[PATCH 3.18 3/7] b43: flush some writes on Broadcom MIPS SoCs

[Florian Fainelli]

2014-07-31 13:23 GMT-07:00 Hauke Mehrtens <hauke at hauke-m.de>:
> On 07/31/2014 09:59 PM, Rafał Miłecki wrote:
>> Access to PHY and radio registers is indirect on Broadcom hardware and
>> it seems that addressing on MIPS SoCs may require flushing. Broadcom
>> code does that unconditionally on MIPS, so let's do the same to make
>> sure hardware won't miss anything important.
>>
>> Signed-off-by: Rafał Miłecki <zajec5 at gmail.com>
>> ---
>>  drivers/net/wireless/b43/b43.h        |  8 ++++++++
>>  drivers/net/wireless/b43/main.c       | 17 ++++++++---------
>>  drivers/net/wireless/b43/phy_a.c      |  4 ++--
>>  drivers/net/wireless/b43/phy_common.c |  4 ++--
>>  drivers/net/wireless/b43/phy_g.c      |  8 ++++----
>>  drivers/net/wireless/b43/phy_ht.c     |  6 +++---
>>  drivers/net/wireless/b43/phy_lcn.c    |  6 +++---
>>  drivers/net/wireless/b43/phy_lp.c     |  6 +++---
>>  drivers/net/wireless/b43/phy_n.c      |  6 +++---
>>  9 files changed, 36 insertions(+), 29 deletions(-)
>>
>
> A comment in brcmsmac says this is only needed to work around a bug in
> the host pcie controller of 4717, 4718 and 4706. I haven't noticed this
> problem with BCM4706, so it could be that it is already fixed in the
> shipped version of bcm4706.

I don't think we have ever needed it on BCM63xx as well.

>
> The comment says the following:
> bcm4716 (which includes 4717 & 4718), plus 4706 on PCIe can reorder
> transactions. As a fix, a read after write is performed on certain
> places in the code. Older chips and the newer 5357 family don't require
> this fix.

The fix in brcmsmac is covering a lot more chips with a __mips__ that
it needs to. Maybe we should have a helper function that returns the
SoC ID so we can selectively apply this write then read flush
workaround?

>
> This commit did the cleanup:
> https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/?id=6a9a25eec0b55ea45e22710a9bcaf9690cb42fe6
>
-- 
Florian