[S390] kvm guest address space mapping

Add code that allows KVM to control the virtual memory layout that
is seen by a guest. The guest address space uses a second page table
that shares the last level pte-tables with the process page table.
If a page is unmapped from the process page table it is automatically
unmapped from the guest page table as well.

The guest address space mapping starts out empty, KVM can map any
individual 1MB segments from the process virtual memory to any 1MB
aligned location in the guest virtual memory. If a target segment in
the process virtual memory does not exist or is unmapped while a
guest mapping exists the desired target address is stored as an
invalid segment table entry in the guest page table.
The population of the guest page table is fault driven.

Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>

4 files changed, 420 insertions, 29 deletions

diff --git a/arch/s390/mm/fault.c b/arch/s390/mm/fault.c
index 095f782a5512..9564fc779b27 100644
--- a/arch/s390/mm/fault.c
+++ b/arch/s390/mm/fault.c
@@ -303,9 +303,24 @@ static inline int do_exception(struct pt_regs *regs, int access,
         flags = FAULT_FLAG_ALLOW_RETRY;
         if (access == VM_WRITE || (trans_exc_code & store_indication) == 0x400)
                 flags |= FAULT_FLAG_WRITE;
-retry:
         down_read(&mm->mmap_sem);
 
+#ifdef CONFIG_PGSTE
+        if (test_tsk_thread_flag(current, TIF_SIE) && S390_lowcore.gmap) {
+                address = gmap_fault(address,
+                                     (struct gmap *) S390_lowcore.gmap);
+                if (address == -EFAULT) {
+                        fault = VM_FAULT_BADMAP;
+                        goto out_up;
+                }
+                if (address == -ENOMEM) {
+                        fault = VM_FAULT_OOM;
+                        goto out_up;
+                }
+        }
+#endif
+
+retry:
         fault = VM_FAULT_BADMAP;
         vma = find_vma(mm, address);
         if (!vma)
@@ -356,6 +371,7 @@ retry:
                         /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
                          * of starvation. */
                         flags &= ~FAULT_FLAG_ALLOW_RETRY;
+                        down_read(&mm->mmap_sem);
                         goto retry;
                 }
         }
diff --git a/arch/s390/mm/hugetlbpage.c b/arch/s390/mm/hugetlbpage.c
index a4d856db9154..597bb2d27c3c 100644
--- a/arch/s390/mm/hugetlbpage.c
+++ b/arch/s390/mm/hugetlbpage.c
@@ -35,7 +35,7 @@ int arch_prepare_hugepage(struct page *page)
         if (MACHINE_HAS_HPAGE)
                 return 0;
 
-        ptep = (pte_t *) pte_alloc_one(&init_mm, address);
+        ptep = (pte_t *) pte_alloc_one(&init_mm, addr);
         if (!ptep)
                 return -ENOMEM;
 
diff --git a/arch/s390/mm/pgtable.c b/arch/s390/mm/pgtable.c
index 37a23c223705..2adb23938a7f 100644
--- a/arch/s390/mm/pgtable.c
+++ b/arch/s390/mm/pgtable.c
@@ -16,6 +16,7 @@
 #include <linux/module.h>
 #include <linux/quicklist.h>
 #include <linux/rcupdate.h>
+#include <linux/slab.h>
 
 #include <asm/system.h>
 #include <asm/pgtable.h>
@@ -133,30 +134,374 @@ void crst_table_downgrade(struct mm_struct *mm, unsigned long limit)
 }
 #endif
 
-static inline unsigned int atomic_xor_bits(atomic_t *v, unsigned int bits)
+#ifdef CONFIG_PGSTE
+
+/**
+ * gmap_alloc - allocate a guest address space
+ * @mm: pointer to the parent mm_struct
+ *
+ * Returns a guest address space structure.
+ */
+struct gmap *gmap_alloc(struct mm_struct *mm)
 {
-        unsigned int old, new;
+        struct gmap *gmap;
+        struct page *page;
+        unsigned long *table;
 
