lists.openwall.net   lists  /  announce  owl-users  owl-dev  john-users  john-dev  passwdqc-users  yescrypt  popa3d-users  /  oss-security  kernel-hardening  musl  sabotage  tlsify  passwords  /  crypt-dev  xvendor  /  Bugtraq  Full-Disclosure  linux-kernel  linux-netdev  linux-ext4  linux-hardening  PHC 
Open Source and information security mailing list archives
 
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date:	Mon, 10 Nov 2014 11:33:29 +0100
From:	Jan Kiszka <jan.kiszka@....de>
To:	Marc Zyngier <marc.zyngier@....com>
CC:	Maxime Ripard <maxime.ripard@...e-electrons.com>,
	"linux-arm-kernel@...ts.infradead.org" 
	<linux-arm-kernel@...ts.infradead.org>,
	"linux-sunxi@...glegroups.com" <linux-sunxi@...glegroups.com>,
	Linux Kernel Mailing List <linux-kernel@...r.kernel.org>
Subject: Re: AMR: sun7i: CPU hotplug support?

On 2014-11-10 10:52, Marc Zyngier wrote:
> On 10/11/14 09:36, Jan Kiszka wrote:
>> On 2014-11-10 10:17, Marc Zyngier wrote:
>>> On 10/11/14 08:25, Jan Kiszka wrote:
>>>> On 2014-11-10 07:03, Jan Kiszka wrote:
>>>>> On 2014-11-10 00:17, Maxime Ripard wrote:
>>>>>> Hi Jan,
>>>>>>
>>>>>> On Sun, Nov 09, 2014 at 08:35:49PM +0100, Jan Kiszka
>>>>>> wrote:
>>>>>>> did anyone already happen to look into enabling CPU
>>>>>>> hotplug for the Allwinner A20 in upstream? I'm currently
>>>>>>> running the sunxi-next branch on Banana Pi, and echo 0 >
>>>>>>> .../cpu1/online just hangs the system. The old 3.4
>>>>>>> LeMaker kernel works fine in this regard. I can try to
>>>>>>> look into details and port things over, just want to
>>>>>>> avoid duplicate efforts.
>>>>>>
>>>>>> Having hotplug support would indeed be very welcome.
>>>>>>
>>>>>> However, it should be done in u-boot, through PSCI, and not
>>>>>> in the kernel itself.
>>>>>>
>>>>>> As far as I'm aware, no one worked actively on it, beside
>>>>>> some WIP commit from Marc a while ago:
>>>>>> https://git.kernel.org/cgit/linux/kernel/git/maz/u-boot.git/commit/?h=wip/psci&id=45379c0f9cf812f0f62722f4015ec907fa5dc144
>>>>>
>>>>>
>>>>>>
>>>
>>>>>>
> OK - I guess I will need a little guidance in then: Is there a good
>>>>> reference board to study and to derive from? And maybe also:
>>>>> What is missing or not working in that u-boot branch? If I
>>>>> get this interface right, I just takes some device tree bits
>>>>> to enable this for the kernel afterward, correct?
>>>>
>>>> Started to play with that patch in naive ways: CPU0 locks up
>>>> when offlining CPU1 - unless I disable the FIQ signal from
>>>> CPU1. Then it "works", both offlining and onlining again.
>>>> However, I suspect that this only parks CPU1 in wfi and does
>>>> not do anything interesting to it.
>>>
>>> Here's how this is supposed to work: - CPU1 sends a FIQ to CPU0,
>>> bringing it into secure mode. - CPU0 then kills CPU1 by doing the
>>> magic incantations on the power controller
>>>
>>> What is missing here is all the cache cleaning before signalling
>>> CPU0. If you add that, things should look a lot better (patches
>>> welcome).
> 
>> Unsure about this, or maybe this was too simplistic: I added calls
>> to u-boot's flush_dcache_all and invalidate_icache_all (right
>> after disabling the cache, just like the vendor kernel does), but
>> CPU0 still locks up. I suspect there is still a bug in the FIQ
>> handling. There is also a suspicious single "@" printed on the
>> console. I'll play with the FIQ handler a bit.
> 
> The '@' is just my own debug stuff, and might be causing issues too.
> 
> Now, you have to realise that by the time you call into this code,
> u-boot itself is long gone. Only the tiny bit of code dealing with
> PSCI still lives in a bank of static, secure memory. So calling into
> u-boot for anything is doomed. You need to actually put the code
> inside the PSCI backend.

OK - seems like quite a bit of code needs to be pulled over...

Meanwhile I'm still starring at psci_fiq_enter:
	...
	movw	r8, #(GICC_BASE & 0xffff)
	movt	r8, #(GICC_BASE >> 16)
	ldr	r9, [r8, #GICC_IAR]
	movw	r10, #0x3ff
	movt	r10, #0
	cmp	r9, r10
	bne	out1
	movw	r10, #0x3fe
	cmp	r9, r10
	bne	out1
	...

How can GICC_IAR be 0x3ff and 0x3fe at the same time? These tests seems
bogus. What was the actual intention here?

Jan


Download attachment "signature.asc" of type "application/pgp-signature" (199 bytes)

Powered by blists - more mailing lists