This commit was manufactured by cvs2svn to create branch 'vserver'.
[linux-2.6.git] / arch / i386 / mm / ioremap-xen.c
diff --git a/arch/i386/mm/ioremap-xen.c b/arch/i386/mm/ioremap-xen.c
new file mode 100644 (file)
index 0000000..3a1c84d
--- /dev/null
@@ -0,0 +1,478 @@
+/*
+ * arch/i386/mm/ioremap.c
+ *
+ * Re-map IO memory to kernel address space so that we can access it.
+ * This is needed for high PCI addresses that aren't mapped in the
+ * 640k-1MB IO memory area on PC's
+ *
+ * (C) Copyright 1995 1996 Linus Torvalds
+ */
+
+#include <linux/vmalloc.h>
+#include <linux/init.h>
+#include <linux/slab.h>
+#include <linux/module.h>
+#include <asm/io.h>
+#include <asm/fixmap.h>
+#include <asm/cacheflush.h>
+#include <asm/tlbflush.h>
+#include <asm/pgtable.h>
+#include <asm/pgalloc.h>
+
+#define ISA_START_ADDRESS      0x0
+#define ISA_END_ADDRESS                0x100000
+
+#if 0 /* not PAE safe */
+/* These hacky macros avoid phys->machine translations. */
+#define __direct_pte(x) ((pte_t) { (x) } )
+#define __direct_mk_pte(page_nr,pgprot) \
+  __direct_pte(((page_nr) << PAGE_SHIFT) | pgprot_val(pgprot))
+#define direct_mk_pte_phys(physpage, pgprot) \
+  __direct_mk_pte((physpage) >> PAGE_SHIFT, pgprot)
+#endif
+
+static int direct_remap_area_pte_fn(pte_t *pte, 
+                                   struct page *pmd_page,
+                                   unsigned long address, 
+                                   void *data)
+{
+       mmu_update_t **v = (mmu_update_t **)data;
+
+       (*v)->ptr = ((u64)pfn_to_mfn(page_to_pfn(pmd_page)) <<
+                    PAGE_SHIFT) | ((unsigned long)pte & ~PAGE_MASK);
+       (*v)++;
+
+       return 0;
+}
+
+static int __direct_remap_pfn_range(struct mm_struct *mm,
+                                   unsigned long address, 
+                                   unsigned long mfn,
+                                   unsigned long size, 
+                                   pgprot_t prot,
+                                   domid_t  domid)
+{
+       int rc;
+       unsigned long i, start_address;
+       mmu_update_t *u, *v, *w;
+
+       u = v = w = (mmu_update_t *)__get_free_page(GFP_KERNEL|__GFP_REPEAT);
+       if (u == NULL)
+               return -ENOMEM;
+
+       start_address = address;
+
+       flush_cache_all();
+
+       for (i = 0; i < size; i += PAGE_SIZE) {
+               if ((v - u) == (PAGE_SIZE / sizeof(mmu_update_t))) {
+                       /* Fill in the PTE pointers. */
+                       rc = apply_to_page_range(mm, start_address, 
+                                                address - start_address,
+                                                direct_remap_area_pte_fn, &w);
+                       if (rc)
+                               goto out;
+                       w = u;
+                       rc = -EFAULT;
+                       if (HYPERVISOR_mmu_update(u, v - u, NULL, domid) < 0)
+                               goto out;
+                       v = u;
+                       start_address = address;
+               }
+
+               /*
+                * Fill in the machine address: PTE ptr is done later by
+                * __direct_remap_area_pages(). 