-        do {
-                old = atomic_read(v);
-                new = old ^ bits;
-        } while (atomic_cmpxchg(v, old, new) != old);
-        return new;
+        gmap = kzalloc(sizeof(struct gmap), GFP_KERNEL);
+        if (!gmap)
+                goto out;
+        INIT_LIST_HEAD(&gmap->crst_list);
+        gmap->mm = mm;
+        page = alloc_pages(GFP_KERNEL, ALLOC_ORDER);
+        if (!page)
+                goto out_free;
+        list_add(&page->lru, &gmap->crst_list);
+        table = (unsigned long *) page_to_phys(page);
+        crst_table_init(table, _REGION1_ENTRY_EMPTY);
+        gmap->table = table;
+        list_add(&gmap->list, &mm->context.gmap_list);
+        return gmap;
+
+out_free:
+        kfree(gmap);
+out:
+        return NULL;
 }
+EXPORT_SYMBOL_GPL(gmap_alloc);
 
-/*
- * page table entry allocation/free routines.
+static int gmap_unlink_segment(struct gmap *gmap, unsigned long *table)
+{
+        struct gmap_pgtable *mp;
+        struct gmap_rmap *rmap;
+        struct page *page;
+
+        if (*table & _SEGMENT_ENTRY_INV)
+                return 0;
+        page = pfn_to_page(*table >> PAGE_SHIFT);
+        mp = (struct gmap_pgtable *) page->index;
+        list_for_each_entry(rmap, &mp->mapper, list) {
+                if (rmap->entry != table)
+                        continue;
+                list_del(&rmap->list);
+                kfree(rmap);
+                break;
+        }
+        *table = _SEGMENT_ENTRY_INV | _SEGMENT_ENTRY_RO | mp->vmaddr;
+        return 1;
+}
+
+static void gmap_flush_tlb(struct gmap *gmap)
+{
+        if (MACHINE_HAS_IDTE)
+                __tlb_flush_idte((unsigned long) gmap->table |
+                                 _ASCE_TYPE_REGION1);
+        else
+                __tlb_flush_global();
+}
+
+/**
+ * gmap_free - free a guest address space
+ * @gmap: pointer to the guest address space structure
  */
-#ifdef CONFIG_PGSTE
-static inline unsigned long *page_table_alloc_pgste(struct mm_struct *mm)
+void gmap_free(struct gmap *gmap)
+{
+        struct page *page, *next;
+        unsigned long *table;
+        int i;
+
+
+        /* Flush tlb. */
+        if (MACHINE_HAS_IDTE)
+                __tlb_flush_idte((unsigned long) gmap->table |
+                                 _ASCE_TYPE_REGION1);
+        else
+                __tlb_flush_global();
+
+        /* Free all segment & region tables. */
+        down_read(&gmap->mm->mmap_sem);
+        list_for_each_entry_safe(page, next, &gmap->crst_list, lru) {
+                table = (unsigned long *) page_to_phys(page);
+                if ((*table & _REGION_ENTRY_TYPE_MASK) == 0)
+                        /* Remove gmap rmap structures for segment table. */
+                        for (i = 0; i < PTRS_PER_PMD; i++, table++)
+                                gmap_unlink_segment(gmap, table);
+                __free_pages(page, ALLOC_ORDER);
+        }
+        up_read(&gmap->mm->mmap_sem);
+        list_del(&gmap->list);
+        kfree(gmap);
+}
+EXPORT_SYMBOL_GPL(gmap_free);
+
+/**
+ * gmap_enable - switch primary space to the guest address space
+ * @gmap: pointer to the guest address space structure
+ */
+void gmap_enable(struct gmap *gmap)
+{
+        /* Load primary space page table origin. */
+        S390_lowcore.user_asce = _ASCE_TYPE_REGION1 | _ASCE_TABLE_LENGTH |
+                                 _ASCE_USER_BITS | __pa(gmap->table);
+        asm volatile("lctlg 1,1,%0\n" : : "m" (S390_lowcore.user_asce) );
+        S390_lowcore.gmap = (unsigned long) gmap;
+}
+EXPORT_SYMBOL_GPL(gmap_enable);
+
+/**
+ * gmap_disable - switch back to the standard primary address space
+ * @gmap: pointer to the guest address space structure
+ */
+void gmap_disable(struct gmap *gmap)
+{
+        /* Load primary space page table origin. */
+        S390_lowcore.user_asce =
+                gmap->mm->context.asce_bits | __pa(gmap->mm->pgd);
+        asm volatile("lctlg 1,1,%0\n" : : "m" (S390_lowcore.user_asce) );
+        S390_lowcore.gmap = 0UL;
+}
+EXPORT_SYMBOL_GPL(gmap_disable);
+
+static int gmap_alloc_table(struct gmap *gmap,
+                               unsigned long *table, unsigned long init)
+{
+        struct page *page;
+        unsigned long *new;
+
+        page = alloc_pages(GFP_KERNEL, ALLOC_ORDER);
+        if (!page)
+                return -ENOMEM;
