aboutsummaryrefslogtreecommitdiffstats
path: root/drivers/block
diff options
context:
space:
mode:
Diffstat (limited to 'drivers/block')
-rw-r--r--drivers/block/aoe/aoechr.c10
-rw-r--r--drivers/block/as-iosched.c27
-rw-r--r--drivers/block/cciss.c67
-rw-r--r--drivers/block/cciss.h4
-rw-r--r--drivers/block/cfq-iosched.c2097
-rw-r--r--drivers/block/deadline-iosched.c15
-rw-r--r--drivers/block/elevator.c9
-rw-r--r--drivers/block/genhd.c27
-rw-r--r--drivers/block/ioctl.c74
-rw-r--r--drivers/block/ll_rw_blk.c346
-rw-r--r--drivers/block/loop.c81
-rw-r--r--drivers/block/paride/pg.c14
-rw-r--r--drivers/block/paride/pt.c20
-rw-r--r--drivers/block/pktcdvd.c39
-rw-r--r--drivers/block/swim3.c10
-rw-r--r--drivers/block/sx8.c11
-rw-r--r--drivers/block/ub.c213
17 files changed, 2023 insertions, 1041 deletions
diff --git a/drivers/block/aoe/aoechr.c b/drivers/block/aoe/aoechr.c
index 14aeca3e2e8c..45a243096187 100644
--- a/drivers/block/aoe/aoechr.c
+++ b/drivers/block/aoe/aoechr.c
@@ -36,7 +36,7 @@ static int emsgs_head_idx, emsgs_tail_idx;
36static struct semaphore emsgs_sema; 36static struct semaphore emsgs_sema;
37static spinlock_t emsgs_lock; 37static spinlock_t emsgs_lock;
38static int nblocked_emsgs_readers; 38static int nblocked_emsgs_readers;
39static struct class_simple *aoe_class; 39static struct class *aoe_class;
40static struct aoe_chardev chardevs[] = { 40static struct aoe_chardev chardevs[] = {
41 { MINOR_ERR, "err" }, 41 { MINOR_ERR, "err" },
42 { MINOR_DISCOVER, "discover" }, 42 { MINOR_DISCOVER, "discover" },
@@ -218,13 +218,13 @@ aoechr_init(void)
218 } 218 }
219 sema_init(&emsgs_sema, 0); 219 sema_init(&emsgs_sema, 0);
220 spin_lock_init(&emsgs_lock); 220 spin_lock_init(&emsgs_lock);
221 aoe_class = class_simple_create(THIS_MODULE, "aoe"); 221 aoe_class = class_create(THIS_MODULE, "aoe");
222 if (IS_ERR(aoe_class)) { 222 if (IS_ERR(aoe_class)) {
223 unregister_chrdev(AOE_MAJOR, "aoechr"); 223 unregister_chrdev(AOE_MAJOR, "aoechr");
224 return PTR_ERR(aoe_class); 224 return PTR_ERR(aoe_class);
225 } 225 }
226 for (i = 0; i < ARRAY_SIZE(chardevs); ++i) 226 for (i = 0; i < ARRAY_SIZE(chardevs); ++i)
227 class_simple_device_add(aoe_class, 227 class_device_create(aoe_class,
228 MKDEV(AOE_MAJOR, chardevs[i].minor), 228 MKDEV(AOE_MAJOR, chardevs[i].minor),
229 NULL, chardevs[i].name); 229 NULL, chardevs[i].name);
230 230
@@ -237,8 +237,8 @@ aoechr_exit(void)
237 int i; 237 int i;
238 238
239 for (i = 0; i < ARRAY_SIZE(chardevs); ++i) 239 for (i = 0; i < ARRAY_SIZE(chardevs); ++i)
240 class_simple_device_remove(MKDEV(AOE_MAJOR, chardevs[i].minor)); 240 class_device_destroy(aoe_class, MKDEV(AOE_MAJOR, chardevs[i].minor));
241 class_simple_destroy(aoe_class); 241 class_destroy(aoe_class);
242 unregister_chrdev(AOE_MAJOR, "aoechr"); 242 unregister_chrdev(AOE_MAJOR, "aoechr");
243} 243}
244 244
diff --git a/drivers/block/as-iosched.c b/drivers/block/as-iosched.c
index a9575bb58a5e..95c0a3690b0f 100644
--- a/drivers/block/as-iosched.c
+++ b/drivers/block/as-iosched.c
@@ -1806,7 +1806,8 @@ static void as_put_request(request_queue_t *q, struct request *rq)
1806 rq->elevator_private = NULL; 1806 rq->elevator_private = NULL;
1807} 1807}
1808 1808
1809static int as_set_request(request_queue_t *q, struct request *rq, int gfp_mask) 1809static int as_set_request(request_queue_t *q, struct request *rq,
1810 struct bio *bio, int gfp_mask)
1810{ 1811{
1811 struct as_data *ad = q->elevator->elevator_data; 1812 struct as_data *ad = q->elevator->elevator_data;
1812 struct as_rq *arq = mempool_alloc(ad->arq_pool, gfp_mask); 1813 struct as_rq *arq = mempool_alloc(ad->arq_pool, gfp_mask);
@@ -1827,7 +1828,7 @@ static int as_set_request(request_queue_t *q, struct request *rq, int gfp_mask)
1827 return 1; 1828 return 1;
1828} 1829}
1829 1830
1830static int as_may_queue(request_queue_t *q, int rw) 1831static int as_may_queue(request_queue_t *q, int rw, struct bio *bio)
1831{ 1832{
1832 int ret = ELV_MQUEUE_MAY; 1833 int ret = ELV_MQUEUE_MAY;
1833 struct as_data *ad = q->elevator->elevator_data; 1834 struct as_data *ad = q->elevator->elevator_data;
@@ -1871,20 +1872,22 @@ static int as_init_queue(request_queue_t *q, elevator_t *e)
1871 if (!arq_pool) 1872 if (!arq_pool)
1872 return -ENOMEM; 1873 return -ENOMEM;
1873 1874
1874 ad = kmalloc(sizeof(*ad), GFP_KERNEL); 1875 ad = kmalloc_node(sizeof(*ad), GFP_KERNEL, q->node);
1875 if (!ad) 1876 if (!ad)
1876 return -ENOMEM; 1877 return -ENOMEM;
1877 memset(ad, 0, sizeof(*ad)); 1878 memset(ad, 0, sizeof(*ad));
1878 1879
1879 ad->q = q; /* Identify what queue the data belongs to */ 1880 ad->q = q; /* Identify what queue the data belongs to */
1880 1881
1881 ad->hash = kmalloc(sizeof(struct list_head)*AS_HASH_ENTRIES,GFP_KERNEL); 1882 ad->hash = kmalloc_node(sizeof(struct list_head)*AS_HASH_ENTRIES,
1883 GFP_KERNEL, q->node);
1882 if (!ad->hash) { 1884 if (!ad->hash) {
1883 kfree(ad); 1885 kfree(ad);
1884 return -ENOMEM; 1886 return -ENOMEM;
1885 } 1887 }
1886 1888
1887 ad->arq_pool = mempool_create(BLKDEV_MIN_RQ, mempool_alloc_slab, mempool_free_slab, arq_pool); 1889 ad->arq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
1890 mempool_free_slab, arq_pool, q->node);
1888 if (!ad->arq_pool) { 1891 if (!ad->arq_pool) {
1889 kfree(ad->hash); 1892 kfree(ad->hash);
1890 kfree(ad); 1893 kfree(ad);
@@ -1932,23 +1935,15 @@ struct as_fs_entry {
1932static ssize_t 1935static ssize_t
1933as_var_show(unsigned int var, char *page) 1936as_var_show(unsigned int var, char *page)
1934{ 1937{
1935 var = (var * 1000) / HZ;
1936 return sprintf(page, "%d\n", var); 1938 return sprintf(page, "%d\n", var);
1937} 1939}
1938 1940
1939static ssize_t 1941static ssize_t
1940as_var_store(unsigned long *var, const char *page, size_t count) 1942as_var_store(unsigned long *var, const char *page, size_t count)
1941{ 1943{
1942 unsigned long tmp;
1943 char *p = (char *) page; 1944 char *p = (char *) page;
1944 1945
1945 tmp = simple_strtoul(p, &p, 10); 1946 *var = simple_strtoul(p, &p, 10);
1946 if (tmp != 0) {
1947 tmp = (tmp * HZ) / 1000;
1948 if (tmp == 0)
1949 tmp = 1;
1950 }
1951 *var = tmp;
1952 return count; 1947 return count;
1953} 1948}
1954 1949
@@ -2044,7 +2039,7 @@ as_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2044 struct as_fs_entry *entry = to_as(attr); 2039 struct as_fs_entry *entry = to_as(attr);
2045 2040
2046 if (!entry->show) 2041 if (!entry->show)
2047 return 0; 2042 return -EIO;
2048 2043
2049 return entry->show(e->elevator_data, page); 2044 return entry->show(e->elevator_data, page);
2050} 2045}
@@ -2057,7 +2052,7 @@ as_attr_store(struct kobject *kobj, struct attribute *attr,
2057 struct as_fs_entry *entry = to_as(attr); 2052 struct as_fs_entry *entry = to_as(attr);
2058 2053
2059 if (!entry->store) 2054 if (!entry->store)
2060 return -EINVAL; 2055 return -EIO;
2061 2056
2062 return entry->store(e->elevator_data, page, length); 2057 return entry->store(e->elevator_data, page, length);
2063} 2058}
diff --git a/drivers/block/cciss.c b/drivers/block/cciss.c
index abde27027c06..418b1469d75d 100644
--- a/drivers/block/cciss.c
+++ b/drivers/block/cciss.c
@@ -1,6 +1,6 @@
1/* 1/*
2 * Disk Array driver for HP SA 5xxx and 6xxx Controllers 2 * Disk Array driver for HP SA 5xxx and 6xxx Controllers
3 * Copyright 2000, 2002 Hewlett-Packard Development Company, L.P. 3 * Copyright 2000, 2005 Hewlett-Packard Development Company, L.P.
4 * 4 *
5 * This program is free software; you can redistribute it and/or modify 5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by 6 * it under the terms of the GNU General Public License as published by
@@ -54,7 +54,7 @@
54MODULE_AUTHOR("Hewlett-Packard Company"); 54MODULE_AUTHOR("Hewlett-Packard Company");
55MODULE_DESCRIPTION("Driver for HP Controller SA5xxx SA6xxx version 2.6.6"); 55MODULE_DESCRIPTION("Driver for HP Controller SA5xxx SA6xxx version 2.6.6");
56MODULE_SUPPORTED_DEVICE("HP SA5i SA5i+ SA532 SA5300 SA5312 SA641 SA642 SA6400" 56MODULE_SUPPORTED_DEVICE("HP SA5i SA5i+ SA532 SA5300 SA5312 SA641 SA642 SA6400"
57 " SA6i P600 P800 E400"); 57 " SA6i P600 P800 E400 E300");
58MODULE_LICENSE("GPL"); 58MODULE_LICENSE("GPL");
59 59
60#include "cciss_cmd.h" 60#include "cciss_cmd.h"
@@ -85,8 +85,10 @@ static const struct pci_device_id cciss_pci_device_id[] = {
85 0x103C, 0x3225, 0, 0, 0}, 85 0x103C, 0x3225, 0, 0, 0},
86 { PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSB, 86 { PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSB,
87 0x103c, 0x3223, 0, 0, 0}, 87 0x103c, 0x3223, 0, 0, 0},
88 { PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSB, 88 { PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSC,
89 0x103c, 0x3231, 0, 0, 0}, 89 0x103c, 0x3231, 0, 0, 0},
90 { PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSC,
91 0x103c, 0x3233, 0, 0, 0},
90 {0,} 92 {0,}
91}; 93};
92MODULE_DEVICE_TABLE(pci, cciss_pci_device_id); 94MODULE_DEVICE_TABLE(pci, cciss_pci_device_id);
@@ -110,6 +112,7 @@ static struct board_type products[] = {
110 { 0x3225103C, "Smart Array P600", &SA5_access}, 112 { 0x3225103C, "Smart Array P600", &SA5_access},
111 { 0x3223103C, "Smart Array P800", &SA5_access}, 113 { 0x3223103C, "Smart Array P800", &SA5_access},
112 { 0x3231103C, "Smart Array E400", &SA5_access}, 114 { 0x3231103C, "Smart Array E400", &SA5_access},
115 { 0x3233103C, "Smart Array E300", &SA5_access},
113}; 116};
114 117
115/* How long to wait (in millesconds) for board to go into simple mode */ 118/* How long to wait (in millesconds) for board to go into simple mode */
@@ -635,6 +638,7 @@ static int cciss_ioctl(struct inode *inode, struct file *filep,
635 cciss_pci_info_struct pciinfo; 638 cciss_pci_info_struct pciinfo;
636 639
637 if (!arg) return -EINVAL; 640 if (!arg) return -EINVAL;
641 pciinfo.domain = pci_domain_nr(host->pdev->bus);
638 pciinfo.bus = host->pdev->bus->number; 642 pciinfo.bus = host->pdev->bus->number;
639 pciinfo.dev_fn = host->pdev->devfn; 643 pciinfo.dev_fn = host->pdev->devfn;
640 pciinfo.board_id = host->board_id; 644 pciinfo.board_id = host->board_id;
@@ -782,18 +786,10 @@ static int cciss_ioctl(struct inode *inode, struct file *filep,
782 786
783 case CCISS_GETLUNINFO: { 787 case CCISS_GETLUNINFO: {
784 LogvolInfo_struct luninfo; 788 LogvolInfo_struct luninfo;
785 int i;
786 789
787 luninfo.LunID = drv->LunID; 790 luninfo.LunID = drv->LunID;
788 luninfo.num_opens = drv->usage_count; 791 luninfo.num_opens = drv->usage_count;
789 luninfo.num_parts = 0; 792 luninfo.num_parts = 0;
790 /* count partitions 1 to 15 with sizes > 0 */
791 for (i = 0; i < MAX_PART - 1; i++) {
792 if (!disk->part[i])
793 continue;
794 if (disk->part[i]->nr_sects != 0)
795 luninfo.num_parts++;
796 }
797 if (copy_to_user(argp, &luninfo, 793 if (copy_to_user(argp, &luninfo,
798 sizeof(LogvolInfo_struct))) 794 sizeof(LogvolInfo_struct)))
799 return -EFAULT; 795 return -EFAULT;
@@ -1139,7 +1135,7 @@ static int revalidate_allvol(ctlr_info_t *host)
1139 /* this is for the online array utilities */ 1135 /* this is for the online array utilities */
1140 if (!drv->heads && i) 1136 if (!drv->heads && i)
1141 continue; 1137 continue;
1142 blk_queue_hardsect_size(host->queue, drv->block_size); 1138 blk_queue_hardsect_size(drv->queue, drv->block_size);
1143 set_capacity(disk, drv->nr_blocks); 1139 set_capacity(disk, drv->nr_blocks);
1144 add_disk(disk); 1140 add_disk(disk);
1145 } 1141 }
@@ -1695,7 +1691,7 @@ static int cciss_revalidate(struct gendisk *disk)
1695 cciss_read_capacity(h->ctlr, logvol, size_buff, 1, &total_size, &block_size); 1691 cciss_read_capacity(h->ctlr, logvol, size_buff, 1, &total_size, &block_size);
1696 cciss_geometry_inquiry(h->ctlr, logvol, 1, total_size, block_size, inq_buff, drv); 1692 cciss_geometry_inquiry(h->ctlr, logvol, 1, total_size, block_size, inq_buff, drv);
1697 1693
1698 blk_queue_hardsect_size(h->queue, drv->block_size); 1694 blk_queue_hardsect_size(drv->queue, drv->block_size);
1699 set_capacity(disk, drv->nr_blocks); 1695 set_capacity(disk, drv->nr_blocks);
1700 1696
1701 kfree(size_buff); 1697 kfree(size_buff);
@@ -2252,12 +2248,12 @@ static irqreturn_t do_cciss_intr(int irq, void *dev_id, struct pt_regs *regs)
2252 * them up. We will also keep track of the next queue to run so 2248 * them up. We will also keep track of the next queue to run so
2253 * that every queue gets a chance to be started first. 2249 * that every queue gets a chance to be started first.
2254 */ 2250 */
2255 for (j=0; j < NWD; j++){ 2251 for (j=0; j < h->highest_lun + 1; j++){
2256 int curr_queue = (start_queue + j) % NWD; 2252 int curr_queue = (start_queue + j) % (h->highest_lun + 1);
2257 /* make sure the disk has been added and the drive is real 2253 /* make sure the disk has been added and the drive is real
2258 * because this can be called from the middle of init_one. 2254 * because this can be called from the middle of init_one.
2259 */ 2255 */
2260 if(!(h->gendisk[curr_queue]->queue) || 2256 if(!(h->drv[curr_queue].queue) ||
2261 !(h->drv[curr_queue].heads)) 2257 !(h->drv[curr_queue].heads))
2262 continue; 2258 continue;
2263 blk_start_queue(h->gendisk[curr_queue]->queue); 2259 blk_start_queue(h->gendisk[curr_queue]->queue);
@@ -2268,14 +2264,14 @@ static irqreturn_t do_cciss_intr(int irq, void *dev_id, struct pt_regs *regs)
2268 if ((find_first_zero_bit(h->cmd_pool_bits, NR_CMDS)) == NR_CMDS) 2264 if ((find_first_zero_bit(h->cmd_pool_bits, NR_CMDS)) == NR_CMDS)
2269 { 2265 {
2270 if (curr_queue == start_queue){ 2266 if (curr_queue == start_queue){
2271 h->next_to_run = (start_queue + 1) % NWD; 2267 h->next_to_run = (start_queue + 1) % (h->highest_lun + 1);
2272 goto cleanup; 2268 goto cleanup;
2273 } else { 2269 } else {
2274 h->next_to_run = curr_queue; 2270 h->next_to_run = curr_queue;
2275 goto cleanup; 2271 goto cleanup;
2276 } 2272 }
2277 } else { 2273 } else {
2278 curr_queue = (curr_queue + 1) % NWD; 2274 curr_queue = (curr_queue + 1) % (h->highest_lun + 1);
2279 } 2275 }
2280 } 2276 }
2281 2277
@@ -2283,7 +2279,6 @@ cleanup:
2283 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags); 2279 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2284 return IRQ_HANDLED; 2280 return IRQ_HANDLED;
2285} 2281}
2286
2287/* 2282/*
2288 * We cannot read the structure directly, for portablity we must use 2283 * We cannot read the structure directly, for portablity we must use
2289 * the io functions. 2284 * the io functions.
@@ -2793,13 +2788,6 @@ static int __devinit cciss_init_one(struct pci_dev *pdev,
2793 } 2788 }
2794 2789
2795 spin_lock_init(&hba[i]->lock); 2790 spin_lock_init(&hba[i]->lock);
2796 q = blk_init_queue(do_cciss_request, &hba[i]->lock);
2797 if (!q)
2798 goto clean4;
2799
2800 q->backing_dev_info.ra_pages = READ_AHEAD;
2801 hba[i]->queue = q;
2802 q->queuedata = hba[i];
2803 2791
2804 /* Initialize the pdev driver private data. 2792 /* Initialize the pdev driver private data.
2805 have it point to hba[i]. */ 2793 have it point to hba[i]. */
@@ -2821,6 +2809,20 @@ static int __devinit cciss_init_one(struct pci_dev *pdev,
2821 2809
2822 cciss_procinit(i); 2810 cciss_procinit(i);
2823 2811
2812 for(j=0; j < NWD; j++) { /* mfm */
2813 drive_info_struct *drv = &(hba[i]->drv[j]);
2814 struct gendisk *disk = hba[i]->gendisk[j];
2815
2816 q = blk_init_queue(do_cciss_request, &hba[i]->lock);
2817 if (!q) {
2818 printk(KERN_ERR
2819 "cciss: unable to allocate queue for disk %d\n",
2820 j);
2821 break;
2822 }
2823 drv->queue = q;
2824
2825 q->backing_dev_info.ra_pages = READ_AHEAD;
2824 blk_queue_bounce_limit(q, hba[i]->pdev->dma_mask); 2826 blk_queue_bounce_limit(q, hba[i]->pdev->dma_mask);
2825 2827
2826 /* This is a hardware imposed limit. */ 2828 /* This is a hardware imposed limit. */
@@ -2831,26 +2833,23 @@ static int __devinit cciss_init_one(struct pci_dev *pdev,
2831 2833
2832 blk_queue_max_sectors(q, 512); 2834 blk_queue_max_sectors(q, 512);
2833 2835
2834 2836 q->queuedata = hba[i];
2835 for(j=0; j<NWD; j++) {
2836 drive_info_struct *drv = &(hba[i]->drv[j]);
2837 struct gendisk *disk = hba[i]->gendisk[j];
2838
2839 sprintf(disk->disk_name, "cciss/c%dd%d", i, j); 2837 sprintf(disk->disk_name, "cciss/c%dd%d", i, j);
2840 sprintf(disk->devfs_name, "cciss/host%d/target%d", i, j); 2838 sprintf(disk->devfs_name, "cciss/host%d/target%d", i, j);
2841 disk->major = hba[i]->major; 2839 disk->major = hba[i]->major;
2842 disk->first_minor = j << NWD_SHIFT; 2840 disk->first_minor = j << NWD_SHIFT;
2843 disk->fops = &cciss_fops; 2841 disk->fops = &cciss_fops;
2844 disk->queue = hba[i]->queue; 2842 disk->queue = q;
2845 disk->private_data = drv; 2843 disk->private_data = drv;
2846 /* we must register the controller even if no disks exist */ 2844 /* we must register the controller even if no disks exist */
2847 /* this is for the online array utilities */ 2845 /* this is for the online array utilities */
2848 if(!drv->heads && j) 2846 if(!drv->heads && j)
2849 continue; 2847 continue;
2850 blk_queue_hardsect_size(hba[i]->queue, drv->block_size); 2848 blk_queue_hardsect_size(q, drv->block_size);
2851 set_capacity(disk, drv->nr_blocks); 2849 set_capacity(disk, drv->nr_blocks);
2852 add_disk(disk); 2850 add_disk(disk);
2853 } 2851 }
2852
2854 return(1); 2853 return(1);
2855 2854
2856clean4: 2855clean4:
@@ -2916,10 +2915,10 @@ static void __devexit cciss_remove_one (struct pci_dev *pdev)
2916 for (j = 0; j < NWD; j++) { 2915 for (j = 0; j < NWD; j++) {
2917 struct gendisk *disk = hba[i]->gendisk[j]; 2916 struct gendisk *disk = hba[i]->gendisk[j];
2918 if (disk->flags & GENHD_FL_UP) 2917 if (disk->flags & GENHD_FL_UP)
2918 blk_cleanup_queue(disk->queue);
2919 del_gendisk(disk); 2919 del_gendisk(disk);
2920 } 2920 }
2921 2921
2922 blk_cleanup_queue(hba[i]->queue);
2923 pci_free_consistent(hba[i]->pdev, NR_CMDS * sizeof(CommandList_struct), 2922 pci_free_consistent(hba[i]->pdev, NR_CMDS * sizeof(CommandList_struct),
2924 hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle); 2923 hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
2925 pci_free_consistent(hba[i]->pdev, NR_CMDS * sizeof( ErrorInfo_struct), 2924 pci_free_consistent(hba[i]->pdev, NR_CMDS * sizeof( ErrorInfo_struct),
diff --git a/drivers/block/cciss.h b/drivers/block/cciss.h
index 8fb19206eddb..566587d0a500 100644
--- a/drivers/block/cciss.h
+++ b/drivers/block/cciss.h
@@ -29,6 +29,7 @@ typedef struct _drive_info_struct
29{ 29{
30 __u32 LunID; 30 __u32 LunID;
31 int usage_count; 31 int usage_count;
32 struct request_queue *queue;
32 sector_t nr_blocks; 33 sector_t nr_blocks;
33 int block_size; 34 int block_size;
34 int heads; 35 int heads;
@@ -72,7 +73,6 @@ struct ctlr_info
72 unsigned int maxQsinceinit; 73 unsigned int maxQsinceinit;
73 unsigned int maxSG; 74 unsigned int maxSG;
74 spinlock_t lock; 75 spinlock_t lock;
75 struct request_queue *queue;
76 76
77 //* pointers to command and error info pool */ 77 //* pointers to command and error info pool */
78 CommandList_struct *cmd_pool; 78 CommandList_struct *cmd_pool;
@@ -260,7 +260,7 @@ struct board_type {
260 struct access_method *access; 260 struct access_method *access;
261}; 261};
262 262
263#define CCISS_LOCK(i) (hba[i]->queue->queue_lock) 263#define CCISS_LOCK(i) (&hba[i]->lock)
264 264
265#endif /* CCISS_H */ 265#endif /* CCISS_H */
266 266
diff --git a/drivers/block/cfq-iosched.c b/drivers/block/cfq-iosched.c
index 2210bacad56a..cd056e7e64ec 100644
--- a/drivers/block/cfq-iosched.c
+++ b/drivers/block/cfq-iosched.c
@@ -21,22 +21,34 @@
21#include <linux/hash.h> 21#include <linux/hash.h>
22#include <linux/rbtree.h> 22#include <linux/rbtree.h>
23#include <linux/mempool.h> 23#include <linux/mempool.h>
24 24#include <linux/ioprio.h>
25static unsigned long max_elapsed_crq; 25#include <linux/writeback.h>
26static unsigned long max_elapsed_dispatch;
27 26
28/* 27/*
29 * tunables 28 * tunables
30 */ 29 */
31static int cfq_quantum = 4; /* max queue in one round of service */ 30static int cfq_quantum = 4; /* max queue in one round of service */
32static int cfq_queued = 8; /* minimum rq allocate limit per-queue*/ 31static int cfq_queued = 8; /* minimum rq allocate limit per-queue*/
33static int cfq_service = HZ; /* period over which service is avg */ 32static int cfq_fifo_expire[2] = { HZ / 4, HZ / 8 };
34static int cfq_fifo_expire_r = HZ / 2; /* fifo timeout for sync requests */
35static int cfq_fifo_expire_w = 5 * HZ; /* fifo timeout for async requests */
36static int cfq_fifo_rate = HZ / 8; /* fifo expiry rate */
37static int cfq_back_max = 16 * 1024; /* maximum backwards seek, in KiB */ 33static int cfq_back_max = 16 * 1024; /* maximum backwards seek, in KiB */
38static int cfq_back_penalty = 2; /* penalty of a backwards seek */ 34static int cfq_back_penalty = 2; /* penalty of a backwards seek */
39 35
36static int cfq_slice_sync = HZ / 10;
37static int cfq_slice_async = HZ / 25;
38static int cfq_slice_async_rq = 2;
39static int cfq_slice_idle = HZ / 100;
40
41#define CFQ_IDLE_GRACE (HZ / 10)
42#define CFQ_SLICE_SCALE (5)
43
44#define CFQ_KEY_ASYNC (0)
45#define CFQ_KEY_ANY (0xffff)
46
47/*
48 * disable queueing at the driver/hardware level
49 */
50static int cfq_max_depth = 2;
51
40/* 52/*
41 * for the hash of cfqq inside the cfqd 53 * for the hash of cfqq inside the cfqd
42 */ 54 */
@@ -55,6 +67,7 @@ static int cfq_back_penalty = 2; /* penalty of a backwards seek */
55#define list_entry_hash(ptr) hlist_entry((ptr), struct cfq_rq, hash) 67#define list_entry_hash(ptr) hlist_entry((ptr), struct cfq_rq, hash)
56 68
57#define list_entry_cfqq(ptr) list_entry((ptr), struct cfq_queue, cfq_list) 69#define list_entry_cfqq(ptr) list_entry((ptr), struct cfq_queue, cfq_list)
70#define list_entry_fifo(ptr) list_entry((ptr), struct request, queuelist)
58 71
59#define RQ_DATA(rq) (rq)->elevator_private 72#define RQ_DATA(rq) (rq)->elevator_private
60 73
@@ -75,78 +88,110 @@ static int cfq_back_penalty = 2; /* penalty of a backwards seek */
75#define rb_entry_crq(node) rb_entry((node), struct cfq_rq, rb_node) 88#define rb_entry_crq(node) rb_entry((node), struct cfq_rq, rb_node)
76#define rq_rb_key(rq) (rq)->sector 89#define rq_rb_key(rq) (rq)->sector
77 90
78/*
79 * threshold for switching off non-tag accounting
80 */
81#define CFQ_MAX_TAG (4)
82
83/*
84 * sort key types and names
85 */
86enum {
87 CFQ_KEY_PGID,
88 CFQ_KEY_TGID,
89 CFQ_KEY_UID,
90 CFQ_KEY_GID,
91 CFQ_KEY_LAST,
92};
93
94static char *cfq_key_types[] = { "pgid", "tgid", "uid", "gid", NULL };
95
96static kmem_cache_t *crq_pool; 91static kmem_cache_t *crq_pool;
97static kmem_cache_t *cfq_pool; 92static kmem_cache_t *cfq_pool;
98static kmem_cache_t *cfq_ioc_pool; 93static kmem_cache_t *cfq_ioc_pool;
99 94
95#define CFQ_PRIO_LISTS IOPRIO_BE_NR
96#define cfq_class_idle(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_IDLE)
97#define cfq_class_be(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_BE)
98#define cfq_class_rt(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_RT)
99
100#define ASYNC (0)
101#define SYNC (1)
102
103#define cfq_cfqq_dispatched(cfqq) \
104 ((cfqq)->on_dispatch[ASYNC] + (cfqq)->on_dispatch[SYNC])
105
106#define cfq_cfqq_class_sync(cfqq) ((cfqq)->key != CFQ_KEY_ASYNC)
107
108#define cfq_cfqq_sync(cfqq) \
109 (cfq_cfqq_class_sync(cfqq) || (cfqq)->on_dispatch[SYNC])
110
111/*
112 * Per block device queue structure
113 */
100struct cfq_data { 114struct cfq_data {
101 struct list_head rr_list; 115 atomic_t ref;
116 request_queue_t *queue;
117
118 /*
119 * rr list of queues with requests and the count of them
120 */
121 struct list_head rr_list[CFQ_PRIO_LISTS];
122 struct list_head busy_rr;
123 struct list_head cur_rr;
124 struct list_head idle_rr;
125 unsigned int busy_queues;
126
127 /*
128 * non-ordered list of empty cfqq's
129 */
102 struct list_head empty_list; 130 struct list_head empty_list;
103 131
132 /*
133 * cfqq lookup hash
134 */
104 struct hlist_head *cfq_hash; 135 struct hlist_head *cfq_hash;
105 struct hlist_head *crq_hash;
106 136
107 /* queues on rr_list (ie they have pending requests */ 137 /*
108 unsigned int busy_queues; 138 * global crq hash for all queues
139 */
140 struct hlist_head *crq_hash;
109 141
110 unsigned int max_queued; 142 unsigned int max_queued;
111 143
112 atomic_t ref; 144 mempool_t *crq_pool;
113 145
114 int key_type; 146 int rq_in_driver;
115 147
116 mempool_t *crq_pool; 148 /*
149 * schedule slice state info
150 */
151 /*
152 * idle window management
153 */
154 struct timer_list idle_slice_timer;
155 struct work_struct unplug_work;
117 156
118 request_queue_t *queue; 157 struct cfq_queue *active_queue;
158 struct cfq_io_context *active_cic;
159 int cur_prio, cur_end_prio;
160 unsigned int dispatch_slice;
161
162 struct timer_list idle_class_timer;
119 163
120 sector_t last_sector; 164 sector_t last_sector;
165 unsigned long last_end_request;
121 166
122 int rq_in_driver; 167 unsigned int rq_starved;
123 168
124 /* 169 /*
125 * tunables, see top of file 170 * tunables, see top of file
126 */ 171 */
127 unsigned int cfq_quantum; 172 unsigned int cfq_quantum;
128 unsigned int cfq_queued; 173 unsigned int cfq_queued;
129 unsigned int cfq_fifo_expire_r; 174 unsigned int cfq_fifo_expire[2];
130 unsigned int cfq_fifo_expire_w;
131 unsigned int cfq_fifo_batch_expire;
132 unsigned int cfq_back_penalty; 175 unsigned int cfq_back_penalty;
133 unsigned int cfq_back_max; 176 unsigned int cfq_back_max;
134 unsigned int find_best_crq; 177 unsigned int cfq_slice[2];
135 178 unsigned int cfq_slice_async_rq;
136 unsigned int cfq_tagged; 179 unsigned int cfq_slice_idle;
180 unsigned int cfq_max_depth;
137}; 181};
138 182
183/*
184 * Per process-grouping structure
185 */
139struct cfq_queue { 186struct cfq_queue {
140 /* reference count */ 187 /* reference count */
141 atomic_t ref; 188 atomic_t ref;
142 /* parent cfq_data */ 189 /* parent cfq_data */
143 struct cfq_data *cfqd; 190 struct cfq_data *cfqd;
144 /* hash of mergeable requests */ 191 /* cfqq lookup hash */
145 struct hlist_node cfq_hash; 192 struct hlist_node cfq_hash;
146 /* hash key */ 193 /* hash key */
147 unsigned long key; 194 unsigned int key;
148 /* whether queue is on rr (or empty) list */
149 int on_rr;
150 /* on either rr or empty list of cfqd */ 195 /* on either rr or empty list of cfqd */
151 struct list_head cfq_list; 196 struct list_head cfq_list;
152 /* sorted list of pending requests */ 197 /* sorted list of pending requests */
@@ -158,21 +203,22 @@ struct cfq_queue {
158 /* currently allocated requests */ 203 /* currently allocated requests */
159 int allocated[2]; 204 int allocated[2];
160 /* fifo list of requests in sort_list */ 205 /* fifo list of requests in sort_list */
161 struct list_head fifo[2]; 206 struct list_head fifo;
162 /* last time fifo expired */
163 unsigned long last_fifo_expire;
164 207
165 int key_type; 208 unsigned long slice_start;
209 unsigned long slice_end;
210 unsigned long slice_left;
211 unsigned long service_last;
166 212
167 unsigned long service_start; 213 /* number of requests that are on the dispatch list */
168 unsigned long service_used; 214 int on_dispatch[2];
169 215
170 unsigned int max_rate; 216 /* io prio of this group */
217 unsigned short ioprio, org_ioprio;
218 unsigned short ioprio_class, org_ioprio_class;
171 219
172 /* number of requests that have been handed to the driver */ 220 /* various state flags, see below */
173 int in_flight; 221 unsigned int flags;
174 /* number of currently allocated requests */
175 int alloc_limit[2];
176}; 222};
177 223
178struct cfq_rq { 224struct cfq_rq {
@@ -184,42 +230,78 @@ struct cfq_rq {
184 struct cfq_queue *cfq_queue; 230 struct cfq_queue *cfq_queue;
185 struct cfq_io_context *io_context; 231 struct cfq_io_context *io_context;
186 232
187 unsigned long service_start; 233 unsigned int crq_flags;
188 unsigned long queue_start; 234};
235
236enum cfqq_state_flags {
237 CFQ_CFQQ_FLAG_on_rr = 0,
238 CFQ_CFQQ_FLAG_wait_request,
239 CFQ_CFQQ_FLAG_must_alloc,
240 CFQ_CFQQ_FLAG_must_alloc_slice,
241 CFQ_CFQQ_FLAG_must_dispatch,
242 CFQ_CFQQ_FLAG_fifo_expire,
243 CFQ_CFQQ_FLAG_idle_window,
244 CFQ_CFQQ_FLAG_prio_changed,
245 CFQ_CFQQ_FLAG_expired,
246};
189 247
190 unsigned int in_flight : 1; 248#define CFQ_CFQQ_FNS(name) \
191 unsigned int accounted : 1; 249static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq) \
192 unsigned int is_sync : 1; 250{ \
193 unsigned int is_write : 1; 251 cfqq->flags |= (1 << CFQ_CFQQ_FLAG_##name); \
252} \
253static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq) \
254{ \
255 cfqq->flags &= ~(1 << CFQ_CFQQ_FLAG_##name); \
256} \
257static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq) \
258{ \
259 return (cfqq->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0; \
260}
261
262CFQ_CFQQ_FNS(on_rr);
263CFQ_CFQQ_FNS(wait_request);
264CFQ_CFQQ_FNS(must_alloc);
265CFQ_CFQQ_FNS(must_alloc_slice);
266CFQ_CFQQ_FNS(must_dispatch);
267CFQ_CFQQ_FNS(fifo_expire);
268CFQ_CFQQ_FNS(idle_window);
269CFQ_CFQQ_FNS(prio_changed);
270CFQ_CFQQ_FNS(expired);
271#undef CFQ_CFQQ_FNS
272
273enum cfq_rq_state_flags {
274 CFQ_CRQ_FLAG_in_flight = 0,
275 CFQ_CRQ_FLAG_in_driver,
276 CFQ_CRQ_FLAG_is_sync,
277 CFQ_CRQ_FLAG_requeued,
194}; 278};
195 279
196static struct cfq_queue *cfq_find_cfq_hash(struct cfq_data *, unsigned long); 280#define CFQ_CRQ_FNS(name) \
281static inline void cfq_mark_crq_##name(struct cfq_rq *crq) \
282{ \
283 crq->crq_flags |= (1 << CFQ_CRQ_FLAG_##name); \
284} \
285static inline void cfq_clear_crq_##name(struct cfq_rq *crq) \
286{ \
287 crq->crq_flags &= ~(1 << CFQ_CRQ_FLAG_##name); \
288} \
289static inline int cfq_crq_##name(const struct cfq_rq *crq) \
290{ \
291 return (crq->crq_flags & (1 << CFQ_CRQ_FLAG_##name)) != 0; \
292}
293
294CFQ_CRQ_FNS(in_flight);
295CFQ_CRQ_FNS(in_driver);
296CFQ_CRQ_FNS(is_sync);
297CFQ_CRQ_FNS(requeued);
298#undef CFQ_CRQ_FNS
299
300static struct cfq_queue *cfq_find_cfq_hash(struct cfq_data *, unsigned int, unsigned short);
197static void cfq_dispatch_sort(request_queue_t *, struct cfq_rq *); 301static void cfq_dispatch_sort(request_queue_t *, struct cfq_rq *);
198static void cfq_update_next_crq(struct cfq_rq *);
199static void cfq_put_cfqd(struct cfq_data *cfqd); 302static void cfq_put_cfqd(struct cfq_data *cfqd);
200 303
201/* 304#define process_sync(tsk) ((tsk)->flags & PF_SYNCWRITE)
202 * what the fairness is based on (ie how processes are grouped and
203 * differentiated)
204 */
205static inline unsigned long
206cfq_hash_key(struct cfq_data *cfqd, struct task_struct *tsk)
207{
208 /*
209 * optimize this so that ->key_type is the offset into the struct
210 */
211 switch (cfqd->key_type) {
212 case CFQ_KEY_PGID:
213 return process_group(tsk);
214 default:
215 case CFQ_KEY_TGID:
216 return tsk->tgid;
217 case CFQ_KEY_UID:
218 return tsk->uid;
219 case CFQ_KEY_GID:
220 return tsk->gid;
221 }
222}
223 305
224/* 306/*
225 * lots of deadline iosched dupes, can be abstracted later... 307 * lots of deadline iosched dupes, can be abstracted later...