+                */
+               v->val = pte_val_ma(pfn_pte_ma(mfn, prot));
+
+               mfn++;
+               address += PAGE_SIZE; 
+               v++;
+       }
+
+       if (v != u) {
+               /* get the ptep's filled in */
+               rc = apply_to_page_range(mm, start_address,
+                                        address - start_address,
+                                        direct_remap_area_pte_fn, &w);
+               if (rc)
+                       goto out;
+               rc = -EFAULT;
+               if (unlikely(HYPERVISOR_mmu_update(u, v - u, NULL, domid) < 0))
+                       goto out;
+       }
+
+       rc = 0;
+
+ out:
+       flush_tlb_all();
+
+       free_page((unsigned long)u);
+
+       return rc;
+}
+
+int direct_remap_pfn_range(struct vm_area_struct *vma,
+                          unsigned long address, 
+                          unsigned long mfn,
+                          unsigned long size, 
+                          pgprot_t prot,
+                          domid_t  domid)
+{
+       /* Same as remap_pfn_range(). */
+       vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP;
+
+       if (domid == DOMID_SELF)
+               return -EINVAL;
+
+       vma->vm_mm->context.has_foreign_mappings = 1;
+
+       return __direct_remap_pfn_range(
+               vma->vm_mm, address, mfn, size, prot, domid);
+}
+EXPORT_SYMBOL(direct_remap_pfn_range);
+
+int direct_kernel_remap_pfn_range(unsigned long address, 
+                                 unsigned long mfn,
+                                 unsigned long size, 
+                                 pgprot_t prot,
+                                 domid_t  domid)
+{
+       return __direct_remap_pfn_range(
+               &init_mm, address, mfn, size, prot, domid);
+}
+EXPORT_SYMBOL(direct_kernel_remap_pfn_range);
+
+static int lookup_pte_fn(
+       pte_t *pte, struct page *pmd_page, unsigned long addr, void *data)
+{
+       uint64_t *ptep = (uint64_t *)data;
+       if (ptep)
+               *ptep = ((uint64_t)pfn_to_mfn(page_to_pfn(pmd_page)) <<
+                        PAGE_SHIFT) | ((unsigned long)pte & ~PAGE_MASK);
+       return 0;
+}
+
+int create_lookup_pte_addr(struct mm_struct *mm, 
+                          unsigned long address,
+                          uint64_t *ptep)
+{
+       return apply_to_page_range(mm, address, PAGE_SIZE,
+                                  lookup_pte_fn, ptep);
+}
+
+EXPORT_SYMBOL(create_lookup_pte_addr);
+
+static int noop_fn(
+       pte_t *pte, struct page *pmd_page, unsigned long addr, void *data)
+{
+       return 0;
+}
+
+int touch_pte_range(struct mm_struct *mm,
+                   unsigned long address,
+                   unsigned long size)
+{
+       return apply_to_page_range(mm, address, size, noop_fn, NULL);
+} 
+
+EXPORT_SYMBOL(touch_pte_range);
+
+void *vm_map_xen_pages (unsigned long maddr, int vm_size, pgprot_t prot)
+{
+       int error;
+       
+       struct vm_struct *vma;
+       vma = get_vm_area (vm_size, VM_IOREMAP);
+      
+       if (vma == NULL) {
+               printk ("ioremap.c,vm_map_xen_pages(): "
+                       "Failed to get VMA area\n");
+               return NULL;
+       }
+
+       error = direct_kernel_remap_pfn_range((unsigned long) vma->addr,
+                                             maddr >> PAGE_SHIFT, vm_size,
+                                             prot, DOMID_SELF );
+       if (error == 0) {
+               return vma->addr;
+       } else {
+               printk ("ioremap.c,vm_map_xen_pages(): "
+                       "Failed to map xen shared pages into kernel space\n");
+               return NULL;
+       }
+}
+EXPORT_SYMBOL(vm_map_xen_pages);
+
+/*
+ * Does @address reside within a non-highmem page that is local to this virtual
+ * machine (i.e., not an I/O page, nor a memory page belonging to another VM).
+ * See the comment that accompanies mfn_to_local_pfn() in page.h to understand
+ * why this works.
+ */
+static inline int is_local_lowmem(unsigned long address)
+{
+       extern unsigned long max_low_pfn;
+       return (mfn_to_local_pfn(address >> PAGE_SHIFT) < max_low_pfn);
+}
+
+/*
+ * Generic mapping function (not visible outside):
+ */
+
+/*
+ * Remap an arbitrary physical address space into the kernel virtual
+ * address space. Needed when the kernel wants to access high addresses
+ * directly.
+ *
+ * NOTE! We need to allow non-page-aligned mappings too: we will obviously
+ * have to convert them into an offset in a page-aligned mapping, but the
+ * caller shouldn't need to know that small detail.
+ */
+void __iomem * __ioremap(unsigned long phys_addr, unsigned long size, unsigned long flags)
+{
+       void __iomem * addr;
+       struct vm_struct * area;
+       unsigned long offset, last_addr;
+       domid_t domid = DOMID_IO;
+
+       /* Don't allow wraparound or zero size */
+       last_addr = phys_addr + size - 1;
+       if (!size || last_addr < phys_addr)
+               return NULL;
+
+       /*
+        * Don't remap the low PCI/ISA area, it's always mapped..