+        new = (unsigned long *) page_to_phys(page);
+        crst_table_init(new, init);
+        down_read(&gmap->mm->mmap_sem);
+        if (*table & _REGION_ENTRY_INV) {
+                list_add(&page->lru, &gmap->crst_list);
+                *table = (unsigned long) new | _REGION_ENTRY_LENGTH |
+                        (*table & _REGION_ENTRY_TYPE_MASK);
+        } else
+                __free_pages(page, ALLOC_ORDER);
+        up_read(&gmap->mm->mmap_sem);
+        return 0;
+}
+
+/**
+ * gmap_unmap_segment - unmap segment from the guest address space
+ * @gmap: pointer to the guest address space structure
+ * @addr: address in the guest address space
+ * @len: length of the memory area to unmap
+ *
+ * Returns 0 if the unmap succeded, -EINVAL if not.
+ */
+int gmap_unmap_segment(struct gmap *gmap, unsigned long to, unsigned long len)
+{
+        unsigned long *table;
+        unsigned long off;
+        int flush;
+
+        if ((to | len) & (PMD_SIZE - 1))
+                return -EINVAL;
+        if (len == 0 || to + len < to)
+                return -EINVAL;
+
+        flush = 0;
+        down_read(&gmap->mm->mmap_sem);
+        for (off = 0; off < len; off += PMD_SIZE) {
+                /* Walk the guest addr space page table */
+                table = gmap->table + (((to + off) >> 53) & 0x7ff);
+                if (*table & _REGION_ENTRY_INV)
+                        return 0;
+                table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
+                table = table + (((to + off) >> 42) & 0x7ff);
+                if (*table & _REGION_ENTRY_INV)
+                        return 0;
+                table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
+                table = table + (((to + off) >> 31) & 0x7ff);
+                if (*table & _REGION_ENTRY_INV)
+                        return 0;
+                table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
+                table = table + (((to + off) >> 20) & 0x7ff);
+
+                /* Clear segment table entry in guest address space. */
+                flush |= gmap_unlink_segment(gmap, table);
+                *table = _SEGMENT_ENTRY_INV;
+        }
+        up_read(&gmap->mm->mmap_sem);
+        if (flush)
+                gmap_flush_tlb(gmap);
+        return 0;
+}
+EXPORT_SYMBOL_GPL(gmap_unmap_segment);
+
+/**
+ * gmap_mmap_segment - map a segment to the guest address space
+ * @gmap: pointer to the guest address space structure
+ * @from: source address in the parent address space
+ * @to: target address in the guest address space
+ *
+ * Returns 0 if the mmap succeded, -EINVAL or -ENOMEM if not.
+ */
+int gmap_map_segment(struct gmap *gmap, unsigned long from,
+                     unsigned long to, unsigned long len)
+{
+        unsigned long *table;
+        unsigned long off;
+        int flush;
+
+        if ((from | to | len) & (PMD_SIZE - 1))
+                return -EINVAL;
+        if (len == 0 || from + len > PGDIR_SIZE ||
+            from + len < from || to + len < to)
+                return -EINVAL;
+
+        flush = 0;
+        down_read(&gmap->mm->mmap_sem);
+        for (off = 0; off < len; off += PMD_SIZE) {
+                /* Walk the gmap address space page table */
+                table = gmap->table + (((to + off) >> 53) & 0x7ff);
+                if ((*table & _REGION_ENTRY_INV) &&
+                    gmap_alloc_table(gmap, table, _REGION2_ENTRY_EMPTY))
+                        goto out_unmap;
+                table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
+                table = table + (((to + off) >> 42) & 0x7ff);
+                if ((*table & _REGION_ENTRY_INV) &&
+                    gmap_alloc_table(gmap, table, _REGION3_ENTRY_EMPTY))
+                        goto out_unmap;
+                table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
+                table = table + (((to + off) >> 31) & 0x7ff);
+                if ((*table & _REGION_ENTRY_INV) &&
+                    gmap_alloc_table(gmap, table, _SEGMENT_ENTRY_EMPTY))
+                        goto out_unmap;
+                table = (unsigned long *) (*table & _REGION_ENTRY_ORIGIN);
+                table = table + (((to + off) >> 20) & 0x7ff);
+
+                /* Store 'from' address in an invalid segment table entry. */