@@ -235,16 +317,12 @@ static void cfq_remove_merge_hints(request_queue_t *q, struct cfq_rq *crq)
235 317
236 if (q->last_merge == crq->request) 318 if (q->last_merge == crq->request)
237 q->last_merge = NULL; 319 q->last_merge = NULL;
238
239 cfq_update_next_crq(crq);
240} 320}
241 321
242static inline void cfq_add_crq_hash(struct cfq_data *cfqd, struct cfq_rq *crq) 322static inline void cfq_add_crq_hash(struct cfq_data *cfqd, struct cfq_rq *crq)
243{ 323{
244 const int hash_idx = CFQ_MHASH_FN(rq_hash_key(crq->request)); 324 const int hash_idx = CFQ_MHASH_FN(rq_hash_key(crq->request));
245 325
246 BUG_ON(!hlist_unhashed(&crq->hash));
247
248 hlist_add_head(&crq->hash, &cfqd->crq_hash[hash_idx]); 326 hlist_add_head(&crq->hash, &cfqd->crq_hash[hash_idx]);
249} 327}
250 328
@@ -257,8 +335,6 @@ static struct request *cfq_find_rq_hash(struct cfq_data *cfqd, sector_t offset)
257 struct cfq_rq *crq = list_entry_hash(entry); 335 struct cfq_rq *crq = list_entry_hash(entry);
258 struct request *__rq = crq->request; 336 struct request *__rq = crq->request;
259 337
260 BUG_ON(hlist_unhashed(&crq->hash));
261
262 if (!rq_mergeable(__rq)) { 338 if (!rq_mergeable(__rq)) {
263 cfq_del_crq_hash(crq); 339 cfq_del_crq_hash(crq);
264 continue; 340 continue;
@@ -271,6 +347,28 @@ static struct request *cfq_find_rq_hash(struct cfq_data *cfqd, sector_t offset)
271 return NULL; 347 return NULL;
272} 348}
273 349
350static inline int cfq_pending_requests(struct cfq_data *cfqd)
351{
352 return !list_empty(&cfqd->queue->queue_head) || cfqd->busy_queues;
353}
354
355/*
356 * scheduler run of queue, if there are requests pending and no one in the
357 * driver that will restart queueing
358 */
359static inline void cfq_schedule_dispatch(struct cfq_data *cfqd)
360{
361 if (!cfqd->rq_in_driver && cfq_pending_requests(cfqd))
362 kblockd_schedule_work(&cfqd->unplug_work);
363}
364
365static int cfq_queue_empty(request_queue_t *q)
366{
367 struct cfq_data *cfqd = q->elevator->elevator_data;
368
369 return !cfq_pending_requests(cfqd);
370}
371
274/* 372/*
275 * Lifted from AS - choose which of crq1 and crq2 that is best served now. 373 * Lifted from AS - choose which of crq1 and crq2 that is best served now.
276 * We choose the request that is closest to the head right now. Distance 374 * We choose the request that is closest to the head right now. Distance
@@ -288,35 +386,21 @@ cfq_choose_req(struct cfq_data *cfqd, struct cfq_rq *crq1, struct cfq_rq *crq2)
288 if (crq2 == NULL) 386 if (crq2 == NULL)
289 return crq1; 387 return crq1;
290 388
389 if (cfq_crq_requeued(crq1) && !cfq_crq_requeued(crq2))
390 return crq1;
391 else if (cfq_crq_requeued(crq2) && !cfq_crq_requeued(crq1))
392 return crq2;
393
394 if (cfq_crq_is_sync(crq1) && !cfq_crq_is_sync(crq2))
395 return crq1;
396 else if (cfq_crq_is_sync(crq2) && !cfq_crq_is_sync(crq1))
397 return crq2;
398
291 s1 = crq1->request->sector; 399 s1 = crq1->request->sector;
292 s2 = crq2->request->sector; 400 s2 = crq2->request->sector;
293 401
294 last = cfqd->last_sector; 402 last = cfqd->last_sector;
295 403
296#if 0
297 if (!list_empty(&cfqd->queue->queue_head)) {
298 struct list_head *entry = &cfqd->queue->queue_head;
299 unsigned long distance = ~0UL;
300 struct request *rq;
301
302 while ((entry = entry->prev) != &cfqd->queue->queue_head) {
303 rq = list_entry_rq(entry);
304
305 if (blk_barrier_rq(rq))
306 break;
307
308 if (distance < abs(s1 - rq->sector + rq->nr_sectors)) {
309 distance = abs(s1 - rq->sector +rq->nr_sectors);
310 last = rq->sector + rq->nr_sectors;
311 }
312 if (distance < abs(s2 - rq->sector + rq->nr_sectors)) {
313 distance = abs(s2 - rq->sector +rq->nr_sectors);
314 last = rq->sector + rq->nr_sectors;
315 }
316 }
317 }
318#endif
319
320 /* 404 /*
321 * by definition, 1KiB is 2 sectors 405 * by definition, 1KiB is 2 sectors
322 */ 406 */
@@ -377,11 +461,14 @@ cfq_find_next_crq(struct cfq_data *cfqd, struct cfq_queue *cfqq,
377 struct cfq_rq *crq_next = NULL, *crq_prev = NULL; 461 struct cfq_rq *crq_next = NULL, *crq_prev = NULL;
378 struct rb_node *rbnext, *rbprev; 462 struct rb_node *rbnext, *rbprev;
379 463
380 if (!ON_RB(&last->rb_node)) 464 rbnext = NULL;
381 return NULL; 465 if (ON_RB(&last->rb_node))
382 466 rbnext = rb_next(&last->rb_node);
383 if ((rbnext = rb_next(&last->rb_node)) == NULL) 467 if (!rbnext) {
384 rbnext = rb_first(&cfqq->sort_list); 468 rbnext = rb_first(&cfqq->sort_list);
469 if (rbnext == &last->rb_node)
470 rbnext = NULL;
471 }
385 472
386 rbprev = rb_prev(&last->rb_node); 473 rbprev = rb_prev(&last->rb_node);
387 474
@@ -401,67 +488,53 @@ static void cfq_update_next_crq(struct cfq_rq *crq)
401 cfqq->next_crq = cfq_find_next_crq(cfqq->cfqd, cfqq, crq); 488 cfqq->next_crq = cfq_find_next_crq(cfqq->cfqd, cfqq, crq);
402} 489}
403 490
404static int cfq_check_sort_rr_list(struct cfq_queue *cfqq) 491static void cfq_resort_rr_list(struct cfq_queue *cfqq, int preempted)
405{ 492{
406 struct list_head *head = &cfqq->cfqd->rr_list; 493 struct cfq_data *cfqd = cfqq->cfqd;
407 struct list_head *next, *prev; 494 struct list_head *list, *entry;
408
409 /*
410 * list might still be ordered
411 */
412 next = cfqq->cfq_list.next;
413 if (next != head) {
414 struct cfq_queue *cnext = list_entry_cfqq(next);
415 495
416 if (cfqq->service_used > cnext->service_used) 496 BUG_ON(!cfq_cfqq_on_rr(cfqq));
417 return 1;
418 }
419 497
420 prev = cfqq->cfq_list.prev; 498 list_del(&cfqq->cfq_list);
421 if (prev != head) {
422 struct cfq_queue *cprev = list_entry_cfqq(prev);
423 499
424 if (cfqq->service_used < cprev->service_used) 500 if (cfq_class_rt(cfqq))
425 return 1; 501 list = &cfqd->cur_rr;
502 else if (cfq_class_idle(cfqq))
503 list = &cfqd->idle_rr;
504 else {
505 /*
506 * if cfqq has requests in flight, don't allow it to be
507 * found in cfq_set_active_queue before it has finished them.
508 * this is done to increase fairness between a process that
509 * has lots of io pending vs one that only generates one
510 * sporadically or synchronously
511 */
512 if (cfq_cfqq_dispatched(cfqq))
513 list = &cfqd->busy_rr;
514 else
515 list = &cfqd->rr_list[cfqq->ioprio];
426 } 516 }
427 517
428 return 0; 518 /*
429} 519 * if queue was preempted, just add to front to be fair. busy_rr
430 520 * isn't sorted.
431static void cfq_sort_rr_list(struct cfq_queue *cfqq, int new_queue) 521 */
432{ 522 if (preempted || list == &cfqd->busy_rr) {
433 struct list_head *entry = &cfqq->cfqd->rr_list; 523 list_add(&cfqq->cfq_list, list);
434
435 if (!cfqq->on_rr)
436 return;
437 if (!new_queue && !cfq_check_sort_rr_list(cfqq))
438 return; 524 return;
439 525 }
440 list_del(&cfqq->cfq_list);
441 526
442 /* 527 /*
443 * sort by our mean service_used, sub-sort by in-flight requests 528 * sort by when queue was last serviced
444 */ 529 */
445 while ((entry = entry->prev) != &cfqq->cfqd->rr_list) { 530 entry = list;
531 while ((entry = entry->prev) != list) {
446 struct cfq_queue *__cfqq = list_entry_cfqq(entry); 532 struct cfq_queue *__cfqq = list_entry_cfqq(entry);
447 533
448 if (cfqq->service_used > __cfqq->service_used) 534 if (!__cfqq->service_last)
535 break;
536 if (time_before(__cfqq->service_last, cfqq->service_last))
449 break; 537 break;
450 else if (cfqq->service_used == __cfqq->service_used) {
451 struct list_head *prv;
452
453 while ((prv = entry->prev) != &cfqq->cfqd->rr_list) {
454 __cfqq = list_entry_cfqq(prv);
455
456 WARN_ON(__cfqq->service_used > cfqq->service_used);
457 if (cfqq->service_used != __cfqq->service_used)
458 break;
459 if (cfqq->in_flight > __cfqq->in_flight)
460 break;
461
462 entry = prv;
463 }
464 }
465 } 538 }
466 539
467 list_add(&cfqq->cfq_list, entry); 540 list_add(&cfqq->cfq_list, entry);
@@ -469,28 +542,24 @@ static void cfq_sort_rr_list(struct cfq_queue *cfqq, int new_queue)
469 542
470/* 543/*
471 * add to busy list of queues for service, trying to be fair in ordering 544 * add to busy list of queues for service, trying to be fair in ordering
472 * the pending list according to requests serviced 545 * the pending list according to last request service
473 */ 546 */
474static inline void 547static inline void
475cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) 548cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq, int requeue)
476{ 549{
477 /* 550 BUG_ON(cfq_cfqq_on_rr(cfqq));
478 * it's currently on the empty list 551 cfq_mark_cfqq_on_rr(cfqq);
479 */
480 cfqq->on_rr = 1;
481 cfqd->busy_queues++; 552 cfqd->busy_queues++;
482 553
483 if (time_after(jiffies, cfqq->service_start + cfq_service)) 554 cfq_resort_rr_list(cfqq, requeue);
484 cfqq->service_used >>= 3;
485
486 cfq_sort_rr_list(cfqq, 1);
487} 555}
488 556
489static inline void 557static inline void
490cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) 558cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
491{ 559{
560 BUG_ON(!cfq_cfqq_on_rr(cfqq));
561 cfq_clear_cfqq_on_rr(cfqq);
492 list_move(&cfqq->cfq_list, &cfqd->empty_list); 562 list_move(&cfqq->cfq_list, &cfqd->empty_list);
493 cfqq->on_rr = 0;
494 563
495 BUG_ON(!cfqd->busy_queues); 564 BUG_ON(!cfqd->busy_queues);
496 cfqd->busy_queues--; 565 cfqd->busy_queues--;
@@ -505,16 +574,17 @@ static inline void cfq_del_crq_rb(struct cfq_rq *crq)
505 574
506 if (ON_RB(&crq->rb_node)) { 575 if (ON_RB(&crq->rb_node)) {
507 struct cfq_data *cfqd = cfqq->cfqd; 576 struct cfq_data *cfqd = cfqq->cfqd;
577 const int sync = cfq_crq_is_sync(crq);
508 578
509 BUG_ON(!cfqq->queued[crq->is_sync]); 579 BUG_ON(!cfqq->queued[sync]);
580 cfqq->queued[sync]--;
510 581
511 cfq_update_next_crq(crq); 582 cfq_update_next_crq(crq);
512 583
513 cfqq->queued[crq->is_sync]--;
514 rb_erase(&crq->rb_node, &cfqq->sort_list); 584 rb_erase(&crq->rb_node, &cfqq->sort_list);
515 RB_CLEAR_COLOR(&crq->rb_node); 585 RB_CLEAR_COLOR(&crq->rb_node);
516 586
517 if (RB_EMPTY(&cfqq->sort_list) && cfqq->on_rr) 587 if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY(&cfqq->sort_list))
518 cfq_del_cfqq_rr(cfqd, cfqq); 588 cfq_del_cfqq_rr(cfqd, cfqq);
519 } 589 }
520} 590}
@@ -550,7 +620,7 @@ static void cfq_add_crq_rb(struct cfq_rq *crq)
550 struct cfq_rq *__alias; 620 struct cfq_rq *__alias;
551 621
552 crq->rb_key = rq_rb_key(rq); 622 crq->rb_key = rq_rb_key(rq);
553 cfqq->queued[crq->is_sync]++; 623 cfqq->queued[cfq_crq_is_sync(crq)]++;
554 624
555 /* 625 /*
556 * looks a little odd, but the first insert might return an alias. 626 * looks a little odd, but the first insert might return an alias.
@@ -561,8 +631,8 @@ static void cfq_add_crq_rb(struct cfq_rq *crq)
561 631
562 rb_insert_color(&crq->rb_node, &cfqq->sort_list); 632 rb_insert_color(&crq->rb_node, &cfqq->sort_list);
563 633
564 if (!cfqq->on_rr) 634 if (!cfq_cfqq_on_rr(cfqq))
565 cfq_add_cfqq_rr(cfqd, cfqq); 635 cfq_add_cfqq_rr(cfqd, cfqq, cfq_crq_requeued(crq));
566 636
567 /* 637 /*
568 * check if this request is a better next-serve candidate 638 * check if this request is a better next-serve candidate
@@ -575,17 +645,16 @@ cfq_reposition_crq_rb(struct cfq_queue *cfqq, struct cfq_rq *crq)
575{ 645{
576 if (ON_RB(&crq->rb_node)) { 646 if (ON_RB(&crq->rb_node)) {
577 rb_erase(&crq->rb_node, &cfqq->sort_list); 647 rb_erase(&crq->rb_node, &cfqq->sort_list);
578 cfqq->queued[crq->is_sync]--; 648 cfqq->queued[cfq_crq_is_sync(crq)]--;
579 } 649 }
580 650
581 cfq_add_crq_rb(crq); 651 cfq_add_crq_rb(crq);
582} 652}
583 653
584static struct request * 654static struct request *cfq_find_rq_rb(struct cfq_data *cfqd, sector_t sector)
585cfq_find_rq_rb(struct cfq_data *cfqd, sector_t sector) 655
586{ 656{
587 const unsigned long key = cfq_hash_key(cfqd, current); 657 struct cfq_queue *cfqq = cfq_find_cfq_hash(cfqd, current->pid, CFQ_KEY_ANY);
588 struct cfq_queue *cfqq = cfq_find_cfq_hash(cfqd, key);
589 struct rb_node *n; 658 struct rb_node *n;
590 659
591 if (!cfqq) 660 if (!cfqq)
@@ -609,20 +678,25 @@ out:
609 678
610static void cfq_deactivate_request(request_queue_t *q, struct request *rq) 679static void cfq_deactivate_request(request_queue_t *q, struct request *rq)
611{ 680{
681 struct cfq_data *cfqd = q->elevator->elevator_data;
612 struct cfq_rq *crq = RQ_DATA(rq); 682 struct cfq_rq *crq = RQ_DATA(rq);
613 683
614 if (crq) { 684 if (crq) {
615 struct cfq_queue *cfqq = crq->cfq_queue; 685 struct cfq_queue *cfqq = crq->cfq_queue;
616 686
617 if (cfqq->cfqd->cfq_tagged) { 687 if (cfq_crq_in_driver(crq)) {
618 cfqq->service_used--; 688 cfq_clear_crq_in_driver(crq);
619 cfq_sort_rr_list(cfqq, 0); 689 WARN_ON(!cfqd->rq_in_driver);
690 cfqd->rq_in_driver--;
620 } 691 }
692 if (cfq_crq_in_flight(crq)) {
693 const int sync = cfq_crq_is_sync(crq);
621 694
622 if (crq->accounted) { 695 cfq_clear_crq_in_flight(crq);
623 crq->accounted = 0; 696 WARN_ON(!cfqq->on_dispatch[sync]);
624 cfqq->cfqd->rq_in_driver--; 697 cfqq->on_dispatch[sync]--;
625 } 698 }
699 cfq_mark_crq_requeued(crq);
626 } 700 }
627} 701}
628 702
@@ -640,11 +714,10 @@ static void cfq_remove_request(request_queue_t *q, struct request *rq)
640 struct cfq_rq *crq = RQ_DATA(rq); 714 struct cfq_rq *crq = RQ_DATA(rq);
641 715
642 if (crq) { 716 if (crq) {
643 cfq_remove_merge_hints(q, crq);
644 list_del_init(&rq->queuelist); 717 list_del_init(&rq->queuelist);
718 cfq_del_crq_rb(crq);
719 cfq_remove_merge_hints(q, crq);
645 720
646 if (crq->cfq_queue)
647 cfq_del_crq_rb(crq);
648 } 721 }
649} 722}
650 723
@@ -662,21 +735,15 @@ cfq_merge(request_queue_t *q, struct request **req, struct bio *bio)
662 } 735 }
663 736
664 __rq = cfq_find_rq_hash(cfqd, bio->bi_sector); 737 __rq = cfq_find_rq_hash(cfqd, bio->bi_sector);
665 if (__rq) { 738 if (__rq && elv_rq_merge_ok(__rq, bio)) {
666 BUG_ON(__rq->sector + __rq->nr_sectors != bio->bi_sector); 739 ret = ELEVATOR_BACK_MERGE;
667 740 goto out;
668 if (elv_rq_merge_ok(__rq, bio)) {
669 ret = ELEVATOR_BACK_MERGE;
670 goto out;
671 }
672 } 741 }
673 742
674 __rq = cfq_find_rq_rb(cfqd, bio->bi_sector + bio_sectors(bio)); 743 __rq = cfq_find_rq_rb(cfqd, bio->bi_sector + bio_sectors(bio));
675 if (__rq) { 744 if (__rq && elv_rq_merge_ok(__rq, bio)) {
676 if (elv_rq_merge_ok(__rq, bio)) { 745 ret = ELEVATOR_FRONT_MERGE;
677 ret = ELEVATOR_FRONT_MERGE; 746 goto out;
678 goto out;
679 }
680 } 747 }
681 748
682 return ELEVATOR_NO_MERGE; 749 return ELEVATOR_NO_MERGE;
@@ -709,20 +776,220 @@ static void
709cfq_merged_requests(request_queue_t *q, struct request *rq, 776cfq_merged_requests(request_queue_t *q, struct request *rq,
710 struct request *next) 777 struct request *next)
711{ 778{
712 struct cfq_rq *crq = RQ_DATA(rq);
713 struct cfq_rq *cnext = RQ_DATA(next);
714
715 cfq_merged_request(q, rq); 779 cfq_merged_request(q, rq);
716 780
717 if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist)) { 781 /*
718 if (time_before(cnext->queue_start, crq->queue_start)) { 782 * reposition in fifo if next is older than rq
719 list_move(&rq->queuelist, &next->queuelist); 783 */
720 crq->queue_start = cnext->queue_start; 784 if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) &&
785 time_before(next->start_time, rq->start_time))
786 list_move(&rq->queuelist, &next->queuelist);
787
788 cfq_remove_request(q, next);
789}
790
791static inline void
792__cfq_set_active_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq)
793{
794 if (cfqq) {
795 /*
796 * stop potential idle class queues waiting service
797 */
798 del_timer(&cfqd->idle_class_timer);
799
800 cfqq->slice_start = jiffies;
801 cfqq->slice_end = 0;
802 cfqq->slice_left = 0;
803 cfq_clear_cfqq_must_alloc_slice(cfqq);
804 cfq_clear_cfqq_fifo_expire(cfqq);
805 cfq_clear_cfqq_expired(cfqq);
806 }
807
808 cfqd->active_queue = cfqq;
809}
810
811/*
812 * 0
813 * 0,1
814 * 0,1,2
815 * 0,1,2,3
816 * 0,1,2,3,4
817 * 0,1,2,3,4,5
818 * 0,1,2,3,4,5,6
819 * 0,1,2,3,4,5,6,7
820 */
821static int cfq_get_next_prio_level(struct cfq_data *cfqd)
822{
823 int prio, wrap;
824
825 prio = -1;
826 wrap = 0;
827 do {
828 int p;
829
830 for (p = cfqd->cur_prio; p <= cfqd->cur_end_prio; p++) {
831 if (!list_empty(&cfqd->rr_list[p])) {
832 prio = p;
833 break;
834 }
835 }
836
837 if (prio != -1)
838 break;
839 cfqd->cur_prio = 0;
840 if (++cfqd->cur_end_prio == CFQ_PRIO_LISTS) {
841 cfqd->cur_end_prio = 0;
842 if (wrap)
843 break;
844 wrap = 1;
721 } 845 }
846 } while (1);
847
848 if (unlikely(prio == -1))
849 return -1;
850
851 BUG_ON(prio >= CFQ_PRIO_LISTS);
852
853 list_splice_init(&cfqd->rr_list[prio], &cfqd->cur_rr);
854
855 cfqd->cur_prio = prio + 1;
856 if (cfqd->cur_prio > cfqd->cur_end_prio) {
857 cfqd->cur_end_prio = cfqd->cur_prio;
858 cfqd->cur_prio = 0;
859 }
860 if (cfqd->cur_end_prio == CFQ_PRIO_LISTS) {
861 cfqd->cur_prio = 0;
862 cfqd->cur_end_prio = 0;
722 } 863 }
723 864
724 cfq_update_next_crq(cnext); 865 return prio;
725 cfq_remove_request(q, next); 866}
867
868static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd)
869{
870 struct cfq_queue *cfqq;
871
872 /*
873 * if current queue is expired but not done with its requests yet,
874 * wait for that to happen
875 */
876 if ((cfqq = cfqd->active_queue) != NULL) {
877 if (cfq_cfqq_expired(cfqq) && cfq_cfqq_dispatched(cfqq))
878 return NULL;
879 }
880
881 /*
882 * if current list is non-empty, grab first entry. if it is empty,
883 * get next prio level and grab first entry then if any are spliced
884 */
885 if (!list_empty(&cfqd->cur_rr) || cfq_get_next_prio_level(cfqd) != -1)
886 cfqq = list_entry_cfqq(cfqd->cur_rr.next);
887
888 /*
889 * if we have idle queues and no rt or be queues had pending
890 * requests, either allow immediate service if the grace period
891 * has passed or arm the idle grace timer
892 */
893 if (!cfqq && !list_empty(&cfqd->idle_rr)) {
894 unsigned long end = cfqd->last_end_request + CFQ_IDLE_GRACE;
895
896 if (time_after_eq(jiffies, end))
897 cfqq = list_entry_cfqq(cfqd->idle_rr.next);
898 else
899 mod_timer(&cfqd->idle_class_timer, end);
900 }
901
902 __cfq_set_active_queue(cfqd, cfqq);
903 return cfqq;
904}
905
906/*
907 * current cfqq expired its slice (or was too idle), select new one
908 */
909static void
910__cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq,
911 int preempted)
912{
913 unsigned long now = jiffies;
914
915 if (cfq_cfqq_wait_request(cfqq))
916 del_timer(&cfqd->idle_slice_timer);
917
918 if (!preempted && !cfq_cfqq_dispatched(cfqq))
919 cfqq->service_last = now;
920
921 cfq_clear_cfqq_must_dispatch(cfqq);
922 cfq_clear_cfqq_wait_request(cfqq);
923
924 /*
925 * store what was left of this slice, if the queue idled out
926 * or was preempted
927 */
928 if (time_after(now, cfqq->slice_end))
929 cfqq->slice_left = now - cfqq->slice_end;
930 else
931 cfqq->slice_left = 0;
932
933 if (cfq_cfqq_on_rr(cfqq))
934 cfq_resort_rr_list(cfqq, preempted);
935
936 if (cfqq == cfqd->active_queue)
937 cfqd->active_queue = NULL;
938
939 if (cfqd->active_cic) {
940 put_io_context(cfqd->active_cic->ioc);
941 cfqd->active_cic = NULL;
942 }
943
944 cfqd->dispatch_slice = 0;
945}
946
947static inline void cfq_slice_expired(struct cfq_data *cfqd, int preempted)
948{
949 struct cfq_queue *cfqq = cfqd->active_queue;
950
951 if (cfqq) {
952 /*
953 * use deferred expiry, if there are requests in progress as
954 * not to disturb the slice of the next queue
955 */
956 if (cfq_cfqq_dispatched(cfqq))
957 cfq_mark_cfqq_expired(cfqq);
958 else
959 __cfq_slice_expired(cfqd, cfqq, preempted);
960 }
961}
962
963static int cfq_arm_slice_timer(struct cfq_data *cfqd, struct cfq_queue *cfqq)
964
965{
966 WARN_ON(!RB_EMPTY(&cfqq->sort_list));
967 WARN_ON(cfqq != cfqd->active_queue);
968
969 /*
970 * idle is disabled, either manually or by past process history
971 */
972 if (!cfqd->cfq_slice_idle)
973 return 0;
974 if (!cfq_cfqq_idle_window(cfqq))
975 return 0;
976 /*
977 * task has exited, don't wait
978 */
979 if (cfqd->active_cic && !cfqd->active_cic->ioc->task)
980 return 0;
981
982 cfq_mark_cfqq_must_dispatch(cfqq);
983 cfq_mark_cfqq_wait_request(cfqq);
984
985 if (!timer_pending(&cfqd->idle_slice_timer)) {
986 unsigned long slice_left = min(cfqq->slice_end - 1, (unsigned long) cfqd->cfq_slice_idle);
987
988 cfqd->idle_slice_timer.expires = jiffies + slice_left;
989 add_timer(&cfqd->idle_slice_timer);
990 }
991
992 return 1;
726} 993}
727 994
728/* 995/*
@@ -738,31 +1005,40 @@ static void cfq_dispatch_sort(request_queue_t *q, struct cfq_rq *crq)
738 struct request *__rq; 1005 struct request *__rq;
739 sector_t last; 1006 sector_t last;
740 1007
741 cfq_del_crq_rb(crq);
742 cfq_remove_merge_hints(q, crq);
743 list_del(&crq->request->queuelist); 1008 list_del(&crq->request->queuelist);
744 1009
745 last = cfqd->last_sector; 1010 last = cfqd->last_sector;
746 while ((entry = entry->prev) != head) { 1011 list_for_each_entry_reverse(__rq, head, queuelist) {
747 __rq = list_entry_rq(entry); 1012 struct cfq_rq *__crq = RQ_DATA(__rq);
748 1013
749 if (blk_barrier_rq(crq->request)) 1014 if (blk_barrier_rq(__rq))
750 break; 1015 break;
751 if (!blk_fs_request(crq->request)) 1016 if (!blk_fs_request(__rq))
1017 break;
1018 if (cfq_crq_requeued(__crq))
752 break; 1019 break;
753 1020
754 if (crq->request->sector > __rq->sector) 1021 if (__rq->sector <= crq->request->sector)
755 break; 1022 break;
756 if (__rq->sector > last && crq->request->sector < last) { 1023 if (__rq->sector > last && crq->request->sector < last) {
757 last = crq->request->sector; 1024 last = crq->request->sector + crq->request->nr_sectors;
758 break; 1025 break;
759 } 1026 }
1027 entry = &__rq->queuelist;
760 } 1028 }
761 1029
762 cfqd->last_sector = last; 1030 cfqd->last_sector = last;
763 crq->in_flight = 1; 1031
764 cfqq->in_flight++; 1032 cfqq->next_crq = cfq_find_next_crq(cfqd, cfqq, crq);
765 list_add(&crq->request->queuelist, entry); 1033
1034 cfq_del_crq_rb(crq);
1035 cfq_remove_merge_hints(q, crq);
1036
1037 cfq_mark_crq_in_flight(crq);
1038 cfq_clear_crq_requeued(crq);
1039
1040 cfqq->on_dispatch[cfq_crq_is_sync(crq)]++;
1041 list_add_tail(&crq->request->queuelist, entry);
766} 1042}
767 1043
768/* 1044/*
@@ -771,173 +1047,225 @@ static void cfq_dispatch_sort(request_queue_t *q, struct cfq_rq *crq)
771static inline struct cfq_rq *cfq_check_fifo(struct cfq_queue *cfqq) 1047static inline struct cfq_rq *cfq_check_fifo(struct cfq_queue *cfqq)
772{ 1048{
773 struct cfq_data *cfqd = cfqq->cfqd; 1049 struct cfq_data *cfqd = cfqq->cfqd;
774 const int reads = !list_empty(&cfqq->fifo[0]); 1050 struct request *rq;
775 const int writes = !list_empty(&cfqq->fifo[1]);
776 unsigned long now = jiffies;
777 struct cfq_rq *crq; 1051 struct cfq_rq *crq;
778 1052
779 if (time_before(now, cfqq->last_fifo_expire + cfqd->cfq_fifo_batch_expire)) 1053 if (cfq_cfqq_fifo_expire(cfqq))
780 return NULL; 1054 return NULL;
781 1055
782 crq = RQ_DATA(list_entry(cfqq->fifo[0].next, struct request, queuelist)); 1056 if (!list_empty(&cfqq->fifo)) {
783 if (reads && time_after(now, crq->queue_start + cfqd->cfq_fifo_expire_r)) { 1057 int fifo = cfq_cfqq_class_sync(cfqq);
784 cfqq->last_fifo_expire = now;
785 return crq;
786 }
787 1058
788 crq = RQ_DATA(list_entry(cfqq->fifo[1].next, struct request, queuelist)); 1059 crq = RQ_DATA(list_entry_fifo(cfqq->fifo.next));
789 if (writes && time_after(now, crq->queue_start + cfqd->cfq_fifo_expire_w)) { 1060 rq = crq->request;
790 cfqq->last_fifo_expire = now; 1061 if (time_after(jiffies, rq->start_time + cfqd->cfq_fifo_expire[fifo])) {
791 return crq; 1062 cfq_mark_cfqq_fifo_expire(cfqq);
1063 return crq;
1064 }
792 } 1065 }
793 1066
794 return NULL; 1067 return NULL;
795} 1068}
796 1069
797/* 1070/*
798 * dispatch a single request from given queue 1071 * Scale schedule slice based on io priority. Use the sync time slice only
1072 * if a queue is marked sync and has sync io queued. A sync queue with async
1073 * io only, should not get full sync slice length.
