[RFC PATCH v2 2/4] arm64: Add support for asymmetric AArch32 EL0 configurations

[Will Deacon]

On Wed, Oct 21, 2020 at 05:52:47PM +0100, Catalin Marinas wrote:
> On Wed, Oct 21, 2020 at 05:21:21PM +0100, Qais Yousef wrote:
> > On 10/21/20 16:39, Will Deacon wrote:
> > > On Wed, Oct 21, 2020 at 11:46:09AM +0100, Qais Yousef wrote:
> > > > When the CONFIG_ASYMMETRIC_AARCH32 option is enabled (EXPERT), the type
> > > > of the ARM64_HAS_32BIT_EL0 capability becomes WEAK_LOCAL_CPU_FEATURE.
> > > > The kernel will now return true for system_supports_32bit_el0() and
> > > > checks 32-bit tasks are affined to AArch32 capable CPUs only in
> > > > do_notify_resume(). If the affinity contains a non-capable AArch32 CPU,
> > > > the tasks will get SIGKILLed. If the last CPU supporting 32-bit is
> > > > offlined, the kernel will SIGKILL any scheduled 32-bit tasks (the
> > > > alternative is to prevent offlining through a new .cpu_disable feature
> > > > entry).
> > > > 
> > > > In addition to the relaxation of the ARM64_HAS_32BIT_EL0 capability,
> > > > this patch factors out the 32-bit cpuinfo and features setting into
> > > > separate functions: __cpuinfo_store_cpu_32bit(),
> > > > init_cpu_32bit_features(). The cpuinfo of the booting CPU
> > > > (boot_cpu_data) is now updated on the first 32-bit capable CPU even if
> > > > it is a secondary one. The ID_AA64PFR0_EL0_64BIT_ONLY feature is relaxed
> > > > to FTR_NONSTRICT and FTR_HIGHER_SAFE when the asymmetric AArch32 support
> > > > is enabled. The compat_elf_hwcaps are only verified for the
> > > > AArch32-capable CPUs to still allow hotplugging AArch64-only CPUs.
> > > > 
> > > > Make sure that KVM never sees the asymmetric 32bit system. Guest can
> > > > still ignore ID registers and force run 32bit at EL0.
> > > > 
> > > > Co-developed-by: Qais Yousef <qais.yousef at arm.com>
> > > > Signed-off-by: Catalin Marinas <catalin.marinas at arm.com>
> > > > Signed-off-by: Qais Yousef <qais.yousef at arm.com>
> > > 
> > > [...]
> > > 
> > > > diff --git a/arch/arm64/include/asm/thread_info.h b/arch/arm64/include/asm/thread_info.h
> > > > index 5e784e16ee89..312974ab2c85 100644
> > > > --- a/arch/arm64/include/asm/thread_info.h
> > > > +++ b/arch/arm64/include/asm/thread_info.h
> > > > @@ -67,6 +67,7 @@ void arch_release_task_struct(struct task_struct *tsk);
> > > >  #define TIF_FOREIGN_FPSTATE	3	/* CPU's FP state is not current's */
> > > >  #define TIF_UPROBE		4	/* uprobe breakpoint or singlestep */
> > > >  #define TIF_FSCHECK		5	/* Check FS is USER_DS on return */
> > > > +#define TIF_CHECK_32BIT_AFFINITY 6	/* Check thread affinity for asymmetric AArch32 */
> > > 
> > > I've looked through the patch and I still can't figure out why this extra
> > > flag is needed. We know if a CPU supports 32-bit EL0, and we know whether
> > > or not a task is 32-bit. So why the extra flag? Is it just a hangover from
> > > the old series?
> > 
> > It did evolve a bit organically.
> > 
> > AFAICS it helps as an optimization to avoid the checks unnecessarily. If it's
> > not expensive to do the checks in the loop in do_notify_resume() we can omit
> > it. We will still protect it with system_supports_asym_32bit_el0() so the check
> > is done on these systems only.
> 
> Ah, I think I remember now. We didn't want ret_to_user (entry.S) to
> always go the work_pending path if there was no context switch for a
> 32-bit task. With the AArch32 check in do_notify_resume(), it would mean
> we add _TIF_32BIT to the _TIF_WORK_MASK.
> 
> However, we could add an asm alternative if AArch32 asym is detected to
> always route TIF_32BIT tasks to work_pending.

Or could we just use TIF_NOTIFY_RESUME, like we for for rseq()?

Will