+                flush |= gmap_unlink_segment(gmap, table);
+                *table = _SEGMENT_ENTRY_INV | _SEGMENT_ENTRY_RO | (from + off);
+        }
+        up_read(&gmap->mm->mmap_sem);
+        if (flush)
+                gmap_flush_tlb(gmap);
+        return 0;
+
+out_unmap:
+        up_read(&gmap->mm->mmap_sem);
+        gmap_unmap_segment(gmap, to, len);
+        return -ENOMEM;
+}
+EXPORT_SYMBOL_GPL(gmap_map_segment);
+
+unsigned long gmap_fault(unsigned long address, struct gmap *gmap)
+{
+        unsigned long *table, vmaddr, segment;
+        struct mm_struct *mm;
+        struct gmap_pgtable *mp;
+        struct gmap_rmap *rmap;
+        struct vm_area_struct *vma;
+        struct page *page;
+        pgd_t *pgd;
+        pud_t *pud;
+        pmd_t *pmd;
+
+        current->thread.gmap_addr = address;
+        mm = gmap->mm;
+        /* Walk the gmap address space page table */
+        table = gmap->table + ((address >> 53) & 0x7ff);
+        if (unlikely(*table & _REGION_ENTRY_INV))
+                return -EFAULT;
+        table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
+        table = table + ((address >> 42) & 0x7ff);
+        if (unlikely(*table & _REGION_ENTRY_INV))
+                return -EFAULT;
+        table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
+        table = table + ((address >> 31) & 0x7ff);
+        if (unlikely(*table & _REGION_ENTRY_INV))
+                return -EFAULT;
+        table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
+        table = table + ((address >> 20) & 0x7ff);
+
+        /* Convert the gmap address to an mm address. */
+        segment = *table;
+        if (likely(!(segment & _SEGMENT_ENTRY_INV))) {
+                page = pfn_to_page(segment >> PAGE_SHIFT);
+                mp = (struct gmap_pgtable *) page->index;
+                return mp->vmaddr | (address & ~PMD_MASK);
+        } else if (segment & _SEGMENT_ENTRY_RO) {
+                vmaddr = segment & _SEGMENT_ENTRY_ORIGIN;
+                vma = find_vma(mm, vmaddr);
+                if (!vma || vma->vm_start > vmaddr)
+                        return -EFAULT;
+
+                /* Walk the parent mm page table */
+                pgd = pgd_offset(mm, vmaddr);
+                pud = pud_alloc(mm, pgd, vmaddr);
+                if (!pud)
+                        return -ENOMEM;
+                pmd = pmd_alloc(mm, pud, vmaddr);
+                if (!pmd)
+                        return -ENOMEM;
+                if (!pmd_present(*pmd) &&
+                    __pte_alloc(mm, vma, pmd, vmaddr))
+                        return -ENOMEM;
+                /* pmd now points to a valid segment table entry. */
+                rmap = kmalloc(sizeof(*rmap), GFP_KERNEL|__GFP_REPEAT);
+                if (!rmap)
+                        return -ENOMEM;
+                /* Link gmap segment table entry location to page table. */
+                page = pmd_page(*pmd);
+                mp = (struct gmap_pgtable *) page->index;
+                rmap->entry = table;
+                list_add(&rmap->list, &mp->mapper);
+                /* Set gmap segment table entry to page table. */
+                *table = pmd_val(*pmd) & PAGE_MASK;
+                return vmaddr | (address & ~PMD_MASK);
+        }
+        return -EFAULT;
+
+}
+EXPORT_SYMBOL_GPL(gmap_fault);
+
+void gmap_unmap_notifier(struct mm_struct *mm, unsigned long *table)
+{
+        struct gmap_rmap *rmap, *next;
+        struct gmap_pgtable *mp;
+        struct page *page;
+        int flush;
+
+        flush = 0;
+        spin_lock(&mm->page_table_lock);
+        page = pfn_to_page(__pa(table) >> PAGE_SHIFT);
+        mp = (struct gmap_pgtable *) page->index;
+        list_for_each_entry_safe(rmap, next, &mp->mapper, list) {
+                *rmap->entry =
+                        _SEGMENT_ENTRY_INV | _SEGMENT_ENTRY_RO | mp->vmaddr;
+                list_del(&rmap->list);
+                kfree(rmap);
+                flush = 1;
+        }
+        spin_unlock(&mm->page_table_lock);
+        if (flush)
+                __tlb_flush_global();
+}
+
+static inline unsigned long *page_table_alloc_pgste(struct mm_struct *mm,
+                                                    unsigned long vmaddr)
 {
         struct page *page;
         unsigned long *table;
+        struct gmap_pgtable *mp;
 