799 */ 1074 */
1075static inline int
1076cfq_prio_to_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq)
1077{
1078 const int base_slice = cfqd->cfq_slice[cfq_cfqq_sync(cfqq)];
1079
1080 WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR);
1081
1082 return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - cfqq->ioprio));
1083}
1084
800static inline void 1085static inline void
801cfq_dispatch_request(request_queue_t *q, struct cfq_data *cfqd, 1086cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq)
802 struct cfq_queue *cfqq)
803{ 1087{
804 struct cfq_rq *crq; 1088 cfqq->slice_end = cfq_prio_to_slice(cfqd, cfqq) + jiffies;
1089}
1090
1091static inline int
1092cfq_prio_to_maxrq(struct cfq_data *cfqd, struct cfq_queue *cfqq)
1093{
1094 const int base_rq = cfqd->cfq_slice_async_rq;
1095
1096 WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR);
1097
1098 return 2 * (base_rq + base_rq * (CFQ_PRIO_LISTS - 1 - cfqq->ioprio));
1099}
1100
1101/*
1102 * get next queue for service
1103 */
1104static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd, int force)
1105{
1106 unsigned long now = jiffies;
1107 struct cfq_queue *cfqq;
1108
1109 cfqq = cfqd->active_queue;
1110 if (!cfqq)
1111 goto new_queue;
1112
1113 if (cfq_cfqq_expired(cfqq))
1114 goto new_queue;
805 1115
806 /* 1116 /*
807 * follow expired path, else get first next available 1117 * slice has expired
808 */ 1118 */
809 if ((crq = cfq_check_fifo(cfqq)) == NULL) { 1119 if (!cfq_cfqq_must_dispatch(cfqq) && time_after(now, cfqq->slice_end))
810 if (cfqd->find_best_crq) 1120 goto expire;
811 crq = cfqq->next_crq;
812 else
813 crq = rb_entry_crq(rb_first(&cfqq->sort_list));
814 }
815
816 cfqd->last_sector = crq->request->sector + crq->request->nr_sectors;
817 1121
818 /* 1122 /*
819 * finally, insert request into driver list 1123 * if queue has requests, dispatch one. if not, check if
1124 * enough slice is left to wait for one
820 */ 1125 */
821 cfq_dispatch_sort(q, crq); 1126 if (!RB_EMPTY(&cfqq->sort_list))
1127 goto keep_queue;
1128 else if (!force && cfq_cfqq_class_sync(cfqq) &&
1129 time_before(now, cfqq->slice_end)) {
1130 if (cfq_arm_slice_timer(cfqd, cfqq))
1131 return NULL;
1132 }
1133
1134expire:
1135 cfq_slice_expired(cfqd, 0);
1136new_queue:
1137 cfqq = cfq_set_active_queue(cfqd);
1138keep_queue:
1139 return cfqq;
822} 1140}
823 1141
824static int cfq_dispatch_requests(request_queue_t *q, int max_dispatch) 1142static int
1143__cfq_dispatch_requests(struct cfq_data *cfqd, struct cfq_queue *cfqq,
1144 int max_dispatch)
825{ 1145{
826 struct cfq_data *cfqd = q->elevator->elevator_data; 1146 int dispatched = 0;
827 struct cfq_queue *cfqq;
828 struct list_head *entry, *tmp;
829 int queued, busy_queues, first_round;
830 1147
831 if (list_empty(&cfqd->rr_list)) 1148 BUG_ON(RB_EMPTY(&cfqq->sort_list));
832 return 0;
833 1149
834 queued = 0; 1150 do {
835 first_round = 1; 1151 struct cfq_rq *crq;
836restart:
837 busy_queues = 0;
838 list_for_each_safe(entry, tmp, &cfqd->rr_list) {
839 cfqq = list_entry_cfqq(entry);
840 1152
841 BUG_ON(RB_EMPTY(&cfqq->sort_list)); 1153 /*
1154 * follow expired path, else get first next available
1155 */
1156 if ((crq = cfq_check_fifo(cfqq)) == NULL)
1157 crq = cfqq->next_crq;
842 1158
843 /* 1159 /*
844 * first round of queueing, only select from queues that 1160 * finally, insert request into driver dispatch list
845 * don't already have io in-flight
846 */ 1161 */
847 if (first_round && cfqq->in_flight) 1162 cfq_dispatch_sort(cfqd->queue, crq);
848 continue;
849 1163
850 cfq_dispatch_request(q, cfqd, cfqq); 1164 cfqd->dispatch_slice++;
1165 dispatched++;
851 1166
852 if (!RB_EMPTY(&cfqq->sort_list)) 1167 if (!cfqd->active_cic) {
853 busy_queues++; 1168 atomic_inc(&crq->io_context->ioc->refcount);
1169 cfqd->active_cic = crq->io_context;
1170 }
854 1171
855 queued++; 1172 if (RB_EMPTY(&cfqq->sort_list))
856 } 1173 break;
1174
1175 } while (dispatched < max_dispatch);
1176
1177 /*
1178 * if slice end isn't set yet, set it. if at least one request was
1179 * sync, use the sync time slice value
1180 */
1181 if (!cfqq->slice_end)
1182 cfq_set_prio_slice(cfqd, cfqq);
1183
1184 /*
1185 * expire an async queue immediately if it has used up its slice. idle
1186 * queue always expire after 1 dispatch round.
1187 */
1188 if ((!cfq_cfqq_sync(cfqq) &&
1189 cfqd->dispatch_slice >= cfq_prio_to_maxrq(cfqd, cfqq)) ||
1190 cfq_class_idle(cfqq))
1191 cfq_slice_expired(cfqd, 0);
1192
1193 return dispatched;
1194}
1195
1196static int
1197cfq_dispatch_requests(request_queue_t *q, int max_dispatch, int force)
1198{
1199 struct cfq_data *cfqd = q->elevator->elevator_data;
1200 struct cfq_queue *cfqq;
1201
1202 if (!cfqd->busy_queues)
1203 return 0;
1204
1205 cfqq = cfq_select_queue(cfqd, force);
1206 if (cfqq) {
1207 cfq_clear_cfqq_must_dispatch(cfqq);
1208 cfq_clear_cfqq_wait_request(cfqq);
1209 del_timer(&cfqd->idle_slice_timer);
857 1210
858 if ((queued < max_dispatch) && (busy_queues || first_round)) { 1211 if (cfq_class_idle(cfqq))
859 first_round = 0; 1212 max_dispatch = 1;
860 goto restart; 1213
1214 return __cfq_dispatch_requests(cfqd, cfqq, max_dispatch);
861 } 1215 }
862 1216
863 return queued; 1217 return 0;
864} 1218}
865 1219
866static inline void cfq_account_dispatch(struct cfq_rq *crq) 1220static inline void cfq_account_dispatch(struct cfq_rq *crq)
867{ 1221{
868 struct cfq_queue *cfqq = crq->cfq_queue; 1222 struct cfq_queue *cfqq = crq->cfq_queue;
869 struct cfq_data *cfqd = cfqq->cfqd; 1223 struct cfq_data *cfqd = cfqq->cfqd;
870 unsigned long now, elapsed;
871 1224
872 if (!blk_fs_request(crq->request)) 1225 if (unlikely(!blk_fs_request(crq->request)))
873 return; 1226 return;
874 1227
875 /* 1228 /*
876 * accounted bit is necessary since some drivers will call 1229 * accounted bit is necessary since some drivers will call
877 * elv_next_request() many times for the same request (eg ide) 1230 * elv_next_request() many times for the same request (eg ide)
878 */ 1231 */
879 if (crq->accounted) 1232 if (cfq_crq_in_driver(crq))
880 return; 1233 return;
881 1234
882 now = jiffies; 1235 cfq_mark_crq_in_driver(crq);
883 if (cfqq->service_start == ~0UL) 1236 cfqd->rq_in_driver++;
884 cfqq->service_start = now;
885
886 /*
887 * on drives with tagged command queueing, command turn-around time
888 * doesn't necessarily reflect the time spent processing this very
889 * command inside the drive. so do the accounting differently there,
890 * by just sorting on the number of requests
891 */
892 if (cfqd->cfq_tagged) {
893 if (time_after(now, cfqq->service_start + cfq_service)) {
894 cfqq->service_start = now;
895 cfqq->service_used /= 10;
896 }
897
898 cfqq->service_used++;
899 cfq_sort_rr_list(cfqq, 0);
900 }
901
902 elapsed = now - crq->queue_start;
903 if (elapsed > max_elapsed_dispatch)
904 max_elapsed_dispatch = elapsed;
905
906 crq->accounted = 1;
907 crq->service_start = now;
908
909 if (++cfqd->rq_in_driver >= CFQ_MAX_TAG && !cfqd->cfq_tagged) {
910 cfqq->cfqd->cfq_tagged = 1;
911 printk("cfq: depth %d reached, tagging now on\n", CFQ_MAX_TAG);
912 }
913} 1237}
914 1238
915static inline void 1239static inline void
916cfq_account_completion(struct cfq_queue *cfqq, struct cfq_rq *crq) 1240cfq_account_completion(struct cfq_queue *cfqq, struct cfq_rq *crq)
917{ 1241{
918 struct cfq_data *cfqd = cfqq->cfqd; 1242 struct cfq_data *cfqd = cfqq->cfqd;
1243 unsigned long now;
919 1244
920 if (!crq->accounted) 1245 if (!cfq_crq_in_driver(crq))
921 return; 1246 return;
922 1247
1248 now = jiffies;
1249
923 WARN_ON(!cfqd->rq_in_driver); 1250 WARN_ON(!cfqd->rq_in_driver);
924 cfqd->rq_in_driver--; 1251 cfqd->rq_in_driver--;
925 1252
926 if (!cfqd->cfq_tagged) { 1253 if (!cfq_class_idle(cfqq))
927 unsigned long now = jiffies; 1254 cfqd->last_end_request = now;
928 unsigned long duration = now - crq->service_start;
929 1255
930 if (time_after(now, cfqq->service_start + cfq_service)) { 1256 if (!cfq_cfqq_dispatched(cfqq)) {
931 cfqq->service_start = now; 1257 if (cfq_cfqq_on_rr(cfqq)) {
932 cfqq->service_used >>= 3; 1258 cfqq->service_last = now;
1259 cfq_resort_rr_list(cfqq, 0);
1260 }
1261 if (cfq_cfqq_expired(cfqq)) {
1262 __cfq_slice_expired(cfqd, cfqq, 0);
1263 cfq_schedule_dispatch(cfqd);
933 } 1264 }
934
935 cfqq->service_used += duration;
936 cfq_sort_rr_list(cfqq, 0);
937
938 if (duration > max_elapsed_crq)
939 max_elapsed_crq = duration;
940 } 1265 }
1266
1267 if (cfq_crq_is_sync(crq))
1268 crq->io_context->last_end_request = now;
941} 1269}
942 1270
943static struct request *cfq_next_request(request_queue_t *q) 1271static struct request *cfq_next_request(request_queue_t *q)
@@ -950,7 +1278,19 @@ static struct request *cfq_next_request(request_queue_t *q)
950dispatch: 1278dispatch:
951 rq = list_entry_rq(q->queue_head.next); 1279 rq = list_entry_rq(q->queue_head.next);
952 1280
953 if ((crq = RQ_DATA(rq)) != NULL) { 1281 crq = RQ_DATA(rq);
1282 if (crq) {
1283 struct cfq_queue *cfqq = crq->cfq_queue;
1284
1285 /*
1286 * if idle window is disabled, allow queue buildup
1287 */
1288 if (!cfq_crq_in_driver(crq) &&
1289 !cfq_cfqq_idle_window(cfqq) &&
1290 !blk_barrier_rq(rq) &&
1291 cfqd->rq_in_driver >= cfqd->cfq_max_depth)
1292 return NULL;
1293
954 cfq_remove_merge_hints(q, crq); 1294 cfq_remove_merge_hints(q, crq);
955 cfq_account_dispatch(crq); 1295 cfq_account_dispatch(crq);
956 } 1296 }
@@ -958,7 +1298,7 @@ dispatch:
958 return rq; 1298 return rq;
959 } 1299 }
960 1300
961 if (cfq_dispatch_requests(q, cfqd->cfq_quantum)) 1301 if (cfq_dispatch_requests(q, cfqd->cfq_quantum, 0))
962 goto dispatch; 1302 goto dispatch;
963 1303
964 return NULL; 1304 return NULL;
@@ -972,13 +1312,21 @@ dispatch:
972 */ 1312 */
973static void cfq_put_queue(struct cfq_queue *cfqq) 1313static void cfq_put_queue(struct cfq_queue *cfqq)
974{ 1314{
975 BUG_ON(!atomic_read(&cfqq->ref)); 1315 struct cfq_data *cfqd = cfqq->cfqd;
1316
1317 BUG_ON(atomic_read(&cfqq->ref) <= 0);
976 1318
977 if (!atomic_dec_and_test(&cfqq->ref)) 1319 if (!atomic_dec_and_test(&cfqq->ref))
978 return; 1320 return;
979 1321
980 BUG_ON(rb_first(&cfqq->sort_list)); 1322 BUG_ON(rb_first(&cfqq->sort_list));
981 BUG_ON(cfqq->on_rr); 1323 BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]);
1324 BUG_ON(cfq_cfqq_on_rr(cfqq));
1325
1326 if (unlikely(cfqd->active_queue == cfqq)) {
1327 __cfq_slice_expired(cfqd, cfqq, 0);
1328 cfq_schedule_dispatch(cfqd);
1329 }
982 1330
983 cfq_put_cfqd(cfqq->cfqd); 1331 cfq_put_cfqd(cfqq->cfqd);
984 1332
@@ -991,15 +1339,17 @@ static void cfq_put_queue(struct cfq_queue *cfqq)
991} 1339}
992 1340
993static inline struct cfq_queue * 1341static inline struct cfq_queue *
994__cfq_find_cfq_hash(struct cfq_data *cfqd, unsigned long key, const int hashval) 1342__cfq_find_cfq_hash(struct cfq_data *cfqd, unsigned int key, unsigned int prio,
1343 const int hashval)
995{ 1344{
996 struct hlist_head *hash_list = &cfqd->cfq_hash[hashval]; 1345 struct hlist_head *hash_list = &cfqd->cfq_hash[hashval];
997 struct hlist_node *entry, *next; 1346 struct hlist_node *entry, *next;
998 1347
999 hlist_for_each_safe(entry, next, hash_list) { 1348 hlist_for_each_safe(entry, next, hash_list) {
1000 struct cfq_queue *__cfqq = list_entry_qhash(entry); 1349 struct cfq_queue *__cfqq = list_entry_qhash(entry);
1350 const unsigned short __p = IOPRIO_PRIO_VALUE(__cfqq->ioprio_class, __cfqq->ioprio);
1001 1351
1002 if (__cfqq->key == key) 1352 if (__cfqq->key == key && (__p == prio || prio == CFQ_KEY_ANY))
1003 return __cfqq; 1353 return __cfqq;
1004 } 1354 }
1005 1355
@@ -1007,94 +1357,220 @@ __cfq_find_cfq_hash(struct cfq_data *cfqd, unsigned long key, const int hashval)
1007} 1357}
1008 1358
1009static struct cfq_queue * 1359static struct cfq_queue *
1010cfq_find_cfq_hash(struct cfq_data *cfqd, unsigned long key) 1360cfq_find_cfq_hash(struct cfq_data *cfqd, unsigned int key, unsigned short prio)
1011{ 1361{
1012 return __cfq_find_cfq_hash(cfqd, key, hash_long(key, CFQ_QHASH_SHIFT)); 1362 return __cfq_find_cfq_hash(cfqd, key, prio, hash_long(key, CFQ_QHASH_SHIFT));
1013} 1363}
1014 1364
1015static inline void 1365static void cfq_free_io_context(struct cfq_io_context *cic)
1016cfq_rehash_cfqq(struct cfq_data *cfqd, struct cfq_queue **cfqq,
1017 struct cfq_io_context *cic)
1018{ 1366{
1019 unsigned long hashkey = cfq_hash_key(cfqd, current); 1367 struct cfq_io_context *__cic;
1020 unsigned long hashval = hash_long(hashkey, CFQ_QHASH_SHIFT); 1368 struct list_head *entry, *next;
1021 struct cfq_queue *__cfqq;
1022 unsigned long flags;
1023
1024 spin_lock_irqsave(cfqd->queue->queue_lock, flags);
1025 1369
1026 hlist_del(&(*cfqq)->cfq_hash); 1370 list_for_each_safe(entry, next, &cic->list) {
1027 1371 __cic = list_entry(entry, struct cfq_io_context, list);
1028 __cfqq = __cfq_find_cfq_hash(cfqd, hashkey, hashval); 1372 kmem_cache_free(cfq_ioc_pool, __cic);
1029 if (!__cfqq || __cfqq == *cfqq) {
1030 __cfqq = *cfqq;
1031 hlist_add_head(&__cfqq->cfq_hash, &cfqd->cfq_hash[hashval]);
1032 __cfqq->key_type = cfqd->key_type;
1033 } else {
1034 atomic_inc(&__cfqq->ref);
1035 cic->cfqq = __cfqq;
1036 cfq_put_queue(*cfqq);
1037 *cfqq = __cfqq;
1038 } 1373 }
1039 1374
1040 cic->cfqq = __cfqq; 1375 kmem_cache_free(cfq_ioc_pool, cic);
1041 spin_unlock_irqrestore(cfqd->queue->queue_lock, flags);
1042} 1376}
1043 1377
1044static void cfq_free_io_context(struct cfq_io_context *cic) 1378/*
1379 * Called with interrupts disabled
1380 */
1381static void cfq_exit_single_io_context(struct cfq_io_context *cic)
1045{ 1382{
1046 kmem_cache_free(cfq_ioc_pool, cic); 1383 struct cfq_data *cfqd = cic->cfqq->cfqd;
1384 request_queue_t *q = cfqd->queue;
1385
1386 WARN_ON(!irqs_disabled());
1387
1388 spin_lock(q->queue_lock);
1389
1390 if (unlikely(cic->cfqq == cfqd->active_queue)) {
1391 __cfq_slice_expired(cfqd, cic->cfqq, 0);
1392 cfq_schedule_dispatch(cfqd);
1393 }
1394
1395 cfq_put_queue(cic->cfqq);
1396 cic->cfqq = NULL;
1397 spin_unlock(q->queue_lock);
1047} 1398}
1048 1399
1049/* 1400/*
1050 * locking hierarchy is: io_context lock -> queue locks 1401 * Another task may update the task cic list, if it is doing a queue lookup
1402 * on its behalf. cfq_cic_lock excludes such concurrent updates
1051 */ 1403 */
1052static void cfq_exit_io_context(struct cfq_io_context *cic) 1404static void cfq_exit_io_context(struct cfq_io_context *cic)
1053{ 1405{
1054 struct cfq_queue *cfqq = cic->cfqq; 1406 struct cfq_io_context *__cic;
1055 struct list_head *entry = &cic->list; 1407 struct list_head *entry;
1056 request_queue_t *q;
1057 unsigned long flags; 1408 unsigned long flags;
1058 1409
1410 local_irq_save(flags);
1411
1059 /* 1412 /*
1060 * put the reference this task is holding to the various queues 1413 * put the reference this task is holding to the various queues
1061 */ 1414 */
1062 spin_lock_irqsave(&cic->ioc->lock, flags); 1415 list_for_each(entry, &cic->list) {
1063 while ((entry = cic->list.next) != &cic->list) {
1064 struct cfq_io_context *__cic;
1065
1066 __cic = list_entry(entry, struct cfq_io_context, list); 1416 __cic = list_entry(entry, struct cfq_io_context, list);
1067 list_del(entry); 1417 cfq_exit_single_io_context(__cic);
1068
1069 q = __cic->cfqq->cfqd->queue;
1070 spin_lock(q->queue_lock);
1071 cfq_put_queue(__cic->cfqq);
1072 spin_unlock(q->queue_lock);
1073 } 1418 }
1074 1419
1075 q = cfqq->cfqd->queue; 1420 cfq_exit_single_io_context(cic);
1076 spin_lock(q->queue_lock); 1421 local_irq_restore(flags);
1077 cfq_put_queue(cfqq);
1078 spin_unlock(q->queue_lock);
1079
1080 cic->cfqq = NULL;
1081 spin_unlock_irqrestore(&cic->ioc->lock, flags);
1082} 1422}
1083 1423
1084static struct cfq_io_context *cfq_alloc_io_context(int gfp_flags) 1424static struct cfq_io_context *
1425cfq_alloc_io_context(struct cfq_data *cfqd, int gfp_mask)
1085{ 1426{
1086 struct cfq_io_context *cic = kmem_cache_alloc(cfq_ioc_pool, gfp_flags); 1427 struct cfq_io_context *cic = kmem_cache_alloc(cfq_ioc_pool, gfp_mask);
1087 1428
1088 if (cic) { 1429 if (cic) {
1089 cic->dtor = cfq_free_io_context;
1090 cic->exit = cfq_exit_io_context;
1091 INIT_LIST_HEAD(&cic->list); 1430 INIT_LIST_HEAD(&cic->list);
1092 cic->cfqq = NULL; 1431 cic->cfqq = NULL;
1432 cic->key = NULL;
1433 cic->last_end_request = jiffies;
1434 cic->ttime_total = 0;
1435 cic->ttime_samples = 0;
1436 cic->ttime_mean = 0;
1437 cic->dtor = cfq_free_io_context;
1438 cic->exit = cfq_exit_io_context;
1093 } 1439 }
1094 1440
1095 return cic; 1441 return cic;
1096} 1442}
1097 1443
1444static void cfq_init_prio_data(struct cfq_queue *cfqq)
1445{
1446 struct task_struct *tsk = current;
1447 int ioprio_class;
1448
1449 if (!cfq_cfqq_prio_changed(cfqq))
1450 return;
1451
1452 ioprio_class = IOPRIO_PRIO_CLASS(tsk->ioprio);
1453 switch (ioprio_class) {
1454 default:
1455 printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class);
1456 case IOPRIO_CLASS_NONE:
1457 /*
1458 * no prio set, place us in the middle of the BE classes
1459 */
1460 cfqq->ioprio = task_nice_ioprio(tsk);
1461 cfqq->ioprio_class = IOPRIO_CLASS_BE;
1462 break;
1463 case IOPRIO_CLASS_RT:
1464 cfqq->ioprio = task_ioprio(tsk);
1465 cfqq->ioprio_class = IOPRIO_CLASS_RT;
1466 break;
1467 case IOPRIO_CLASS_BE:
1468 cfqq->ioprio = task_ioprio(tsk);
1469 cfqq->ioprio_class = IOPRIO_CLASS_BE;
1470 break;
1471 case IOPRIO_CLASS_IDLE:
1472 cfqq->ioprio_class = IOPRIO_CLASS_IDLE;
1473 cfqq->ioprio = 7;
1474 cfq_clear_cfqq_idle_window(cfqq);
1475 break;
1476 }
1477
1478 /*
1479 * keep track of original prio settings in case we have to temporarily
1480 * elevate the priority of this queue
1481 */
1482 cfqq->org_ioprio = cfqq->ioprio;
1483 cfqq->org_ioprio_class = cfqq->ioprio_class;
1484
1485 if (cfq_cfqq_on_rr(cfqq))
1486 cfq_resort_rr_list(cfqq, 0);
1487
1488 cfq_clear_cfqq_prio_changed(cfqq);
1489}
1490
1491static inline void changed_ioprio(struct cfq_queue *cfqq)
1492{
1493 if (cfqq) {
1494 struct cfq_data *cfqd = cfqq->cfqd;
1495
1496 spin_lock(cfqd->queue->queue_lock);
1497 cfq_mark_cfqq_prio_changed(cfqq);
1498 cfq_init_prio_data(cfqq);
1499 spin_unlock(cfqd->queue->queue_lock);
1500 }
1501}
1502
1503/*
1504 * callback from sys_ioprio_set, irqs are disabled
1505 */
1506static int cfq_ioc_set_ioprio(struct io_context *ioc, unsigned int ioprio)
1507{
1508 struct cfq_io_context *cic = ioc->cic;
1509
1510 changed_ioprio(cic->cfqq);
1511
1512 list_for_each_entry(cic, &cic->list, list)
1513 changed_ioprio(cic->cfqq);
1514
1515 return 0;
1516}
1517
1518static struct cfq_queue *
1519cfq_get_queue(struct cfq_data *cfqd, unsigned int key, unsigned short ioprio,
1520 int gfp_mask)
1521{
1522 const int hashval = hash_long(key, CFQ_QHASH_SHIFT);
1523 struct cfq_queue *cfqq, *new_cfqq = NULL;
1524
1525retry:
1526 cfqq = __cfq_find_cfq_hash(cfqd, key, ioprio, hashval);
1527
1528 if (!cfqq) {
1529 if (new_cfqq) {
1530 cfqq = new_cfqq;
1531 new_cfqq = NULL;
1532 } else if (gfp_mask & __GFP_WAIT) {
1533 spin_unlock_irq(cfqd->queue->queue_lock);
1534 new_cfqq = kmem_cache_alloc(cfq_pool, gfp_mask);
1535 spin_lock_irq(cfqd->queue->queue_lock);
1536 goto retry;
1537 } else {
1538 cfqq = kmem_cache_alloc(cfq_pool, gfp_mask);
1539 if (!cfqq)
1540 goto out;
1541 }
1542
1543 memset(cfqq, 0, sizeof(*cfqq));
1544
1545 INIT_HLIST_NODE(&cfqq->cfq_hash);
1546 INIT_LIST_HEAD(&cfqq->cfq_list);
1547 RB_CLEAR_ROOT(&cfqq->sort_list);
1548 INIT_LIST_HEAD(&cfqq->fifo);
1549
1550 cfqq->key = key;
1551 hlist_add_head(&cfqq->cfq_hash, &cfqd->cfq_hash[hashval]);
1552 atomic_set(&cfqq->ref, 0);
1553 cfqq->cfqd = cfqd;
1554 atomic_inc(&cfqd->ref);
1555 cfqq->service_last = 0;
1556 /*
1557 * set ->slice_left to allow preemption for a new process
1558 */
1559 cfqq->slice_left = 2 * cfqd->cfq_slice_idle;
1560 cfq_mark_cfqq_idle_window(cfqq);
1561 cfq_mark_cfqq_prio_changed(cfqq);
1562 cfq_init_prio_data(cfqq);
1563 }
1564
1565 if (new_cfqq)
1566 kmem_cache_free(cfq_pool, new_cfqq);
1567
1568 atomic_inc(&cfqq->ref);
1569out:
1570 WARN_ON((gfp_mask & __GFP_WAIT) && !cfqq);
1571 return cfqq;
1572}
1573
1098/* 1574/*
1099 * Setup general io context and cfq io context. There can be several cfq 1575 * Setup general io context and cfq io context. There can be several cfq
1100 * io contexts per general io context, if this process is doing io to more 1576 * io contexts per general io context, if this process is doing io to more
@@ -1102,39 +1578,39 @@ static struct cfq_io_context *cfq_alloc_io_context(int gfp_flags)
1102 * cfqq, so we don't need to worry about it disappearing 1578 * cfqq, so we don't need to worry about it disappearing
1103 */ 1579 */
1104static struct cfq_io_context * 1580static struct cfq_io_context *
1105cfq_get_io_context(struct cfq_queue **cfqq, int gfp_flags) 1581cfq_get_io_context(struct cfq_data *cfqd, pid_t pid, int gfp_mask)
1106{ 1582{
1107 struct cfq_data *cfqd = (*cfqq)->cfqd; 1583 struct io_context *ioc = NULL;
1108 struct cfq_queue *__cfqq = *cfqq;
1109 struct cfq_io_context *cic; 1584 struct cfq_io_context *cic;
1110 struct io_context *ioc;
1111 1585
1112 might_sleep_if(gfp_flags & __GFP_WAIT); 1586 might_sleep_if(gfp_mask & __GFP_WAIT);
1113 1587
1114 ioc = get_io_context(gfp_flags); 1588 ioc = get_io_context(gfp_mask);
1115 if (!ioc) 1589 if (!ioc)
1116 return NULL; 1590 return NULL;
1117 1591
1118 if ((cic = ioc->cic) == NULL) { 1592 if ((cic = ioc->cic) == NULL) {
1119 cic = cfq_alloc_io_context(gfp_flags); 1593 cic = cfq_alloc_io_context(cfqd, gfp_mask);
1120 1594
1121 if (cic == NULL) 1595 if (cic == NULL)
1122 goto err; 1596 goto err;
1123 1597
1598 /*
1599 * manually increment generic io_context usage count, it
1600 * cannot go away since we are already holding one ref to it
1601 */
1124 ioc->cic = cic; 1602 ioc->cic = cic;
1603 ioc->set_ioprio = cfq_ioc_set_ioprio;
1125 cic->ioc = ioc; 1604 cic->ioc = ioc;
1126 cic->cfqq = __cfqq; 1605 cic->key = cfqd;
1127 atomic_inc(&__cfqq->ref); 1606 atomic_inc(&cfqd->ref);
1128 } else { 1607 } else {
1129 struct cfq_io_context *__cic; 1608 struct cfq_io_context *__cic;
1130 unsigned long flags;
1131 1609
1132 /* 1610 /*
1133 * since the first cic on the list is actually the head 1611 * the first cic on the list is actually the head itself
1134 * itself, need to check this here or we'll duplicate an
1135 * cic per ioc for no reason
1136 */ 1612 */
1137 if (cic->cfqq == __cfqq) 1613 if (cic->key == cfqd)
1138 goto out; 1614 goto out;
1139 1615
1140 /* 1616 /*
@@ -1142,152 +1618,255 @@ cfq_get_io_context(struct cfq_queue **cfqq, int gfp_flags)
1142 * should be ok here, the list will usually not be more than 1618 * should be ok here, the list will usually not be more than
1143 * 1 or a few entries long 1619 * 1 or a few entries long
1144 */ 1620 */
1145 spin_lock_irqsave(&ioc->lock, flags);
1146 list_for_each_entry(__cic, &cic->list, list) { 1621 list_for_each_entry(__cic, &cic->list, list) {
1147 /* 1622 /*
1148 * this process is already holding a reference to 1623 * this process is already holding a reference to
1149 * this queue, so no need to get one more 1624 * this queue, so no need to get one more
1150 */ 1625 */
1151 if (__cic->cfqq == __cfqq) { 1626 if (__cic->key == cfqd) {
1152 cic = __cic; 1627 cic = __cic;
1153 spin_unlock_irqrestore(&ioc->lock, flags);
1154 goto out; 1628 goto out;
1155 } 1629 }
1156 } 1630 }
1157 spin_unlock_irqrestore(&ioc->lock, flags);
1158 1631
1159 /* 1632 /*
1160 * nope, process doesn't have a cic assoicated with this 1633 * nope, process doesn't have a cic assoicated with this
1161 * cfqq yet. get a new one and add to list 1634 * cfqq yet. get a new one and add to list
1162 */ 1635 */
1163 __cic = cfq_alloc_io_context(gfp_flags); 1636 __cic = cfq_alloc_io_context(cfqd, gfp_mask);
1164 if (__cic == NULL) 1637 if (__cic == NULL)
1165 goto err; 1638 goto err;
1166 1639
1167 __cic->ioc = ioc; 1640 __cic->ioc = ioc;
1168 __cic->cfqq = __cfqq; 1641 __cic->key = cfqd;
1169 atomic_inc(&__cfqq->ref); 1642 atomic_inc(&cfqd->ref);
1170 spin_lock_irqsave(&ioc->lock, flags);
1171 list_add(&__cic->list, &cic->list); 1643 list_add(&__cic->list, &cic->list);
1172 spin_unlock_irqrestore(&ioc->lock, flags);
1173
1174 cic = __cic; 1644 cic = __cic;
1175 *cfqq = __cfqq;
1176 } 1645 }
1177 1646
1178out: 1647out:
1179 /*
1180 * if key_type has been changed on the fly, we lazily rehash
1181 * each queue at lookup time
1182 */
1183 if ((*cfqq)->key_type != cfqd->key_type)
1184 cfq_rehash_cfqq(cfqd, cfqq, cic);
1185
1186 return cic; 1648 return cic;
1187err: 1649err:
1188 put_io_context(ioc); 1650 put_io_context(ioc);
1189 return NULL; 1651 return NULL;
1190} 1652}
1191 1653
1192static struct cfq_queue * 1654static void
1193__cfq_get_queue(struct cfq_data *cfqd, unsigned long key, int gfp_mask) 1655cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_io_context *cic)
1194{ 1656{
1195 const int hashval = hash_long(key, CFQ_QHASH_SHIFT); 1657 unsigned long elapsed, ttime;
1196 struct cfq_queue *cfqq, *new_cfqq = NULL;
1197
1198retry:
1199 cfqq = __cfq_find_cfq_hash(cfqd, key, hashval);
1200 1658
1201 if (!cfqq) { 1659 /*
1202 if (new_cfqq) { 1660 * if this context already has stuff queued, thinktime is from
1203 cfqq = new_cfqq; 1661 * last queue not last end
1204 new_cfqq = NULL; 1662 */
1205 } else { 1663#if 0
1206 spin_unlock_irq(cfqd->queue->queue_lock); 1664 if (time_after(cic->last_end_request, cic->last_queue))
1207 new_cfqq = kmem_cache_alloc(cfq_pool, gfp_mask); 1665 elapsed = jiffies - cic->last_end_request;
1208 spin_lock_irq(cfqd->queue->queue_lock); 1666 else
1667 elapsed = jiffies - cic->last_queue;
1668#else
1669 elapsed = jiffies - cic->last_end_request;
1670#endif
1209 1671
1210 if (!new_cfqq && !(gfp_mask & __GFP_WAIT)) 1672 ttime = min(elapsed, 2UL * cfqd->cfq_slice_idle);
1211 goto out;
1212 1673
1213 goto retry; 1674 cic->ttime_samples = (7*cic->ttime_samples + 256) / 8;
1214 } 1675 cic->ttime_total = (7*cic->ttime_total + 256*ttime) / 8;
1676 cic->ttime_mean = (cic->ttime_total + 128) / cic->ttime_samples;
1677}
1215 1678
1216 memset(cfqq, 0, sizeof(*cfqq)); 1679#define sample_valid(samples) ((samples) > 80)
1217 1680
1218 INIT_HLIST_NODE(&cfqq->cfq_hash); 1681/*
1219 INIT_LIST_HEAD(&cfqq->cfq_list); 1682 * Disable idle window if the process thinks too long or seeks so much that
1220 RB_CLEAR_ROOT(&cfqq->sort_list); 1683 * it doesn't matter
1221 INIT_LIST_HEAD(&cfqq->fifo[0]); 1684 */
1222 INIT_LIST_HEAD(&cfqq->fifo[1]); 1685static void
1686cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq,
1687 struct cfq_io_context *cic)
1688{
1689 int enable_idle = cfq_cfqq_idle_window(cfqq);
1223 1690
1224 cfqq->key = key; 1691 if (!cic->ioc->task || !cfqd->cfq_slice_idle)
1225 hlist_add_head(&cfqq->cfq_hash, &cfqd->cfq_hash[hashval]); 1692 enable_idle = 0;
1226 atomic_set(&cfqq->ref, 0); 1693 else if (sample_valid(cic->ttime_samples)) {
1227 cfqq->cfqd = cfqd; 1694 if (cic->ttime_mean > cfqd->cfq_slice_idle)
1228 atomic_inc(&cfqd->ref); 1695 enable_idle = 0;
1229 cfqq->key_type = cfqd->key_type; 1696 else
1230 cfqq->service_start = ~0UL; 1697 enable_idle = 1;
1231 } 1698 }
1232 1699
1233 if (new_cfqq) 1700 if (enable_idle)
1234 kmem_cache_free(cfq_pool, new_cfqq); 1701 cfq_mark_cfqq_idle_window(cfqq);
1702 else
1703 cfq_clear_cfqq_idle_window(cfqq);
1704}
1235 1705
1236 atomic_inc(&cfqq->ref); 1706
1237out: 1707/*
1238 WARN_ON((gfp_mask & __GFP_WAIT) && !cfqq); 1708 * Check if new_cfqq should preempt the currently active queue. Return 0 for
1239 return cfqq; 1709 * no or if we aren't sure, a 1 will cause a preempt.
