This is the mail archive of the
mailing list for the glibc project.
Re: [PATCH 11/27] [AARCH64] Syscalls for ILP32 are passed always via 64bit values.
- From: Yury Norov <ynorov at caviumnetworks dot com>
- To: Catalin Marinas <catalin dot marinas at arm dot com>
- Cc: <libc-alpha at sourceware dot org>, <linux-kernel at vger dot kernel dot org>, <arnd at arndb dot de>, <marcus dot shawcroft at arm dot com>, <philb at gnu dot org>, <davem at davemloft dot net>, <szabolcs dot nagy at arm dot com>, <maxim dot kuvyrkov at linaro dot org>, <joseph at codesourcery dot com>, <pinskia at gmail dot com>, Andrew Pinski <apinski at cavium dot com>
- Date: Thu, 23 Jun 2016 09:32:32 +0300
- Subject: Re: [PATCH 11/27] [AARCH64] Syscalls for ILP32 are passed always via 64bit values.
- Authentication-results: sourceware.org; auth=none
- Authentication-results: spf=none (sender IP is ) smtp dot mailfrom=Yuri dot Norov at caviumnetworks dot com;
- References: <1466485631-3532-1-git-send-email-ynorov at caviumnetworks dot com> <1466485631-3532-13-git-send-email-ynorov at caviumnetworks dot com> <20160622154951 dot GD6521 at e104818-lin dot cambridge dot arm dot com>
- Spamdiagnosticmetadata: NSPM
- Spamdiagnosticoutput: 1:99
On Wed, Jun 22, 2016 at 04:49:52PM +0100, Catalin Marinas wrote:
> On Tue, Jun 21, 2016 at 08:06:55AM +0300, Yury Norov wrote:
> > This patch adds support for ILP32 syscalls, sign and zero extending
> > where needed. Unlike LP64, pointers are 32bit and need to be zero
> > extended rather than the standard sign extend that the code would do.
> > We take advatage of ssize_t being long rather than int for ILP32,
> > to get this correct.
> Unless I misunderstand what this patch does, I thought we agreed that
> 32-bit arguments are passed as 32-bit values and it is the kernel's
> responsibility to zero or sign-extend the upper half (IOW, assume
> garbage just like the PCS ABI).
> We are still debating whether 64-bit values are passed in one or two
> registers but this doesn't change the situation of 32-bit values like
I dropped this patch with no regressions. It might be needed it we
choose s390-like wrappers, but this series is not about it anyway.
BTW, what about 64-bit values passing? Any new comments?