         page = alloc_page(GFP_KERNEL|__GFP_REPEAT);
         if (!page)
                 return NULL;
+        mp = kmalloc(sizeof(*mp), GFP_KERNEL|__GFP_REPEAT);
+        if (!mp) {
+                __free_page(page);
+                return NULL;
+        }
         pgtable_page_ctor(page);
+        mp->vmaddr = vmaddr & PMD_MASK;
+        INIT_LIST_HEAD(&mp->mapper);
+        page->index = (unsigned long) mp;
         atomic_set(&page->_mapcount, 3);
         table = (unsigned long *) page_to_phys(page);
         clear_table(table, _PAGE_TYPE_EMPTY, PAGE_SIZE/2);
@@ -167,24 +512,57 @@ static inline unsigned long *page_table_alloc_pgste(struct mm_struct *mm)
 static inline void page_table_free_pgste(unsigned long *table)
 {
         struct page *page;
+        struct gmap_pgtable *mp;
 
         page = pfn_to_page(__pa(table) >> PAGE_SHIFT);
+        mp = (struct gmap_pgtable *) page->index;
+        BUG_ON(!list_empty(&mp->mapper));
         pgtable_page_ctor(page);
         atomic_set(&page->_mapcount, -1);
+        kfree(mp);
         __free_page(page);
 }
-#endif
 
-unsigned long *page_table_alloc(struct mm_struct *mm)
+#else /* CONFIG_PGSTE */
+
+static inline unsigned long *page_table_alloc_pgste(struct mm_struct *mm,
+                                                    unsigned long vmaddr)
+{
+}
+
+static inline void page_table_free_pgste(unsigned long *table)
+{
+}
+
+static inline void gmap_unmap_notifier(struct mm_struct *mm,
+                                          unsigned long *table)
+{
+}
+
+#endif /* CONFIG_PGSTE */
+
+static inline unsigned int atomic_xor_bits(atomic_t *v, unsigned int bits)
+{
+        unsigned int old, new;
+
+        do {
+                old = atomic_read(v);
+                new = old ^ bits;
+        } while (atomic_cmpxchg(v, old, new) != old);
+        return new;
+}
+
+/*
+ * page table entry allocation/free routines.
+ */
+unsigned long *page_table_alloc(struct mm_struct *mm, unsigned long vmaddr)
 {
         struct page *page;
         unsigned long *table;
         unsigned int mask, bit;
 