1710 */
1711static int
1712cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq,
1713 struct cfq_rq *crq)
1714{
1715 struct cfq_queue *cfqq = cfqd->active_queue;
1716
1717 if (cfq_class_idle(new_cfqq))
1718 return 0;
1719
1720 if (!cfqq)
1721 return 1;
1722
1723 if (cfq_class_idle(cfqq))
1724 return 1;
1725 if (!cfq_cfqq_wait_request(new_cfqq))
1726 return 0;
1727 /*
1728 * if it doesn't have slice left, forget it
1729 */
1730 if (new_cfqq->slice_left < cfqd->cfq_slice_idle)
1731 return 0;
1732 if (cfq_crq_is_sync(crq) && !cfq_cfqq_sync(cfqq))
1733 return 1;
1734
1735 return 0;
1736}
1737
1738/*
1739 * cfqq preempts the active queue. if we allowed preempt with no slice left,
1740 * let it have half of its nominal slice.
1741 */
1742static void cfq_preempt_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq)
1743{
1744 struct cfq_queue *__cfqq, *next;
1745
1746 list_for_each_entry_safe(__cfqq, next, &cfqd->cur_rr, cfq_list)
1747 cfq_resort_rr_list(__cfqq, 1);
1748
1749 if (!cfqq->slice_left)
1750 cfqq->slice_left = cfq_prio_to_slice(cfqd, cfqq) / 2;
1751
1752 cfqq->slice_end = cfqq->slice_left + jiffies;
1753 __cfq_slice_expired(cfqd, cfqq, 1);
1754 __cfq_set_active_queue(cfqd, cfqq);
1755}
1756
1757/*
1758 * should really be a ll_rw_blk.c helper
1759 */
1760static void cfq_start_queueing(struct cfq_data *cfqd, struct cfq_queue *cfqq)
1761{
1762 request_queue_t *q = cfqd->queue;
1763
1764 if (!blk_queue_plugged(q))
1765 q->request_fn(q);
1766 else
1767 __generic_unplug_device(q);
1768}
1769
1770/*
1771 * Called when a new fs request (crq) is added (to cfqq). Check if there's
1772 * something we should do about it
1773 */
1774static void
1775cfq_crq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq,
1776 struct cfq_rq *crq)
1777{
1778 struct cfq_io_context *cic;
1779
1780 cfqq->next_crq = cfq_choose_req(cfqd, cfqq->next_crq, crq);
1781
1782 /*
1783 * we never wait for an async request and we don't allow preemption
1784 * of an async request. so just return early
1785 */
1786 if (!cfq_crq_is_sync(crq))
1787 return;
1788
1789 cic = crq->io_context;
1790
1791 cfq_update_io_thinktime(cfqd, cic);
1792 cfq_update_idle_window(cfqd, cfqq, cic);
1793
1794 cic->last_queue = jiffies;
1795
1796 if (cfqq == cfqd->active_queue) {
1797 /*
1798 * if we are waiting for a request for this queue, let it rip
1799 * immediately and flag that we must not expire this queue
1800 * just now
1801 */
1802 if (cfq_cfqq_wait_request(cfqq)) {
1803 cfq_mark_cfqq_must_dispatch(cfqq);
1804 del_timer(&cfqd->idle_slice_timer);
1805 cfq_start_queueing(cfqd, cfqq);
1806 }
1807 } else if (cfq_should_preempt(cfqd, cfqq, crq)) {
1808 /*
1809 * not the active queue - expire current slice if it is
1810 * idle and has expired it's mean thinktime or this new queue
1811 * has some old slice time left and is of higher priority
1812 */
1813 cfq_preempt_queue(cfqd, cfqq);
1814 cfq_mark_cfqq_must_dispatch(cfqq);
1815 cfq_start_queueing(cfqd, cfqq);
1816 }
1240} 1817}
1241 1818
1242static void cfq_enqueue(struct cfq_data *cfqd, struct cfq_rq *crq) 1819static void cfq_enqueue(struct cfq_data *cfqd, struct request *rq)
1243{ 1820{
1244 crq->is_sync = 0; 1821 struct cfq_rq *crq = RQ_DATA(rq);
1245 if (rq_data_dir(crq->request) == READ || current->flags & PF_SYNCWRITE) 1822 struct cfq_queue *cfqq = crq->cfq_queue;
1246 crq->is_sync = 1; 1823
1824 cfq_init_prio_data(cfqq);
1247 1825
1248 cfq_add_crq_rb(crq); 1826 cfq_add_crq_rb(crq);
1249 crq->queue_start = jiffies;
1250 1827
1251 list_add_tail(&crq->request->queuelist, &crq->cfq_queue->fifo[crq->is_sync]); 1828 list_add_tail(&rq->queuelist, &cfqq->fifo);
1829
1830 if (rq_mergeable(rq)) {
1831 cfq_add_crq_hash(cfqd, crq);
1832
1833 if (!cfqd->queue->last_merge)
1834 cfqd->queue->last_merge = rq;
1835 }
1836
1837 cfq_crq_enqueued(cfqd, cfqq, crq);
1252} 1838}
1253 1839
1254static void 1840static void
1255cfq_insert_request(request_queue_t *q, struct request *rq, int where) 1841cfq_insert_request(request_queue_t *q, struct request *rq, int where)
1256{ 1842{
1257 struct cfq_data *cfqd = q->elevator->elevator_data; 1843 struct cfq_data *cfqd = q->elevator->elevator_data;
1258 struct cfq_rq *crq = RQ_DATA(rq);
1259 1844
1260 switch (where) { 1845 switch (where) {
1261 case ELEVATOR_INSERT_BACK: 1846 case ELEVATOR_INSERT_BACK:
1262 while (cfq_dispatch_requests(q, cfqd->cfq_quantum)) 1847 while (cfq_dispatch_requests(q, INT_MAX, 1))
1263 ; 1848 ;
1264 list_add_tail(&rq->queuelist, &q->queue_head); 1849 list_add_tail(&rq->queuelist, &q->queue_head);
1850 /*
1851 * If we were idling with pending requests on
1852 * inactive cfqqs, force dispatching will
1853 * remove the idle timer and the queue won't
1854 * be kicked by __make_request() afterward.
1855 * Kick it here.
1856 */
1857 cfq_schedule_dispatch(cfqd);
1265 break; 1858 break;
1266 case ELEVATOR_INSERT_FRONT: 1859 case ELEVATOR_INSERT_FRONT:
1267 list_add(&rq->queuelist, &q->queue_head); 1860 list_add(&rq->queuelist, &q->queue_head);
1268 break; 1861 break;
1269 case ELEVATOR_INSERT_SORT: 1862 case ELEVATOR_INSERT_SORT:
1270 BUG_ON(!blk_fs_request(rq)); 1863 BUG_ON(!blk_fs_request(rq));
1271 cfq_enqueue(cfqd, crq); 1864 cfq_enqueue(cfqd, rq);
1272 break; 1865 break;
1273 default: 1866 default:
1274 printk("%s: bad insert point %d\n", __FUNCTION__,where); 1867 printk("%s: bad insert point %d\n", __FUNCTION__,where);
1275 return; 1868 return;
1276 } 1869 }
1277
1278 if (rq_mergeable(rq)) {
1279 cfq_add_crq_hash(cfqd, crq);
1280
1281 if (!q->last_merge)
1282 q->last_merge = rq;
1283 }
1284}
1285
1286static int cfq_queue_empty(request_queue_t *q)
1287{
1288 struct cfq_data *cfqd = q->elevator->elevator_data;
1289
1290 return list_empty(&q->queue_head) && list_empty(&cfqd->rr_list);
1291} 1870}
1292 1871
1293static void cfq_completed_request(request_queue_t *q, struct request *rq) 1872static void cfq_completed_request(request_queue_t *q, struct request *rq)
@@ -1300,9 +1879,11 @@ static void cfq_completed_request(request_queue_t *q, struct request *rq)
1300 1879
1301 cfqq = crq->cfq_queue; 1880 cfqq = crq->cfq_queue;
1302 1881
1303 if (crq->in_flight) { 1882 if (cfq_crq_in_flight(crq)) {
1304 WARN_ON(!cfqq->in_flight); 1883 const int sync = cfq_crq_is_sync(crq);
1305 cfqq->in_flight--; 1884
1885 WARN_ON(!cfqq->on_dispatch[sync]);
1886 cfqq->on_dispatch[sync]--;
1306 } 1887 }
1307 1888
1308 cfq_account_completion(cfqq, crq); 1889 cfq_account_completion(cfqq, crq);
@@ -1332,51 +1913,136 @@ cfq_latter_request(request_queue_t *q, struct request *rq)
1332 return NULL; 1913 return NULL;
1333} 1914}
1334 1915
1335static int cfq_may_queue(request_queue_t *q, int rw) 1916/*
1917 * we temporarily boost lower priority queues if they are holding fs exclusive
1918 * resources. they are boosted to normal prio (CLASS_BE/4)
1919 */
1920static void cfq_prio_boost(struct cfq_queue *cfqq)
1336{ 1921{
1337 struct cfq_data *cfqd = q->elevator->elevator_data; 1922 const int ioprio_class = cfqq->ioprio_class;
1338 struct cfq_queue *cfqq; 1923 const int ioprio = cfqq->ioprio;
1339 int ret = ELV_MQUEUE_MAY;
1340 1924
1341 if (current->flags & PF_MEMALLOC) 1925 if (has_fs_excl()) {
1342 return ELV_MQUEUE_MAY; 1926 /*
1927 * boost idle prio on transactions that would lock out other
1928 * users of the filesystem
1929 */
1930 if (cfq_class_idle(cfqq))
1931 cfqq->ioprio_class = IOPRIO_CLASS_BE;
1932 if (cfqq->ioprio > IOPRIO_NORM)
1933 cfqq->ioprio = IOPRIO_NORM;
1934 } else {
1935 /*
1936 * check if we need to unboost the queue
1937 */
1938 if (cfqq->ioprio_class != cfqq->org_ioprio_class)
1939 cfqq->ioprio_class = cfqq->org_ioprio_class;
1940 if (cfqq->ioprio != cfqq->org_ioprio)
1941 cfqq->ioprio = cfqq->org_ioprio;
1942 }
1343 1943
1344 cfqq = cfq_find_cfq_hash(cfqd, cfq_hash_key(cfqd, current)); 1944 /*
1345 if (cfqq) { 1945 * refile between round-robin lists if we moved the priority class
1346 int limit = cfqd->max_queued; 1946 */
1947 if ((ioprio_class != cfqq->ioprio_class || ioprio != cfqq->ioprio) &&
1948 cfq_cfqq_on_rr(cfqq))
1949 cfq_resort_rr_list(cfqq, 0);
1950}
1347 1951
1348 if (cfqq->allocated[rw] < cfqd->cfq_queued) 1952static inline pid_t cfq_queue_pid(struct task_struct *task, int rw)
1349 return ELV_MQUEUE_MUST; 1953{
1954 if (rw == READ || process_sync(task))
1955 return task->pid;
1350 1956
1351 if (cfqd->busy_queues) 1957 return CFQ_KEY_ASYNC;
1352 limit = q->nr_requests / cfqd->busy_queues; 1958}
1353 1959
1354 if (limit < cfqd->cfq_queued) 1960static inline int
1355 limit = cfqd->cfq_queued; 1961__cfq_may_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq,
1356 else if (limit > cfqd->max_queued) 1962 struct task_struct *task, int rw)
1357 limit = cfqd->max_queued; 1963{
1964#if 1
1965 if ((cfq_cfqq_wait_request(cfqq) || cfq_cfqq_must_alloc(cfqq)) &&
1966 !cfq_cfqq_must_alloc_slice(cfqq)) {
1967 cfq_mark_cfqq_must_alloc_slice(cfqq);
1968 return ELV_MQUEUE_MUST;
1969 }
1358 1970
1359 if (cfqq->allocated[rw] >= limit) { 1971 return ELV_MQUEUE_MAY;
1360 if (limit > cfqq->alloc_limit[rw]) 1972#else
1361 cfqq->alloc_limit[rw] = limit; 1973 if (!cfqq || task->flags & PF_MEMALLOC)
1974 return ELV_MQUEUE_MAY;
1975 if (!cfqq->allocated[rw] || cfq_cfqq_must_alloc(cfqq)) {
1976 if (cfq_cfqq_wait_request(cfqq))
1977 return ELV_MQUEUE_MUST;
1362 1978
1363 ret = ELV_MQUEUE_NO; 1979 /*
1980 * only allow 1 ELV_MQUEUE_MUST per slice, otherwise we
1981 * can quickly flood the queue with writes from a single task
1982 */
1983 if (rw == READ || !cfq_cfqq_must_alloc_slice(cfqq)) {
1984 cfq_mark_cfqq_must_alloc_slice(cfqq);
1985 return ELV_MQUEUE_MUST;
1364 } 1986 }
1987
1988 return ELV_MQUEUE_MAY;
1365 } 1989 }
1990 if (cfq_class_idle(cfqq))
1991 return ELV_MQUEUE_NO;
1992 if (cfqq->allocated[rw] >= cfqd->max_queued) {
1993 struct io_context *ioc = get_io_context(GFP_ATOMIC);
1994 int ret = ELV_MQUEUE_NO;
1366 1995
1367 return ret; 1996 if (ioc && ioc->nr_batch_requests)
1997 ret = ELV_MQUEUE_MAY;
1998
1999 put_io_context(ioc);
2000 return ret;
2001 }
2002
2003 return ELV_MQUEUE_MAY;
2004#endif
2005}
2006
2007static int cfq_may_queue(request_queue_t *q, int rw, struct bio *bio)
2008{
2009 struct cfq_data *cfqd = q->elevator->elevator_data;
2010 struct task_struct *tsk = current;
2011 struct cfq_queue *cfqq;
2012
2013 /*
2014 * don't force setup of a queue from here, as a call to may_queue
2015 * does not necessarily imply that a request actually will be queued.
2016 * so just lookup a possibly existing queue, or return 'may queue'
2017 * if that fails
2018 */
2019 cfqq = cfq_find_cfq_hash(cfqd, cfq_queue_pid(tsk, rw), tsk->ioprio);
2020 if (cfqq) {
2021 cfq_init_prio_data(cfqq);
2022 cfq_prio_boost(cfqq);
2023
2024 return __cfq_may_queue(cfqd, cfqq, tsk, rw);
2025 }
2026
2027 return ELV_MQUEUE_MAY;
1368} 2028}
1369 2029
1370static void cfq_check_waiters(request_queue_t *q, struct cfq_queue *cfqq) 2030static void cfq_check_waiters(request_queue_t *q, struct cfq_queue *cfqq)
1371{ 2031{
2032 struct cfq_data *cfqd = q->elevator->elevator_data;
1372 struct request_list *rl = &q->rq; 2033 struct request_list *rl = &q->rq;
1373 const int write = waitqueue_active(&rl->wait[WRITE]);
1374 const int read = waitqueue_active(&rl->wait[READ]);
1375 2034
1376 if (read && cfqq->allocated[READ] < cfqq->alloc_limit[READ]) 2035 if (cfqq->allocated[READ] <= cfqd->max_queued || cfqd->rq_starved) {
1377 wake_up(&rl->wait[READ]); 2036 smp_mb();
1378 if (write && cfqq->allocated[WRITE] < cfqq->alloc_limit[WRITE]) 2037 if (waitqueue_active(&rl->wait[READ]))
1379 wake_up(&rl->wait[WRITE]); 2038 wake_up(&rl->wait[READ]);
2039 }
2040
2041 if (cfqq->allocated[WRITE] <= cfqd->max_queued || cfqd->rq_starved) {
2042 smp_mb();
2043 if (waitqueue_active(&rl->wait[WRITE]))
2044 wake_up(&rl->wait[WRITE]);
2045 }
1380} 2046}
1381 2047
1382/* 2048/*
@@ -1389,69 +2055,61 @@ static void cfq_put_request(request_queue_t *q, struct request *rq)
1389 2055
1390 if (crq) { 2056 if (crq) {
1391 struct cfq_queue *cfqq = crq->cfq_queue; 2057 struct cfq_queue *cfqq = crq->cfq_queue;
2058 const int rw = rq_data_dir(rq);
1392 2059
1393 BUG_ON(q->last_merge == rq); 2060 BUG_ON(!cfqq->allocated[rw]);
1394 BUG_ON(!hlist_unhashed(&crq->hash)); 2061 cfqq->allocated[rw]--;
1395 2062
1396 if (crq->io_context) 2063 put_io_context(crq->io_context->ioc);
1397 put_io_context(crq->io_context->ioc);
1398
1399 BUG_ON(!cfqq->allocated[crq->is_write]);
1400 cfqq->allocated[crq->is_write]--;
1401 2064
1402 mempool_free(crq, cfqd->crq_pool); 2065 mempool_free(crq, cfqd->crq_pool);
1403 rq->elevator_private = NULL; 2066 rq->elevator_private = NULL;
1404 2067
1405 smp_mb();
1406 cfq_check_waiters(q, cfqq); 2068 cfq_check_waiters(q, cfqq);
1407 cfq_put_queue(cfqq); 2069 cfq_put_queue(cfqq);
1408 } 2070 }
1409} 2071}
1410 2072
1411/* 2073/*
1412 * Allocate cfq data structures associated with this request. A queue and 2074 * Allocate cfq data structures associated with this request.
1413 */ 2075 */
1414static int cfq_set_request(request_queue_t *q, struct request *rq, int gfp_mask) 2076static int
2077cfq_set_request(request_queue_t *q, struct request *rq, struct bio *bio,
2078 int gfp_mask)
1415{ 2079{
1416 struct cfq_data *cfqd = q->elevator->elevator_data; 2080 struct cfq_data *cfqd = q->elevator->elevator_data;
2081 struct task_struct *tsk = current;
1417 struct cfq_io_context *cic; 2082 struct cfq_io_context *cic;
1418 const int rw = rq_data_dir(rq); 2083 const int rw = rq_data_dir(rq);
1419 struct cfq_queue *cfqq, *saved_cfqq; 2084 pid_t key = cfq_queue_pid(tsk, rw);
2085 struct cfq_queue *cfqq;
1420 struct cfq_rq *crq; 2086 struct cfq_rq *crq;
1421 unsigned long flags; 2087 unsigned long flags;
1422 2088
1423 might_sleep_if(gfp_mask & __GFP_WAIT); 2089 might_sleep_if(gfp_mask & __GFP_WAIT);
1424 2090
2091 cic = cfq_get_io_context(cfqd, key, gfp_mask);
2092
1425 spin_lock_irqsave(q->queue_lock, flags); 2093 spin_lock_irqsave(q->queue_lock, flags);
1426 2094
1427 cfqq = __cfq_get_queue(cfqd, cfq_hash_key(cfqd, current), gfp_mask); 2095 if (!cic)
1428 if (!cfqq) 2096 goto queue_fail;
1429 goto out_lock; 2097
2098 if (!cic->cfqq) {
2099 cfqq = cfq_get_queue(cfqd, key, tsk->ioprio, gfp_mask);
2100 if (!cfqq)
2101 goto queue_fail;
1430 2102
1431repeat: 2103 cic->cfqq = cfqq;
1432 if (cfqq->allocated[rw] >= cfqd->max_queued) 2104 } else
1433 goto out_lock; 2105 cfqq = cic->cfqq;
1434 2106
1435 cfqq->allocated[rw]++; 2107 cfqq->allocated[rw]++;
2108 cfq_clear_cfqq_must_alloc(cfqq);
2109 cfqd->rq_starved = 0;
2110 atomic_inc(&cfqq->ref);
1436 spin_unlock_irqrestore(q->queue_lock, flags); 2111 spin_unlock_irqrestore(q->queue_lock, flags);
1437 2112
1438 /*
1439 * if hashing type has changed, the cfq_queue might change here.
1440 */
1441 saved_cfqq = cfqq;
1442 cic = cfq_get_io_context(&cfqq, gfp_mask);
1443 if (!cic)
1444 goto err;
1445
1446 /*
1447 * repeat allocation checks on queue change
1448 */
1449 if (unlikely(saved_cfqq != cfqq)) {
1450 spin_lock_irqsave(q->queue_lock, flags);
1451 saved_cfqq->allocated[rw]--;
1452 goto repeat;
1453 }
1454
1455 crq = mempool_alloc(cfqd->crq_pool, gfp_mask); 2113 crq = mempool_alloc(cfqd->crq_pool, gfp_mask);
1456 if (crq) { 2114 if (crq) {
1457 RB_CLEAR(&crq->rb_node); 2115 RB_CLEAR(&crq->rb_node);
@@ -1460,24 +2118,141 @@ repeat:
1460 INIT_HLIST_NODE(&crq->hash); 2118 INIT_HLIST_NODE(&crq->hash);
1461 crq->cfq_queue = cfqq; 2119 crq->cfq_queue = cfqq;
1462 crq->io_context = cic; 2120 crq->io_context = cic;
1463 crq->service_start = crq->queue_start = 0; 2121 cfq_clear_crq_in_flight(crq);
1464 crq->in_flight = crq->accounted = crq->is_sync = 0; 2122 cfq_clear_crq_in_driver(crq);
1465 crq->is_write = rw; 2123 cfq_clear_crq_requeued(crq);
2124
2125 if (rw == READ || process_sync(tsk))
2126 cfq_mark_crq_is_sync(crq);
2127 else
2128 cfq_clear_crq_is_sync(crq);
2129
1466 rq->elevator_private = crq; 2130 rq->elevator_private = crq;
1467 cfqq->alloc_limit[rw] = 0;
1468 return 0; 2131 return 0;
1469 } 2132 }
1470 2133
1471 put_io_context(cic->ioc);
1472err:
1473 spin_lock_irqsave(q->queue_lock, flags); 2134 spin_lock_irqsave(q->queue_lock, flags);
1474 cfqq->allocated[rw]--; 2135 cfqq->allocated[rw]--;
2136 if (!(cfqq->allocated[0] + cfqq->allocated[1]))
2137 cfq_mark_cfqq_must_alloc(cfqq);
1475 cfq_put_queue(cfqq); 2138 cfq_put_queue(cfqq);
1476out_lock: 2139queue_fail:
2140 if (cic)
2141 put_io_context(cic->ioc);
2142 /*
2143 * mark us rq allocation starved. we need to kickstart the process
2144 * ourselves if there are no pending requests that can do it for us.
