On 02/14/2010 03:53 AM, Thomas Backlund wrote:
quoted text
> Hi,
> (please cc me on replies)
> 
> We have a user that tried out b43, but got this in the logs:
> 
> --- cut ---
> 65858:Feb  9 22:05:16 elmo kernel: b43-phy2 ERROR: This device does not
> support DMA on your system. Please use PIO instead. 65859:Feb  9
> 22:05:16 elmo kernel: b43-phy2 ERROR: CONFIG_B43_FORCE_PIO must
> be set in your kernel configuration.
> 65860:Feb  9 22:05:16 elmo kernel: b43-phy2 debug: Adding Interface type 2
> 65861:Feb  9 22:05:16 elmo kernel: b43-phy2 ERROR: Fatal DMA error:
> 0x00000400, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000
> 
> 65862:Feb  9 22:05:16 elmo kernel: b43-phy2 ERROR: This device does not
> support DMA on your system. Please use PIO instead.
> 65863:Feb  9 22:05:16 elmo kernel: b43-phy2 ERROR: CONFIG_B43_FORCE_PIO
> must be set in your kernel configuration.
> --- cut ---
> 
> 
> 
> But reading the Kconfig help, it states:
> --- cut ---
> config B43_FORCE_PIO
>         bool "Force usage of PIO instead of DMA"
>         depends on B43 && B43_DEBUG
>         ---help---
>           This will disable DMA and always enable PIO instead.
> 
>           Say N!
>           This is only for debugging the PIO engine code. You do
>           _NOT_ want to enable this.
> --- cut ---
> 
> 
> So,
> wich one is it ?
> 
> Do I belive the dmesg output, or the Kconfig ?
> 
> Note,
> the b43 works for the user if he enable the CONFIG_B43_FORCE_PIO.
> 
> But I'm thinking of this problem from a distro point of view.
> Will it break for others if I enable it ?

From a distro point of view, you would not want to set FORCE_PIO as the
performance penalty would be very large.

You do not give the specific details on the problem system; however, it is
probably a BCM4312 802.11 b/g device with PCI ID 14e4:4315 being used with an
Atom processor in a netbook. We have no fix.

In the 2.6.34 code, b43 will be changed to allow the selection of PIO mode at
run time rather than compile time. For a distro, this method is clearly
superior. Those users that need PIO can select it without forcing the
performance penalty on everyone. The patch was too late for the 2.6.33 merge,
and too intrusive to be applied to 2.6.33-rcX. It has been present in the
wireless-testing code base for 2 months with no trouble reported, thus it should
be safe for inclusion in your kernels. The relevant patch is attached.

Larry