This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
[RFC] Expand available hwcap bits with AT_HWCAP2 in auxv
- From: Ryan Arnold <rsa at us dot ibm dot com>
- To: "linux-kernel at vger dot kernel dot org" <linux-kernel at vger dot kernel dot org>
- Cc: "libc-alpha at sourceware dot org" <libc-alpha at sourceware dot org>, BenjaminHerrenschmidt <benh at kernel dot crashing dot org>
- Date: Fri, 16 Nov 2012 16:53:54 -0600
- Subject: [RFC] Expand available hwcap bits with AT_HWCAP2 in auxv
- Reply-to: rsa at us dot ibm dot com
Hi,
We're out of bits in the AT_HWCAP entry for PowerPC. Per the following
discussion on libc-alpha, I'd like to propose the expansion of the auxv
to make additional space available for further system feature flags.
http://www.sourceware.org/ml/libc-alpha/2012-10/msg00438.html
There are a few ways to do this (as pointed out by Ben Herrenschmidt).
glibc would prefer to parse an AT_HWCAP2 entry from the auxv and compose
those bits into the high 32-bits of the existing glibc uint64_t hwcap
field.
Currently, the high 32-bits of this field are only available in 64-bit
userspace, and only for platforms where the kernel designates the auxv
hwcap as unsigned long int (example: s390). Platforms like PowerPC, arm,
and x86 maintain parity between 64-bit and 32-bit userspace and thus
only allow 32-bits for capabilities.
I propose that the kernel shift capabilities designated in the high
32-bits into an AT_HWCAP2 auxv entry. In userspace, glibc will compose
that into the available high 32-bits of glibc's hwcap field making the
rest of it available to all platforms.
The beauty of this approach is that if either the Kernel or glibc lack
the AT_HWCAP2 improvement the high 32-bits of glibc's hwcap field will
remain empty, i.e., feature not available.
With this in mind, I need to reserve an identifier in
include/linux/auxvec.h for AT_HWCAP2. I was thinking:
#define AT_HWCAP 26
GLIBC will have parity with this identifier in elf/elf.h.
I think Ben H. had planned to work on the kernel side of this and I have
a GLIBC patch in process.
Comments?
Ryan S. Arnold
IBM Linux Technology Center
glibc PowerPC maintainer