+        */
+       if (xen_start_info->flags & SIF_PRIVILEGED &&
+           phys_addr >= ISA_START_ADDRESS && last_addr < ISA_END_ADDRESS)
+               return (void __iomem *) isa_bus_to_virt(phys_addr);
+
+       /*
+        * Don't allow anybody to remap normal RAM that we're using..
+        */
+       if (is_local_lowmem(phys_addr)) {
+               char *t_addr, *t_end;
+               struct page *page;
+
+               t_addr = bus_to_virt(phys_addr);
+               t_end = t_addr + (size - 1);
+          
+               for(page = virt_to_page(t_addr); page <= virt_to_page(t_end); page++)
+                       if(!PageReserved(page))
+                               return NULL;
+
+               domid = DOMID_SELF;
+       }
+
+       /*
+        * Mappings have to be page-aligned
+        */
+       offset = phys_addr & ~PAGE_MASK;
+       phys_addr &= PAGE_MASK;
+       size = PAGE_ALIGN(last_addr+1) - phys_addr;
+
+       /*
+        * Ok, go for it..
+        */
+       area = get_vm_area(size, VM_IOREMAP | (flags << 20));
+       if (!area)
+               return NULL;
+       area->phys_addr = phys_addr;
+       addr = (void __iomem *) area->addr;
+       flags |= _PAGE_PRESENT | _PAGE_RW | _PAGE_DIRTY | _PAGE_ACCESSED;
+#ifdef __x86_64__
+       flags |= _PAGE_USER;
+#endif
+       if (__direct_remap_pfn_range(&init_mm, (unsigned long)addr,
+                                    phys_addr>>PAGE_SHIFT,
+                                    size, __pgprot(flags), domid)) {
+               vunmap((void __force *) addr);
+               return NULL;
+       }
+       return (void __iomem *) (offset + (char __iomem *)addr);
+}
+EXPORT_SYMBOL(__ioremap);
+
+/**
+ * ioremap_nocache     -   map bus memory into CPU space
+ * @offset:    bus address of the memory
+ * @size:      size of the resource to map
+ *
+ * ioremap_nocache performs a platform specific sequence of operations to
+ * make bus memory CPU accessible via the readb/readw/readl/writeb/
+ * writew/writel functions and the other mmio helpers. The returned
+ * address is not guaranteed to be usable directly as a virtual
+ * address. 
+ *
+ * This version of ioremap ensures that the memory is marked uncachable
+ * on the CPU as well as honouring existing caching rules from things like
+ * the PCI bus. Note that there are other caches and buffers on many 
+ * busses. In particular driver authors should read up on PCI writes
+ *
+ * It's useful if some control registers are in such an area and
+ * write combining or read caching is not desirable:
+ * 
+ * Must be freed with iounmap.
+ */
+
+void __iomem *ioremap_nocache (unsigned long phys_addr, unsigned long size)
+{
+       unsigned long last_addr;
+       void __iomem *p = __ioremap(phys_addr, size, _PAGE_PCD);
+       if (!p) 
+               return p; 
+
+       /* Guaranteed to be > phys_addr, as per __ioremap() */
+       last_addr = phys_addr + size - 1;
+
+       if (is_local_lowmem(last_addr)) { 
+               struct page *ppage = virt_to_page(bus_to_virt(phys_addr));
+               unsigned long npages;
+
+               phys_addr &= PAGE_MASK;
+
+               /* This might overflow and become zero.. */
+               last_addr = PAGE_ALIGN(last_addr);
+
+               /* .. but that's ok, because modulo-2**n arithmetic will make
+               * the page-aligned "last - first" come out right.
+               */
+               npages = (last_addr - phys_addr) >> PAGE_SHIFT;
+
+               if (change_page_attr(ppage, npages, PAGE_KERNEL_NOCACHE) < 0) { 
+                       iounmap(p); 
+                       p = NULL;
+               }
+               global_flush_tlb();
+       }
+
+       return p;                                       
+}
+EXPORT_SYMBOL(ioremap_nocache);
+
+/**
+ * iounmap - Free a IO remapping
+ * @addr: virtual address from ioremap_*
+ *
+ * Caller must ensure there is only one unmapping for the same pointer.