-#ifdef CONFIG_PGSTE
         if (mm_has_pgste(mm))
-                return page_table_alloc_pgste(mm);
-#endif
+                return page_table_alloc_pgste(mm, vmaddr);
         /* Allocate fragments of a 4K page as 1K/2K page table */
         spin_lock_bh(&mm->context.list_lock);
         mask = FRAG_MASK;
@@ -222,10 +600,10 @@ void page_table_free(struct mm_struct *mm, unsigned long *table)
         struct page *page;
         unsigned int bit, mask;
 
-#ifdef CONFIG_PGSTE
-        if (mm_has_pgste(mm))
+        if (mm_has_pgste(mm)) {
+                gmap_unmap_notifier(mm, table);
                 return page_table_free_pgste(table);
-#endif
+        }
         /* Free 1K/2K page table fragment of a 4K page */
         page = pfn_to_page(__pa(table) >> PAGE_SHIFT);
         bit = 1 << ((__pa(table) & ~PAGE_MASK)/(PTRS_PER_PTE*sizeof(pte_t)));
@@ -249,10 +627,8 @@ static void __page_table_free_rcu(void *table, unsigned bit)
 {
         struct page *page;
 
-#ifdef CONFIG_PGSTE
         if (bit == FRAG_MASK)
                 return page_table_free_pgste(table);
-#endif
         /* Free 1K/2K page table fragment of a 4K page */
         page = pfn_to_page(__pa(table) >> PAGE_SHIFT);
         if (atomic_xor_bits(&page->_mapcount, bit) == 0) {
@@ -269,13 +645,12 @@ void page_table_free_rcu(struct mmu_gather *tlb, unsigned long *table)
         unsigned int bit, mask;
 
         mm = tlb->mm;
-#ifdef CONFIG_PGSTE
         if (mm_has_pgste(mm)) {
+                gmap_unmap_notifier(mm, table);
                 table = (unsigned long *) (__pa(table) | FRAG_MASK);
                 tlb_remove_table(tlb, table);
                 return;
         }
-#endif
         bit = 1 << ((__pa(table) & ~PAGE_MASK) / (PTRS_PER_PTE*sizeof(pte_t)));
         page = pfn_to_page(__pa(table) >> PAGE_SHIFT);
         spin_lock_bh(&mm->context.list_lock);
diff --git a/arch/s390/mm/vmem.c b/arch/s390/mm/vmem.c
index 8c1970d1dd91..781ff5169560 100644
--- a/arch/s390/mm/vmem.c
+++ b/arch/s390/mm/vmem.c
@@ -61,12 +61,12 @@ static inline pmd_t *vmem_pmd_alloc(void)
         return pmd;
 }
 
-static pte_t __ref *vmem_pte_alloc(void)
+static pte_t __ref *vmem_pte_alloc(unsigned long address)
 {
         pte_t *pte;
 
         if (slab_is_available())
-                pte = (pte_t *) page_table_alloc(&init_mm);
+                pte = (pte_t *) page_table_alloc(&init_mm, address);
         else
                 pte = alloc_bootmem(PTRS_PER_PTE * sizeof(pte_t));
         if (!pte)
@@ -120,7 +120,7 @@ static int vmem_add_mem(unsigned long start, unsigned long size, int ro)
                 }
 #endif
                 if (pmd_none(*pm_dir)) {
-                        pt_dir = vmem_pte_alloc();
+                        pt_dir = vmem_pte_alloc(address);
                         if (!pt_dir)
                                 goto out;
                         pmd_populate(&init_mm, pm_dir, pt_dir);
@@ -205,7 +205,7 @@ int __meminit vmemmap_populate(struct page *start, unsigned long nr, int node)
 
                 pm_dir = pmd_offset(pu_dir, address);
                 if (pmd_none(*pm_dir)) {
-                        pt_dir = vmem_pte_alloc();
+                        pt_dir = vmem_pte_alloc(address);
                         if (!pt_dir)
                                 goto out;
                         pmd_populate(&init_mm, pm_dir, pt_dir);