On 17.07.2018 14:44, Janosch Frank wrote:
> Let's allow huge pmd linking when enabled through the
> KVM_CAP_S390_HPAGE_1M capability. Also we can now restrict gmap
> invalidation and notification to the cases where the capability has
> been activated and save some cycles when that's not the case.
>
> Signed-off-by: Janosch Frank <frankja@xxxxxxxxxxxxx>
> ---
> arch/s390/include/asm/mmu.h | 2 ++
> arch/s390/include/asm/mmu_context.h | 1 +
> arch/s390/mm/gmap.c | 9 ++++++---
> arch/s390/mm/pgtable.c | 12 ++++++------
> 4 files changed, 15 insertions(+), 9 deletions(-)
>
> diff --git a/arch/s390/include/asm/mmu.h b/arch/s390/include/asm/mmu.h
> index f5ff9dbad8ac..fad2ec2c1fd1 100644
> --- a/arch/s390/include/asm/mmu.h
> +++ b/arch/s390/include/asm/mmu.h
> @@ -24,6 +24,8 @@ typedef struct {
> unsigned int uses_skeys:1;
> /* The mmu context uses CMM. */
> unsigned int uses_cmm:1;
> + /* The gmap associated with this context uses huge pages. */
... are allowed to use ...
> + unsigned int allow_gmap_hpage_1m:1;
> } mm_context_t;
>
> #define INIT_MM_CONTEXT(name) \
> diff --git a/arch/s390/include/asm/mmu_context.h b/arch/s390/include/asm/mmu_context.h
> index d16bc79c30bb..0717ee76885d 100644
> --- a/arch/s390/include/asm/mmu_context.h
> +++ b/arch/s390/include/asm/mmu_context.h
> @@ -32,6 +32,7 @@ static inline int init_new_context(struct task_struct *tsk,
> mm->context.has_pgste = 0;
> mm->context.uses_skeys = 0;
> mm->context.uses_cmm = 0;
> + mm->context.allow_gmap_hpage_1m = 0;
> #endif
> switch (mm->context.asce_limit) {
> case _REGION2_SIZE:
> diff --git a/arch/s390/mm/gmap.c b/arch/s390/mm/gmap.c
> index a5a60afae8df..16878dacaeab 100644
> --- a/arch/s390/mm/gmap.c
> +++ b/arch/s390/mm/gmap.c
> @@ -2,8 +2,10 @@
> /*
> * KVM guest address space mapping code
> *
> - * Copyright IBM Corp. 2007, 2016
> + * Copyright IBM Corp. 2007, 2016, 2018
> * Author(s): Martin Schwidefsky <schwidefsky@xxxxxxxxxx>
> + * David Hildenbrand <david@xxxxxxxxxx>
> + * Janosch Frank <frankja@xxxxxxxxxxxxxxxxxx>
> */
>
> #include <linux/kernel.h>
> @@ -589,8 +591,8 @@ int __gmap_link(struct gmap *gmap, unsigned long gaddr, unsigned long vmaddr)
> return -EFAULT;
> pmd = pmd_offset(pud, vmaddr);
> VM_BUG_ON(pmd_none(*pmd));
> - /* large pmds cannot yet be handled */
> - if (pmd_large(*pmd))
> + /* Are we allowed to use huge pages? */
> + if (pmd_large(*pmd) && !gmap->mm->context.allow_gmap_hpage_1m)
> return -EFAULT;
> /* Link gmap segment table entry location to page table. */
> rc = radix_tree_preload(GFP_KERNEL);
> @@ -1634,6 +1636,7 @@ struct gmap *gmap_shadow(struct gmap *parent, unsigned long asce,
> unsigned long limit;
> int rc;
>
> + BUG_ON(parent->mm->context.allow_gmap_hpage_1m);
> BUG_ON(gmap_is_shadow(parent));
> spin_lock(&parent->shadow_lock);
> sg = gmap_find_shadow(parent, asce, edat_level);
> diff --git a/arch/s390/mm/pgtable.c b/arch/s390/mm/pgtable.c
> index 147380c0b5d3..8595c37e0370 100644
> --- a/arch/s390/mm/pgtable.c
> +++ b/arch/s390/mm/pgtable.c
> @@ -348,7 +348,7 @@ static inline void pmdp_idte_local(struct mm_struct *mm,
> mm->context.asce, IDTE_LOCAL);
> else
> __pmdp_idte(addr, pmdp, 0, 0, IDTE_LOCAL);
> - if (mm_has_pgste(mm))
> + if (mm_has_pgste(mm) && mm->context.allow_gmap_hpage_1m)
> gmap_pmdp_idte_local(mm, addr);
> }
>
> @@ -358,15 +358,15 @@ static inline void pmdp_idte_global(struct mm_struct *mm,
> if (MACHINE_HAS_TLB_GUEST) {
> __pmdp_idte(addr, pmdp, IDTE_NODAT | IDTE_GUEST_ASCE,
> mm->context.asce, IDTE_GLOBAL);
> - if (mm_has_pgste(mm))
> + if (mm_has_pgste(mm) && mm->context.allow_gmap_hpage_1m)
> gmap_pmdp_idte_global(mm, addr);
> } else if (MACHINE_HAS_IDTE) {
> __pmdp_idte(addr, pmdp, 0, 0, IDTE_GLOBAL);
> - if (mm_has_pgste(mm))
> + if (mm_has_pgste(mm) && mm->context.allow_gmap_hpage_1m)
> gmap_pmdp_idte_global(mm, addr);
> } else {
> __pmdp_csp(pmdp);
> - if (mm_has_pgste(mm))
> + if (mm_has_pgste(mm) && mm->context.allow_gmap_hpage_1m)
> gmap_pmdp_csp(mm, addr);
> }
> }
> @@ -435,7 +435,7 @@ pmd_t pmdp_xchg_direct(struct mm_struct *mm, unsigned long addr,
> pmd_t old;
>
> preempt_disable();
> - if (mm_has_pgste(mm))
> + if (mm_has_pgste(mm) && mm->context.allow_gmap_hpage_1m)
> pmdp_clear_skeys(mm, pmdp, new);
> old = pmdp_flush_direct(mm, addr, pmdp);
> *pmdp = new;
> @@ -450,7 +450,7 @@ pmd_t pmdp_xchg_lazy(struct mm_struct *mm, unsigned long addr,
> pmd_t old;
>
> preempt_disable();
> - if (mm_has_pgste(mm))
> + if (mm_has_pgste(mm) && mm->context.allow_gmap_hpage_1m)
> pmdp_clear_skeys(mm, pmdp, new);
> old = pmdp_flush_lazy(mm, addr, pmdp);
> *pmdp = new;
>
Not sure if all of these extra checks really make sense (performance
wise), but I guess it would be helpful if some part of e.g. QEMU (shared
libraries?) are based on huge pages, while the GMAP does not contain
huge pages.
--
Thanks,
David / dhildenb