2145 * that would be an extremely rare OOM situation
2146 */
2147 cfqd->rq_starved = 1;
2148 cfq_schedule_dispatch(cfqd);
1477 spin_unlock_irqrestore(q->queue_lock, flags); 2149 spin_unlock_irqrestore(q->queue_lock, flags);
1478 return 1; 2150 return 1;
1479} 2151}
1480 2152
2153static void cfq_kick_queue(void *data)
2154{
2155 request_queue_t *q = data;
2156 struct cfq_data *cfqd = q->elevator->elevator_data;
2157 unsigned long flags;
2158
2159 spin_lock_irqsave(q->queue_lock, flags);
2160
2161 if (cfqd->rq_starved) {
2162 struct request_list *rl = &q->rq;
2163
2164 /*
2165 * we aren't guaranteed to get a request after this, but we
2166 * have to be opportunistic
2167 */
2168 smp_mb();
2169 if (waitqueue_active(&rl->wait[READ]))
2170 wake_up(&rl->wait[READ]);
2171 if (waitqueue_active(&rl->wait[WRITE]))
2172 wake_up(&rl->wait[WRITE]);
2173 }
2174
2175 blk_remove_plug(q);
2176 q->request_fn(q);
2177 spin_unlock_irqrestore(q->queue_lock, flags);
2178}
2179
2180/*
2181 * Timer running if the active_queue is currently idling inside its time slice
2182 */
2183static void cfq_idle_slice_timer(unsigned long data)
2184{
2185 struct cfq_data *cfqd = (struct cfq_data *) data;
2186 struct cfq_queue *cfqq;
2187 unsigned long flags;
2188
2189 spin_lock_irqsave(cfqd->queue->queue_lock, flags);
2190
2191 if ((cfqq = cfqd->active_queue) != NULL) {
2192 unsigned long now = jiffies;
2193
2194 /*
2195 * expired
2196 */
2197 if (time_after(now, cfqq->slice_end))
2198 goto expire;
2199
2200 /*
2201 * only expire and reinvoke request handler, if there are
2202 * other queues with pending requests
2203 */
2204 if (!cfq_pending_requests(cfqd)) {
2205 cfqd->idle_slice_timer.expires = min(now + cfqd->cfq_slice_idle, cfqq->slice_end);
2206 add_timer(&cfqd->idle_slice_timer);
2207 goto out_cont;
2208 }
2209
2210 /*
2211 * not expired and it has a request pending, let it dispatch
2212 */
2213 if (!RB_EMPTY(&cfqq->sort_list)) {
2214 cfq_mark_cfqq_must_dispatch(cfqq);
2215 goto out_kick;
2216 }
2217 }
2218expire:
2219 cfq_slice_expired(cfqd, 0);
2220out_kick:
2221 cfq_schedule_dispatch(cfqd);
2222out_cont:
2223 spin_unlock_irqrestore(cfqd->queue->queue_lock, flags);
2224}
2225
2226/*
2227 * Timer running if an idle class queue is waiting for service
2228 */
2229static void cfq_idle_class_timer(unsigned long data)
2230{
2231 struct cfq_data *cfqd = (struct cfq_data *) data;
2232 unsigned long flags, end;
2233
2234 spin_lock_irqsave(cfqd->queue->queue_lock, flags);
2235
2236 /*
2237 * race with a non-idle queue, reset timer
2238 */
2239 end = cfqd->last_end_request + CFQ_IDLE_GRACE;
2240 if (!time_after_eq(jiffies, end)) {
2241 cfqd->idle_class_timer.expires = end;
2242 add_timer(&cfqd->idle_class_timer);
2243 } else
2244 cfq_schedule_dispatch(cfqd);
2245
2246 spin_unlock_irqrestore(cfqd->queue->queue_lock, flags);
2247}
2248
2249static void cfq_shutdown_timer_wq(struct cfq_data *cfqd)
2250{
2251 del_timer_sync(&cfqd->idle_slice_timer);
2252 del_timer_sync(&cfqd->idle_class_timer);
2253 blk_sync_queue(cfqd->queue);
2254}
2255
1481static void cfq_put_cfqd(struct cfq_data *cfqd) 2256static void cfq_put_cfqd(struct cfq_data *cfqd)
1482{ 2257{
1483 request_queue_t *q = cfqd->queue; 2258 request_queue_t *q = cfqd->queue;
@@ -1487,6 +2262,9 @@ static void cfq_put_cfqd(struct cfq_data *cfqd)
1487 2262
1488 blk_put_queue(q); 2263 blk_put_queue(q);
1489 2264
2265 cfq_shutdown_timer_wq(cfqd);
2266 q->elevator->elevator_data = NULL;
2267
1490 mempool_destroy(cfqd->crq_pool); 2268 mempool_destroy(cfqd->crq_pool);
1491 kfree(cfqd->crq_hash); 2269 kfree(cfqd->crq_hash);
1492 kfree(cfqd->cfq_hash); 2270 kfree(cfqd->cfq_hash);
@@ -1495,7 +2273,10 @@ static void cfq_put_cfqd(struct cfq_data *cfqd)
1495 2273
1496static void cfq_exit_queue(elevator_t *e) 2274static void cfq_exit_queue(elevator_t *e)
1497{ 2275{
1498 cfq_put_cfqd(e->elevator_data); 2276 struct cfq_data *cfqd = e->elevator_data;
2277
2278 cfq_shutdown_timer_wq(cfqd);
2279 cfq_put_cfqd(cfqd);
1499} 2280}
1500 2281
1501static int cfq_init_queue(request_queue_t *q, elevator_t *e) 2282static int cfq_init_queue(request_queue_t *q, elevator_t *e)
@@ -1508,7 +2289,13 @@ static int cfq_init_queue(request_queue_t *q, elevator_t *e)
1508 return -ENOMEM; 2289 return -ENOMEM;
1509 2290
1510 memset(cfqd, 0, sizeof(*cfqd)); 2291 memset(cfqd, 0, sizeof(*cfqd));
1511 INIT_LIST_HEAD(&cfqd->rr_list); 2292
2293 for (i = 0; i < CFQ_PRIO_LISTS; i++)
2294 INIT_LIST_HEAD(&cfqd->rr_list[i]);
2295
2296 INIT_LIST_HEAD(&cfqd->busy_rr);
2297 INIT_LIST_HEAD(&cfqd->cur_rr);
2298 INIT_LIST_HEAD(&cfqd->idle_rr);
1512 INIT_LIST_HEAD(&cfqd->empty_list); 2299 INIT_LIST_HEAD(&cfqd->empty_list);
1513 2300
1514 cfqd->crq_hash = kmalloc(sizeof(struct hlist_head) * CFQ_MHASH_ENTRIES, GFP_KERNEL); 2301 cfqd->crq_hash = kmalloc(sizeof(struct hlist_head) * CFQ_MHASH_ENTRIES, GFP_KERNEL);
@@ -1533,24 +2320,32 @@ static int cfq_init_queue(request_queue_t *q, elevator_t *e)
1533 cfqd->queue = q; 2320 cfqd->queue = q;
1534 atomic_inc(&q->refcnt); 2321 atomic_inc(&q->refcnt);
1535 2322
1536 /* 2323 cfqd->max_queued = q->nr_requests / 4;
1537 * just set it to some high value, we want anyone to be able to queue
1538 * some requests. fairness is handled differently
1539 */
1540 q->nr_requests = 1024;
1541 cfqd->max_queued = q->nr_requests / 16;
1542 q->nr_batching = cfq_queued; 2324 q->nr_batching = cfq_queued;
1543 cfqd->key_type = CFQ_KEY_TGID; 2325
1544 cfqd->find_best_crq = 1; 2326 init_timer(&cfqd->idle_slice_timer);
2327 cfqd->idle_slice_timer.function = cfq_idle_slice_timer;
2328 cfqd->idle_slice_timer.data = (unsigned long) cfqd;
2329
2330 init_timer(&cfqd->idle_class_timer);
2331 cfqd->idle_class_timer.function = cfq_idle_class_timer;
2332 cfqd->idle_class_timer.data = (unsigned long) cfqd;
2333
2334 INIT_WORK(&cfqd->unplug_work, cfq_kick_queue, q);
2335
1545 atomic_set(&cfqd->ref, 1); 2336 atomic_set(&cfqd->ref, 1);
1546 2337
1547 cfqd->cfq_queued = cfq_queued; 2338 cfqd->cfq_queued = cfq_queued;
1548 cfqd->cfq_quantum = cfq_quantum; 2339 cfqd->cfq_quantum = cfq_quantum;
1549 cfqd->cfq_fifo_expire_r = cfq_fifo_expire_r; 2340 cfqd->cfq_fifo_expire[0] = cfq_fifo_expire[0];
1550 cfqd->cfq_fifo_expire_w = cfq_fifo_expire_w; 2341 cfqd->cfq_fifo_expire[1] = cfq_fifo_expire[1];
1551 cfqd->cfq_fifo_batch_expire = cfq_fifo_rate;
1552 cfqd->cfq_back_max = cfq_back_max; 2342 cfqd->cfq_back_max = cfq_back_max;
1553 cfqd->cfq_back_penalty = cfq_back_penalty; 2343 cfqd->cfq_back_penalty = cfq_back_penalty;
2344 cfqd->cfq_slice[0] = cfq_slice_async;
2345 cfqd->cfq_slice[1] = cfq_slice_sync;
2346 cfqd->cfq_slice_async_rq = cfq_slice_async_rq;
2347 cfqd->cfq_slice_idle = cfq_slice_idle;
2348 cfqd->cfq_max_depth = cfq_max_depth;
1554 2349
1555 return 0; 2350 return 0;
1556out_crqpool: 2351out_crqpool:
@@ -1595,7 +2390,6 @@ fail:
1595 return -ENOMEM; 2390 return -ENOMEM;
1596} 2391}
1597 2392
1598
1599/* 2393/*
1600 * sysfs parts below --> 2394 * sysfs parts below -->
1601 */ 2395 */
@@ -1620,45 +2414,6 @@ cfq_var_store(unsigned int *var, const char *page, size_t count)
1620 return count; 2414 return count;
1621} 2415}
1622 2416
1623static ssize_t
1624cfq_clear_elapsed(struct cfq_data *cfqd, const char *page, size_t count)
1625{
1626 max_elapsed_dispatch = max_elapsed_crq = 0;
1627 return count;
1628}
1629
1630static ssize_t
1631cfq_set_key_type(struct cfq_data *cfqd, const char *page, size_t count)
1632{
1633 spin_lock_irq(cfqd->queue->queue_lock);
1634 if (!strncmp(page, "pgid", 4))
1635 cfqd->key_type = CFQ_KEY_PGID;
1636 else if (!strncmp(page, "tgid", 4))
1637 cfqd->key_type = CFQ_KEY_TGID;
1638 else if (!strncmp(page, "uid", 3))
1639 cfqd->key_type = CFQ_KEY_UID;
1640 else if (!strncmp(page, "gid", 3))
1641 cfqd->key_type = CFQ_KEY_GID;
1642 spin_unlock_irq(cfqd->queue->queue_lock);
1643 return count;
1644}
1645
1646static ssize_t
1647cfq_read_key_type(struct cfq_data *cfqd, char *page)
1648{
1649 ssize_t len = 0;
1650 int i;
1651
1652 for (i = CFQ_KEY_PGID; i < CFQ_KEY_LAST; i++) {
1653 if (cfqd->key_type == i)
1654 len += sprintf(page+len, "[%s] ", cfq_key_types[i]);
1655 else
1656 len += sprintf(page+len, "%s ", cfq_key_types[i]);
1657 }
1658 len += sprintf(page+len, "\n");
1659 return len;
1660}
1661
1662#define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \ 2417#define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \
1663static ssize_t __FUNC(struct cfq_data *cfqd, char *page) \ 2418static ssize_t __FUNC(struct cfq_data *cfqd, char *page) \
1664{ \ 2419{ \
@@ -1669,12 +2424,15 @@ static ssize_t __FUNC(struct cfq_data *cfqd, char *page) \
1669} 2424}
1670SHOW_FUNCTION(cfq_quantum_show, cfqd->cfq_quantum, 0); 2425SHOW_FUNCTION(cfq_quantum_show, cfqd->cfq_quantum, 0);
1671SHOW_FUNCTION(cfq_queued_show, cfqd->cfq_queued, 0); 2426SHOW_FUNCTION(cfq_queued_show, cfqd->cfq_queued, 0);
1672SHOW_FUNCTION(cfq_fifo_expire_r_show, cfqd->cfq_fifo_expire_r, 1); 2427SHOW_FUNCTION(cfq_fifo_expire_sync_show, cfqd->cfq_fifo_expire[1], 1);
1673SHOW_FUNCTION(cfq_fifo_expire_w_show, cfqd->cfq_fifo_expire_w, 1); 2428SHOW_FUNCTION(cfq_fifo_expire_async_show, cfqd->cfq_fifo_expire[0], 1);
1674SHOW_FUNCTION(cfq_fifo_batch_expire_show, cfqd->cfq_fifo_batch_expire, 1);
1675SHOW_FUNCTION(cfq_find_best_show, cfqd->find_best_crq, 0);
1676SHOW_FUNCTION(cfq_back_max_show, cfqd->cfq_back_max, 0); 2429SHOW_FUNCTION(cfq_back_max_show, cfqd->cfq_back_max, 0);
1677SHOW_FUNCTION(cfq_back_penalty_show, cfqd->cfq_back_penalty, 0); 2430SHOW_FUNCTION(cfq_back_penalty_show, cfqd->cfq_back_penalty, 0);
2431SHOW_FUNCTION(cfq_slice_idle_show, cfqd->cfq_slice_idle, 1);
2432SHOW_FUNCTION(cfq_slice_sync_show, cfqd->cfq_slice[1], 1);
2433SHOW_FUNCTION(cfq_slice_async_show, cfqd->cfq_slice[0], 1);
2434SHOW_FUNCTION(cfq_slice_async_rq_show, cfqd->cfq_slice_async_rq, 0);
2435SHOW_FUNCTION(cfq_max_depth_show, cfqd->cfq_max_depth, 0);
1678#undef SHOW_FUNCTION 2436#undef SHOW_FUNCTION
1679 2437
1680#define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \ 2438#define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \
@@ -1694,12 +2452,15 @@ static ssize_t __FUNC(struct cfq_data *cfqd, const char *page, size_t count) \
1694} 2452}
1695STORE_FUNCTION(cfq_quantum_store, &cfqd->cfq_quantum, 1, UINT_MAX, 0); 2453STORE_FUNCTION(cfq_quantum_store, &cfqd->cfq_quantum, 1, UINT_MAX, 0);
1696STORE_FUNCTION(cfq_queued_store, &cfqd->cfq_queued, 1, UINT_MAX, 0); 2454STORE_FUNCTION(cfq_queued_store, &cfqd->cfq_queued, 1, UINT_MAX, 0);
1697STORE_FUNCTION(cfq_fifo_expire_r_store, &cfqd->cfq_fifo_expire_r, 1, UINT_MAX, 1); 2455STORE_FUNCTION(cfq_fifo_expire_sync_store, &cfqd->cfq_fifo_expire[1], 1, UINT_MAX, 1);
1698STORE_FUNCTION(cfq_fifo_expire_w_store, &cfqd->cfq_fifo_expire_w, 1, UINT_MAX, 1); 2456STORE_FUNCTION(cfq_fifo_expire_async_store, &cfqd->cfq_fifo_expire[0], 1, UINT_MAX, 1);
1699STORE_FUNCTION(cfq_fifo_batch_expire_store, &cfqd->cfq_fifo_batch_expire, 0, UINT_MAX, 1);
1700STORE_FUNCTION(cfq_find_best_store, &cfqd->find_best_crq, 0, 1, 0);
1701STORE_FUNCTION(cfq_back_max_store, &cfqd->cfq_back_max, 0, UINT_MAX, 0); 2457STORE_FUNCTION(cfq_back_max_store, &cfqd->cfq_back_max, 0, UINT_MAX, 0);
1702STORE_FUNCTION(cfq_back_penalty_store, &cfqd->cfq_back_penalty, 1, UINT_MAX, 0); 2458STORE_FUNCTION(cfq_back_penalty_store, &cfqd->cfq_back_penalty, 1, UINT_MAX, 0);
2459STORE_FUNCTION(cfq_slice_idle_store, &cfqd->cfq_slice_idle, 0, UINT_MAX, 1);
2460STORE_FUNCTION(cfq_slice_sync_store, &cfqd->cfq_slice[1], 1, UINT_MAX, 1);
2461STORE_FUNCTION(cfq_slice_async_store, &cfqd->cfq_slice[0], 1, UINT_MAX, 1);
2462STORE_FUNCTION(cfq_slice_async_rq_store, &cfqd->cfq_slice_async_rq, 1, UINT_MAX, 0);
2463STORE_FUNCTION(cfq_max_depth_store, &cfqd->cfq_max_depth, 1, UINT_MAX, 0);
1703#undef STORE_FUNCTION 2464#undef STORE_FUNCTION
1704 2465
1705static struct cfq_fs_entry cfq_quantum_entry = { 2466static struct cfq_fs_entry cfq_quantum_entry = {
@@ -1712,25 +2473,15 @@ static struct cfq_fs_entry cfq_queued_entry = {
1712 .show = cfq_queued_show, 2473 .show = cfq_queued_show,
1713 .store = cfq_queued_store, 2474 .store = cfq_queued_store,
1714}; 2475};
1715static struct cfq_fs_entry cfq_fifo_expire_r_entry = { 2476static struct cfq_fs_entry cfq_fifo_expire_sync_entry = {
1716 .attr = {.name = "fifo_expire_sync", .mode = S_IRUGO | S_IWUSR }, 2477 .attr = {.name = "fifo_expire_sync", .mode = S_IRUGO | S_IWUSR },
1717 .show = cfq_fifo_expire_r_show, 2478 .show = cfq_fifo_expire_sync_show,
1718 .store = cfq_fifo_expire_r_store, 2479 .store = cfq_fifo_expire_sync_store,
1719}; 2480};
1720static struct cfq_fs_entry cfq_fifo_expire_w_entry = { 2481static struct cfq_fs_entry cfq_fifo_expire_async_entry = {
1721 .attr = {.name = "fifo_expire_async", .mode = S_IRUGO | S_IWUSR }, 2482 .attr = {.name = "fifo_expire_async", .mode = S_IRUGO | S_IWUSR },
1722 .show = cfq_fifo_expire_w_show, 2483 .show = cfq_fifo_expire_async_show,
1723 .store = cfq_fifo_expire_w_store, 2484 .store = cfq_fifo_expire_async_store,
1724};
1725static struct cfq_fs_entry cfq_fifo_batch_expire_entry = {
1726 .attr = {.name = "fifo_batch_expire", .mode = S_IRUGO | S_IWUSR },
1727 .show = cfq_fifo_batch_expire_show,
1728 .store = cfq_fifo_batch_expire_store,
1729};
1730static struct cfq_fs_entry cfq_find_best_entry = {
1731 .attr = {.name = "find_best_crq", .mode = S_IRUGO | S_IWUSR },
1732 .show = cfq_find_best_show,
1733 .store = cfq_find_best_store,
1734}; 2485};
1735static struct cfq_fs_entry cfq_back_max_entry = { 2486static struct cfq_fs_entry cfq_back_max_entry = {
1736 .attr = {.name = "back_seek_max", .mode = S_IRUGO | S_IWUSR }, 2487 .attr = {.name = "back_seek_max", .mode = S_IRUGO | S_IWUSR },
@@ -1742,27 +2493,44 @@ static struct cfq_fs_entry cfq_back_penalty_entry = {
1742 .show = cfq_back_penalty_show, 2493 .show = cfq_back_penalty_show,
1743 .store = cfq_back_penalty_store, 2494 .store = cfq_back_penalty_store,
1744}; 2495};
1745static struct cfq_fs_entry cfq_clear_elapsed_entry = { 2496static struct cfq_fs_entry cfq_slice_sync_entry = {
1746 .attr = {.name = "clear_elapsed", .mode = S_IWUSR }, 2497 .attr = {.name = "slice_sync", .mode = S_IRUGO | S_IWUSR },
1747 .store = cfq_clear_elapsed, 2498 .show = cfq_slice_sync_show,
2499 .store = cfq_slice_sync_store,
2500};
2501static struct cfq_fs_entry cfq_slice_async_entry = {
2502 .attr = {.name = "slice_async", .mode = S_IRUGO | S_IWUSR },
2503 .show = cfq_slice_async_show,
2504 .store = cfq_slice_async_store,
1748}; 2505};
1749static struct cfq_fs_entry cfq_key_type_entry = { 2506static struct cfq_fs_entry cfq_slice_async_rq_entry = {
1750 .attr = {.name = "key_type", .mode = S_IRUGO | S_IWUSR }, 2507 .attr = {.name = "slice_async_rq", .mode = S_IRUGO | S_IWUSR },
1751 .show = cfq_read_key_type, 2508 .show = cfq_slice_async_rq_show,
1752 .store = cfq_set_key_type, 2509 .store = cfq_slice_async_rq_store,
2510};
2511static struct cfq_fs_entry cfq_slice_idle_entry = {
2512 .attr = {.name = "slice_idle", .mode = S_IRUGO | S_IWUSR },
2513 .show = cfq_slice_idle_show,
2514 .store = cfq_slice_idle_store,
2515};
2516static struct cfq_fs_entry cfq_max_depth_entry = {
2517 .attr = {.name = "max_depth", .mode = S_IRUGO | S_IWUSR },
2518 .show = cfq_max_depth_show,
2519 .store = cfq_max_depth_store,
1753}; 2520};
1754 2521
1755static struct attribute *default_attrs[] = { 2522static struct attribute *default_attrs[] = {
1756 &cfq_quantum_entry.attr, 2523 &cfq_quantum_entry.attr,
1757 &cfq_queued_entry.attr, 2524 &cfq_queued_entry.attr,
1758 &cfq_fifo_expire_r_entry.attr, 2525 &cfq_fifo_expire_sync_entry.attr,
1759 &cfq_fifo_expire_w_entry.attr, 2526 &cfq_fifo_expire_async_entry.attr,
1760 &cfq_fifo_batch_expire_entry.attr,
1761 &cfq_key_type_entry.attr,
1762 &cfq_find_best_entry.attr,
1763 &cfq_back_max_entry.attr, 2527 &cfq_back_max_entry.attr,
1764 &cfq_back_penalty_entry.attr, 2528 &cfq_back_penalty_entry.attr,
1765 &cfq_clear_elapsed_entry.attr, 2529 &cfq_slice_sync_entry.attr,
2530 &cfq_slice_async_entry.attr,
2531 &cfq_slice_async_rq_entry.attr,
2532 &cfq_slice_idle_entry.attr,
2533 &cfq_max_depth_entry.attr,
1766 NULL, 2534 NULL,
1767}; 2535};
1768 2536
@@ -1775,7 +2543,7 @@ cfq_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1775 struct cfq_fs_entry *entry = to_cfq(attr); 2543 struct cfq_fs_entry *entry = to_cfq(attr);
1776 2544
1777 if (!entry->show) 2545 if (!entry->show)
1778 return 0; 2546 return -EIO;
1779 2547
1780 return entry->show(e->elevator_data, page); 2548 return entry->show(e->elevator_data, page);
1781} 2549}
@@ -1788,7 +2556,7 @@ cfq_attr_store(struct kobject *kobj, struct attribute *attr,
1788 struct cfq_fs_entry *entry = to_cfq(attr); 2556 struct cfq_fs_entry *entry = to_cfq(attr);
1789 2557
1790 if (!entry->store) 2558 if (!entry->store)
1791 return -EINVAL; 2559 return -EIO;
1792 2560
1793 return entry->store(e->elevator_data, page, length); 2561 return entry->store(e->elevator_data, page, length);
1794} 2562}
@@ -1832,21 +2600,46 @@ static int __init cfq_init(void)
1832{ 2600{
1833 int ret; 2601 int ret;
1834 2602
2603 /*
2604 * could be 0 on HZ < 1000 setups
2605 */
2606 if (!cfq_slice_async)
2607 cfq_slice_async = 1;
2608 if (!cfq_slice_idle)
2609 cfq_slice_idle = 1;
2610
1835 if (cfq_slab_setup()) 2611 if (cfq_slab_setup())
1836 return -ENOMEM; 2612 return -ENOMEM;
1837 2613
1838 ret = elv_register(&iosched_cfq); 2614 ret = elv_register(&iosched_cfq);
1839 if (!ret) { 2615 if (ret)
1840 __module_get(THIS_MODULE); 2616 cfq_slab_kill();
1841 return 0;
1842 }
1843 2617
1844 cfq_slab_kill();
1845 return ret; 2618 return ret;
1846} 2619}
1847 2620
1848static void __exit cfq_exit(void) 2621static void __exit cfq_exit(void)
1849{ 2622{
2623 struct task_struct *g, *p;
2624 unsigned long flags;
2625
2626 read_lock_irqsave(&tasklist_lock, flags);
2627
2628 /*
2629 * iterate each process in the system, removing our io_context
2630 */
2631 do_each_thread(g, p) {
2632 struct io_context *ioc = p->io_context;
2633
2634 if (ioc && ioc->cic) {
2635 ioc->cic->exit(ioc->cic);
2636 cfq_free_io_context(ioc->cic);
2637 ioc->cic = NULL;
2638 }
2639 } while_each_thread(g, p);
2640
2641 read_unlock_irqrestore(&tasklist_lock, flags);
2642
1850 cfq_slab_kill(); 2643 cfq_slab_kill();
1851 elv_unregister(&iosched_cfq); 2644 elv_unregister(&iosched_cfq);
1852} 2645}
diff --git a/drivers/block/deadline-iosched.c b/drivers/block/deadline-iosched.c
index d63d34c671f7..ff5201e02153 100644
--- a/drivers/block/deadline-iosched.c
+++ b/drivers/block/deadline-iosched.c
@@ -711,18 +711,20 @@ static int deadline_init_queue(request_queue_t *q, elevator_t *e)
711 if (!drq_pool) 711 if (!drq_pool)
712 return -ENOMEM; 712 return -ENOMEM;
713 713
714 dd = kmalloc(sizeof(*dd), GFP_KERNEL); 714 dd = kmalloc_node(sizeof(*dd), GFP_KERNEL, q->node);
715 if (!dd) 715 if (!dd)
716 return -ENOMEM; 716 return -ENOMEM;
717 memset(dd, 0, sizeof(*dd)); 717 memset(dd, 0, sizeof(*dd));
718 718
719 dd->hash = kmalloc(sizeof(struct list_head)*DL_HASH_ENTRIES,GFP_KERNEL); 719 dd->hash = kmalloc_node(sizeof(struct list_head)*DL_HASH_ENTRIES,
720 GFP_KERNEL, q->node);
720 if (!dd->hash) { 721 if (!dd->hash) {
721 kfree(dd); 722 kfree(dd);
722 return -ENOMEM; 723 return -ENOMEM;
723 } 724 }
724 725
725 dd->drq_pool = mempool_create(BLKDEV_MIN_RQ, mempool_alloc_slab, mempool_free_slab, drq_pool); 726 dd->drq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
727 mempool_free_slab, drq_pool, q->node);
726 if (!dd->drq_pool) { 728 if (!dd->drq_pool) {
727 kfree(dd->hash); 729 kfree(dd->hash);
728 kfree(dd); 730 kfree(dd);
@@ -758,7 +760,8 @@ static void deadline_put_request(request_queue_t *q, struct request *rq)
758} 760}
759 761
760static int 762static int
761deadline_set_request(request_queue_t *q, struct request *rq, int gfp_mask) 763deadline_set_request(request_queue_t *q, struct request *rq, struct bio *bio,
764 int gfp_mask)
762{ 765{
763 struct deadline_data *dd = q->elevator->elevator_data; 766 struct deadline_data *dd = q->elevator->elevator_data;
764 struct deadline_rq *drq; 767 struct deadline_rq *drq;
@@ -886,7 +889,7 @@ deadline_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
886 struct deadline_fs_entry *entry = to_deadline(attr); 889 struct deadline_fs_entry *entry = to_deadline(attr);
887 890
888 if (!entry->show) 891 if (!entry->show)
889 return 0; 892 return -EIO;
890 893
891 return entry->show(e->elevator_data, page); 894 return entry->show(e->elevator_data, page);
892} 895}
@@ -899,7 +902,7 @@ deadline_attr_store(struct kobject *kobj, struct attribute *attr,
899 struct deadline_fs_entry *entry = to_deadline(attr); 902 struct deadline_fs_entry *entry = to_deadline(attr);
900 903
901 if (!entry->store) 904 if (!entry->store)
902 return -EINVAL; 905 return -EIO;
903 906
904 return entry->store(e->elevator_data, page, length); 907 return entry->store(e->elevator_data, page, length);
905} 908}
diff --git a/drivers/block/elevator.c b/drivers/block/elevator.c
index f831f08f839c..98f0126a2deb 100644
--- a/drivers/block/elevator.c
+++ b/drivers/block/elevator.c
@@ -486,12 +486,13 @@ struct request *elv_former_request(request_queue_t *q, struct request *rq)
486 return NULL; 486 return NULL;
487} 487}
488 488
489int elv_set_request(request_queue_t *q, struct request *rq, int gfp_mask) 489int elv_set_request(request_queue_t *q, struct request *rq, struct bio *bio,
490 int gfp_mask)
490{ 491{
491 elevator_t *e = q->elevator; 492 elevator_t *e = q->elevator;
492 493
493 if (e->ops->elevator_set_req_fn) 494 if (e->ops->elevator_set_req_fn)
494 return e->ops->elevator_set_req_fn(q, rq, gfp_mask); 495 return e->ops->elevator_set_req_fn(q, rq, bio, gfp_mask);
495 496
496 rq->elevator_private = NULL; 497 rq->elevator_private = NULL;
497 return 0; 498 return 0;
@@ -505,12 +506,12 @@ void elv_put_request(request_queue_t *q, struct request *rq)
505 e->ops->elevator_put_req_fn(q, rq); 506 e->ops->elevator_put_req_fn(q, rq);
506} 507}
507 508
508int elv_may_queue(request_queue_t *q, int rw) 509int elv_may_queue(request_queue_t *q, int rw, struct bio *bio)
509{ 510{
510 elevator_t *e = q->elevator; 511 elevator_t *e = q->elevator;
511 512
512 if (e->ops->elevator_may_queue_fn) 513 if (e->ops->elevator_may_queue_fn)
513 return e->ops->elevator_may_queue_fn(q, rw); 514 return e->ops->elevator_may_queue_fn(q, rw, bio);
514 515
515 return ELV_MQUEUE_MAY; 516 return ELV_MQUEUE_MAY;
516} 517}
diff --git a/drivers/block/genhd.c b/drivers/block/genhd.c
index 8bbe01d4b487..47fd3659a061 100644
--- a/drivers/block/genhd.c
+++ b/drivers/block/genhd.c
@@ -40,7 +40,7 @@ static inline int major_to_index(int major)
40 40
41#ifdef CONFIG_PROC_FS 41#ifdef CONFIG_PROC_FS
42/* get block device names in somewhat random order */ 42/* get block device names in somewhat random order */
43int get_blkdev_list(char *p) 43int get_blkdev_list(char *p, int used)
44{ 44{
45 struct blk_major_name *n; 45 struct blk_major_name *n;
46 int i, len; 46 int i, len;
@@ -49,10 +49,18 @@ int get_blkdev_list(char *p)
49 49
50 down(&block_subsys_sem); 50 down(&block_subsys_sem);
51 for (i = 0; i < ARRAY_SIZE(major_names); i++) { 51 for (i = 0; i < ARRAY_SIZE(major_names); i++) {
52 for (n = major_names[i]; n; n = n->next) 52 for (n = major_names[i]; n; n = n->next) {
53 /*
54 * If the curent string plus the 5 extra characters
55 * in the line would run us off the page, then we're done
56 */
57 if ((len + used + strlen(n->name) + 5) >= PAGE_SIZE)
58 goto page_full;
53 len += sprintf(p+len, "%3d %s\n", 59 len += sprintf(p+len, "%3d %s\n",
54 n->major, n->name); 60 n->major, n->name);
61 }
55 } 62 }
63page_full:
56 up(&block_subsys_sem); 64 up(&block_subsys_sem);
57 65
58 return len; 66 return len;
@@ -322,7 +330,7 @@ static ssize_t disk_attr_show(struct kobject *kobj, struct attribute *attr,
322 struct gendisk *disk = to_disk(kobj); 330 struct gendisk *disk = to_disk(kobj);
323 struct disk_attribute *disk_attr = 331 struct disk_attribute *disk_attr =
324 container_of(attr,struct disk_attribute,attr); 332 container_of(attr,struct disk_attribute,attr);
325 ssize_t ret = 0; 333 ssize_t ret = -EIO;
326 334
327 if (disk_attr->show) 335 if (disk_attr->show)
328 ret = disk_attr->show(disk,page); 336 ret = disk_attr->show(disk,page);
@@ -582,10 +590,16 @@ struct seq_operations diskstats_op = {
582 .show = diskstats_show 590 .show = diskstats_show
583}; 591};
584 592
585
586struct gendisk *alloc_disk(int minors) 593struct gendisk *alloc_disk(int minors)
587{ 594{
588 struct gendisk *disk = kmalloc(sizeof(struct gendisk), GFP_KERNEL); 595 return alloc_disk_node(minors, -1);
596}
597
598struct gendisk *alloc_disk_node(int minors, int node_id)
599{
600 struct gendisk *disk;
601
602 disk = kmalloc_node(sizeof(struct gendisk), GFP_KERNEL, node_id);
589 if (disk) { 603 if (disk) {
590 memset(disk, 0, sizeof(struct gendisk)); 604 memset(disk, 0, sizeof(struct gendisk));
591 if (!init_disk_stats(disk)) { 605 if (!init_disk_stats(disk)) {
@@ -594,7 +608,7 @@ struct gendisk *alloc_disk(int minors)
594 } 608 }
595 if (minors > 1) { 609 if (minors > 1) {
596 int size = (minors - 1) * sizeof(struct hd_struct *); 610 int size = (minors - 1) * sizeof(struct hd_struct *);
597 disk->part = kmalloc(size, GFP_KERNEL); 611 disk->part = kmalloc_node(size, GFP_KERNEL, node_id);
598 if (!disk->part) { 612 if (!disk->part) {
599 kfree(disk); 613 kfree(disk);
600 return NULL; 614 return NULL;
@@ -610,6 +624,7 @@ struct gendisk *alloc_disk(int minors)
610} 624}
611 625
612EXPORT_SYMBOL(alloc_disk); 626EXPORT_SYMBOL(alloc_disk);
627EXPORT_SYMBOL(alloc_disk_node);
613 628
614struct kobject *get_disk(struct gendisk *disk) 629struct kobject *get_disk(struct gendisk *disk)
615{ 630{
diff --git a/drivers/block/ioctl.c b/drivers/block/ioctl.c
index 6d7bcc9da9e7..6e278474f9a8 100644
--- a/drivers/block/ioctl.c
+++ b/drivers/block/ioctl.c
@@ -133,11 +133,9 @@ static int put_u64(unsigned long arg, u64 val)
133 return put_user(val, (u64 __user *)arg); 133 return put_user(val, (u64 __user *)arg);
134} 134}
135 135
136int blkdev_ioctl(struct inode *inode, struct file *file, unsigned cmd, 136static int blkdev_locked_ioctl(struct file *file, struct block_device *bdev,
137 unsigned long arg) 137 unsigned cmd, unsigned long arg)
138{ 138{
139 struct block_device *bdev = inode->i_bdev;
140 struct gendisk *disk = bdev->bd_disk;
141 struct backing_dev_info *bdi; 139 struct backing_dev_info *bdi;
142 int ret, n; 140 int ret, n;
143 141
@@ -190,36 +188,72 @@ int blkdev_ioctl(struct inode *inode, struct file *file, unsigned cmd,
190 return put_ulong(arg, bdev->bd_inode->i_size >> 9); 188 return put_ulong(arg, bdev->bd_inode->i_size >> 9);
191 case BLKGETSIZE64: 189 case BLKGETSIZE64:
192 return put_u64(arg, bdev->bd_inode->i_size); 190 return put_u64(arg, bdev->bd_inode->i_size);
191 }
192 return -ENOIOCTLCMD;
193}
194
195static int blkdev_driver_ioctl(struct inode *inode, struct file *file,
196 struct gendisk *disk, unsigned cmd, unsigned long arg)
197{
198 int ret;
199 if (disk->fops->unlocked_ioctl)
200 return disk->fops->unlocked_ioctl(file, cmd, arg);
201
202 if (disk->fops->ioctl) {
203 lock_kernel();
204 ret = disk->fops->ioctl(inode, file, cmd, arg);
205 unlock_kernel();
206 return ret;
207 }
208
209 return -ENOTTY;
210}
211
212int blkdev_ioctl(struct inode *inode, struct file *file, unsigned cmd,
213 unsigned long arg)
214{
215 struct block_device *bdev = inode->i_bdev;
216 struct gendisk *disk = bdev->bd_disk;
217 int ret, n;
218
219 switch(cmd) {
193 case BLKFLSBUF: 220 case BLKFLSBUF:
194 if (!capable(CAP_SYS_ADMIN)) 221 if (!capable(CAP_SYS_ADMIN))
195 return -EACCES; 222 return -EACCES;
196 if (disk->fops->ioctl) { 223
197 ret = disk->fops->ioctl(inode, file, cmd, arg); 224 ret = blkdev_driver_ioctl(inode, file, disk, cmd, arg);
198 /* -EINVAL to handle old uncorrected drivers */ 225 /* -EINVAL to handle old uncorrected drivers */
199 if (ret != -EINVAL && ret != -ENOTTY) 226 if (ret != -EINVAL && ret != -ENOTTY)
200 return ret; 227 return ret;
201 } 228
229 lock_kernel();
202 fsync_bdev(bdev); 230 fsync_bdev(bdev);
203 invalidate_bdev(bdev, 0); 231 invalidate_bdev(bdev, 0);
232 unlock_kernel();
204 return 0; 233 return 0;
234
205 case BLKROSET: 235 case BLKROSET:
206 if (disk->fops->ioctl) { 236 ret = blkdev_driver_ioctl(inode, file, disk, cmd, arg);
207 ret = disk->fops->ioctl(inode, file, cmd, arg); 237 /* -EINVAL to handle old uncorrected drivers */
208 /* -EINVAL to handle old uncorrected drivers */ 238 if (ret != -EINVAL && ret != -ENOTTY)
209 if (ret != -EINVAL && ret != -ENOTTY) 239 return ret;
210 return ret;
211 }
212 if (!capable(CAP_SYS_ADMIN)) 240 if (!capable(CAP_SYS_ADMIN))
213 return -EACCES; 241 return -EACCES;
214 if (get_user(n, (int __user *)(arg))) 242 if (get_user(n, (int __user *)(arg)))
215 return -EFAULT; 243 return -EFAULT;
244 lock_kernel();
216 set_device_ro(bdev, n); 245 set_device_ro(bdev, n);
246 unlock_kernel();
217 return 0; 247 return 0;
218 default:
219 if (disk->fops->ioctl)
220 return disk->fops->ioctl(inode, file, cmd, arg);
221 } 248 }
222 return -ENOTTY; 249
250 lock_kernel();
251 ret = blkdev_locked_ioctl(file, bdev, cmd, arg);
252 unlock_kernel();
253 if (ret != -ENOIOCTLCMD)
254 return ret;
255
256 return blkdev_driver_ioctl(inode, file, disk, cmd, arg);
223} 257}
224 258
225/* Most of the generic ioctls are handled in the normal fallback path. 259/* Most of the generic ioctls are handled in the normal fallback path.
