ide-tape: improve buffer allocation strategy
authorBorislav Petkov <petkovbb@googlemail.com>
Sun, 27 Apr 2008 13:38:32 +0000 (15:38 +0200)
committerBartlomiej Zolnierkiewicz <bzolnier@gmail.com>
Sun, 27 Apr 2008 13:38:32 +0000 (15:38 +0200)
Instead of allocating pages for the buffer one by one, take advantage of the
buddy alloc system and request them 2^order at a time. This increases the chance
for bigger buffer parts to be contigious and reduces loop iteration count. While
at it, rename function __idetape_kmalloc_stage() to ide_tape_kmalloc_buffer().

[bart: fold with "ide-tape: fix mem leak" patch to preserve bisectability]

Signed-off-by: Borislav Petkov <petkovbb@gmail.com>
Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>
drivers/ide/ide-tape.c

index 2275cd22902cabdd7efb6ff70ac4f8e3d871d38f..dc7abb25a8ebf3d87f85c8795add3cc7d93f1aba 100644 (file)
@@ -1290,20 +1290,20 @@ out:
 }
 
 /*
- * The function below uses __get_free_page to allocate a pipeline stage, along
- * with all the necessary small buffers which together make a buffer of size
+ * The function below uses __get_free_pages to allocate a data buffer of size
  * tape->stage_size (or a bit more). We attempt to combine sequential pages as
  * much as possible.
  *
- * It returns a pointer to the new allocated stage, or NULL if we can't (or
- * don't want to) allocate a stage.
+ * It returns a pointer to the newly allocated buffer, or NULL in case of
+ * failure.
  */
-static idetape_stage_t *__idetape_kmalloc_stage(idetape_tape_t *tape, int full,
+static idetape_stage_t *ide_tape_kmalloc_buffer(idetape_tape_t *tape, int full,
                                                int clear)
 {
        idetape_stage_t *stage;
        struct idetape_bh *prev_bh, *bh;
        int pages = tape->pages_per_stage;
+       unsigned int order, b_allocd;
        char *b_data = NULL;
 
        stage = kmalloc(sizeof(idetape_stage_t), GFP_KERNEL);
@@ -1315,46 +1315,60 @@ static idetape_stage_t *__idetape_kmalloc_stage(idetape_tape_t *tape, int full,
        bh = stage->bh;
        if (bh == NULL)
                goto abort;
-       bh->b_reqnext = NULL;
-       bh->b_data = (char *) __get_free_page(GFP_KERNEL);
+
+       order = fls(pages) - 1;
+       bh->b_data = (char *) __get_free_pages(GFP_KERNEL, order);
        if (!bh->b_data)
                goto abort;
+       b_allocd = (1 << order) * PAGE_SIZE;
+       pages &= (order-1);
+
        if (clear)
-               memset(bh->b_data, 0, PAGE_SIZE);
-       bh->b_size = PAGE_SIZE;
+               memset(bh->b_data, 0, b_allocd);
+       bh->b_reqnext = NULL;
+       bh->b_size = b_allocd;
        atomic_set(&bh->b_count, full ? bh->b_size : 0);
 
-       while (--pages) {
-               b_data = (char *) __get_free_page(GFP_KERNEL);
+       while (pages) {
+               order = fls(pages) - 1;
+               b_data = (char *) __get_free_pages(GFP_KERNEL, order);
                if (!b_data)
                        goto abort;
+               b_allocd = (1 << order) * PAGE_SIZE;
+
                if (clear)
-                       memset(b_data, 0, PAGE_SIZE);
-               if (bh->b_data == b_data + PAGE_SIZE) {
-                       bh->b_size += PAGE_SIZE;
-                       bh->b_data -= PAGE_SIZE;
+                       memset(b_data, 0, b_allocd);
+
+               /* newly allocated page frames below buffer header or ...*/
+               if (bh->b_data == b_data + b_allocd) {
+                       bh->b_size += b_allocd;
+                       bh->b_data -= b_allocd;
                        if (full)
-                               atomic_add(PAGE_SIZE, &bh->b_count);
+                               atomic_add(b_allocd, &bh->b_count);
                        continue;
                }
+               /* they are above the header */
                if (b_data == bh->b_data + bh->b_size) {
-                       bh->b_size += PAGE_SIZE;
+                       bh->b_size += b_allocd;
                        if (full)
-                               atomic_add(PAGE_SIZE, &bh->b_count);
+                               atomic_add(b_allocd, &bh->b_count);
                        continue;
                }
                prev_bh = bh;
                bh = kmalloc(sizeof(struct idetape_bh), GFP_KERNEL);
                if (!bh) {
-                       free_page((unsigned long) b_data);
+                       free_pages((unsigned long) b_data, order);
                        goto abort;
                }
                bh->b_reqnext = NULL;
                bh->b_data = b_data;
-               bh->b_size = PAGE_SIZE;
+               bh->b_size = b_allocd;
                atomic_set(&bh->b_count, full ? bh->b_size : 0);
                prev_bh->b_reqnext = bh;
+
+               pages &= (order-1);
        }
+
        bh->b_size -= tape->excess_bh_size;
        if (full)
                atomic_sub(tape->excess_bh_size, &bh->b_count);
@@ -1837,7 +1851,7 @@ static int idetape_init_read(ide_drive_t *drive)
                                         " 0 now\n");
                        tape->merge_stage_size = 0;
                }
-               tape->merge_stage = __idetape_kmalloc_stage(tape, 0, 0);
+               tape->merge_stage = ide_tape_kmalloc_buffer(tape, 0, 0);
                if (!tape->merge_stage)
                        return -ENOMEM;
                tape->chrdev_dir = IDETAPE_DIR_READ;
@@ -2115,7 +2129,7 @@ static ssize_t idetape_chrdev_write(struct file *file, const char __user *buf,
                                "should be 0 now\n");
                        tape->merge_stage_size = 0;
                }
-               tape->merge_stage = __idetape_kmalloc_stage(tape, 0, 0);
+               tape->merge_stage = ide_tape_kmalloc_buffer(tape, 0, 0);
                if (!tape->merge_stage)
                        return -ENOMEM;
                tape->chrdev_dir = IDETAPE_DIR_WRITE;
@@ -2495,7 +2509,7 @@ static void idetape_write_release(ide_drive_t *drive, unsigned int minor)
        idetape_tape_t *tape = drive->driver_data;
 
        idetape_empty_write_pipeline(drive);
-       tape->merge_stage = __idetape_kmalloc_stage(tape, 1, 0);
+       tape->merge_stage = ide_tape_kmalloc_buffer(tape, 1, 0);
        if (tape->merge_stage != NULL) {
                idetape_pad_zeros(drive, tape->blk_size *
                                (tape->user_bs_factor - 1));