[v2,10/28] arm64/sve: Low-level CPU setup

Message ID 1504198860-12951-11-git-send-email-Dave.Martin@arm.com
State New
Headers show
Series
  • ARM Scalable Vector Extension (SVE)
Related show

Commit Message

Dave Martin Aug. 31, 2017, 5 p.m.
To enable the kernel to use SVE, all SVE traps from EL1 must be
disabled.  To take maximum advantage of the hardware, the full
available vector length also needs to be enabled for EL1 by
programming ZCR_EL2.LEN.  (The kernel will program ZCR_EL1.LEN as
required, but this cannot override the limit set by ZCR_EL2.)

In advance of full SVE support being implemented for userspace, it
also necessary to ensure that SVE traps from EL0 are enabled.

This patch makes the appropriate changes to the primary and
secondary CPU initialisation code.

Signed-off-by: Dave Martin <Dave.Martin@arm.com>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
---
 arch/arm64/kernel/head.S | 13 ++++++++++++-
 arch/arm64/mm/proc.S     | 14 ++++++++++++--
 2 files changed, 24 insertions(+), 3 deletions(-)

Comments

Catalin Marinas Sept. 13, 2017, 1:32 p.m. | #1
On Thu, Aug 31, 2017 at 06:00:42PM +0100, Dave P Martin wrote:
> diff --git a/arch/arm64/mm/proc.S b/arch/arm64/mm/proc.S
> index 877d42f..dd22ef2 100644
> --- a/arch/arm64/mm/proc.S
> +++ b/arch/arm64/mm/proc.S
> @@ -27,6 +27,7 @@
>  #include <asm/pgtable-hwdef.h>
>  #include <asm/cpufeature.h>
>  #include <asm/alternative.h>
> +#include <asm/sysreg.h>
>  
>  #ifdef CONFIG_ARM64_64K_PAGES
>  #define TCR_TG_FLAGS	TCR_TG0_64K | TCR_TG1_64K
> @@ -186,8 +187,17 @@ ENTRY(__cpu_setup)
>  	tlbi	vmalle1				// Invalidate local TLB
>  	dsb	nsh
>  
> -	mov	x0, #3 << 20
> -	msr	cpacr_el1, x0			// Enable FP/ASIMD
> +	mov	x0, #3 << 20			// FEN
> +
> +	/* SVE */
> +	mrs	x5, id_aa64pfr0_el1
> +	ubfx	x5, x5, #ID_AA64PFR0_SVE_SHIFT, #4
> +	cbz	x5, 1f
> +
> +	bic	x0, x0, #CPACR_EL1_ZEN
> +	orr	x0, x0, #CPACR_EL1_ZEN_EL1EN	// SVE: trap for EL0, not EL1
> +1:	msr	cpacr_el1, x0			// Enable FP/ASIMD

For EL1, I wonder whether we could move this later to cpufeature.c. IIRC
I tried to do the same with FPSIMD but hit an issue with EFI run-time
services (I may be wrong though).
Dave Martin Sept. 13, 2017, 7:21 p.m. | #2
On Wed, Sep 13, 2017 at 06:32:06AM -0700, Catalin Marinas wrote:
> On Thu, Aug 31, 2017 at 06:00:42PM +0100, Dave P Martin wrote:
> > diff --git a/arch/arm64/mm/proc.S b/arch/arm64/mm/proc.S
> > index 877d42f..dd22ef2 100644
> > --- a/arch/arm64/mm/proc.S
> > +++ b/arch/arm64/mm/proc.S
> > @@ -27,6 +27,7 @@
> >  #include <asm/pgtable-hwdef.h>
> >  #include <asm/cpufeature.h>
> >  #include <asm/alternative.h>
> > +#include <asm/sysreg.h>
> >  
> >  #ifdef CONFIG_ARM64_64K_PAGES
> >  #define TCR_TG_FLAGS	TCR_TG0_64K | TCR_TG1_64K
> > @@ -186,8 +187,17 @@ ENTRY(__cpu_setup)
> >  	tlbi	vmalle1				// Invalidate local TLB
> >  	dsb	nsh
> >  
> > -	mov	x0, #3 << 20
> > -	msr	cpacr_el1, x0			// Enable FP/ASIMD
> > +	mov	x0, #3 << 20			// FEN
> > +
> > +	/* SVE */
> > +	mrs	x5, id_aa64pfr0_el1
> > +	ubfx	x5, x5, #ID_AA64PFR0_SVE_SHIFT, #4
> > +	cbz	x5, 1f
> > +
> > +	bic	x0, x0, #CPACR_EL1_ZEN
> > +	orr	x0, x0, #CPACR_EL1_ZEN_EL1EN	// SVE: trap for EL0, not EL1
> > +1:	msr	cpacr_el1, x0			// Enable FP/ASIMD
> 
> For EL1, I wonder whether we could move this later to cpufeature.c. IIRC
> I tried to do the same with FPSIMD but hit an issue with EFI run-time
> services (I may be wrong though).