diff --git a/drivers/block/ll_rw_blk.c b/drivers/block/ll_rw_blk.c
index f6fda036b4ae..0c7599563b65 100644
--- a/drivers/block/ll_rw_blk.c
+++ b/drivers/block/ll_rw_blk.c
@@ -28,6 +28,7 @@
28#include <linux/slab.h> 28#include <linux/slab.h>
29#include <linux/swap.h> 29#include <linux/swap.h>
30#include <linux/writeback.h> 30#include <linux/writeback.h>
31#include <linux/blkdev.h>
31 32
32/* 33/*
33 * for max sense size 34 * for max sense size
@@ -36,6 +37,7 @@
36 37
37static void blk_unplug_work(void *data); 38static void blk_unplug_work(void *data);
38static void blk_unplug_timeout(unsigned long data); 39static void blk_unplug_timeout(unsigned long data);
40static void drive_stat_acct(struct request *rq, int nr_sectors, int new_io);
39 41
40/* 42/*
41 * For the allocated request tables 43 * For the allocated request tables
@@ -274,6 +276,7 @@ static inline void rq_init(request_queue_t *q, struct request *rq)
274 rq->errors = 0; 276 rq->errors = 0;
275 rq->rq_status = RQ_ACTIVE; 277 rq->rq_status = RQ_ACTIVE;
276 rq->bio = rq->biotail = NULL; 278 rq->bio = rq->biotail = NULL;
279 rq->ioprio = 0;
277 rq->buffer = NULL; 280 rq->buffer = NULL;
278 rq->ref_count = 1; 281 rq->ref_count = 1;
279 rq->q = q; 282 rq->q = q;
@@ -775,9 +778,9 @@ EXPORT_SYMBOL(blk_queue_free_tags);
775static int 778static int
776init_tag_map(request_queue_t *q, struct blk_queue_tag *tags, int depth) 779init_tag_map(request_queue_t *q, struct blk_queue_tag *tags, int depth)
777{ 780{
778 int bits, i;
779 struct request **tag_index; 781 struct request **tag_index;
780 unsigned long *tag_map; 782 unsigned long *tag_map;
783 int nr_ulongs;
781 784
782 if (depth > q->nr_requests * 2) { 785 if (depth > q->nr_requests * 2) {
783 depth = q->nr_requests * 2; 786 depth = q->nr_requests * 2;
@@ -789,24 +792,18 @@ init_tag_map(request_queue_t *q, struct blk_queue_tag *tags, int depth)
789 if (!tag_index) 792 if (!tag_index)
790 goto fail; 793 goto fail;
791 794
792 bits = (depth / BLK_TAGS_PER_LONG) + 1; 795 nr_ulongs = ALIGN(depth, BITS_PER_LONG) / BITS_PER_LONG;
793 tag_map = kmalloc(bits * sizeof(unsigned long), GFP_ATOMIC); 796 tag_map = kmalloc(nr_ulongs * sizeof(unsigned long), GFP_ATOMIC);
794 if (!tag_map) 797 if (!tag_map)
795 goto fail; 798 goto fail;
796 799
797 memset(tag_index, 0, depth * sizeof(struct request *)); 800 memset(tag_index, 0, depth * sizeof(struct request *));
798 memset(tag_map, 0, bits * sizeof(unsigned long)); 801 memset(tag_map, 0, nr_ulongs * sizeof(unsigned long));
802 tags->real_max_depth = depth;
799 tags->max_depth = depth; 803 tags->max_depth = depth;
800 tags->real_max_depth = bits * BITS_PER_LONG;
801 tags->tag_index = tag_index; 804 tags->tag_index = tag_index;
802 tags->tag_map = tag_map; 805 tags->tag_map = tag_map;
803 806
804 /*
805 * set the upper bits if the depth isn't a multiple of the word size
806 */
807 for (i = depth; i < bits * BLK_TAGS_PER_LONG; i++)
808 __set_bit(i, tag_map);
809
810 return 0; 807 return 0;
811fail: 808fail:
812 kfree(tag_index); 809 kfree(tag_index);
@@ -871,13 +868,16 @@ int blk_queue_resize_tags(request_queue_t *q, int new_depth)
871 struct blk_queue_tag *bqt = q->queue_tags; 868 struct blk_queue_tag *bqt = q->queue_tags;
872 struct request **tag_index; 869 struct request **tag_index;
873 unsigned long *tag_map; 870 unsigned long *tag_map;
874 int bits, max_depth; 871 int max_depth, nr_ulongs;
875 872
876 if (!bqt) 873 if (!bqt)
877 return -ENXIO; 874 return -ENXIO;
878 875
879 /* 876 /*
880 * don't bother sizing down 877 * if we already have large enough real_max_depth. just
878 * adjust max_depth. *NOTE* as requests with tag value
879 * between new_depth and real_max_depth can be in-flight, tag
880 * map can not be shrunk blindly here.
881 */ 881 */
882 if (new_depth <= bqt->real_max_depth) { 882 if (new_depth <= bqt->real_max_depth) {
883 bqt->max_depth = new_depth; 883 bqt->max_depth = new_depth;
@@ -895,8 +895,8 @@ int blk_queue_resize_tags(request_queue_t *q, int new_depth)
895 return -ENOMEM; 895 return -ENOMEM;
896 896
897 memcpy(bqt->tag_index, tag_index, max_depth * sizeof(struct request *)); 897 memcpy(bqt->tag_index, tag_index, max_depth * sizeof(struct request *));
898 bits = max_depth / BLK_TAGS_PER_LONG; 898 nr_ulongs = ALIGN(max_depth, BITS_PER_LONG) / BITS_PER_LONG;
899 memcpy(bqt->tag_map, tag_map, bits * sizeof(unsigned long)); 899 memcpy(bqt->tag_map, tag_map, nr_ulongs * sizeof(unsigned long));
900 900
901 kfree(tag_index); 901 kfree(tag_index);
902 kfree(tag_map); 902 kfree(tag_map);
@@ -927,10 +927,15 @@ void blk_queue_end_tag(request_queue_t *q, struct request *rq)
927 BUG_ON(tag == -1); 927 BUG_ON(tag == -1);
928 928
929 if (unlikely(tag >= bqt->real_max_depth)) 929 if (unlikely(tag >= bqt->real_max_depth))
930 /*
931 * This can happen after tag depth has been reduced.
932 * FIXME: how about a warning or info message here?
933 */
930 return; 934 return;
931 935
932 if (unlikely(!__test_and_clear_bit(tag, bqt->tag_map))) { 936 if (unlikely(!__test_and_clear_bit(tag, bqt->tag_map))) {
933 printk("attempt to clear non-busy tag (%d)\n", tag); 937 printk(KERN_ERR "%s: attempt to clear non-busy tag (%d)\n",
938 __FUNCTION__, tag);
934 return; 939 return;
935 } 940 }
936 941
@@ -939,7 +944,8 @@ void blk_queue_end_tag(request_queue_t *q, struct request *rq)
939 rq->tag = -1; 944 rq->tag = -1;
940 945
941 if (unlikely(bqt->tag_index[tag] == NULL)) 946 if (unlikely(bqt->tag_index[tag] == NULL))
942 printk("tag %d is missing\n", tag); 947 printk(KERN_ERR "%s: tag %d is missing\n",
948 __FUNCTION__, tag);
943 949
944 bqt->tag_index[tag] = NULL; 950 bqt->tag_index[tag] = NULL;
945 bqt->busy--; 951 bqt->busy--;
@@ -968,24 +974,20 @@ EXPORT_SYMBOL(blk_queue_end_tag);
968int blk_queue_start_tag(request_queue_t *q, struct request *rq) 974int blk_queue_start_tag(request_queue_t *q, struct request *rq)
969{ 975{
970 struct blk_queue_tag *bqt = q->queue_tags; 976 struct blk_queue_tag *bqt = q->queue_tags;
971 unsigned long *map = bqt->tag_map; 977 int tag;
972 int tag = 0;
973 978
974 if (unlikely((rq->flags & REQ_QUEUED))) { 979 if (unlikely((rq->flags & REQ_QUEUED))) {
975 printk(KERN_ERR 980 printk(KERN_ERR
976 "request %p for device [%s] already tagged %d", 981 "%s: request %p for device [%s] already tagged %d",
977 rq, rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->tag); 982 __FUNCTION__, rq,
983 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->tag);
978 BUG(); 984 BUG();
979 } 985 }
980 986
981 for (map = bqt->tag_map; *map == -1UL; map++) { 987 tag = find_first_zero_bit(bqt->tag_map, bqt->max_depth);
982 tag += BLK_TAGS_PER_LONG; 988 if (tag >= bqt->max_depth)
983 989 return 1;
984 if (tag >= bqt->max_depth)
985 return 1;
986 }
987 990
988 tag += ffz(*map);
989 __set_bit(tag, bqt->tag_map); 991 __set_bit(tag, bqt->tag_map);
990 992
991 rq->flags |= REQ_QUEUED; 993 rq->flags |= REQ_QUEUED;
@@ -1021,7 +1023,8 @@ void blk_queue_invalidate_tags(request_queue_t *q)
1021 rq = list_entry_rq(tmp); 1023 rq = list_entry_rq(tmp);
1022 1024
1023 if (rq->tag == -1) { 1025 if (rq->tag == -1) {
1024 printk("bad tag found on list\n"); 1026 printk(KERN_ERR
1027 "%s: bad tag found on list\n", __FUNCTION__);
1025 list_del_init(&rq->queuelist); 1028 list_del_init(&rq->queuelist);
1026 rq->flags &= ~REQ_QUEUED; 1029 rq->flags &= ~REQ_QUEUED;
1027 } else 1030 } else
@@ -1149,7 +1152,7 @@ new_hw_segment:
1149} 1152}
1150 1153
1151 1154
1152int blk_phys_contig_segment(request_queue_t *q, struct bio *bio, 1155static int blk_phys_contig_segment(request_queue_t *q, struct bio *bio,
1153 struct bio *nxt) 1156 struct bio *nxt)
1154{ 1157{
1155 if (!(q->queue_flags & (1 << QUEUE_FLAG_CLUSTER))) 1158 if (!(q->queue_flags & (1 << QUEUE_FLAG_CLUSTER)))
@@ -1170,9 +1173,7 @@ int blk_phys_contig_segment(request_queue_t *q, struct bio *bio,
1170 return 0; 1173 return 0;
1171} 1174}
1172 1175
1173EXPORT_SYMBOL(blk_phys_contig_segment); 1176static int blk_hw_contig_segment(request_queue_t *q, struct bio *bio,
1174
1175int blk_hw_contig_segment(request_queue_t *q, struct bio *bio,
1176 struct bio *nxt) 1177 struct bio *nxt)
1177{ 1178{
1178 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) 1179 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
@@ -1188,8 +1189,6 @@ int blk_hw_contig_segment(request_queue_t *q, struct bio *bio,
1188 return 1; 1189 return 1;
1189} 1190}
1190 1191
1191EXPORT_SYMBOL(blk_hw_contig_segment);
1192
1193/* 1192/*
1194 * map a request to scatterlist, return number of sg entries setup. Caller 1193 * map a request to scatterlist, return number of sg entries setup. Caller
1195 * must make sure sg can hold rq->nr_phys_segments entries 1194 * must make sure sg can hold rq->nr_phys_segments entries
@@ -1359,8 +1358,8 @@ static int ll_front_merge_fn(request_queue_t *q, struct request *req,
1359static int ll_merge_requests_fn(request_queue_t *q, struct request *req, 1358static int ll_merge_requests_fn(request_queue_t *q, struct request *req,
1360 struct request *next) 1359 struct request *next)
1361{ 1360{
1362 int total_phys_segments = req->nr_phys_segments +next->nr_phys_segments; 1361 int total_phys_segments;
1363 int total_hw_segments = req->nr_hw_segments + next->nr_hw_segments; 1362 int total_hw_segments;
1364 1363
1365 /* 1364 /*
1366 * First check if the either of the requests are re-queued 1365 * First check if the either of the requests are re-queued
@@ -1370,7 +1369,7 @@ static int ll_merge_requests_fn(request_queue_t *q, struct request *req,
1370 return 0; 1369 return 0;
1371 1370
1372 /* 1371 /*
1373 * Will it become to large? 1372 * Will it become too large?
1374 */ 1373 */
1375 if ((req->nr_sectors + next->nr_sectors) > q->max_sectors) 1374 if ((req->nr_sectors + next->nr_sectors) > q->max_sectors)
1376 return 0; 1375 return 0;
@@ -1451,17 +1450,13 @@ EXPORT_SYMBOL(blk_remove_plug);
1451 */ 1450 */
1452void __generic_unplug_device(request_queue_t *q) 1451void __generic_unplug_device(request_queue_t *q)
1453{ 1452{
1454 if (test_bit(QUEUE_FLAG_STOPPED, &q->queue_flags)) 1453 if (unlikely(test_bit(QUEUE_FLAG_STOPPED, &q->queue_flags)))
1455 return; 1454 return;
1456 1455
1457 if (!blk_remove_plug(q)) 1456 if (!blk_remove_plug(q))
1458 return; 1457 return;
1459 1458
1460 /* 1459 q->request_fn(q);
1461 * was plugged, fire request_fn if queue has stuff to do
1462 */
1463 if (elv_next_request(q))
1464 q->request_fn(q);
1465} 1460}
1466EXPORT_SYMBOL(__generic_unplug_device); 1461EXPORT_SYMBOL(__generic_unplug_device);
1467 1462
@@ -1646,7 +1641,8 @@ static int blk_init_free_list(request_queue_t *q)
1646 init_waitqueue_head(&rl->wait[WRITE]); 1641 init_waitqueue_head(&rl->wait[WRITE]);
1647 init_waitqueue_head(&rl->drain); 1642 init_waitqueue_head(&rl->drain);
1648 1643
1649 rl->rq_pool = mempool_create(BLKDEV_MIN_RQ, mempool_alloc_slab, mempool_free_slab, request_cachep); 1644 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
1645 mempool_free_slab, request_cachep, q->node);
1650 1646
1651 if (!rl->rq_pool) 1647 if (!rl->rq_pool)
1652 return -ENOMEM; 1648 return -ENOMEM;
@@ -1658,8 +1654,15 @@ static int __make_request(request_queue_t *, struct bio *);
1658 1654
1659request_queue_t *blk_alloc_queue(int gfp_mask) 1655request_queue_t *blk_alloc_queue(int gfp_mask)
1660{ 1656{
1661 request_queue_t *q = kmem_cache_alloc(requestq_cachep, gfp_mask); 1657 return blk_alloc_queue_node(gfp_mask, -1);
1658}
1659EXPORT_SYMBOL(blk_alloc_queue);
1660
1661request_queue_t *blk_alloc_queue_node(int gfp_mask, int node_id)
1662{
1663 request_queue_t *q;
1662 1664
1665 q = kmem_cache_alloc_node(requestq_cachep, gfp_mask, node_id);
1663 if (!q) 1666 if (!q)
1664 return NULL; 1667 return NULL;
1665 1668
@@ -1672,8 +1675,7 @@ request_queue_t *blk_alloc_queue(int gfp_mask)
1672 1675
1673 return q; 1676 return q;
1674} 1677}
1675 1678EXPORT_SYMBOL(blk_alloc_queue_node);
1676EXPORT_SYMBOL(blk_alloc_queue);
1677 1679
1678/** 1680/**
1679 * blk_init_queue - prepare a request queue for use with a block device 1681 * blk_init_queue - prepare a request queue for use with a block device
@@ -1706,13 +1708,22 @@ EXPORT_SYMBOL(blk_alloc_queue);
1706 * blk_init_queue() must be paired with a blk_cleanup_queue() call 1708 * blk_init_queue() must be paired with a blk_cleanup_queue() call
1707 * when the block device is deactivated (such as at module unload). 1709 * when the block device is deactivated (such as at module unload).
1708 **/ 1710 **/
1711
1709request_queue_t *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock) 1712request_queue_t *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
1710{ 1713{
1711 request_queue_t *q = blk_alloc_queue(GFP_KERNEL); 1714 return blk_init_queue_node(rfn, lock, -1);
1715}
1716EXPORT_SYMBOL(blk_init_queue);
1717
1718request_queue_t *
1719blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
1720{
1721 request_queue_t *q = blk_alloc_queue_node(GFP_KERNEL, node_id);
1712 1722
1713 if (!q) 1723 if (!q)
1714 return NULL; 1724 return NULL;
1715 1725
1726 q->node = node_id;
1716 if (blk_init_free_list(q)) 1727 if (blk_init_free_list(q))
1717 goto out_init; 1728 goto out_init;
1718 1729
@@ -1755,12 +1766,11 @@ out_init:
1755 kmem_cache_free(requestq_cachep, q); 1766 kmem_cache_free(requestq_cachep, q);
1756 return NULL; 1767 return NULL;
1757} 1768}
1758 1769EXPORT_SYMBOL(blk_init_queue_node);
1759EXPORT_SYMBOL(blk_init_queue);
1760 1770
1761int blk_get_queue(request_queue_t *q) 1771int blk_get_queue(request_queue_t *q)
1762{ 1772{
1763 if (!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)) { 1773 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
1764 atomic_inc(&q->refcnt); 1774 atomic_inc(&q->refcnt);
1765 return 0; 1775 return 0;
1766 } 1776 }
@@ -1776,8 +1786,8 @@ static inline void blk_free_request(request_queue_t *q, struct request *rq)
1776 mempool_free(rq, q->rq.rq_pool); 1786 mempool_free(rq, q->rq.rq_pool);
1777} 1787}
1778 1788
1779static inline struct request *blk_alloc_request(request_queue_t *q, int rw, 1789static inline struct request *
1780 int gfp_mask) 1790blk_alloc_request(request_queue_t *q, int rw, struct bio *bio, int gfp_mask)
1781{ 1791{
1782 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask); 1792 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
1783 1793
@@ -1790,7 +1800,7 @@ static inline struct request *blk_alloc_request(request_queue_t *q, int rw,
1790 */ 1800 */
1791 rq->flags = rw; 1801 rq->flags = rw;
1792 1802
1793 if (!elv_set_request(q, rq, gfp_mask)) 1803 if (!elv_set_request(q, rq, bio, gfp_mask))
1794 return rq; 1804 return rq;
1795 1805
1796 mempool_free(rq, q->rq.rq_pool); 1806 mempool_free(rq, q->rq.rq_pool);
@@ -1822,7 +1832,7 @@ static inline int ioc_batching(request_queue_t *q, struct io_context *ioc)
1822 * is the behaviour we want though - once it gets a wakeup it should be given 1832 * is the behaviour we want though - once it gets a wakeup it should be given
1823 * a nice run. 1833 * a nice run.
1824 */ 1834 */
1825void ioc_set_batching(request_queue_t *q, struct io_context *ioc) 1835static void ioc_set_batching(request_queue_t *q, struct io_context *ioc)
1826{ 1836{
1827 if (!ioc || ioc_batching(q, ioc)) 1837 if (!ioc || ioc_batching(q, ioc))
1828 return; 1838 return;
@@ -1839,7 +1849,6 @@ static void __freed_request(request_queue_t *q, int rw)
1839 clear_queue_congested(q, rw); 1849 clear_queue_congested(q, rw);
1840 1850
1841 if (rl->count[rw] + 1 <= q->nr_requests) { 1851 if (rl->count[rw] + 1 <= q->nr_requests) {
1842 smp_mb();
1843 if (waitqueue_active(&rl->wait[rw])) 1852 if (waitqueue_active(&rl->wait[rw]))
1844 wake_up(&rl->wait[rw]); 1853 wake_up(&rl->wait[rw]);
1845 1854
@@ -1871,18 +1880,20 @@ static void freed_request(request_queue_t *q, int rw)
1871 1880
1872#define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist) 1881#define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
1873/* 1882/*
1874 * Get a free request, queue_lock must not be held 1883 * Get a free request, queue_lock must be held.
1884 * Returns NULL on failure, with queue_lock held.
1885 * Returns !NULL on success, with queue_lock *not held*.
1875 */ 1886 */
1876static struct request *get_request(request_queue_t *q, int rw, int gfp_mask) 1887static struct request *get_request(request_queue_t *q, int rw, struct bio *bio,
1888 int gfp_mask)
1877{ 1889{
1878 struct request *rq = NULL; 1890 struct request *rq = NULL;
1879 struct request_list *rl = &q->rq; 1891 struct request_list *rl = &q->rq;
1880 struct io_context *ioc = get_io_context(gfp_mask); 1892 struct io_context *ioc = current_io_context(GFP_ATOMIC);
1881 1893
1882 if (unlikely(test_bit(QUEUE_FLAG_DRAIN, &q->queue_flags))) 1894 if (unlikely(test_bit(QUEUE_FLAG_DRAIN, &q->queue_flags)))
1883 goto out; 1895 goto out;
1884 1896
1885 spin_lock_irq(q->queue_lock);
1886 if (rl->count[rw]+1 >= q->nr_requests) { 1897 if (rl->count[rw]+1 >= q->nr_requests) {
1887 /* 1898 /*
1888 * The queue will fill after this allocation, so set it as 1899 * The queue will fill after this allocation, so set it as
@@ -1896,7 +1907,7 @@ static struct request *get_request(request_queue_t *q, int rw, int gfp_mask)
1896 } 1907 }
1897 } 1908 }
1898 1909
1899 switch (elv_may_queue(q, rw)) { 1910 switch (elv_may_queue(q, rw, bio)) {
1900 case ELV_MQUEUE_NO: 1911 case ELV_MQUEUE_NO:
1901 goto rq_starved; 1912 goto rq_starved;
1902 case ELV_MQUEUE_MAY: 1913 case ELV_MQUEUE_MAY:
@@ -1910,18 +1921,25 @@ static struct request *get_request(request_queue_t *q, int rw, int gfp_mask)
1910 * The queue is full and the allocating process is not a 1921 * The queue is full and the allocating process is not a
1911 * "batcher", and not exempted by the IO scheduler 1922 * "batcher", and not exempted by the IO scheduler
1912 */ 1923 */
1913 spin_unlock_irq(q->queue_lock);
1914 goto out; 1924 goto out;
1915 } 1925 }
1916 1926
1917get_rq: 1927get_rq:
1928 /*
1929 * Only allow batching queuers to allocate up to 50% over the defined
1930 * limit of requests, otherwise we could have thousands of requests
1931 * allocated with any setting of ->nr_requests
1932 */
1933 if (rl->count[rw] >= (3 * q->nr_requests / 2))
1934 goto out;
1935
1918 rl->count[rw]++; 1936 rl->count[rw]++;
1919 rl->starved[rw] = 0; 1937 rl->starved[rw] = 0;
1920 if (rl->count[rw] >= queue_congestion_on_threshold(q)) 1938 if (rl->count[rw] >= queue_congestion_on_threshold(q))
1921 set_queue_congested(q, rw); 1939 set_queue_congested(q, rw);
1922 spin_unlock_irq(q->queue_lock); 1940 spin_unlock_irq(q->queue_lock);
1923 1941
1924 rq = blk_alloc_request(q, rw, gfp_mask); 1942 rq = blk_alloc_request(q, rw, bio, gfp_mask);
1925 if (!rq) { 1943 if (!rq) {
1926 /* 1944 /*
1927 * Allocation failed presumably due to memory. Undo anything 1945 * Allocation failed presumably due to memory. Undo anything
@@ -1944,7 +1962,6 @@ rq_starved:
1944 if (unlikely(rl->count[rw] == 0)) 1962 if (unlikely(rl->count[rw] == 0))
1945 rl->starved[rw] = 1; 1963 rl->starved[rw] = 1;
1946 1964
1947 spin_unlock_irq(q->queue_lock);
1948 goto out; 1965 goto out;
1949 } 1966 }
1950 1967
@@ -1954,31 +1971,35 @@ rq_starved:
1954 rq_init(q, rq); 1971 rq_init(q, rq);
1955 rq->rl = rl; 1972 rq->rl = rl;
1956out: 1973out:
1957 put_io_context(ioc);
1958 return rq; 1974 return rq;
1959} 1975}
1960 1976
1961/* 1977/*
1962 * No available requests for this queue, unplug the device and wait for some 1978 * No available requests for this queue, unplug the device and wait for some
1963 * requests to become available. 1979 * requests to become available.
1980 *
1981 * Called with q->queue_lock held, and returns with it unlocked.
1964 */ 1982 */
1965static struct request *get_request_wait(request_queue_t *q, int rw) 1983static struct request *get_request_wait(request_queue_t *q, int rw,
1984 struct bio *bio)
1966{ 1985{
1967 DEFINE_WAIT(wait);
1968 struct request *rq; 1986 struct request *rq;
1969 1987
1970 generic_unplug_device(q); 1988 rq = get_request(q, rw, bio, GFP_NOIO);
1971 do { 1989 while (!rq) {
1990 DEFINE_WAIT(wait);
1972 struct request_list *rl = &q->rq; 1991 struct request_list *rl = &q->rq;
1973 1992
1974 prepare_to_wait_exclusive(&rl->wait[rw], &wait, 1993 prepare_to_wait_exclusive(&rl->wait[rw], &wait,
1975 TASK_UNINTERRUPTIBLE); 1994 TASK_UNINTERRUPTIBLE);
1976 1995
1977 rq = get_request(q, rw, GFP_NOIO); 1996 rq = get_request(q, rw, bio, GFP_NOIO);
1978 1997
1979 if (!rq) { 1998 if (!rq) {
1980 struct io_context *ioc; 1999 struct io_context *ioc;
1981 2000
2001 __generic_unplug_device(q);
2002 spin_unlock_irq(q->queue_lock);
1982 io_schedule(); 2003 io_schedule();
1983 2004
1984 /* 2005 /*
@@ -1987,12 +2008,13 @@ static struct request *get_request_wait(request_queue_t *q, int rw)
1987 * up to a big batch of them for a small period time. 2008 * up to a big batch of them for a small period time.