+ */
+void iounmap(volatile void __iomem *addr)
+{
+       struct vm_struct *p, *o;
+
+       if ((void __force *)addr <= high_memory)
+               return;
+
+       /*
+        * __ioremap special-cases the PCI/ISA range by not instantiating a
+        * vm_area and by simply returning an address into the kernel mapping
+        * of ISA space.   So handle that here.
+        */
+       if ((unsigned long) addr >= fix_to_virt(FIX_ISAMAP_BEGIN))
+               return;
+
+       addr = (volatile void __iomem *)(PAGE_MASK & (unsigned long __force)addr);
+
+       /* Use the vm area unlocked, assuming the caller
+          ensures there isn't another iounmap for the same address
+          in parallel. Reuse of the virtual address is prevented by
+          leaving it in the global lists until we're done with it.
+          cpa takes care of the direct mappings. */
+       read_lock(&vmlist_lock);
+       for (p = vmlist; p; p = p->next) {
+               if (p->addr == addr)
+                       break;
+       }
+       read_unlock(&vmlist_lock);
+
+       if (!p) {
+               printk("iounmap: bad address %p\n", addr);
+               dump_stack();
+               return;
+       }
+
+       /* Reset the direct mapping. Can block */
+       if ((p->flags >> 20) && is_local_lowmem(p->phys_addr)) {
+               /* p->size includes the guard page, but cpa doesn't like that */
+               change_page_attr(virt_to_page(bus_to_virt(p->phys_addr)),
+                                (p->size - PAGE_SIZE) >> PAGE_SHIFT,
+                                PAGE_KERNEL);
+               global_flush_tlb();
+       } 
+
+       /* Finally remove it */
+       o = remove_vm_area((void *)addr);
+       BUG_ON(p != o || o == NULL);
+       kfree(p); 
+}
+EXPORT_SYMBOL(iounmap);
+
+void __init *bt_ioremap(unsigned long phys_addr, unsigned long size)
+{
+       unsigned long offset, last_addr;
+       unsigned int nrpages;
+       enum fixed_addresses idx;
+
+       /* Don't allow wraparound or zero size */
+       last_addr = phys_addr + size - 1;
+       if (!size || last_addr < phys_addr)
+               return NULL;
+
+       /*
+        * Don't remap the low PCI/ISA area, it's always mapped..
+        */
+       if (xen_start_info->flags & SIF_PRIVILEGED &&
+           phys_addr >= ISA_START_ADDRESS && last_addr < ISA_END_ADDRESS)
+               return isa_bus_to_virt(phys_addr);
+
+       /*
+        * Mappings have to be page-aligned
+        */
+       offset = phys_addr & ~PAGE_MASK;
+       phys_addr &= PAGE_MASK;
+       size = PAGE_ALIGN(last_addr) - phys_addr;
+
+       /*
+        * Mappings have to fit in the FIX_BTMAP area.
+        */
+       nrpages = size >> PAGE_SHIFT;
+       if (nrpages > NR_FIX_BTMAPS)
+               return NULL;
+
+       /*
+        * Ok, go for it..
+        */
+       idx = FIX_BTMAP_BEGIN;
+       while (nrpages > 0) {
+               set_fixmap(idx, phys_addr);
+               phys_addr += PAGE_SIZE;
+               --idx;
+               --nrpages;
+       }
+       return (void*) (offset + fix_to_virt(FIX_BTMAP_BEGIN));
+}
+
+void __init bt_iounmap(void *addr, unsigned long size)
+{
+       unsigned long virt_addr;
+       unsigned long offset;
+       unsigned int nrpages;
+       enum fixed_addresses idx;
+
+       virt_addr = (unsigned long)addr;
+       if (virt_addr < fix_to_virt(FIX_BTMAP_BEGIN))
+               return;
+       if (virt_addr >= fix_to_virt(FIX_ISAMAP_BEGIN))
+               return;
+       offset = virt_addr & ~PAGE_MASK;
+       nrpages = PAGE_ALIGN(offset + size - 1) >> PAGE_SHIFT;
+
+       idx = FIX_BTMAP_BEGIN;
+       while (nrpages > 0) {
+               clear_fixmap(idx);
+               --idx;
+               --nrpages;
+       }
+}