I'll take a look at this -- I believe it should be safe to disable this
trap for EL1 relatively late.  This is needed before probing for
available vector lengths, but apart from that the kernel shouldn't touch
SVE until/unless some user task uses SVE.

This would change if we eventually enable kernel-mode SVE, but I wouldn't
expect that to get used in early boot before the cpufeatures code runs.

Ard may have a view on this.

Cheers
---Dave
Dave Martin Oct. 5, 2017, 10:47 a.m. | #3
On Wed, Sep 13, 2017 at 08:21:11PM +0100, Dave Martin wrote:
> On Wed, Sep 13, 2017 at 06:32:06AM -0700, Catalin Marinas wrote:
> > On Thu, Aug 31, 2017 at 06:00:42PM +0100, Dave P Martin wrote:
> > > diff --git a/arch/arm64/mm/proc.S b/arch/arm64/mm/proc.S
> > > index 877d42f..dd22ef2 100644
> > > --- a/arch/arm64/mm/proc.S
> > > +++ b/arch/arm64/mm/proc.S
> > > @@ -27,6 +27,7 @@
> > >  #include <asm/pgtable-hwdef.h>
> > >  #include <asm/cpufeature.h>
> > >  #include <asm/alternative.h>
> > > +#include <asm/sysreg.h>
> > >  
> > >  #ifdef CONFIG_ARM64_64K_PAGES
> > >  #define TCR_TG_FLAGS	TCR_TG0_64K | TCR_TG1_64K
> > > @@ -186,8 +187,17 @@ ENTRY(__cpu_setup)
> > >  	tlbi	vmalle1				// Invalidate local TLB
> > >  	dsb	nsh
> > >  
> > > -	mov	x0, #3 << 20
> > > -	msr	cpacr_el1, x0			// Enable FP/ASIMD
> > > +	mov	x0, #3 << 20			// FEN
> > > +
> > > +	/* SVE */
> > > +	mrs	x5, id_aa64pfr0_el1
> > > +	ubfx	x5, x5, #ID_AA64PFR0_SVE_SHIFT, #4
> > > +	cbz	x5, 1f
> > > +
> > > +	bic	x0, x0, #CPACR_EL1_ZEN
> > > +	orr	x0, x0, #CPACR_EL1_ZEN_EL1EN	// SVE: trap for EL0, not EL1
> > > +1:	msr	cpacr_el1, x0			// Enable FP/ASIMD
> > 
> > For EL1, I wonder whether we could move this later to cpufeature.c. IIRC
> > I tried to do the same with FPSIMD but hit an issue with EFI run-time
> > services (I may be wrong though).
> 
> I'll take a look at this -- I believe it should be safe to disable this
> trap for EL1 relatively late.  This is needed before probing for
> available vector lengths, but apart from that the kernel shouldn't touch
> SVE until/unless some user task uses SVE.
> 
> This would change if we eventually enable kernel-mode SVE, but I wouldn't
> expect that to get used in early boot before the cpufeatures code runs.
> 
> Ard may have a view on this.

I've had a go at this, but there's a lot of splatter.

I can add a helper el1_enable_sve() say, and call it before probing
ZCR_EL1 in the cpufeatures code.

This makes enabling SVE a side-effect of the cpufeatures code, which
I'm not that comfortable with -- it feels like something that later
refactoring could easily break.

I can also add an explicit call to el1_enable_sve() in sve_setup(),
but this only works on the boot cpu.

For secondaries, I could add something in secondary_start_kernel(),
but this looks incongruous since there's no other call to do something
similar yet.


** Suzuki, do we have any other case where the trap for a CPU feature is
turned off by the cpufeatures code?  If there's already a template for
this then I'm happy to follow it.

Otherwise, maybe it's simpler to keep it in __cpu_setup after all
since that's a common path that all CPUs pass through.

