Re: [PATCH RESEND 0/4] SMP support for Broadcom NSP

[Kapil Hali <kapilh@broadcom.com>]

On 11/6/2015 1:55 AM, Hauke Mehrtens wrote:
> On 11/05/2015 10:34 AM, Russell King - ARM Linux wrote:
>> On Thu, Nov 05, 2015 at 12:51:17AM -0500, Kapil Hali wrote:
>>> Hi,
>>>
>>> This series adds SMP support for Broadcom's Northstar Plus SoC.
>>>
>>> There are similar SMP enablement methods for many ARMv7 bsed SoCs.
>>> BCM NSP SoC, has a typical such mechanism - after power-on, the
>>> secondary core is held in a standby state, primary core provides a
>>> startup address for the secondary core and wakes it up. Booting of
>>> the secondary core is serialized using pen_release global variable.
>>
>> Why do you need the pen_release stuff?  The above implies that you
>> have only one secondary core, and you can control when it comes out
>> of standby state.
>>
>> Please, don't assume that the pen_release stuff is any kind of recommended
>> or standardised system.  It isn't.  It's a hack for ARMs evaluation
>> platforms.
>>
> 
> I tried to remove the pen code because I also thought/hoped that it is
> useless, but the 2. CPU did not boot any more after I removed it. I do
> not know the internals of SoC, but it looks like this is needed.
> 
> I described it here:
> http://www.spinics.net/lists/arm-kernel/msg452178.html
> 
> I removed this comparison and the jump afterwards ("cmp     r7, r0") and
> the 2. CPU did not boot any more. Is this pen stuff some kind of
> workaround for some bug in the silicon?
> 
pen stuff is not a work around for any bug in BCM NSP silicon. It was the
mechanism to bring-up SMP on some of the ARM based SoCs and many silicon
vendors seem to consider it as a generic mechanism for controlled bring-up
of SMP. But as is clear from Russel King's comments, it is not a 
standardized recommended method.
Also, I removed the pen_release method and it works on BCM NSP SoCs. I 
will add the changes in the next patch set.
> Hauke
> 
Thanks,
Kapil Hali