1988 * See ioc_batching, ioc_set_batching 2009 * See ioc_batching, ioc_set_batching
1989 */ 2010 */
1990 ioc = get_io_context(GFP_NOIO); 2011 ioc = current_io_context(GFP_NOIO);
1991 ioc_set_batching(q, ioc); 2012 ioc_set_batching(q, ioc);
1992 put_io_context(ioc); 2013
2014 spin_lock_irq(q->queue_lock);
1993 } 2015 }
1994 finish_wait(&rl->wait[rw], &wait); 2016 finish_wait(&rl->wait[rw], &wait);
1995 } while (!rq); 2017 }
1996 2018
1997 return rq; 2019 return rq;
1998} 2020}
@@ -2003,14 +2025,18 @@ struct request *blk_get_request(request_queue_t *q, int rw, int gfp_mask)
2003 2025
2004 BUG_ON(rw != READ && rw != WRITE); 2026 BUG_ON(rw != READ && rw != WRITE);
2005 2027
2006 if (gfp_mask & __GFP_WAIT) 2028 spin_lock_irq(q->queue_lock);
2007 rq = get_request_wait(q, rw); 2029 if (gfp_mask & __GFP_WAIT) {
2008 else 2030 rq = get_request_wait(q, rw, NULL);
2009 rq = get_request(q, rw, gfp_mask); 2031 } else {
2032 rq = get_request(q, rw, NULL, gfp_mask);
2033 if (!rq)
2034 spin_unlock_irq(q->queue_lock);
2035 }
2036 /* q->queue_lock is unlocked at this point */
2010 2037
2011 return rq; 2038 return rq;
2012} 2039}
2013
2014EXPORT_SYMBOL(blk_get_request); 2040EXPORT_SYMBOL(blk_get_request);
2015 2041
2016/** 2042/**
@@ -2385,7 +2411,7 @@ int blkdev_scsi_issue_flush_fn(request_queue_t *q, struct gendisk *disk,
2385 2411
2386EXPORT_SYMBOL(blkdev_scsi_issue_flush_fn); 2412EXPORT_SYMBOL(blkdev_scsi_issue_flush_fn);
2387 2413
2388void drive_stat_acct(struct request *rq, int nr_sectors, int new_io) 2414static void drive_stat_acct(struct request *rq, int nr_sectors, int new_io)
2389{ 2415{
2390 int rw = rq_data_dir(rq); 2416 int rw = rq_data_dir(rq);
2391 2417
@@ -2467,7 +2493,6 @@ static void __blk_put_request(request_queue_t *q, struct request *req)
2467 return; 2493 return;
2468 2494
2469 req->rq_status = RQ_INACTIVE; 2495 req->rq_status = RQ_INACTIVE;
2470 req->q = NULL;
2471 req->rl = NULL; 2496 req->rl = NULL;
2472 2497
2473 /* 2498 /*
@@ -2596,6 +2621,8 @@ static int attempt_merge(request_queue_t *q, struct request *req,
2596 req->rq_disk->in_flight--; 2621 req->rq_disk->in_flight--;
2597 } 2622 }
2598 2623
2624 req->ioprio = ioprio_best(req->ioprio, next->ioprio);
2625
2599 __blk_put_request(q, next); 2626 __blk_put_request(q, next);
2600 return 1; 2627 return 1;
2601} 2628}
@@ -2644,25 +2671,17 @@ void blk_attempt_remerge(request_queue_t *q, struct request *rq)
2644 2671
2645EXPORT_SYMBOL(blk_attempt_remerge); 2672EXPORT_SYMBOL(blk_attempt_remerge);
2646 2673
2647/*
2648 * Non-locking blk_attempt_remerge variant.
2649 */
2650void __blk_attempt_remerge(request_queue_t *q, struct request *rq)
2651{
2652 attempt_back_merge(q, rq);
2653}
2654
2655EXPORT_SYMBOL(__blk_attempt_remerge);
2656
2657static int __make_request(request_queue_t *q, struct bio *bio) 2674static int __make_request(request_queue_t *q, struct bio *bio)
2658{ 2675{
2659 struct request *req, *freereq = NULL; 2676 struct request *req;
2660 int el_ret, rw, nr_sectors, cur_nr_sectors, barrier, err, sync; 2677 int el_ret, rw, nr_sectors, cur_nr_sectors, barrier, err, sync;
2678 unsigned short prio;
2661 sector_t sector; 2679 sector_t sector;
2662 2680
2663 sector = bio->bi_sector; 2681 sector = bio->bi_sector;
2664 nr_sectors = bio_sectors(bio); 2682 nr_sectors = bio_sectors(bio);
2665 cur_nr_sectors = bio_cur_sectors(bio); 2683 cur_nr_sectors = bio_cur_sectors(bio);
2684 prio = bio_prio(bio);
2666 2685
2667 rw = bio_data_dir(bio); 2686 rw = bio_data_dir(bio);
2668 sync = bio_sync(bio); 2687 sync = bio_sync(bio);
@@ -2677,19 +2696,14 @@ static int __make_request(request_queue_t *q, struct bio *bio)
2677 spin_lock_prefetch(q->queue_lock); 2696 spin_lock_prefetch(q->queue_lock);
2678 2697
2679 barrier = bio_barrier(bio); 2698 barrier = bio_barrier(bio);
2680 if (barrier && (q->ordered == QUEUE_ORDERED_NONE)) { 2699 if (unlikely(barrier) && (q->ordered == QUEUE_ORDERED_NONE)) {
2681 err = -EOPNOTSUPP; 2700 err = -EOPNOTSUPP;
2682 goto end_io; 2701 goto end_io;
2683 } 2702 }
2684 2703
2685again:
2686 spin_lock_irq(q->queue_lock); 2704 spin_lock_irq(q->queue_lock);
2687 2705
2688 if (elv_queue_empty(q)) { 2706 if (unlikely(barrier) || elv_queue_empty(q))
2689 blk_plug_device(q);
2690 goto get_rq;
2691 }
2692 if (barrier)
2693 goto get_rq; 2707 goto get_rq;
2694 2708
2695 el_ret = elv_merge(q, &req, bio); 2709 el_ret = elv_merge(q, &req, bio);
@@ -2703,6 +2717,7 @@ again:
2703 req->biotail->bi_next = bio; 2717 req->biotail->bi_next = bio;
2704 req->biotail = bio; 2718 req->biotail = bio;
2705 req->nr_sectors = req->hard_nr_sectors += nr_sectors; 2719 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
2720 req->ioprio = ioprio_best(req->ioprio, prio);
2706 drive_stat_acct(req, nr_sectors, 0); 2721 drive_stat_acct(req, nr_sectors, 0);
2707 if (!attempt_back_merge(q, req)) 2722 if (!attempt_back_merge(q, req))
2708 elv_merged_request(q, req); 2723 elv_merged_request(q, req);
@@ -2727,45 +2742,30 @@ again:
2727 req->hard_cur_sectors = cur_nr_sectors; 2742 req->hard_cur_sectors = cur_nr_sectors;
2728 req->sector = req->hard_sector = sector; 2743 req->sector = req->hard_sector = sector;
2729 req->nr_sectors = req->hard_nr_sectors += nr_sectors; 2744 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
2745 req->ioprio = ioprio_best(req->ioprio, prio);
2730 drive_stat_acct(req, nr_sectors, 0); 2746 drive_stat_acct(req, nr_sectors, 0);
2731 if (!attempt_front_merge(q, req)) 2747 if (!attempt_front_merge(q, req))
2732 elv_merged_request(q, req); 2748 elv_merged_request(q, req);
2733 goto out; 2749 goto out;
2734 2750
2735 /* 2751 /* ELV_NO_MERGE: elevator says don't/can't merge. */
2736 * elevator says don't/can't merge. get new request
2737 */
2738 case ELEVATOR_NO_MERGE:
2739 break;
2740
2741 default: 2752 default:
2742 printk("elevator returned crap (%d)\n", el_ret); 2753 ;
2743 BUG();
2744 } 2754 }
2745 2755
2756get_rq:
2746 /* 2757 /*
2747 * Grab a free request from the freelist - if that is empty, check 2758 * Grab a free request. This is might sleep but can not fail.
2748 * if we are doing read ahead and abort instead of blocking for 2759 * Returns with the queue unlocked.
2749 * a free slot. 2760 */
2761 req = get_request_wait(q, rw, bio);
2762
2763 /*
2764 * After dropping the lock and possibly sleeping here, our request
2765 * may now be mergeable after it had proven unmergeable (above).
2766 * We don't worry about that case for efficiency. It won't happen
2767 * often, and the elevators are able to handle it.
2750 */ 2768 */
2751get_rq:
2752 if (freereq) {
2753 req = freereq;
2754 freereq = NULL;
2755 } else {
2756 spin_unlock_irq(q->queue_lock);
2757 if ((freereq = get_request(q, rw, GFP_ATOMIC)) == NULL) {
2758 /*
2759 * READA bit set
2760 */
2761 err = -EWOULDBLOCK;
2762 if (bio_rw_ahead(bio))
2763 goto end_io;
2764
2765 freereq = get_request_wait(q, rw);
2766 }
2767 goto again;
2768 }
2769 2769
2770 req->flags |= REQ_CMD; 2770 req->flags |= REQ_CMD;
2771 2771
@@ -2778,7 +2778,7 @@ get_rq:
2778 /* 2778 /*
2779 * REQ_BARRIER implies no merging, but lets make it explicit 2779 * REQ_BARRIER implies no merging, but lets make it explicit
2780 */ 2780 */
2781 if (barrier) 2781 if (unlikely(barrier))
2782 req->flags |= (REQ_HARDBARRIER | REQ_NOMERGE); 2782 req->flags |= (REQ_HARDBARRIER | REQ_NOMERGE);
2783 2783
2784 req->errors = 0; 2784 req->errors = 0;
@@ -2790,13 +2790,15 @@ get_rq:
2790 req->buffer = bio_data(bio); /* see ->buffer comment above */ 2790 req->buffer = bio_data(bio); /* see ->buffer comment above */
2791 req->waiting = NULL; 2791 req->waiting = NULL;
2792 req->bio = req->biotail = bio; 2792 req->bio = req->biotail = bio;
2793 req->ioprio = prio;
2793 req->rq_disk = bio->bi_bdev->bd_disk; 2794 req->rq_disk = bio->bi_bdev->bd_disk;
2794 req->start_time = jiffies; 2795 req->start_time = jiffies;
2795 2796
2797 spin_lock_irq(q->queue_lock);
2798 if (elv_queue_empty(q))
2799 blk_plug_device(q);
2796 add_request(q, req); 2800 add_request(q, req);
2797out: 2801out:
2798 if (freereq)
2799 __blk_put_request(q, freereq);
2800 if (sync) 2802 if (sync)
2801 __generic_unplug_device(q); 2803 __generic_unplug_device(q);
2802 2804
@@ -2818,7 +2820,7 @@ static inline void blk_partition_remap(struct bio *bio)
2818 if (bdev != bdev->bd_contains) { 2820 if (bdev != bdev->bd_contains) {
2819 struct hd_struct *p = bdev->bd_part; 2821 struct hd_struct *p = bdev->bd_part;
2820 2822
2821 switch (bio->bi_rw) { 2823 switch (bio_data_dir(bio)) {
2822 case READ: 2824 case READ:
2823 p->read_sectors += bio_sectors(bio); 2825 p->read_sectors += bio_sectors(bio);
2824 p->reads++; 2826 p->reads++;
@@ -2837,6 +2839,7 @@ void blk_finish_queue_drain(request_queue_t *q)
2837{ 2839{
2838 struct request_list *rl = &q->rq; 2840 struct request_list *rl = &q->rq;
2839 struct request *rq; 2841 struct request *rq;
2842 int requeued = 0;
2840 2843
2841 spin_lock_irq(q->queue_lock); 2844 spin_lock_irq(q->queue_lock);
2842 clear_bit(QUEUE_FLAG_DRAIN, &q->queue_flags); 2845 clear_bit(QUEUE_FLAG_DRAIN, &q->queue_flags);
@@ -2845,9 +2848,13 @@ void blk_finish_queue_drain(request_queue_t *q)
2845 rq = list_entry_rq(q->drain_list.next); 2848 rq = list_entry_rq(q->drain_list.next);
2846 2849
2847 list_del_init(&rq->queuelist); 2850 list_del_init(&rq->queuelist);
2848 __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 1); 2851 elv_requeue_request(q, rq);
2852 requeued++;
2849 } 2853 }
2850 2854
2855 if (requeued)
2856 q->request_fn(q);
2857
2851 spin_unlock_irq(q->queue_lock); 2858 spin_unlock_irq(q->queue_lock);
2852 2859
2853 wake_up(&rl->wait[0]); 2860 wake_up(&rl->wait[0]);
@@ -2902,7 +2909,7 @@ static inline void block_wait_queue_running(request_queue_t *q)
2902{ 2909{
2903 DEFINE_WAIT(wait); 2910 DEFINE_WAIT(wait);
2904 2911
2905 while (test_bit(QUEUE_FLAG_DRAIN, &q->queue_flags)) { 2912 while (unlikely(test_bit(QUEUE_FLAG_DRAIN, &q->queue_flags))) {
2906 struct request_list *rl = &q->rq; 2913 struct request_list *rl = &q->rq;
2907 2914
2908 prepare_to_wait_exclusive(&rl->drain, &wait, 2915 prepare_to_wait_exclusive(&rl->drain, &wait,
@@ -3011,7 +3018,7 @@ end_io:
3011 goto end_io; 3018 goto end_io;
3012 } 3019 }
3013 3020
3014 if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)) 3021 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
3015 goto end_io; 3022 goto end_io;
3016 3023
3017 block_wait_queue_running(q); 3024 block_wait_queue_running(q);
@@ -3044,7 +3051,7 @@ void submit_bio(int rw, struct bio *bio)
3044 3051
3045 BIO_BUG_ON(!bio->bi_size); 3052 BIO_BUG_ON(!bio->bi_size);
3046 BIO_BUG_ON(!bio->bi_io_vec); 3053 BIO_BUG_ON(!bio->bi_io_vec);
3047 bio->bi_rw = rw; 3054 bio->bi_rw |= rw;
3048 if (rw & WRITE) 3055 if (rw & WRITE)
3049 mod_page_state(pgpgout, count); 3056 mod_page_state(pgpgout, count);
3050 else 3057 else
@@ -3064,7 +3071,7 @@ void submit_bio(int rw, struct bio *bio)
3064 3071
3065EXPORT_SYMBOL(submit_bio); 3072EXPORT_SYMBOL(submit_bio);
3066 3073
3067void blk_recalc_rq_segments(struct request *rq) 3074static void blk_recalc_rq_segments(struct request *rq)
3068{ 3075{
3069 struct bio *bio, *prevbio = NULL; 3076 struct bio *bio, *prevbio = NULL;
3070 int nr_phys_segs, nr_hw_segs; 3077 int nr_phys_segs, nr_hw_segs;
@@ -3106,7 +3113,7 @@ void blk_recalc_rq_segments(struct request *rq)
3106 rq->nr_hw_segments = nr_hw_segs; 3113 rq->nr_hw_segments = nr_hw_segs;
3107} 3114}
3108 3115
3109void blk_recalc_rq_sectors(struct request *rq, int nsect) 3116static void blk_recalc_rq_sectors(struct request *rq, int nsect)
3110{ 3117{
3111 if (blk_fs_request(rq)) { 3118 if (blk_fs_request(rq)) {
3112 rq->hard_sector += nsect; 3119 rq->hard_sector += nsect;
@@ -3401,8 +3408,11 @@ void exit_io_context(void)
3401 struct io_context *ioc; 3408 struct io_context *ioc;
3402 3409
3403 local_irq_save(flags); 3410 local_irq_save(flags);
3411 task_lock(current);
3404 ioc = current->io_context; 3412 ioc = current->io_context;
3405 current->io_context = NULL; 3413 current->io_context = NULL;
3414 ioc->task = NULL;
3415 task_unlock(current);
3406 local_irq_restore(flags); 3416 local_irq_restore(flags);
3407 3417
3408 if (ioc->aic && ioc->aic->exit) 3418 if (ioc->aic && ioc->aic->exit)
@@ -3415,53 +3425,49 @@ void exit_io_context(void)
3415 3425
3416/* 3426/*
3417 * If the current task has no IO context then create one and initialise it. 3427 * If the current task has no IO context then create one and initialise it.
3418 * If it does have a context, take a ref on it. 3428 * Otherwise, return its existing IO context.
3419 * 3429 *
3420 * This is always called in the context of the task which submitted the I/O. 3430 * This returned IO context doesn't have a specifically elevated refcount,
3421 * But weird things happen, so we disable local interrupts to ensure exclusive 3431 * but since the current task itself holds a reference, the context can be
3422 * access to *current. 3432 * used in general code, so long as it stays within `current` context.
3423 */ 3433 */
3424struct io_context *get_io_context(int gfp_flags) 3434struct io_context *current_io_context(int gfp_flags)
3425{ 3435{
3426 struct task_struct *tsk = current; 3436 struct task_struct *tsk = current;
3427 unsigned long flags;
3428 struct io_context *ret; 3437 struct io_context *ret;
3429 3438
3430 local_irq_save(flags);
3431 ret = tsk->io_context; 3439 ret = tsk->io_context;
3432 if (ret) 3440 if (likely(ret))
3433 goto out; 3441 return ret;
3434
3435 local_irq_restore(flags);
3436 3442
3437 ret = kmem_cache_alloc(iocontext_cachep, gfp_flags); 3443 ret = kmem_cache_alloc(iocontext_cachep, gfp_flags);
3438 if (ret) { 3444 if (ret) {
3439 atomic_set(&ret->refcount, 1); 3445 atomic_set(&ret->refcount, 1);
3440 ret->pid = tsk->pid; 3446 ret->task = current;
3447 ret->set_ioprio = NULL;
3441 ret->last_waited = jiffies; /* doesn't matter... */ 3448 ret->last_waited = jiffies; /* doesn't matter... */
3442 ret->nr_batch_requests = 0; /* because this is 0 */ 3449 ret->nr_batch_requests = 0; /* because this is 0 */
3443 ret->aic = NULL; 3450 ret->aic = NULL;
3444 ret->cic = NULL; 3451 ret->cic = NULL;
3445 spin_lock_init(&ret->lock); 3452 tsk->io_context = ret;
3446 3453 }
3447 local_irq_save(flags);
3448 3454
3449 /* 3455 return ret;
3450 * very unlikely, someone raced with us in setting up the task 3456}
3451 * io context. free new context and just grab a reference. 3457EXPORT_SYMBOL(current_io_context);
3452 */
3453 if (!tsk->io_context)
3454 tsk->io_context = ret;
3455 else {
3456 kmem_cache_free(iocontext_cachep, ret);
3457 ret = tsk->io_context;
3458 }
3459 3458
3460out: 3459/*
3460 * If the current task has no IO context then create one and initialise it.
3461 * If it does have a context, take a ref on it.
3462 *
3463 * This is always called in the context of the task which submitted the I/O.
3464 */
3465struct io_context *get_io_context(int gfp_flags)
3466{
3467 struct io_context *ret;
3468 ret = current_io_context(gfp_flags);
3469 if (likely(ret))
3461 atomic_inc(&ret->refcount); 3470 atomic_inc(&ret->refcount);
3462 local_irq_restore(flags);
3463 }
3464
3465 return ret; 3471 return ret;
3466} 3472}
3467EXPORT_SYMBOL(get_io_context); 3473EXPORT_SYMBOL(get_io_context);
@@ -3670,7 +3676,7 @@ queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3670 3676
3671 q = container_of(kobj, struct request_queue, kobj); 3677 q = container_of(kobj, struct request_queue, kobj);
3672 if (!entry->show) 3678 if (!entry->show)
3673 return 0; 3679 return -EIO;
3674 3680
3675 return entry->show(q, page); 3681 return entry->show(q, page);
3676} 3682}
@@ -3684,7 +3690,7 @@ queue_attr_store(struct kobject *kobj, struct attribute *attr,
3684 3690
3685 q = container_of(kobj, struct request_queue, kobj); 3691 q = container_of(kobj, struct request_queue, kobj);
3686 if (!entry->store) 3692 if (!entry->store)
3687 return -EINVAL; 3693 return -EIO;
3688 3694
3689 return entry->store(q, page, length); 3695 return entry->store(q, page, length);
3690} 3696}
@@ -3694,7 +3700,7 @@ static struct sysfs_ops queue_sysfs_ops = {
3694 .store = queue_attr_store, 3700 .store = queue_attr_store,
3695}; 3701};
3696 3702
3697struct kobj_type queue_ktype = { 3703static struct kobj_type queue_ktype = {
3698 .sysfs_ops = &queue_sysfs_ops, 3704 .sysfs_ops = &queue_sysfs_ops,
3699 .default_attrs = default_attrs, 3705 .default_attrs = default_attrs,
3700}; 3706};
diff --git a/drivers/block/loop.c b/drivers/block/loop.c
index 6f011d0d8e97..b35e08876dd4 100644
--- a/drivers/block/loop.c
+++ b/drivers/block/loop.c
@@ -472,17 +472,11 @@ static int do_bio_filebacked(struct loop_device *lo, struct bio *bio)
472 */ 472 */
473static void loop_add_bio(struct loop_device *lo, struct bio *bio) 473static void loop_add_bio(struct loop_device *lo, struct bio *bio)
474{ 474{
475 unsigned long flags;
476
477 spin_lock_irqsave(&lo->lo_lock, flags);
478 if (lo->lo_biotail) { 475 if (lo->lo_biotail) {
479 lo->lo_biotail->bi_next = bio; 476 lo->lo_biotail->bi_next = bio;
480 lo->lo_biotail = bio; 477 lo->lo_biotail = bio;
481 } else 478 } else
482 lo->lo_bio = lo->lo_biotail = bio; 479 lo->lo_bio = lo->lo_biotail = bio;
483 spin_unlock_irqrestore(&lo->lo_lock, flags);
484
485 up(&lo->lo_bh_mutex);
486} 480}
487 481
488/* 482/*
@@ -492,14 +486,12 @@ static struct bio *loop_get_bio(struct loop_device *lo)
492{ 486{
493 struct bio *bio; 487 struct bio *bio;
494 488
495 spin_lock_irq(&lo->lo_lock);
496 if ((bio = lo->lo_bio)) { 489 if ((bio = lo->lo_bio)) {
497 if (bio == lo->lo_biotail) 490 if (bio == lo->lo_biotail)
498 lo->lo_biotail = NULL; 491 lo->lo_biotail = NULL;
499 lo->lo_bio = bio->bi_next; 492 lo->lo_bio = bio->bi_next;
500 bio->bi_next = NULL; 493 bio->bi_next = NULL;
501 } 494 }
502 spin_unlock_irq(&lo->lo_lock);
503 495
504 return bio; 496 return bio;
505} 497}
@@ -509,35 +501,28 @@ static int loop_make_request(request_queue_t *q, struct bio *old_bio)
509 struct loop_device *lo = q->queuedata; 501 struct loop_device *lo = q->queuedata;
510 int rw = bio_rw(old_bio); 502 int rw = bio_rw(old_bio);
511 503
512 if (!lo) 504 if (rw == READA)
513 goto out; 505 rw = READ;
506
507 BUG_ON(!lo || (rw != READ && rw != WRITE));
514 508
515 spin_lock_irq(&lo->lo_lock); 509 spin_lock_irq(&lo->lo_lock);
516 if (lo->lo_state != Lo_bound) 510 if (lo->lo_state != Lo_bound)
517 goto inactive; 511 goto out;
518 atomic_inc(&lo->lo_pending); 512 if (unlikely(rw == WRITE && (lo->lo_flags & LO_FLAGS_READ_ONLY)))
519 spin_unlock_irq(&lo->lo_lock); 513 goto out;
520 514 lo->lo_pending++;
521 if (rw == WRITE) {
522 if (lo->lo_flags & LO_FLAGS_READ_ONLY)
523 goto err;
524 } else if (rw == READA) {
525 rw = READ;
526 } else if (rw != READ) {
527 printk(KERN_ERR "loop: unknown command (%x)\n", rw);
528 goto err;
529 }
530 loop_add_bio(lo, old_bio); 515 loop_add_bio(lo, old_bio);
516 spin_unlock_irq(&lo->lo_lock);
517 up(&lo->lo_bh_mutex);
531 return 0; 518 return 0;
532err: 519
533 if (atomic_dec_and_test(&lo->lo_pending))
534 up(&lo->lo_bh_mutex);
535out: 520out:
521 if (lo->lo_pending == 0)
522 up(&lo->lo_bh_mutex);
523 spin_unlock_irq(&lo->lo_lock);
536 bio_io_error(old_bio, old_bio->bi_size); 524 bio_io_error(old_bio, old_bio->bi_size);
537 return 0; 525 return 0;
538inactive:
539 spin_unlock_irq(&lo->lo_lock);
540 goto out;
541} 526}
542 527
543/* 528/*
@@ -560,13 +545,11 @@ static void do_loop_switch(struct loop_device *, struct switch_request *);
560 545
561static inline void loop_handle_bio(struct loop_device *lo, struct bio *bio) 546static inline void loop_handle_bio(struct loop_device *lo, struct bio *bio)
562{ 547{
563 int ret;
564
565 if (unlikely(!bio->bi_bdev)) { 548 if (unlikely(!bio->bi_bdev)) {
566 do_loop_switch(lo, bio->bi_private); 549 do_loop_switch(lo, bio->bi_private);
567 bio_put(bio); 550 bio_put(bio);
568 } else { 551 } else {
569 ret = do_bio_filebacked(lo, bio); 552 int ret = do_bio_filebacked(lo, bio);
570 bio_endio(bio, bio->bi_size, ret); 553 bio_endio(bio, bio->bi_size, ret);
571 } 554 }
572} 555}
@@ -594,7 +577,7 @@ static int loop_thread(void *data)
594 set_user_nice(current, -20); 577 set_user_nice(current, -20);
595 578
596 lo->lo_state = Lo_bound; 579 lo->lo_state = Lo_bound;
597 atomic_inc(&lo->lo_pending); 580 lo->lo_pending = 1;
598 581
599 /* 582 /*
600 * up sem, we are running 583 * up sem, we are running
@@ -602,26 +585,37 @@ static int loop_thread(void *data)
602 up(&lo->lo_sem); 585 up(&lo->lo_sem);
603 586
604 for (;;) { 587 for (;;) {
605 down_interruptible(&lo->lo_bh_mutex); 588 int pending;
589
606 /* 590 /*
607 * could be upped because of tear-down, not because of 591 * interruptible just to not contribute to load avg
608 * pending work
609 */ 592 */
610 if (!atomic_read(&lo->lo_pending)) 593 if (down_interruptible(&lo->lo_bh_mutex))
594 continue;
595
596 spin_lock_irq(&lo->lo_lock);
597
598 /*
599 * could be upped because of tear-down, not pending work
600 */
601 if (unlikely(!lo->lo_pending)) {
602 spin_unlock_irq(&lo->lo_lock);
611 break; 603 break;
604 }
612 605
613 bio = loop_get_bio(lo); 606 bio = loop_get_bio(lo);
614 if (!bio) { 607 lo->lo_pending--;
615 printk("loop: missing bio\n"); 608 pending = lo->lo_pending;
616 continue; 609 spin_unlock_irq(&lo->lo_lock);
617 } 610
611 BUG_ON(!bio);
618 loop_handle_bio(lo, bio); 612 loop_handle_bio(lo, bio);
619 613
620 /* 614 /*
621 * upped both for pending work and tear-down, lo_pending 615 * upped both for pending work and tear-down, lo_pending
622 * will hit zero then 616 * will hit zero then
623 */ 617 */
624 if (atomic_dec_and_test(&lo->lo_pending)) 618 if (unlikely(!pending))
625 break; 619 break;
626 } 620 }
627 621
@@ -900,7 +894,8 @@ static int loop_clr_fd(struct loop_device *lo, struct block_device *bdev)
900 894
901 spin_lock_irq(&lo->lo_lock); 895 spin_lock_irq(&lo->lo_lock);
902 lo->lo_state = Lo_rundown; 896 lo->lo_state = Lo_rundown;
903 if (atomic_dec_and_test(&lo->lo_pending)) 897 lo->lo_pending--;
898 if (!lo->lo_pending)
904 up(&lo->lo_bh_mutex); 899 up(&lo->lo_bh_mutex);
905 spin_unlock_irq(&lo->lo_lock); 900 spin_unlock_irq(&lo->lo_lock);
906 901
diff --git a/drivers/block/paride/pg.c b/drivers/block/paride/pg.c
index dbeb107bb971..84d8e291ed96 100644
--- a/drivers/block/paride/pg.c
+++ b/drivers/block/paride/pg.c
@@ -222,7 +222,7 @@ static int pg_identify(struct pg *dev, int log);
222 222
223static char pg_scratch[512]; /* scratch block buffer */ 223static char pg_scratch[512]; /* scratch block buffer */
224 224
225static struct class_simple *pg_class; 225static struct class *pg_class;
226 226
227/* kernel glue structures */ 227/* kernel glue structures */
228 228
@@ -666,7 +666,7 @@ static int __init pg_init(void)
666 err = -1; 666 err = -1;
667 goto out; 667 goto out;
668 } 668 }
669 pg_class = class_simple_create(THIS_MODULE, "pg"); 669 pg_class = class_create(THIS_MODULE, "pg");
670 if (IS_ERR(pg_class)) { 670 if (IS_ERR(pg_class)) {
671 err = PTR_ERR(pg_class); 671 err = PTR_ERR(pg_class);
672 goto out_chrdev; 672 goto out_chrdev;
@@ -675,7 +675,7 @@ static int __init pg_init(void)
675 for (unit = 0; unit < PG_UNITS; unit++) { 675 for (unit = 0; unit < PG_UNITS; unit++) {
676 struct pg *dev = &devices[unit]; 676 struct pg *dev = &devices[unit];
677 if (dev->present) { 677 if (dev->present) {
678 class_simple_device_add(pg_class, MKDEV(major, unit), 678 class_device_create(pg_class, MKDEV(major, unit),
679 NULL, "pg%u", unit); 679 NULL, "pg%u", unit);
680 err = devfs_mk_cdev(MKDEV(major, unit), 680 err = devfs_mk_cdev(MKDEV(major, unit),
681 S_IFCHR | S_IRUSR | S_IWUSR, "pg/%u", 681 S_IFCHR | S_IRUSR | S_IWUSR, "pg/%u",
@@ -688,8 +688,8 @@ static int __init pg_init(void)
688 goto out; 688 goto out;
689 689
690out_class: 690out_class:
691 class_simple_device_remove(MKDEV(major, unit)); 691 class_device_destroy(pg_class, MKDEV(major, unit));
692 class_simple_destroy(pg_class); 692 class_destroy(pg_class);
693out_chrdev: 693out_chrdev:
694 unregister_chrdev(major, "pg"); 694 unregister_chrdev(major, "pg");
695out: 695out:
@@ -703,11 +703,11 @@ static void __exit pg_exit(void)
703 for (unit = 0; unit < PG_UNITS; unit++) { 703 for (unit = 0; unit < PG_UNITS; unit++) {
704 struct pg *dev = &devices[unit]; 704 struct pg *dev = &devices[unit];
705 if (dev->present) { 705 if (dev->present) {
706 class_simple_device_remove(MKDEV(major, unit)); 706 class_device_destroy(pg_class, MKDEV(major, unit));
707 devfs_remove("pg/%u", unit); 707 devfs_remove("pg/%u", unit);
708 } 708 }
709 } 709 }
710 class_simple_destroy(pg_class); 710 class_destroy(pg_class);
711 devfs_remove("pg"); 711 devfs_remove("pg");
712 unregister_chrdev(major, name); 712 unregister_chrdev(major, name);
713 713
diff --git a/drivers/block/paride/pt.c b/drivers/block/paride/pt.c
index 8fbd6922fe0d..5fe8ee86f095 100644
--- a/drivers/block/paride/pt.c
+++ b/drivers/block/paride/pt.c
@@ -242,7 +242,7 @@ static struct file_operations pt_fops = {
242}; 242};
243 243
244/* sysfs class support */ 244/* sysfs class support */
245static struct class_simple *pt_class; 245static struct class *pt_class;
246 246
247static inline int status_reg(struct pi_adapter *pi) 247static inline int status_reg(struct pi_adapter *pi)
248{ 248{
@@ -963,7 +963,7 @@ static int __init pt_init(void)
963 err = -1; 963 err = -1;
964 goto out; 964 goto out;
965 } 965 }
966 pt_class = class_simple_create(THIS_MODULE, "pt"); 966 pt_class = class_create(THIS_MODULE, "pt");
967 if (IS_ERR(pt_class)) { 967 if (IS_ERR(pt_class)) {
968 err = PTR_ERR(pt_class); 968 err = PTR_ERR(pt_class);
969 goto out_chrdev; 969 goto out_chrdev;
@@ -972,29 +972,29 @@ static int __init pt_init(void)
972 devfs_mk_dir("pt"); 972 devfs_mk_dir("pt");
973 for (unit = 0; unit < PT_UNITS; unit++) 973 for (unit = 0; unit < PT_UNITS; unit++)
974 if (pt[unit].present) { 974 if (pt[unit].present) {
975 class_simple_device_add(pt_class, MKDEV(major, unit), 975 class_device_create(pt_class, MKDEV(major, unit),
976 NULL, "pt%d", unit); 976 NULL, "pt%d", unit);
977 err = devfs_mk_cdev(MKDEV(major, unit), 977 err = devfs_mk_cdev(MKDEV(major, unit),
978 S_IFCHR | S_IRUSR | S_IWUSR, 978 S_IFCHR | S_IRUSR | S_IWUSR,
979 "pt/%d", unit); 979 "pt/%d", unit);
980 if (err) { 980 if (err) {
981 class_simple_device_remove(MKDEV(major, unit)); 981 class_device_destroy(pt_class, MKDEV(major, unit));
982 goto out_class; 982 goto out_class;
983 } 983 }
984 class_simple_device_add(pt_class, MKDEV(major, unit + 128), 984 class_device_create(pt_class, MKDEV(major, unit + 128),
985 NULL, "pt%dn", unit); 985 NULL, "pt%dn", unit);
986 err = devfs_mk_cdev(MKDEV(major, unit + 128), 986 err = devfs_mk_cdev(MKDEV(major, unit + 128),
987 S_IFCHR | S_IRUSR | S_IWUSR, 987 S_IFCHR | S_IRUSR | S_IWUSR,
988 "pt/%dn", unit); 988 "pt/%dn", unit);
989 if (err) { 989 if (err) {
990 class_simple_device_remove(MKDEV(major, unit + 128)); 990 class_device_destroy(pt_class, MKDEV(major, unit + 128));
991 goto out_class; 991 goto out_class;
992 } 992 }
993 } 993 }
994 goto out; 994 goto out;
995 995
996out_class: 996out_class:
997 class_simple_destroy(pt_class); 997 class_destroy(pt_class);
998out_chrdev: 998out_chrdev:
999 unregister_chrdev(major, "pt"); 999 unregister_chrdev(major, "pt");
1000out: 1000out:
@@ -1006,12 +1006,12 @@ static void __exit pt_exit(void)
1006 int unit; 1006 int unit;
1007 for (unit = 0; unit < PT_UNITS; unit++) 1007 for (unit = 0; unit < PT_UNITS; unit++)
1008 if (pt[unit].present) { 1008 if (pt[unit].present) {
1009 class_simple_device_remove(MKDEV(major, unit)); 1009 class_device_destroy(pt_class, MKDEV(major, unit));
1010 devfs_remove("pt/%d", unit); 1010 devfs_remove("pt/%d", unit);
1011 class_simple_device_remove(MKDEV(major, unit + 128)); 1011 class_device_destroy(pt_class, MKDEV(major, unit + 128));
1012 devfs_remove("pt/%dn", unit); 1012 devfs_remove("pt/%dn", unit);
1013 } 1013 }
1014 class_simple_destroy(pt_class); 1014 class_destroy(pt_class);
1015 devfs_remove("pt"); 1015 devfs_remove("pt");
1016 unregister_chrdev(major, name); 1016 unregister_chrdev(major, name);
1017 for (unit = 0; unit < PT_UNITS; unit++) 1017 for (unit = 0; unit < PT_UNITS; unit++)
diff --git a/drivers/block/pktcdvd.c b/drivers/block/pktcdvd.c
index bc56770bcc90..7b838342f0a3 100644
--- a/drivers/block/pktcdvd.c
+++ b/drivers/block/pktcdvd.c
@@ -467,14 +467,12 @@ static int pkt_set_speed(struct pktcdvd_device *pd, unsigned write_speed, unsign
467 * Queue a bio for processing by the low-level CD device. Must be called 467 * Queue a bio for processing by the low-level CD device. Must be called
468 * from process context. 468 * from process context.