Cheers
---Dave
Suzuki K Poulose Oct. 5, 2017, 11:04 a.m. | #4
On 05/10/17 11:47, Dave Martin wrote:
> On Wed, Sep 13, 2017 at 08:21:11PM +0100, Dave Martin wrote:
>> On Wed, Sep 13, 2017 at 06:32:06AM -0700, Catalin Marinas wrote:
>>> On Thu, Aug 31, 2017 at 06:00:42PM +0100, Dave P Martin wrote:
>>>> diff --git a/arch/arm64/mm/proc.S b/arch/arm64/mm/proc.S
>>>> index 877d42f..dd22ef2 100644
>>>> --- a/arch/arm64/mm/proc.S
>>>> +++ b/arch/arm64/mm/proc.S
>>>> @@ -27,6 +27,7 @@
>>>>   #include <asm/pgtable-hwdef.h>
>>>>   #include <asm/cpufeature.h>
>>>>   #include <asm/alternative.h>
>>>> +#include <asm/sysreg.h>
>>>>   
>>>>   #ifdef CONFIG_ARM64_64K_PAGES
>>>>   #define TCR_TG_FLAGS	TCR_TG0_64K | TCR_TG1_64K
>>>> @@ -186,8 +187,17 @@ ENTRY(__cpu_setup)
>>>>   	tlbi	vmalle1				// Invalidate local TLB
>>>>   	dsb	nsh
>>>>   
>>>> -	mov	x0, #3 << 20
>>>> -	msr	cpacr_el1, x0			// Enable FP/ASIMD
>>>> +	mov	x0, #3 << 20			// FEN
>>>> +
>>>> +	/* SVE */
>>>> +	mrs	x5, id_aa64pfr0_el1
>>>> +	ubfx	x5, x5, #ID_AA64PFR0_SVE_SHIFT, #4
>>>> +	cbz	x5, 1f
>>>> +
>>>> +	bic	x0, x0, #CPACR_EL1_ZEN
>>>> +	orr	x0, x0, #CPACR_EL1_ZEN_EL1EN	// SVE: trap for EL0, not EL1
>>>> +1:	msr	cpacr_el1, x0			// Enable FP/ASIMD
>>>
>>> For EL1, I wonder whether we could move this later to cpufeature.c. IIRC
>>> I tried to do the same with FPSIMD but hit an issue with EFI run-time
>>> services (I may be wrong though).
>>
>> I'll take a look at this -- I believe it should be safe to disable this
>> trap for EL1 relatively late.  This is needed before probing for
>> available vector lengths, but apart from that the kernel shouldn't touch
>> SVE until/unless some user task uses SVE.
>>
>> This would change if we eventually enable kernel-mode SVE, but I wouldn't
>> expect that to get used in early boot before the cpufeatures code runs.
>>
>> Ard may have a view on this.
> 
> I've had a go at this, but there's a lot of splatter.
> 
> I can add a helper el1_enable_sve() say, and call it before probing
> ZCR_EL1 in the cpufeatures code.
> 
> This makes enabling SVE a side-effect of the cpufeatures code, which
> I'm not that comfortable with -- it feels like something that later
> refactoring could easily break.
> 
> I can also add an explicit call to el1_enable_sve() in sve_setup(),
> but this only works on the boot cpu.
> 
> For secondaries, I could add something in secondary_start_kernel(),
> but this looks incongruous since there's no other call to do something
> similar yet.
> 
> 
> ** Suzuki, do we have any other case where the trap for a CPU feature is
> turned off by the cpufeatures code?  If there's already a template for
> this then I'm happy to follow it.

The closest thing you have is an "enable" callback for each "available"
capability, which gets invoked on all CPUs (boot time active CPUs and the
ones which are brought up later). This is used by things like, PAN to
do some CPU specific setups.

See  :
    enable_cpu_capabilities()  // For all boot time active CPUs
    and
    verify_local_cpu_features()  // For CPUs brought up later

Cheers
Suzuki
Dave Martin Oct. 5, 2017, 11:22 a.m. | #5
On Thu, Oct 05, 2017 at 12:04:12PM +0100, Suzuki K Poulose wrote:
> On 05/10/17 11:47, Dave Martin wrote:

[...]

> >>>On Thu, Aug 31, 2017 at 06:00:42PM +0100, Dave P Martin wrote:
> >>>>diff --git a/arch/arm64/mm/proc.S b/arch/arm64/mm/proc.S
> >>>>index 877d42f..dd22ef2 100644
> >>>>--- a/arch/arm64/mm/proc.S
> >>>>+++ b/arch/arm64/mm/proc.S
> >>>>@@ -27,6 +27,7 @@
> >>>>  #include <asm/pgtable-hwdef.h>
> >>>>  #include <asm/cpufeature.h>
> >>>>  #include <asm/alternative.h>
> >>>>+#include <asm/sysreg.h>
> >>>>  #ifdef CONFIG_ARM64_64K_PAGES
> >>>>  #define TCR_TG_FLAGS	TCR_TG0_64K | TCR_TG1_64K
> >>>>@@ -186,8 +187,17 @@ ENTRY(__cpu_setup)
> >>>>  	tlbi	vmalle1				// Invalidate local TLB
> >>>>  	dsb	nsh
> >>>>-	mov	x0, #3 << 20
> >>>>-	msr	cpacr_el1, x0			// Enable FP/ASIMD
> >>>>+	mov	x0, #3 << 20			// FEN
> >>>>+
> >>>>+	/* SVE */
> >>>>+	mrs	x5, id_aa64pfr0_el1
> >>>>+	ubfx	x5, x5, #ID_AA64PFR0_SVE_SHIFT, #4
> >>>>+	cbz	x5, 1f
> >>>>+
> >>>>+	bic	x0, x0, #CPACR_EL1_ZEN
> >>>>+	orr	x0, x0, #CPACR_EL1_ZEN_EL1EN	// SVE: trap for EL0, not EL1
> >>>>+1:	msr	cpacr_el1, x0			// Enable FP/ASIMD

[..]

