Re: [PATCH 05/19] arm64: rename COMPAT to AARCH32_EL0 in Kconfig

[Catalin Marinas <catalin.marinas@arm.com>]

On Thu, Aug 11, 2016 at 05:16:45PM +0200, Arnd Bergmann wrote:
> On Thursday, August 11, 2016 3:50:00 PM CEST Catalin Marinas wrote:
> > On Thu, Aug 11, 2016 at 10:53:01AM +0200, Arnd Bergmann wrote:
> > > On Thursday, August 11, 2016 3:35:01 PM CEST Zhangjian (Bamvor) wrote:
> > > > On 2016/6/18 7:54, Yury Norov wrote:
> > > > > From: Andrew Pinski <apinski@cavium.com>
> > > > >
> > > > > In this patchset  ILP32 ABI support is added. Additionally to AARCH32,
> > > > > which is binary-compatible with ARM, ILP32 is (mostly) ABI-compatible.
> > > > >
> > > > >  From now, AARCH32_EL0 (former COMPAT) config option means the support of
> > > > > AARCH32 userspace, ARM64_ILP32 - support of ILP32 ABI (see next patches),
> > > > > and COMPAT indicates that one of them, or both, is enabled.
> > > > >
> > > > > Where needed, CONFIG_COMPAT is changed over to use CONFIG_AARCH32_EL0 instead
> > > > >
> > > > > Reviewed-by: David Daney <ddaney@caviumnetworks.com>
> > > > > Signed-off-by: Andrew Pinski <Andrew.Pinski@caviumnetworks.com>
> > > > > Signed-off-by: Philipp Tomsich <philipp.tomsich@theobroma-systems.com>
> > > > > Signed-off-by: Christoph Muellner <christoph.muellner@theobroma-systems.com>
> > > > > Signed-off-by: Bamvor Jian Zhang <bamvor.zhangjian@linaro.org>
> > > > > Signed-off-by: Yury Norov <ynorov@caviumnetworks.com>
> > > > ...
> > > > > diff --git a/arch/arm64/kernel/cpuinfo.c b/arch/arm64/kernel/cpuinfo.c
> > > > > index c173d32..af200a8 100644
> > > > > --- a/arch/arm64/kernel/cpuinfo.c
> > > > > +++ b/arch/arm64/kernel/cpuinfo.c
> > > > > @@ -134,15 +134,17 @@ static int c_show(struct seq_file *m, void *v)
> > > > >   		 */
> > > > >   		seq_puts(m, "Features\t:");
> > > > >   		if (compat) {
> > > > > -#ifdef CONFIG_COMPAT
> > > > > -			for (j = 0; compat_hwcap_str[j]; j++)
> > > > > -				if (compat_elf_hwcap & (1 << j))
> > > > > -					seq_printf(m, " %s", compat_hwcap_str[j]);
> > > > > -
> > > > > -			for (j = 0; compat_hwcap2_str[j]; j++)
> > > > > -				if (compat_elf_hwcap2 & (1 << j))
> > > > > -					seq_printf(m, " %s", compat_hwcap2_str[j]);
> > > > > -#endif /* CONFIG_COMPAT */
> > > > > +#ifdef CONFIG_AARCH32_EL0
> > > > I saw that compat_hwcap_str and compat_hwcap2_str is defined when
> > > > "CONFIG_COMPAT" is true. Why we only change it to CONFIG_AARCH32_EL0
> > > > in c show()?
> > > > > +			if (personality(current->personality) == PER_LINUX32) {
> > > > And "compat" is "personality(current->personality) == PER_LINUX32;",
> > > > it seems that there is no need to add this twice.
> > > 
> > > I think it would be best to remove the #ifdef here completely,
> > > the PER_LINUX32 concept is not strictly tied to the emulation
> > > of ARM binaries, it literally just changes the output of
> > > /proc/cpuinfo and 'uname',
> > 
> > It's not strictly related to ARM binaries, however it is related to
> > AArch32 CPU features being supported and detected by the kernel.
> > Currently, with CONFIG_COMPAT disabled, we won't have access to a
> > (meaningful) compat_elf_hwcap.
> 
> Ah, makes sense. In that case, using CONFIG_AARCH32_EL0 sounds like
> the right thing to do here, though I guess we can just drop the
> "if (compat)" check, as we specifically want to print the supported
> features of the CPU, and they are still present even if a
> process with PER_LINUX reads them.

Do you mean always printing both compat and native hwcaps in
/proc/cpuinfo? We discussed this in the past and it's not something we
can easily fix at this stage without breaking the ABI. If we noticed
this before, we could have used distinct feature strings for AArch32 and
AArch64 (e.g. crypto stuff like aes32 and aes64 but we only have aes for
both).

> > > and you can have ARM binaries with
> > > PER_LINUX (using the arm64 uname) just like you can have
> > > arm64 binaries running with PER_LINUX32.
> > 
> > I was actually looking to enforce the 32-bit binaries to only see
> > PER_LINUX32, though with a risk of breaking the ABI. OTOH, people are
> > abusing this and write 32-bit apps relying on the 64-bit /proc/cpuinfo:
> > 
> > http://lkml.kernel.org/g/1464706504-25224-3-git-send-email-catalin.marinas@arm.com
> > 
> > (you were summoned on that discussion couple of times ;))
> 
> Hmm, I thought I saw the thread and didn't have any good idea for
> the uname information, but didn't notice it was for /proc/cpuinfo.
> 
> What's wrong with always showing both the 32-bit and the 64-bit
> hwcap strings here (minus the duplicates, which hopefully have
> the same meaning here)?

As I said above, some of them have the same name (which may be a good
thing at a first look) but we don't have an architecture guarantee that
the feature is present in both AArch32 and AArch64 modes (e.g. AES may
only be available in AArch64).

-- 
Catalin