Re: [PATCH 09/14] KVM: arm64/sve: Simplify KVM_REG_ARM64_SVE_VLS array sizing

[Dave Martin <Dave.Martin@arm.com>]

On Tue, Apr 16, 2019 at 03:00:54PM +0200, Andrew Jones wrote:
> On Tue, Apr 16, 2019 at 01:41:42PM +0100, Dave Martin wrote:
> > On Mon, Apr 15, 2019 at 05:20:37PM +0200, Andrew Jones wrote:
> > > On Fri, Apr 12, 2019 at 05:28:13PM +0100, Dave Martin wrote:
> > > > A complicated DIV_ROUND_UP() expression is currently written out
> > > > explicitly in multiple places in order to specify the size of the
> > > > bitmap exchanged with userspace to represent the value of the
> > > > KVM_REG_ARM64_SVE_VLS pseudo-register.
> > > > 
> > > > To make this more readable, this patch replaces these with a single
> > > > define.
> > > > 
> > > > Since the number of words in a bitmap is just the index of the last
> > > > word used + 1, this patch expresses the bound that way instead.
> > > > This should make it clearer what is being expressed.
> > > > 
> > > > Since use of DIV_ROUND_UP() was the only reason for including
> > > > <linux/kernel.h> in guest.c, this patch removes that #include.
> > > > 
> > > > No functional change.
> > > > 
> > > > Suggested-by: Andrew Jones <drjones@redhat.com>
> > > > Signed-off-by: Dave Martin <Dave.Martin@arm.com>
> > > > ---
> > > >  arch/arm64/kvm/guest.c | 9 +++++----
> > > >  1 file changed, 5 insertions(+), 4 deletions(-)
> > > > 
> > > > diff --git a/arch/arm64/kvm/guest.c b/arch/arm64/kvm/guest.c
> > > > index 73044e3..f025a2f 100644
> > > > --- a/arch/arm64/kvm/guest.c
> > > > +++ b/arch/arm64/kvm/guest.c
> > > > @@ -23,7 +23,6 @@
> > > >  #include <linux/errno.h>
> > > >  #include <linux/err.h>
> > > >  #include <linux/nospec.h>
> > > > -#include <linux/kernel.h>
> > > >  #include <linux/kvm_host.h>
> > > >  #include <linux/module.h>
> > > >  #include <linux/stddef.h>
> > > > @@ -209,8 +208,10 @@ static int set_core_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
> > > >  #define vq_word(vq) (((vq) - SVE_VQ_MIN) / 64)
> > > >  #define vq_mask(vq) ((u64)1 << ((vq) - SVE_VQ_MIN) % 64)
> > > >  
> > > > +#define SVE_VLS_WORDS (vq_word(SVE_VQ_MAX) + 1)
> > > > +
> > > >  static bool vq_present(
> > > > -	const u64 (*const vqs)[DIV_ROUND_UP(SVE_VQ_MAX - SVE_VQ_MIN + 1, 64)],
> > > > +	const u64 (*const vqs)[SVE_VLS_WORDS],
> > > >  	unsigned int vq)
> > > >  {
> > > >  	return (*vqs)[vq_word(vq)] & vq_mask(vq);
> > > > @@ -219,7 +220,7 @@ static bool vq_present(
> > > >  static int get_sve_vls(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
> > > >  {
> > > >  	unsigned int max_vq, vq;
> > > > -	u64 vqs[DIV_ROUND_UP(SVE_VQ_MAX - SVE_VQ_MIN + 1, 64)];
> > > > +	u64 vqs[SVE_VLS_WORDS];
> > > >  
> > > >  	if (!vcpu_has_sve(vcpu))
> > > >  		return -ENOENT;
> > > > @@ -243,7 +244,7 @@ static int get_sve_vls(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
> > > >  static int set_sve_vls(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
> > > >  {
> > > >  	unsigned int max_vq, vq;
> > > > -	u64 vqs[DIV_ROUND_UP(SVE_VQ_MAX - SVE_VQ_MIN + 1, 64)];
> > > > +	u64 vqs[SVE_VLS_WORDS];
> > > >  
> > > >  	if (!vcpu_has_sve(vcpu))
> > > >  		return -ENOENT;
> > > > -- 
> > > > 2.1.4
> > > >
> > > 
> > > This is good, but I wonder if we could define the number of VLS words in
> > > the documentation in terms of SVE_VQ_MAX too. Currently it's just the
> > > hard coded 8 ("__u64 vector_lengths[8]").
> > > 
> > > Reviewed-by: Andrew Jones <drjones@redhat.com>
> > 
> > I see your point, but SVE_VQ_MAX isn't really part of the KVM API, so I
> > was avoiding it here.
> 
> It's not part of the KVM API, but it is uapi (asm/sigcontext.h). I'd
> prefer we use it than to encourage KVM userspace to scatter 8's around.
> 
> > 
> > [8] is at least impossible to misinterpret, even if it's not the most
> > self-explanatory option.
> 
> It's impossible to misinterpret, but also begs the questions of where
> it comes from and if it will always be that way forever and ever.

It is sized to the maximum that makes sense without a major API
redesign, so the expectation is that it will never change (and userspace
should assume that it does not).

I sympathise though, since this case is less intuitive than most.

We could add some dedicated defines for this:

arch/arm64/include/uapi/asm/kvm.h:

#include <asm/sve_context.h> /* which we already have anyway */

#define KVM_ARM64_SVE_VQ_MAX __SVE_VQ_MAX
#define KVM_ARM64_SVE_VQ_MIN __SVE_VQ_MIN

/* Vector lengths pseudo-register: */
#define KVM_REG_ARM64_SVE_VLS		(KVM_REG_ARM64 | KVM_REG_ARM64_SVE | \
					 KVM_REG_SIZE_U512 | 0xffff)
#define KVM_ARM64_SVE_VLS_WORDS	\
	((KVM_ARM64_SVE_VQ_MAX - KVM_ARM64_SVE_VQ_MIN) / 64 + 1)


Then document as follows:
Documentation/virtual/kvm/api.txt:

__u64 vector_lengths[KVM_ARM64_SVE_VLS_WORDS];

if (vq >= KVM_ARM64_SVE_VQ_MIN && vq <= KVM_ARM64_SVE_VQ_MAX)
    (vector_lengths[(vq - KVM_ARM64_SVE_VQ_MIN) / 64] >>
		((vq - KVM_ARM64_SVE_VQ_MIN) % 64)) & 1)
	/* Vector length vq * 16 bytes supported */
else
	/* Vector length vq * 16 bytes not supported */


Does that work for you?

Doing things this way avoids people having to include <asm/sigcontext.h>
(and the consequent clashes with glibc headers).

Cheers
---Dave