Re: [PATCH] pkeys: Introduce PKEY_ALLOC_SIGNALINHERIT and change signal semantics

From: Dave Hansen <dave.hansen@intel.com>
To: Andy Lutomirski <luto@kernel.org>, Florian Weimer <fweimer@redhat.com>
Cc: Linux-MM <linux-mm@kvack.org>,
	Linux API <linux-api@vger.kernel.org>,
	linux-x86_64@vger.kernel.org,
	linux-arch <linux-arch@vger.kernel.org>, X86 ML <x86@kernel.org>,
	linuxram@us.ibm.com
Subject: Re: [PATCH] pkeys: Introduce PKEY_ALLOC_SIGNALINHERIT and change signal semantics
Date: Wed, 2 May 2018 15:08:27 -0700	[thread overview]
Message-ID: <a37b7deb-7f5a-3dfa-f360-956cab8a813a@intel.com> (raw)
In-Reply-To: <CALCETrW_Dt-HoG4keFJd8DSD=tvyR+bBCFrBDYdym4GQbfng4A@mail.gmail.com>

On 05/02/2018 02:23 PM, Andy Lutomirski wrote:
>> The kernel could do *something*, probably along the membarrier system
>> call.  I mean, I could implement a reasonable close approximation in
>> userspace, via the setxid mechanism in glibc (but I really don't want to).
> 
> I beg to differ.
> 
> Thread A:
> old = RDPKRU();
> WRPKRU(old & ~3);
> ...
> WRPKRU(old);
> 
> Thread B:
> pkey_alloc().
> 
> If pkey_alloc() happens while thread A is in the ... part, you lose.  It
> makes no difference what the kernel does.  The problem is that the WRPKRU
> instruction itself is designed incorrectly.

Yes, *if* we define pkey_alloc() to be implicitly changing other
threads' PKRU value.

Let's say we go to the hardware guys and ask for a new instruction to
fix this.  We're going to have to make a pretty good case that this is
either impossible or really hard to do in software.

Surely we have the locking to tell another thread that we want its PKRU
value to change without actively going out and having the kernel poke a
new value in.