Re: [PATCH] arm64: Enable MRS emulation early

From: Suzuki K Poulose
Date: Wed Oct 04 2017 - 09:00:46 EST


On 04/10/17 12:32, Mark Rutland wrote:
On Wed, Oct 04, 2017 at 12:10:40PM +0100, Catalin Marinas wrote:
On Wed, Oct 04, 2017 at 11:14:26AM +0100, Mark Rutland wrote:
On Wed, Oct 04, 2017 at 10:48:05AM +0100, Suzuki K Poulose wrote:
Make sure the MRS emulation is enabled early enough, such that the
early userspace applications (e.g, those run from initrd) could
use the facility without crashing them.

Fixes: commit 77c97b4ee2129 ("arm64: cpufeature: Expose CPUID registers by emulation")
Reported-by: Matwey V. Kornilov <matwey.kornilov@xxxxxxxxx>
Cc: James Morse <james.morse@xxxxxxx>
Cc: Dave Martin <Dave.martin@xxxxxxx>
Cc: Catalin Marinas <catalin.marinas@xxxxxxx>
Cc: Will Deacon <will.deacon@xxxxxxx>
Cc: stable@xxxxxxxxxxxxxxx
Cc: Mark Rutland <mark.rutland@xxxxxxx>
Signed-off-by: Suzuki K Poulose <suzuki.poulose@xxxxxxx>

This looks sensible, but shouldn't we do the same for other
late_inicalls can affect initrd userspace?

e.g. armv8_deprecated_init, fpsimd_init, sys_reg_genericv8_init?

I think we should, though not all of them are concerned with the user
code. For example, fpsimd_init() takes care of the pm/hotplug aspect and
nothing to do with user space.

My worry was that without the pm/hotplug notifiers, things could go
wrong during the initrd. e.g. we could lose userspace fp state without
the pm notifier, or userspace could trigger hotplpug that we wouldn't
manage correctly

So even if it's not directly userspace related, it can affect (or can be
affected by) initrd userspace.

You're right. In fact, I had a version of the patch which enables the emulation
as soon as we have initialised the ELF_HWCAPs from setup_cpu_features(), rather
than depending on an initcall. But that requires moving the setup_cpu_features()
to the bottom, which makes the hunk look a bit more complex than it is.

And similarly, we should be able to do the fpsimd_init from setup_cpu_features(),
as we have finalised the HWCAPs and pm_register_notifier any adds the entry to
a static list of notifiers ( even though the cpu_pm callbacks are registered as
a core_initcall() ).

Similarly for sys_reg_genericv8_init & armv8_deprecated_init could be made a core
initcall.

I think it would be good to separate them out.

i.e, enable_mrs_emulation & fpsimd_init from setup_cpu_features()
and the other two promoted as core_initcalls.

Thoughts ?

Suzuki