i915: use io-mapping interfaces instead of a variety of mapping kludges

Impact: optimize/clean-up the IO mapping implementation of the i915 DRM driver

Switch the i915 device aperture mapping to the io-mapping interface, taking
advantage of the cleaner API to extend it across all of the mapping uses,
including both pwrite and relocation updates.

This dramatically improves performance on 64-bit kernels which were using
the same slow path as 32-bit non-HIGHMEM kernels prior to this patch.

Signed-off-by: Keith Packard <keithp@keithp.com>
Signed-off-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

2 files changed, 83 insertions, 94 deletions

diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index f20ffe17df71..126b2f13258c 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -31,6 +31,7 @@
 #define _I915_DRV_H_
 
 #include "i915_reg.h"
+#include <linux/io-mapping.h>
 
 /* General customization:
  */
@@ -246,6 +247,8 @@ typedef struct drm_i915_private {
         struct {
                 struct drm_mm gtt_space;
 
+                struct io_mapping *gtt_mapping;
+
                 /**
                  * List of objects currently involved in rendering from the
                  * ringbuffer.
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 17ae330ff269..61183b95b108 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -171,35 +171,50 @@ i915_gem_pread_ioctl(struct drm_device *dev, void *data,
         return 0;
 }
 
-/*
- * Try to write quickly with an atomic kmap. Return true on success.
- *
- * If this fails (which includes a partial write), we'll redo the whole
- * thing with the slow version.
- *
- * This is a workaround for the low performance of iounmap (approximate
- * 10% cpu cost on normal 3D workloads).  kmap_atomic on HIGHMEM kernels
- * happens to let us map card memory without taking IPIs.  When the vmap
- * rework lands we should be able to dump this hack.
+/* This is the fast write path which cannot handle
+ * page faults in the source data
  */
-static inline int fast_user_write(unsigned long pfn, char __user *user_data,
-                                  int l, int o)
+
+static inline int
+fast_user_write(struct io_mapping *mapping,
+                loff_t page_base, int page_offset,
+                char __user *user_data,
+                int length)
 {
-#ifdef CONFIG_HIGHMEM
-        unsigned long unwritten;
         char *vaddr_atomic;
+        unsigned long unwritten;
 
-        vaddr_atomic = kmap_atomic_pfn(pfn, KM_USER0);
-#if WATCH_PWRITE
-        DRM_INFO("pwrite i %d o %d l %d pfn %ld vaddr %p\n",
-                 i, o, l, pfn, vaddr_atomic);
-#endif
-        unwritten = __copy_from_user_inatomic_nocache(vaddr_atomic + o, user_data, l);
-        kunmap_atomic(vaddr_atomic, KM_USER0);
-        return !unwritten;
-#else
+        vaddr_atomic = io_mapping_map_atomic_wc(mapping, page_base);
+        unwritten = __copy_from_user_inatomic_nocache(vaddr_atomic + page_offset,
+                                                      user_data, length);
+        io_mapping_unmap_atomic(vaddr_atomic);
+        if (unwritten)
+                return -EFAULT;
+        return 0;
+}
+
+/* Here's the write path which can sleep for
+ * page faults
+ */
+
+static inline int
+slow_user_write(struct io_mapping *mapping,
+                loff_t page_base, int page_offset,
+                char __user *user_data,
+                int length)
+{
+        char __iomem *vaddr;
+        unsigned long unwritten;
+
+        vaddr = io_mapping_map_wc(mapping, page_base);
+        if (vaddr == NULL)
+                return -EFAULT;
+        unwritten = __copy_from_user(vaddr + page_offset,
+                                     user_data, length);
+        io_mapping_unmap(vaddr);
+        if (unwritten)
+                return -EFAULT;
         return 0;
-#endif
 }
 
 static int
@@ -208,10 +223,12 @@ i915_gem_gtt_pwrite(struct drm_device *dev, struct drm_gem_object *obj,
                     struct drm_file *file_priv)
 {
         struct drm_i915_gem_object *obj_priv = obj->driver_private;
+        drm_i915_private_t *dev_priv = dev->dev_private;
         ssize_t remain;
-        loff_t offset;
+        loff_t offset, page_base;
         char __user *user_data;
-        int ret = 0;
+        int page_offset, page_length;
+        int ret;
 
         user_data = (char __user *) (uintptr_t) args->data_ptr;
         remain = args->size;
@@ -235,57 +252,37 @@ i915_gem_gtt_pwrite(struct drm_device *dev, struct drm_gem_object *obj,
         obj_priv->dirty = 1;
 
         while (remain > 0) {
-                unsigned long pfn;
-                int i, o, l;
-
                 /* Operation in this page
                  *
-                 * i = page number
-                 * o = offset within page
-                 * l = bytes to copy
+                 * page_base = page offset within aperture
+                 * page_offset = offset within page
+                 * page_length = bytes to copy for this page
                  */
-                i = offset >> PAGE_SHIFT;
-                o = offset & (PAGE_SIZE-1);
-                l = remain;
-                if ((o + l) > PAGE_SIZE)
-                        l = PAGE_SIZE - o;
-
-                pfn = (dev->agp->base >> PAGE_SHIFT) + i;
-
-                if (!fast_user_write(pfn, user_data, l, o)) {
-                        unsigned long unwritten;
-                        char __iomem *vaddr;
-
-                        vaddr = ioremap_wc(pfn << PAGE_SHIFT, PAGE_SIZE);
-#if WATCH_PWRITE
-                        DRM_INFO("pwrite slow i %d o %d l %d "
-                                 "pfn %ld vaddr %p\n",
-                                 i, o, l, pfn, vaddr);
-#endif
-                        if (vaddr == NULL) {
-                                ret = -EFAULT;
-                                goto fail;
-                        }
-                        unwritten = __copy_from_user(vaddr + o, user_data, l);
-#if WATCH_PWRITE
-                        DRM_INFO("unwritten %ld\n", unwritten);
-#endif
-                        iounmap(vaddr);
-                        if (unwritten) {
-                                ret = -EFAULT;
+                page_base = (offset & ~(PAGE_SIZE-1));
+                page_offset = offset & (PAGE_SIZE-1);
+                page_length = remain;
+                if ((page_offset + remain) > PAGE_SIZE)
+                        page_length = PAGE_SIZE - page_offset;
+
+                ret = fast_user_write (dev_priv->mm.gtt_mapping, page_base,
+                                       page_offset, user_data, page_length);
+
+                /* If we get a fault while copying data, then (presumably) our
+                 * source page isn't available. In this case, use the
+                 * non-atomic function
+                 */
+                if (ret) {
+                        ret = slow_user_write (dev_priv->mm.gtt_mapping,
+                                               page_base, page_offset,
+                                               user_data, page_length);
+                        if (ret)
                                 goto fail;
-                        }
                 }
 
-                remain -= l;
-                user_data += l;
-                offset += l;
+                remain -= page_length;
+                user_data += page_length;
+                offset += page_length;
         }