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

From: George Kashperko
Date: Tue Apr 12 2011 - 15:37:36 EST



> 2011/4/12 George Kashperko <george@xxxxxxxxxxx>:
> >
> >> 2011/4/12 George Kashperko <george@xxxxxxxxxxx>:
> >> >
> >> >> Hi,
> >> >>
> >> >> On Tue, Apr 12, 2011 at 01:57:07AM +0200, RafaÅ MiÅecki wrote:
> >> >> > Cc: Michael BÃsch <mb@xxxxxxxxx>
> >> >> > Cc: Larry Finger <Larry.Finger@xxxxxxxxxxxx>
> >> >> > Cc: George Kashperko <george@xxxxxxxxxxx>
> >> >> > Cc: Arend van Spriel <arend@xxxxxxxxxxxx>
> >> >> > Cc: linux-arm-kernel@xxxxxxxxxxxxxxxxxxx
> >> >> > Cc: Russell King <rmk@xxxxxxxxxxxxxxxx>
> >> >> > Cc: Arnd Bergmann <arnd@xxxxxxxx>
> >> >> > Cc: Andy Botting <andy@xxxxxxxxxxxxxxx>
> >> >> > Cc: linuxdriverproject <devel@xxxxxxxxxxxxxxxxxxxxxx>
> >> >> > Cc: linux-kernel@xxxxxxxxxxxxxxx <linux-kernel@xxxxxxxxxxxxxxx>
> >> >> > Signed-off-by: RafaÅ MiÅecki <zajec5@xxxxxxxxx>
> >> >> > ---
> >> >> > V2: Rename to axi
> >> >> > Use DEFINE_PCI_DEVICE_TABLE in bridge
> >> >> > Make use of pr_fmt and pr_*
> >> >> > Store core class
> >> >> > Rename bridge to not b43 specific
> >> >> > Replace magic 0x1000 with BCMAI_CORE_SIZE
> >> >> > Remove some old "ssb" names and defines
> >> >> > Move BCMAI_ADDR_BASE def
> >> >> > Add drvdata field
> >> >> > V3: Fix reloading (kfree issue)
> >> >> > Add 14e4:0x4331
> >> >> > Fix non-initialized struct issue
> >> >> > Drop useless inline functions wrappers for pci core drv
> >> >> > Proper pr_* usage
> >> >> > V3.1: Include forgotten changes (pr_* and include related)
> >> >> > Explain why we dare to implement empty release function
> >> >>
> >> >> I'm not sure we need this. If you have an IP Core which talks AXI and
> >> >> you want to put it on a PCI bus, you will have a PCI Bus wrapper around
> >> >> that IP Core, so you should go and let the kernel know about that. See
> >> >> [1] for a core IP which talks AXI and [2] for a PCI bus glue layer.
> >> >>
> >> >> Besides, if you introduce this bus layer, it'll be more difficult for
> >> >> other licensees of the same core to re-use the same driver, since it's
> >> >> now talking a PCI emulated on top of AXI. The same can be achieved with
> >> >> the platform_bus which is more widely used, specially on ARM SoCs.
> >> >>
> >> >> [1] http://gitorious.org/usb/usb/blobs/dwc3/drivers/usb/dwc3/core.c
> >> >> [2] http://gitorious.org/usb/usb/blobs/dwc3/drivers/usb/dwc3/dwc3-haps.c
> >> >>
> >> >
> >> > Already noticed earlier that AXI isnt really good name for
> >> > Broadcom-specific axi bus customization. As of tech docs available from
> >> > arm, corelink AXI cores use own identification registers which feature
> >> > different format and layout comparing to that we use for Broadcom cores.
> >> >
> >> > Maybe there is something "standartized" by the DMP specs? If so I'm
> >> > curious if that DMP is obligatory for every axi bus ?
> >> >
> >> > Naming particular Broadcom's implementation just axi limits other
> >> > licensees in reusing axi bus name/code or will require hacks/workarounds
> >> > from them to fit Broadcom-like core scanning/identificating techniques.
> >> > You use bus named AXI to group and manage Broadcom cores, while never
> >> > even publish device records for native axi cores Broadcom use to talk to
> >> > the interconnect through. Yet again, something like bcmb/bcmai looks
> >> > like better name for this bus.
> >>
> >> I don't know, I'm really tired of this. Earlier I was told to not use
> >> anything like bcmai, because it is not Broadcom specific. Now it seems
> >> (and I'm afraid I agree) there is quite a lot of Broadcom specific
> >> stuff.
> > Well, _if_ that "magic" EROM core layout is arm's "standard" for axi
> > ports identification _and_ _if_ that EROM core is obligatory axi
> > component then sure axi name is good one as soon as you consider
> > registering master port (agent) cores with device subsystem as well.
> > I have no clue here about how resolve those _if_'s, hopefully Broadcom
> > guys can enlighten us on the subject.
>
> Do you think that in my code only scanning is Broadcom specific? In
> such a case we could keep it "axi", and just s/scan/bcmscan/. This is
> only correct choice if the rest (addressing, core enabling, host
> management) is AXI specific.
We rely on core identification following its id/ven/rev/class layout
from EROM. Corelink cores use different layout/location. Also core
enabling/disabling are DMP related. Another _if_ here -- is DMP
obligatory for all axi buses?

>
>
> >> > Also can't figure out how is this implementation supposed to manage
> >> > multicore devices.
> >>
> >> We have got ideas, but let's first find (wait for) such a device ;)
> > bcm4716 usb host ? Don't really know if there are any other multicores.
>
> This is SoC, we do not have any problems supporting multiple cores on
> SoCs. All cores are available at the same time, without any sliding
> windows.
>
Multicores have multiple slave ports bound to single master port
(agent). Individual core driver must be confident of its "partner"
device/driver and both must know who is permitted to
enable/disable/reset and who is not or there should be some other
technique for agent management.

Have nice day,
George


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/