xen: avoid hypercalls when updating unpinned pud/pmd

When operating on an unpinned pagetable (ie, one under construction or
destruction), it isn't necessary to use a hypercall to update a
pud/pmd entry.  Jan Beulich observed that a similar optimisation
avoided many thousands of hypercalls while doing a kernel build.

One tricky part is that early in the kernel boot there's no page
structure, so we can't check to see if the page is pinned.  In that
case, we just always use the hypercall.

Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Cc: Jan Beulich <jbeulich@novell.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

3 files changed, 50 insertions, 11 deletions

diff --git a/arch/x86/xen/enlighten.c b/arch/x86/xen/enlighten.c
index b94f63ac228b..ed9f04b3836d 100644
--- a/arch/x86/xen/enlighten.c
+++ b/arch/x86/xen/enlighten.c
@@ -901,6 +901,14 @@ static __init void xen_pagetable_setup_done(pgd_t *base)
         pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(base)));
 }
 
+static __init void xen_post_allocator_init(void)
+{
+        pv_mmu_ops.set_pmd = xen_set_pmd;
+        pv_mmu_ops.set_pud = xen_set_pud;
+
+        xen_mark_init_mm_pinned();
+}
+
 /* This is called once we have the cpu_possible_map */
 void xen_setup_vcpu_info_placement(void)
 {
@@ -988,7 +996,7 @@ static const struct pv_init_ops xen_init_ops __initdata = {
         .banner = xen_banner,
         .memory_setup = xen_memory_setup,
         .arch_setup = xen_arch_setup,
-        .post_allocator_init = xen_mark_init_mm_pinned,
+        .post_allocator_init = xen_post_allocator_init,
 };
 
 static const struct pv_time_ops xen_time_ops __initdata = {
@@ -1100,7 +1108,7 @@ static const struct pv_mmu_ops xen_mmu_ops __initdata = {
 
         .set_pte = NULL,        /* see xen_pagetable_setup_* */
         .set_pte_at = xen_set_pte_at,
-        .set_pmd = xen_set_pmd,
+        .set_pmd = xen_set_pmd_hyper,
 
         .pte_val = xen_pte_val,
         .pte_flags = native_pte_val,
@@ -1111,7 +1119,7 @@ static const struct pv_mmu_ops xen_mmu_ops __initdata = {
 
         .set_pte_atomic = xen_set_pte_atomic,
         .set_pte_present = xen_set_pte_at,
-        .set_pud = xen_set_pud,
+        .set_pud = xen_set_pud_hyper,
         .pte_clear = xen_pte_clear,
         .pmd_clear = xen_pmd_clear,
 
diff --git a/arch/x86/xen/mmu.c b/arch/x86/xen/mmu.c
index 17f374eb1faa..4fa0934db925 100644
--- a/arch/x86/xen/mmu.c
+++ b/arch/x86/xen/mmu.c
@@ -223,7 +223,14 @@ void make_lowmem_page_readwrite(void *vaddr)
 }
 
 
-void xen_set_pmd(pmd_t *ptr, pmd_t val)
+static bool page_pinned(void *ptr)
+{
+        struct page *page = virt_to_page(ptr);
+
+        return PagePinned(page);
+}
+
+void xen_set_pmd_hyper(pmd_t *ptr, pmd_t val)
 {
         struct multicall_space mcs;
         struct mmu_update *u;
@@ -241,6 +248,18 @@ void xen_set_pmd(pmd_t *ptr, pmd_t val)
         preempt_enable();
 }
 
+void xen_set_pmd(pmd_t *ptr, pmd_t val)
+{
+        /* If page is not pinned, we can just update the entry
+           directly */
+        if (!page_pinned(ptr)) {
+                *ptr = val;
+                return;
+        }
+
+        xen_set_pmd_hyper(ptr, val);
+}
+
 /*
  * Associate a virtual page frame with a given physical page frame
  * and protection flags for that frame.
@@ -348,7 +367,7 @@ pmdval_t xen_pmd_val(pmd_t pmd)
         return ret;
 }
 
-void xen_set_pud(pud_t *ptr, pud_t val)
+void xen_set_pud_hyper(pud_t *ptr, pud_t val)
 {
         struct multicall_space mcs;
         struct mmu_update *u;
@@ -366,6 +385,18 @@ void xen_set_pud(pud_t *ptr, pud_t val)
         preempt_enable();
 }
 
+void xen_set_pud(pud_t *ptr, pud_t val)
+{
+        /* If page is not pinned, we can just update the entry
+           directly */
+        if (!page_pinned(ptr)) {
+                *ptr = val;
+                return;
+        }
+
+        xen_set_pud_hyper(ptr, val);
+}
+
 void xen_set_pte(pte_t *ptep, pte_t pte)
 {
         ptep->pte_high = pte.pte_high;
@@ -387,7 +418,7 @@ void xen_pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
 
 void xen_pmd_clear(pmd_t *pmdp)
 {
-        xen_set_pmd(pmdp, __pmd(0));
+        set_pmd(pmdp, __pmd(0));
 }
 
 pmd_t xen_make_pmd(pmdval_t pmd)
@@ -758,7 +789,7 @@ void xen_exit_mmap(struct mm_struct *mm)
         spin_lock(&mm->page_table_lock);
 
         /* pgd may not be pinned in the error exit path of execve */
-        if (PagePinned(virt_to_page(mm->pgd)))
+        if (page_pinned(mm->pgd))
                 xen_pgd_unpin(mm->pgd);
 
         spin_unlock(&mm->page_table_lock);
diff --git a/arch/x86/xen/mmu.h b/arch/x86/xen/mmu.h
index 5fe961caffd4..e3dd09e25c63 100644
--- a/arch/x86/xen/mmu.h
+++ b/arch/x86/xen/mmu.h
@@ -25,10 +25,6 @@ enum pt_level {
 
 void set_pte_mfn(unsigned long vaddr, unsigned long pfn, pgprot_t flags);
 
-void xen_set_pte(pte_t *ptep, pte_t pteval);
-void xen_set_pte_at(struct mm_struct *mm, unsigned long addr,
-                    pte_t *ptep, pte_t pteval);
-void xen_set_pmd(pmd_t *pmdp, pmd_t pmdval);
 
 void xen_activate_mm(struct mm_struct *prev, struct mm_struct *next);
 void xen_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm);
@@ -45,10 +41,14 @@ pte_t xen_make_pte(pteval_t);
 pmd_t xen_make_pmd(pmdval_t);
 pgd_t xen_make_pgd(pgdval_t);
 
+void xen_set_pte(pte_t *ptep, pte_t pteval);
 void xen_set_pte_at(struct mm_struct *mm, unsigned long addr,
                     pte_t *ptep, pte_t pteval);
 void xen_set_pte_atomic(pte_t *ptep, pte_t pte);
+void xen_set_pmd(pmd_t *pmdp, pmd_t pmdval);
 void xen_set_pud(pud_t *ptr, pud_t val);
+void xen_set_pmd_hyper(pmd_t *pmdp, pmd_t pmdval);
+void xen_set_pud_hyper(pud_t *ptr, pud_t val);
 void xen_pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep);
 void xen_pmd_clear(pmd_t *pmdp);