linux-kernel - Re: [PATCH v5 02/19] ARM: LPAE: add ISBs around MMU enabling code

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 linux-cve-announce 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]

Message-ID: <1304956930.7658.84.camel@e102109-lin.cambridge.arm.com>
Date:	Mon, 09 May 2011 17:02:10 +0100
From:	Catalin Marinas <catalin.marinas@....com>
To:	Russell King - ARM Linux <linux@....linux.org.uk>
Cc:	linux-arm-kernel@...ts.infradead.org, linux-kernel@...r.kernel.org,
	Will Deacon <Will.Deacon@....com>
Subject: Re: [PATCH v5 02/19] ARM: LPAE: add ISBs around MMU enabling code

On Mon, 2011-05-09 at 16:48 +0100, Russell King - ARM Linux wrote:
> On Mon, May 09, 2011 at 04:34:16PM +0100, Russell King - ARM Linux wrote:
> > On Mon, May 09, 2011 at 04:01:56PM +0100, Catalin Marinas wrote:
> > > This doesn't work. From the ARM ARM (B1.3.3):
> > >
> > >         The execution state bits are the IT[7:0], J, E, and T bits. In
> > >         exception modes you can read or write these bits in the current
> > >         SPSR.
> > >         In the CPSR, unless the processor is in Debug state:
> > >         • The execution state bits, other than the E bit, are RAZ when
> > >         read by an MRS instruction.
> > >
> > > So reading the CPSR doesn't copy the T and E bits. Of course, we could
> > > set them explicitly but I find the ISB much simpler (and in practice we
> > > only need it for ARMv7 onwards but adding the ARMv6 in case we have a
> > > kernel compiled for both).
> >
> > Err.  If that's correct then the Linux kernel is totally broken, and
> > that's an incompatible change to the behaviour of the MRS and MSR
> > instructions which has gone unnoticed.
> >
> > We use "MRS reg, cpsr" for saving the IRQ state in SVC mode and
> > "MSR cpsr, reg" to restore the interrupt state.  If the T bit gets
> > reset by that, then Thumb kernels can never work.
> >
> > What you've just said tells me that our implementation of:
> > - arch_local_irq_save()
> > - arch_local_save_flags()
> > - arch_local_irq_restore()
> > won't work because we can't read or write the I and F bits using
> > MSR/MRS, even in SVC mode.
> >
> > What is the replacement method for doing this?
> >
> > If there isn't a replacement method...
> 
> And actually, that whole paragraph in the ARM ARM seems to be inconsistent
> with other bits of the ARM ARM.  For example:
> 
> In the CPSR, unless the processor is in Debug state:
> • The execution state bits, other than the E bit, are RAZ when read by
>   an MRS instruction.
> • Writes to the execution state bits, other than the E bit, by an MSR
>   instruction are:
>   — For ARMv7 and ARMv6T2, ignored in all modes.
>   — For architecture variants before ARMv6T2, ignored in User mode and
>     required to write zeros in privileged modes. If a nonzero value is
>     written in a privileged mode, behavior is UNPREDICTABLE.
> 
> Now, G.5.1 says this about ARMv6 (which is an 'architecture variants before
> ARMv6T2'):
> 
> ARMv6 and ARMv6K have the following differences:
> • Bits[26:25] are RAZ/WI.
> • Bits[15:10] are reserved.
> • The J and T bits of the CPSR must not be changed when the CPSR is written
>   by an MSR instruction, or else the behavior is UNPREDICTABLE. MSR
>   instructions exist only in ARM state in these architecture variants, so
>   this is equivalent to saying the MSR instructions in privileged modes must
>   treat these bits as SBZP. MSR instructions in User mode still ignore
>   writes to these bits.
> 
> The thing is, if you write zeros into the mode bits from supervisor mode
> (as required by B1.3.3), you're going to take the CPU back to 26-bit user
> mode, which on many CPUs is an undefined mode.

I still don't get what you mean. B1.3.3 paragraph above refers to the
"execution state bits, other than the E bit", IOW - IT[7:0], J and T.
>>From the same B1.3.3:

        The execution state bits are the IT[7:0], J, E, and T bits.

The mode bits you are talking about are called M[4:0]. These are note
"execution state bits" and can be read by MRS.

-- 
Catalin


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