA too). 
 
> - to which precision is it possible to drive packet transmission?
> e.g., is it possible to start sending a packet a chosen interval after the previous?
Sorry, but my documentation doesn't cover the topic.
You have to ask Atheros.

> - is it possible to send packet independently of CCA, e.g., starting
> sending a packet even if the channel is busy?
hmm, there's a CCA threshold register somewhere.
But no details in the docs, so you might want to ask Atheros.

> - is it possible to build a packet internally and decide to send
> it independently of the host?
kind of. Take a look at handle_bar() and wlan_send_buffered_ba().

> - is it possible to modify packets sent by the host before
> they leave the adapter?
sure, you can probably do that. but keep in mind that device
uses vectored-i/o and the frame data is split across 320 bytes
chunks.

Regards,
	Christian

^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2010-10-08 18:43 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2010-10-08 16:31 Questions on carl9170 architecture Francesco Gringoli
2010-10-08 18:42 ` Christian Lamparter

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.