Re: Fixing rt2500pci

[Ivo van Doorn <ivdoorn@gmail.com>]

On Wednesday 30 July 2008, Ivo van Doorn wrote:
> On Wednesday 30 July 2008, Peter Chubb wrote:
> > 
> > In kernel version 2.6.26-rc9 my wireless LAN card worked; but in the
> > released 2.6.26, my RaLink rt2500 card wouldn't associate.
> > 
> > Git-bisect led me to this patch:
> > 
> > 	61486e0f68d1f8966c09b734566a187d42d65c54
> > 	rt2x00: Remove ieee80211_tx_control argument from write_tx_desc()
> > 
> > I believe that there are two problems with that patch.  Setting the Tx
> > Length has been removed inadvertently, and one of the conditions for
> > when to set ifs to IFS_SIFS has also been removed.
> > 
> > This patch fixes those things, and with it my card works again.
> > 
> > Signed-off-by: Peter Chubb <peterc@gelato.unsw.edu.au>
> > 
> > diff --git a/drivers/net/wireless/rt2x00/rt2500pci.c b/drivers/net/wireless/rt2x00/rt2500pci.c
> > index 1724ce9..4ba2165 100644
> > --- a/drivers/net/wireless/rt2x00/rt2500pci.c
> > +++ b/drivers/net/wireless/rt2x00/rt2500pci.c
> > @@ -1198,6 +1198,7 @@ static void rt2500pci_write_tx_desc(struct rt2x00_dev *rt2x00dev,
> >  	rt2x00_set_field32(&word, TXD_W0_IFS, txdesc->ifs);
> >  	rt2x00_set_field32(&word, TXD_W0_RETRY_MODE,
> >  			   test_bit(ENTRY_TXD_RETRY_MODE, &txdesc->flags));
> > +	rt2x00_set_field32(&word, TXD_W0_DATABYTE_COUNT, skbdesc->data_len);
> >  	rt2x00_set_field32(&word, TXD_W0_CIPHER_ALG, CIPHER_NONE);
> >  	rt2x00_desc_write(txd, 0, word);
> >  }
> 
> How could that have disappeared? good catch.
> The above change is:
> 
> Acked-by: Ivo van Doorn <IvDoorn@gmail.com>
> 
> But:
> 
> > diff --git a/drivers/net/wireless/rt2x00/rt2x00dev.c b/drivers/net/wireless/rt2x00/rt2x00dev.c
> > index e1368f7..a11d6ff 100644
> > --- a/drivers/net/wireless/rt2x00/rt2x00dev.c
> > +++ b/drivers/net/wireless/rt2x00/rt2x00dev.c
> > @@ -691,7 +691,8 @@ void rt2x00lib_write_tx_desc(struct rt2x00_dev *rt2x00dev,
> >  	 * Set ifs to IFS_SIFS when the this is not the first fragment,
> >  	 * or this fragment came after RTS/CTS.
> >  	 */
> > -	if (test_bit(ENTRY_TXD_RTS_FRAME, &txdesc.flags)) {
> > +	if ((seq_ctrl & IEEE80211_SCTL_FRAG)  ||
> > +		test_bit(ENTRY_TXD_RTS_FRAME, &txdesc.flags)) {
> >  		txdesc.ifs = IFS_SIFS;
> >  	} else if (control->flags & IEEE80211_TXCTL_FIRST_FRAGMENT) {
> >  		__set_bit(ENTRY_TXD_FIRST_FRAGMENT, &txdesc.flags);
> 
> The above does not correspond with legacy driver, so I am hesitant in accepting this change.
> Unless you are _absolutely_ sure this change does fix the issue, and the databate count
> initialization is not sufficient on its own. If that is not the case, please resend the patch with the
> rt2x00dev.c change ommitted and my ack added. (Please add John Linville to the CC list as well,
> so it can be merged faster :) )

Correction, the code does correspond with the legacy driver, but the codechange in:
61486e0f68d1f8966c09b734566a187d42d65c54 was correct since it still applies IFS_SIFS to
frames with fragments (excluding the first fragment).

So please resend the patch with only the rt2500pci change.

Ivo