[RFC][PATCH] axi: add AXI bus driver

Rafał Miłecki zajec5 at gmail.com
Wed Apr 13 17:05:51 EDT 2011

2011/4/13 George Kashperko <george at znau.edu.ua>:
>> 2011/4/13 Arend van Spriel <arend at broadcom.com>:
>> > On Tue, 12 Apr 2011 22:44:01 +0200, Rafał Miłecki <zajec5 at gmail.com> wrote:
>> >
>> >> 2011/4/12 George Kashperko <george at znau.edu.ua>:
>> >>>>
>> >>>> 2011/4/12 Rafał Miłecki <zajec5 at gmail.com>:
>> >>>> That way I see really low (or not at all) relation between out
>> >>>> (not)Broadcom bus and present AMBA bus.
>> >>>>
>> >>> Agree.
>> >>
>> >> Ehh, sounds like one more renaming to functions, defines, prefixes.
>> >>
>> >> Let's wait for Arend comments, he was the one voting for not-bcm-specific
>> >> name.
>> >>
>> >
>> > Hi Rafał,
>> >
>> > Still think its better to stick with a generic name even if currently you
>> > only come across this in Broadcom chips right now. I do agree that the term
>> > 'axi' is implying something else than what this bus driver is providing. The
>> > name 'axi-dmp' I gave earlier may be more to the point.
>> >
>> > I also looked at the amba driver after receiving comments on the brcmaxi
>> > library module (this is what Hauke Mehrtens referred to) and came to similar
>> > conclusion as you did. It does however support PM properly so you may want
>> > to get inspiration in that area. I also noticed a reference to AMBA term APB
>> > which is a different bus in the AMBA family. AXI was introduced later as
>> > higher performance bus (in AMBA rev3 if I remember correctly).
>> I don't focus on PM yet, do not consider it a problem, it just needs some time.
>> Note for not involved: AMBA is family with few buses/interfaces
>> possible: AXI, AHB, ASB, APB, ATB [1].
>> So what are you saying is that drivers/amba/ is for AMBA APB? OK, I
>> can accept such a explanation and it makes me even more sure we need
>> another AMBA driver (this time: AMBA AXI).
> AMBA is AMBA. axi/apb/ahb etc are all subsets of AMBA and as of current
> all fit to what already is in drivers/amba.
>> The left question is: how much of the implemented code is AMBA AXI
>> specific and how much is Broadcom specific?
> Answers to 1. and 2. are there in drivers/amba/bus.c
> Look _probe and _register fn for more details.

Not able to. There is some call to amba_get_enable_vcore in probe. It
calls regulator_enable. Which calls some
rdev->desc->ops->enable(rdev). I managed to track enable is part of
struct regulator_ops but I'm not brave enough to track where does it
come from, etc. There are many documented regulator ops, but not
general description. I would have to spend some time on it but I'm not
sure if it's worth that. Not to mention there is some additional
amba_get_enable_pclk which calls some more functions I didn't date to

>> 1) Does AMBA AXI identify cores by manuf, id, rev and class? Is this
>> really AMBA AXI specific and a evolution from simple "id" in AMBA APB?
> During amba device registration amba bus code read component/peripheral
> id registers from known locations, exactly where
> peripherialid/componentid registers of master port (agent) core are.
> Look brcm80211/inclue/aidmp.h for those.

I guess you mean:
pid |= (readl(tmp + size - 0x20 + 4 * i) & 255) << (i * 8);
cid |= (readl(tmp + size - 0x10 + 4 * i) & 255) << (i * 8);
Well, OK, it proves Broadcom uses AMBA somehow, but not more than that
(at least for me).

>> 2) Is this standard for AMBA AXI to keep list of available cores in
>> some specific memory (is this always EPROM like in case of Bcm?)?
>> 3) Does every AMBA AXI device need enabling/disabling/resetting
>> routine we implemented? Is that really Bcm independent?
> Enable/disable abstracted by clocks interface.

Will comment on that later.


More information about the b43-dev mailing list