[PATCH 1/2] arm/mm: L2CC shared mutex with ARM TZ

[Etienne CARRIERE ST]

Sorry, I don't catch "swizzle your implementation" ?

You are referring to the raw_spinlock LOCKBREAK stuf ?
I will look deeper into linux code to check is some deadlock could appear.

The shared mutex arch spinlock is locked after raw_spinlick locking and released before l2x0 native raw_spinlock releasing.
It is intended to leave kernel preemption stuff inside the linux world.

etienne

-----Original Message-----
From: Will Deacon [mailto:will.deacon at arm.com] 
Sent: Wednesday, November 14, 2012 10:51 AM
To: Etienne CARRIERE ST
Cc: Russell King - ARM Linux; linux-arm-kernel at lists.infradead.org; Marc Zyngier; Catalin Marinas; Linus Walleij; Rabin VINCENT; Srinidhi KASAGAR
Subject: Re: [PATCH 1/2] arm/mm: L2CC shared mutex with ARM TZ

On Wed, Nov 14, 2012 at 09:18:31AM +0000, Etienne CARRIERE ST wrote:
> This is why the mutex added for TZ shared of L2 cache maintenance is a "arch_spinlock". It assumes linux handles the spinlock as a native ARM native mutex DDR cell, handled from the ARMv7-A LDREX/STREX.
> I think this is the weakest part of the driver since they is no insurance that linux pristine arch_spinlock will not evolve into something else (i.e features in raw_spinclock).

How does this work with our ticket lock approach? Can you swizzle your locking implementation by inspecting the kernel version of the code running on the non-secure side? (I sincerely hope not...)

Will