> >I can add a helper el1_enable_sve() say, and call it before probing
> >ZCR_EL1 in the cpufeatures code.
> >
> >This makes enabling SVE a side-effect of the cpufeatures code, which
> >I'm not that comfortable with -- it feels like something that later
> >refactoring could easily break.
> >
> >I can also add an explicit call to el1_enable_sve() in sve_setup(),
> >but this only works on the boot cpu.
> >
> >For secondaries, I could add something in secondary_start_kernel(),
> >but this looks incongruous since there's no other call to do something
> >similar yet.
> >
> >
> >** Suzuki, do we have any other case where the trap for a CPU feature is
> >turned off by the cpufeatures code?  If there's already a template for
> >this then I'm happy to follow it.
> 
> The closest thing you have is an "enable" callback for each "available"
> capability, which gets invoked on all CPUs (boot time active CPUs and the
> ones which are brought up later). This is used by things like, PAN to
> do some CPU specific setups.
> 
> See  :
>    enable_cpu_capabilities()  // For all boot time active CPUs
>    and
>    verify_local_cpu_features()  // For CPUs brought up later

That may allow me to do something tidier, provided the enable method is
called early enough.

I'll take a look.

Cheers
---Dave

Patch

diff --git a/arch/arm64/kernel/head.S b/arch/arm64/kernel/head.S
index 7434ec0..f411f71 100644
--- a/arch/arm64/kernel/head.S
+++ b/arch/arm64/kernel/head.S
@@ -516,8 +516,19 @@  CPU_LE(	movk	x0, #0x30d0, lsl #16	)	// Clear EE and E0E on LE systems
 	mov	x0, #0x33ff
 	msr	cptr_el2, x0			// Disable copro. traps to EL2
 
+	/* SVE register access */
+	mrs	x1, id_aa64pfr0_el1
+	ubfx	x1, x1, #ID_AA64PFR0_SVE_SHIFT, #4
+	cbz	x1, 7f
+
+	bic	x0, x0, #CPTR_EL2_TZ		// Also disable SVE traps
+	msr	cptr_el2, x0			// Disable copro. traps to EL2
+	isb
+	mov	x1, #ZCR_ELx_LEN_MASK		// SVE: Enable full vector
+	msr_s	SYS_ZCR_EL2, x1			// length for EL1.
+
 	/* Hypervisor stub */
-	adr_l	x0, __hyp_stub_vectors
+7:	adr_l	x0, __hyp_stub_vectors
 	msr	vbar_el2, x0
 
 	/* spsr */
diff --git a/arch/arm64/mm/proc.S b/arch/arm64/mm/proc.S
index 877d42f..dd22ef2 100644
--- a/arch/arm64/mm/proc.S
+++ b/arch/arm64/mm/proc.S
@@ -27,6 +27,7 @@ 
 #include <asm/pgtable-hwdef.h>
 #include <asm/cpufeature.h>
 #include <asm/alternative.h>
+#include <asm/sysreg.h>
 
 #ifdef CONFIG_ARM64_64K_PAGES
 #define TCR_TG_FLAGS	TCR_TG0_64K | TCR_TG1_64K
@@ -186,8 +187,17 @@  ENTRY(__cpu_setup)
 	tlbi	vmalle1				// Invalidate local TLB
 	dsb	nsh
 
-	mov	x0, #3 << 20
-	msr	cpacr_el1, x0			// Enable FP/ASIMD
+	mov	x0, #3 << 20			// FEN
+
+	/* SVE */
+	mrs	x5, id_aa64pfr0_el1
+	ubfx	x5, x5, #ID_AA64PFR0_SVE_SHIFT, #4
+	cbz	x5, 1f
+
+	bic	x0, x0, #CPACR_EL1_ZEN
+	orr	x0, x0, #CPACR_EL1_ZEN_EL1EN	// SVE: trap for EL0, not EL1
+1:	msr	cpacr_el1, x0			// Enable FP/ASIMD
+
 	mov	x0, #1 << 12			// Reset mdscr_el1 and disable
 	msr	mdscr_el1, x0			// access to the DCC from EL0
 	isb					// Unmask debug exceptions now,