Re: [rt2x00-users] [PATCH 2/4] rt2x00: move disabling of DMA before loading firmware

[Jakub Kicinski <kubakici@wp.pl>]

On Wed, 28 Mar 2012 14:14:29 +0200
Stanislaw Gruszka <sgruszka@redhat.com> wrote:

> Hello,
> 
> On Tue, Mar 27, 2012 at 08:11:07PM +0200, Gertjan van Wingerde wrote:
> > >  	/*
> > > -	 * Disable DMA, will be reenabled later when enabling
> > > -	 * the radio.
> > > -	 */
> > > -	rt2800_register_read(rt2x00dev, WPDMA_GLO_CFG, &reg);
> > > -	rt2x00_set_field32(&reg, WPDMA_GLO_CFG_ENABLE_TX_DMA, 0);
> > > -	rt2x00_set_field32(&reg, WPDMA_GLO_CFG_ENABLE_RX_DMA, 0);
> > > -	rt2800_register_write(rt2x00dev, WPDMA_GLO_CFG, reg);
> > > -
> > > -	/*
> > >  	 * Initialize firmware.
> > >  	 */
> > >  	rt2800_register_write(rt2x00dev, H2M_BBP_AGENT, 0);
> > 
> > This basically seems to be a revert of commit
> > 4ed1dd2a7ec880b301beb61cbc1e08811ec340e4 in the wireless-testing tree.
> > That commit was submitted by Stanislaw, which basically does the reverse
> > of what you are doing here.
> > 
> > I leave it up to you and Stanislaw to decide on what the state is that
> > we need to be in.
> 
> When DMA was disabled before FW load, I observed frame receives before
> ->enable_radio, so looks like WPDMA_GLO_CFG register is overwritten by
> F/W load. Although received frames did not make any damage, I prefer
> not to allow hw to send data to driver, before driver will be really
> ready to receive them.

Hm. I obviously did check whether FW load does overwrite WPDMA_GLO_CFG
and it didn't. Maybe it's device dependent? What bothers me is that now
we have a (very) short window between FW load and disabling DMA.

  -- Kuba