469 */ 469 */
470static void pkt_queue_bio(struct pktcdvd_device *pd, struct bio *bio, int high_prio_read) 470static void pkt_queue_bio(struct pktcdvd_device *pd, struct bio *bio)
471{ 471{
472 spin_lock(&pd->iosched.lock); 472 spin_lock(&pd->iosched.lock);
473 if (bio_data_dir(bio) == READ) { 473 if (bio_data_dir(bio) == READ) {
474 pkt_add_list_last(bio, &pd->iosched.read_queue, 474 pkt_add_list_last(bio, &pd->iosched.read_queue,
475 &pd->iosched.read_queue_tail); 475 &pd->iosched.read_queue_tail);
476 if (high_prio_read)
477 pd->iosched.high_prio_read = 1;
478 } else { 476 } else {
479 pkt_add_list_last(bio, &pd->iosched.write_queue, 477 pkt_add_list_last(bio, &pd->iosched.write_queue,
480 &pd->iosched.write_queue_tail); 478 &pd->iosched.write_queue_tail);
@@ -490,15 +488,16 @@ static void pkt_queue_bio(struct pktcdvd_device *pd, struct bio *bio, int high_p
490 * requirements for CDRW drives: 488 * requirements for CDRW drives:
491 * - A cache flush command must be inserted before a read request if the 489 * - A cache flush command must be inserted before a read request if the
492 * previous request was a write. 490 * previous request was a write.
493 * - Switching between reading and writing is slow, so don't it more often 491 * - Switching between reading and writing is slow, so don't do it more often
494 * than necessary. 492 * than necessary.
493 * - Optimize for throughput at the expense of latency. This means that streaming
494 * writes will never be interrupted by a read, but if the drive has to seek
495 * before the next write, switch to reading instead if there are any pending
496 * read requests.
495 * - Set the read speed according to current usage pattern. When only reading 497 * - Set the read speed according to current usage pattern. When only reading
496 * from the device, it's best to use the highest possible read speed, but 498 * from the device, it's best to use the highest possible read speed, but
497 * when switching often between reading and writing, it's better to have the 499 * when switching often between reading and writing, it's better to have the
498 * same read and write speeds. 500 * same read and write speeds.
499 * - Reads originating from user space should have higher priority than reads
500 * originating from pkt_gather_data, because some process is usually waiting
501 * on reads of the first kind.
502 */ 501 */
503static void pkt_iosched_process_queue(struct pktcdvd_device *pd) 502static void pkt_iosched_process_queue(struct pktcdvd_device *pd)
504{ 503{
@@ -512,21 +511,24 @@ static void pkt_iosched_process_queue(struct pktcdvd_device *pd)
512 511
513 for (;;) { 512 for (;;) {
514 struct bio *bio; 513 struct bio *bio;
515 int reads_queued, writes_queued, high_prio_read; 514 int reads_queued, writes_queued;
516 515
517 spin_lock(&pd->iosched.lock); 516 spin_lock(&pd->iosched.lock);
518 reads_queued = (pd->iosched.read_queue != NULL); 517 reads_queued = (pd->iosched.read_queue != NULL);
519 writes_queued = (pd->iosched.write_queue != NULL); 518 writes_queued = (pd->iosched.write_queue != NULL);
520 if (!reads_queued)
521 pd->iosched.high_prio_read = 0;
522 high_prio_read = pd->iosched.high_prio_read;
523 spin_unlock(&pd->iosched.lock); 519 spin_unlock(&pd->iosched.lock);
524 520
525 if (!reads_queued && !writes_queued) 521 if (!reads_queued && !writes_queued)
526 break; 522 break;
527 523
528 if (pd->iosched.writing) { 524 if (pd->iosched.writing) {
529 if (high_prio_read || (!writes_queued && reads_queued)) { 525 int need_write_seek = 1;
526 spin_lock(&pd->iosched.lock);
527 bio = pd->iosched.write_queue;
528 spin_unlock(&pd->iosched.lock);
529 if (bio && (bio->bi_sector == pd->iosched.last_write))
530 need_write_seek = 0;
531 if (need_write_seek && reads_queued) {
530 if (atomic_read(&pd->cdrw.pending_bios) > 0) { 532 if (atomic_read(&pd->cdrw.pending_bios) > 0) {
531 VPRINTK("pktcdvd: write, waiting\n"); 533 VPRINTK("pktcdvd: write, waiting\n");
532 break; 534 break;
@@ -559,8 +561,10 @@ static void pkt_iosched_process_queue(struct pktcdvd_device *pd)
559 561
560 if (bio_data_dir(bio) == READ) 562 if (bio_data_dir(bio) == READ)
561 pd->iosched.successive_reads += bio->bi_size >> 10; 563 pd->iosched.successive_reads += bio->bi_size >> 10;
562 else 564 else {
563 pd->iosched.successive_reads = 0; 565 pd->iosched.successive_reads = 0;
566 pd->iosched.last_write = bio->bi_sector + bio_sectors(bio);
567 }
564 if (pd->iosched.successive_reads >= HI_SPEED_SWITCH) { 568 if (pd->iosched.successive_reads >= HI_SPEED_SWITCH) {
565 if (pd->read_speed == pd->write_speed) { 569 if (pd->read_speed == pd->write_speed) {
566 pd->read_speed = MAX_SPEED; 570 pd->read_speed = MAX_SPEED;
@@ -765,7 +769,7 @@ static void pkt_gather_data(struct pktcdvd_device *pd, struct packet_data *pkt)
765 769
766 atomic_inc(&pkt->io_wait); 770 atomic_inc(&pkt->io_wait);
767 bio->bi_rw = READ; 771 bio->bi_rw = READ;
768 pkt_queue_bio(pd, bio, 0); 772 pkt_queue_bio(pd, bio);
769 frames_read++; 773 frames_read++;
770 } 774 }
771 775
@@ -1062,7 +1066,7 @@ static void pkt_start_write(struct pktcdvd_device *pd, struct packet_data *pkt)
1062 1066
1063 atomic_set(&pkt->io_wait, 1); 1067 atomic_set(&pkt->io_wait, 1);
1064 pkt->w_bio->bi_rw = WRITE; 1068 pkt->w_bio->bi_rw = WRITE;
1065 pkt_queue_bio(pd, pkt->w_bio, 0); 1069 pkt_queue_bio(pd, pkt->w_bio);
1066} 1070}
1067 1071
1068static void pkt_finish_packet(struct packet_data *pkt, int uptodate) 1072static void pkt_finish_packet(struct packet_data *pkt, int uptodate)
@@ -1247,8 +1251,7 @@ static int kcdrwd(void *foobar)
1247 VPRINTK("kcdrwd: wake up\n"); 1251 VPRINTK("kcdrwd: wake up\n");
1248 1252
1249 /* make swsusp happy with our thread */ 1253 /* make swsusp happy with our thread */
1250 if (current->flags & PF_FREEZE) 1254 try_to_freeze();
1251 refrigerator(PF_FREEZE);
1252 1255
1253 list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) { 1256 list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
1254 if (!pkt->sleep_time) 1257 if (!pkt->sleep_time)
@@ -2120,7 +2123,7 @@ static int pkt_make_request(request_queue_t *q, struct bio *bio)
2120 cloned_bio->bi_private = psd; 2123 cloned_bio->bi_private = psd;
2121 cloned_bio->bi_end_io = pkt_end_io_read_cloned; 2124 cloned_bio->bi_end_io = pkt_end_io_read_cloned;
2122 pd->stats.secs_r += bio->bi_size >> 9; 2125 pd->stats.secs_r += bio->bi_size >> 9;
2123 pkt_queue_bio(pd, cloned_bio, 1); 2126 pkt_queue_bio(pd, cloned_bio);
2124 return 0; 2127 return 0;
2125 } 2128 }
2126 2129
diff --git a/drivers/block/swim3.c b/drivers/block/swim3.c
index 5b09cf154ac7..e5f7494c00ee 100644
--- a/drivers/block/swim3.c
+++ b/drivers/block/swim3.c
@@ -253,7 +253,7 @@ static int floppy_revalidate(struct gendisk *disk);
253static int swim3_add_device(struct device_node *swims); 253static int swim3_add_device(struct device_node *swims);
254int swim3_init(void); 254int swim3_init(void);
255 255
256#ifndef CONFIG_PMAC_PBOOK 256#ifndef CONFIG_PMAC_MEDIABAY
257#define check_media_bay(which, what) 1 257#define check_media_bay(which, what) 1
258#endif 258#endif
259 259
@@ -297,9 +297,11 @@ static void do_fd_request(request_queue_t * q)
297 int i; 297 int i;
298 for(i=0;i<floppy_count;i++) 298 for(i=0;i<floppy_count;i++)
299 { 299 {
300#ifdef CONFIG_PMAC_MEDIABAY
300 if (floppy_states[i].media_bay && 301 if (floppy_states[i].media_bay &&
301 check_media_bay(floppy_states[i].media_bay, MB_FD)) 302 check_media_bay(floppy_states[i].media_bay, MB_FD))
302 continue; 303 continue;
304#endif /* CONFIG_PMAC_MEDIABAY */
303 start_request(&floppy_states[i]); 305 start_request(&floppy_states[i]);
304 } 306 }
305 sti(); 307 sti();
@@ -856,8 +858,10 @@ static int floppy_ioctl(struct inode *inode, struct file *filp,
856 if ((cmd & 0x80) && !capable(CAP_SYS_ADMIN)) 858 if ((cmd & 0x80) && !capable(CAP_SYS_ADMIN))
857 return -EPERM; 859 return -EPERM;
858 860
861#ifdef CONFIG_PMAC_MEDIABAY
859 if (fs->media_bay && check_media_bay(fs->media_bay, MB_FD)) 862 if (fs->media_bay && check_media_bay(fs->media_bay, MB_FD))
860 return -ENXIO; 863 return -ENXIO;
864#endif
861 865
862 switch (cmd) { 866 switch (cmd) {
863 case FDEJECT: 867 case FDEJECT:
@@ -881,8 +885,10 @@ static int floppy_open(struct inode *inode, struct file *filp)
881 int n, err = 0; 885 int n, err = 0;
882 886
883 if (fs->ref_count == 0) { 887 if (fs->ref_count == 0) {
888#ifdef CONFIG_PMAC_MEDIABAY
884 if (fs->media_bay && check_media_bay(fs->media_bay, MB_FD)) 889 if (fs->media_bay && check_media_bay(fs->media_bay, MB_FD))
885 return -ENXIO; 890 return -ENXIO;
891#endif
886 out_8(&sw->setup, S_IBM_DRIVE | S_FCLK_DIV2); 892 out_8(&sw->setup, S_IBM_DRIVE | S_FCLK_DIV2);
887 out_8(&sw->control_bic, 0xff); 893 out_8(&sw->control_bic, 0xff);
888 out_8(&sw->mode, 0x95); 894 out_8(&sw->mode, 0x95);
@@ -967,8 +973,10 @@ static int floppy_revalidate(struct gendisk *disk)
967 struct swim3 __iomem *sw; 973 struct swim3 __iomem *sw;
968 int ret, n; 974 int ret, n;
969 975
976#ifdef CONFIG_PMAC_MEDIABAY
970 if (fs->media_bay && check_media_bay(fs->media_bay, MB_FD)) 977 if (fs->media_bay && check_media_bay(fs->media_bay, MB_FD))
971 return -ENXIO; 978 return -ENXIO;
979#endif
972 980
973 sw = fs->swim3; 981 sw = fs->swim3;
974 grab_drive(fs, revalidating, 0); 982 grab_drive(fs, revalidating, 0);
diff --git a/drivers/block/sx8.c b/drivers/block/sx8.c
index 5ed3a6379452..d57007b92f77 100644
--- a/drivers/block/sx8.c
+++ b/drivers/block/sx8.c
@@ -26,6 +26,7 @@
26#include <linux/delay.h> 26#include <linux/delay.h>
27#include <linux/time.h> 27#include <linux/time.h>
28#include <linux/hdreg.h> 28#include <linux/hdreg.h>
29#include <linux/dma-mapping.h>
29#include <asm/io.h> 30#include <asm/io.h>
30#include <asm/semaphore.h> 31#include <asm/semaphore.h>
31#include <asm/uaccess.h> 32#include <asm/uaccess.h>
@@ -1581,10 +1582,10 @@ static int carm_init_one (struct pci_dev *pdev, const struct pci_device_id *ent)
1581 if (rc) 1582 if (rc)
1582 goto err_out; 1583 goto err_out;
1583 1584
1584#if IF_64BIT_DMA_IS_POSSIBLE /* grrrr... */ 1585#ifdef IF_64BIT_DMA_IS_POSSIBLE /* grrrr... */
1585 rc = pci_set_dma_mask(pdev, 0xffffffffffffffffULL); 1586 rc = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
1586 if (!rc) { 1587 if (!rc) {
1587 rc = pci_set_consistent_dma_mask(pdev, 0xffffffffffffffffULL); 1588 rc = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK);
1588 if (rc) { 1589 if (rc) {
1589 printk(KERN_ERR DRV_NAME "(%s): consistent DMA mask failure\n", 1590 printk(KERN_ERR DRV_NAME "(%s): consistent DMA mask failure\n",
1590 pci_name(pdev)); 1591 pci_name(pdev));
@@ -1593,14 +1594,14 @@ static int carm_init_one (struct pci_dev *pdev, const struct pci_device_id *ent)
1593 pci_dac = 1; 1594 pci_dac = 1;
1594 } else { 1595 } else {
1595#endif 1596#endif
1596 rc = pci_set_dma_mask(pdev, 0xffffffffULL); 1597 rc = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
1597 if (rc) { 1598 if (rc) {
1598 printk(KERN_ERR DRV_NAME "(%s): DMA mask failure\n", 1599 printk(KERN_ERR DRV_NAME "(%s): DMA mask failure\n",
1599 pci_name(pdev)); 1600 pci_name(pdev));
1600 goto err_out_regions; 1601 goto err_out_regions;
1601 } 1602 }
1602 pci_dac = 0; 1603 pci_dac = 0;
1603#if IF_64BIT_DMA_IS_POSSIBLE /* grrrr... */ 1604#ifdef IF_64BIT_DMA_IS_POSSIBLE /* grrrr... */
1604 } 1605 }
1605#endif 1606#endif
1606 1607
diff --git a/drivers/block/ub.c b/drivers/block/ub.c
index 19c5e59bcfa8..a026567f5d18 100644
--- a/drivers/block/ub.c
+++ b/drivers/block/ub.c
@@ -23,6 +23,7 @@
23 * -- Exterminate P3 printks 23 * -- Exterminate P3 printks
24 * -- Resove XXX's 24 * -- Resove XXX's
25 * -- Redo "benh's retries", perhaps have spin-up code to handle them. V:D=? 25 * -- Redo "benh's retries", perhaps have spin-up code to handle them. V:D=?
26 * -- CLEAR, CLR2STS, CLRRS seem to be ripe for refactoring.
26 */ 27 */
27#include <linux/kernel.h> 28#include <linux/kernel.h>
28#include <linux/module.h> 29#include <linux/module.h>
@@ -38,6 +39,73 @@
38#define UB_MAJOR 180 39#define UB_MAJOR 180
39 40
40/* 41/*
42 * The command state machine is the key model for understanding of this driver.
43 *
44 * The general rule is that all transitions are done towards the bottom
45 * of the diagram, thus preventing any loops.
46 *
47 * An exception to that is how the STAT state is handled. A counter allows it
48 * to be re-entered along the path marked with [C].
49 *
50 * +--------+
51 * ! INIT !
52 * +--------+
53 * !
54 * ub_scsi_cmd_start fails ->--------------------------------------\
55 * ! !
56 * V !
57 * +--------+ !
58 * ! CMD ! !
59 * +--------+ !
60 * ! +--------+ !
61 * was -EPIPE -->-------------------------------->! CLEAR ! !
62 * ! +--------+ !
63 * ! ! !
64 * was error -->------------------------------------- ! --------->\
65 * ! ! !
66 * /--<-- cmd->dir == NONE ? ! !
67 * ! ! ! !
68 * ! V ! !
69 * ! +--------+ ! !
70 * ! ! DATA ! ! !
71 * ! +--------+ ! !
72 * ! ! +---------+ ! !
73 * ! was -EPIPE -->--------------->! CLR2STS ! ! !
74 * ! ! +---------+ ! !
75 * ! ! ! ! !
76 * ! ! was error -->---- ! --------->\
77 * ! was error -->--------------------- ! ------------- ! --------->\
78 * ! ! ! ! !
79 * ! V ! ! !
80 * \--->+--------+ ! ! !
81 * ! STAT !<--------------------------/ ! !
82 * /--->+--------+ ! !
83 * ! ! ! !
84 * [C] was -EPIPE -->-----------\ ! !
85 * ! ! ! ! !
86 * +<---- len == 0 ! ! !
87 * ! ! ! ! !
88 * ! was error -->--------------------------------------!---------->\
89 * ! ! ! ! !
90 * +<---- bad CSW ! ! !
91 * +<---- bad tag ! ! !
92 * ! ! V ! !
93 * ! ! +--------+ ! !
94 * ! ! ! CLRRS ! ! !
95 * ! ! +--------+ ! !
96 * ! ! ! ! !
97 * \------- ! --------------------[C]--------\ ! !
98 * ! ! ! !
99 * cmd->error---\ +--------+ ! !
100 * ! +--------------->! SENSE !<----------/ !
101 * STAT_FAIL----/ +--------+ !
102 * ! ! V
103 * ! V +--------+
104 * \--------------------------------\--------------------->! DONE !
105 * +--------+
106 */
107
108/*
41 * Definitions which have to be scattered once we understand the layout better. 109 * Definitions which have to be scattered once we understand the layout better.
42 */ 110 */
43 111
@@ -91,8 +159,6 @@ struct bulk_cs_wrap {
91 159
92#define US_BULK_CS_WRAP_LEN 13 160#define US_BULK_CS_WRAP_LEN 13
93#define US_BULK_CS_SIGN 0x53425355 /* spells out 'USBS' */ 161#define US_BULK_CS_SIGN 0x53425355 /* spells out 'USBS' */
94/* This is for Olympus Camedia digital cameras */
95#define US_BULK_CS_OLYMPUS_SIGN 0x55425355 /* spells out 'USBU' */
96#define US_BULK_STAT_OK 0 162#define US_BULK_STAT_OK 0
97#define US_BULK_STAT_FAIL 1 163#define US_BULK_STAT_FAIL 1
98#define US_BULK_STAT_PHASE 2 164#define US_BULK_STAT_PHASE 2
@@ -135,6 +201,7 @@ enum ub_scsi_cmd_state {
135 UB_CMDST_CLR2STS, /* Clearing before requesting status */ 201 UB_CMDST_CLR2STS, /* Clearing before requesting status */
136 UB_CMDST_STAT, /* Status phase */ 202 UB_CMDST_STAT, /* Status phase */
137 UB_CMDST_CLEAR, /* Clearing a stall (halt, actually) */ 203 UB_CMDST_CLEAR, /* Clearing a stall (halt, actually) */
204 UB_CMDST_CLRRS, /* Clearing before retrying status */
138 UB_CMDST_SENSE, /* Sending Request Sense */ 205 UB_CMDST_SENSE, /* Sending Request Sense */
139 UB_CMDST_DONE /* Final state */ 206 UB_CMDST_DONE /* Final state */
140}; 207};
@@ -146,6 +213,7 @@ static char *ub_scsi_cmd_stname[] = {
146 "c2s", 213 "c2s",
147 "sts", 214 "sts",
148 "clr", 215 "clr",
216 "crs",
149 "Sen", 217 "Sen",
150 "fin" 218 "fin"
151}; 219};
@@ -316,6 +384,7 @@ struct ub_dev {
316 struct urb work_urb; 384 struct urb work_urb;
317 struct timer_list work_timer; 385 struct timer_list work_timer;
318 int last_pipe; /* What might need clearing */ 386 int last_pipe; /* What might need clearing */
387 __le32 signature; /* Learned signature */
319 struct bulk_cb_wrap work_bcb; 388 struct bulk_cb_wrap work_bcb;
320 struct bulk_cs_wrap work_bcs; 389 struct bulk_cs_wrap work_bcs;
321 struct usb_ctrlrequest work_cr; 390 struct usb_ctrlrequest work_cr;
@@ -339,8 +408,9 @@ static void ub_scsi_action(unsigned long _dev);
339static void ub_scsi_dispatch(struct ub_dev *sc); 408static void ub_scsi_dispatch(struct ub_dev *sc);
340static void ub_scsi_urb_compl(struct ub_dev *sc, struct ub_scsi_cmd *cmd); 409static void ub_scsi_urb_compl(struct ub_dev *sc, struct ub_scsi_cmd *cmd);
341static void ub_state_done(struct ub_dev *sc, struct ub_scsi_cmd *cmd, int rc); 410static void ub_state_done(struct ub_dev *sc, struct ub_scsi_cmd *cmd, int rc);
342static void __ub_state_stat(struct ub_dev *sc, struct ub_scsi_cmd *cmd); 411static int __ub_state_stat(struct ub_dev *sc, struct ub_scsi_cmd *cmd);
343static void ub_state_stat(struct ub_dev *sc, struct ub_scsi_cmd *cmd); 412static void ub_state_stat(struct ub_dev *sc, struct ub_scsi_cmd *cmd);
413static void ub_state_stat_counted(struct ub_dev *sc, struct ub_scsi_cmd *cmd);
344static void ub_state_sense(struct ub_dev *sc, struct ub_scsi_cmd *cmd); 414static void ub_state_sense(struct ub_dev *sc, struct ub_scsi_cmd *cmd);
345static int ub_submit_clear_stall(struct ub_dev *sc, struct ub_scsi_cmd *cmd, 415static int ub_submit_clear_stall(struct ub_dev *sc, struct ub_scsi_cmd *cmd,
346 int stalled_pipe); 416 int stalled_pipe);
@@ -430,7 +500,7 @@ static void ub_cmdtr_sense(struct ub_dev *sc, struct ub_scsi_cmd *cmd,
430 } 500 }
431} 501}
432 502
433static ssize_t ub_diag_show(struct device *dev, char *page) 503static ssize_t ub_diag_show(struct device *dev, struct device_attribute *attr, char *page)
434{ 504{
435 struct usb_interface *intf; 505 struct usb_interface *intf;
436 struct ub_dev *sc; 506 struct ub_dev *sc;
@@ -1085,6 +1155,28 @@ static void ub_scsi_urb_compl(struct ub_dev *sc, struct ub_scsi_cmd *cmd)
1085 1155
1086 ub_state_stat(sc, cmd); 1156 ub_state_stat(sc, cmd);
1087 1157
1158 } else if (cmd->state == UB_CMDST_CLRRS) {
1159 if (urb->status == -EPIPE) {
1160 /*
1161 * STALL while clearning STALL.
1162 * The control pipe clears itself - nothing to do.
1163 * XXX Might try to reset the device here and retry.
1164 */
1165 printk(KERN_NOTICE "%s: stall on control pipe\n",
1166 sc->name);
1167 goto Bad_End;
1168 }
1169
1170 /*
1171 * We ignore the result for the halt clear.
1172 */
1173
1174 /* reset the endpoint toggle */
1175 usb_settoggle(sc->dev, usb_pipeendpoint(sc->last_pipe),
1176 usb_pipeout(sc->last_pipe), 0);
1177
1178 ub_state_stat_counted(sc, cmd);
1179
1088 } else if (cmd->state == UB_CMDST_CMD) { 1180 } else if (cmd->state == UB_CMDST_CMD) {
1089 if (urb->status == -EPIPE) { 1181 if (urb->status == -EPIPE) {
1090 rc = ub_submit_clear_stall(sc, cmd, sc->last_pipe); 1182 rc = ub_submit_clear_stall(sc, cmd, sc->last_pipe);
@@ -1190,52 +1282,57 @@ static void ub_scsi_urb_compl(struct ub_dev *sc, struct ub_scsi_cmd *cmd)
1190 */ 1282 */
1191 goto Bad_End; 1283 goto Bad_End;
1192 } 1284 }
1193 cmd->state = UB_CMDST_CLEAR; 1285
1286 /*
1287 * Having a stall when getting CSW is an error, so
1288 * make sure uppper levels are not oblivious to it.
1289 */
1290 cmd->error = -EIO; /* A cheap trick... */
1291
1292 cmd->state = UB_CMDST_CLRRS;
1194 ub_cmdtr_state(sc, cmd); 1293 ub_cmdtr_state(sc, cmd);
1195 return; 1294 return;
1196 } 1295 }
1296 if (urb->status == -EOVERFLOW) {
1297 /*
1298 * XXX We are screwed here. Retrying is pointless,
1299 * because the pipelined data will not get in until
1300 * we read with a big enough buffer. We must reset XXX.
1301 */
1302 goto Bad_End;
1303 }
1197 if (urb->status != 0) 1304 if (urb->status != 0)
1198 goto Bad_End; 1305 goto Bad_End;
1199 1306
1200 if (urb->actual_length == 0) { 1307 if (urb->actual_length == 0) {
1201 /* 1308 ub_state_stat_counted(sc, cmd);
1202 * Some broken devices add unnecessary zero-length
1203 * packets to the end of their data transfers.
1204 * Such packets show up as 0-length CSWs. If we
1205 * encounter such a thing, try to read the CSW again.
1206 */
1207 if (++cmd->stat_count >= 4) {
1208 printk(KERN_NOTICE "%s: unable to get CSW\n",
1209 sc->name);
1210 goto Bad_End;
1211 }
1212 __ub_state_stat(sc, cmd);
1213 return; 1309 return;
1214 } 1310 }
1215 1311
1216 /* 1312 /*
1217 * Check the returned Bulk protocol status. 1313 * Check the returned Bulk protocol status.
1314 * The status block has to be validated first.
1218 */ 1315 */
1219 1316
1220 bcs = &sc->work_bcs; 1317 bcs = &sc->work_bcs;
1221 rc = le32_to_cpu(bcs->Residue); 1318
1222 if (rc != cmd->len - cmd->act_len) { 1319 if (sc->signature == cpu_to_le32(0)) {
1223 /* 1320 /*
1224 * It is all right to transfer less, the caller has 1321 * This is the first reply, so do not perform the check.
1225 * to check. But it's not all right if the device 1322 * Instead, remember the signature the device uses
1226 * counts disagree with our counts. 1323 * for future checks. But do not allow a nul.
1227 */ 1324 */
1228 /* P3 */ printk("%s: resid %d len %d act %d\n", 1325 sc->signature = bcs->Signature;
1229 sc->name, rc, cmd->len, cmd->act_len); 1326 if (sc->signature == cpu_to_le32(0)) {
1230 goto Bad_End; 1327 ub_state_stat_counted(sc, cmd);
1231 } 1328 return;
1232 1329 }
1233#if 0 1330 } else {
1234 if (bcs->Signature != cpu_to_le32(US_BULK_CS_SIGN) && 1331 if (bcs->Signature != sc->signature) {
1235 bcs->Signature != cpu_to_le32(US_BULK_CS_OLYMPUS_SIGN)) { 1332 ub_state_stat_counted(sc, cmd);
1236 /* Windows ignores signatures, so do we. */ 1333 return;
1334 }
1237 } 1335 }
1238#endif
1239 1336
1240 if (bcs->Tag != cmd->tag) { 1337 if (bcs->Tag != cmd->tag) {
1241 /* 1338 /*
@@ -1245,16 +1342,22 @@ static void ub_scsi_urb_compl(struct ub_dev *sc, struct ub_scsi_cmd *cmd)
1245 * commands and reply at commands we timed out before. 1342 * commands and reply at commands we timed out before.
1246 * Without flushing these replies we loop forever. 1343 * Without flushing these replies we loop forever.
1247 */ 1344 */
1248 if (++cmd->stat_count >= 4) { 1345 ub_state_stat_counted(sc, cmd);
1249 printk(KERN_NOTICE "%s: "
1250 "tag mismatch orig 0x%x reply 0x%x\n",
1251 sc->name, cmd->tag, bcs->Tag);
1252 goto Bad_End;
1253 }
1254 __ub_state_stat(sc, cmd);
1255 return; 1346 return;
1256 } 1347 }
1257 1348
1349 rc = le32_to_cpu(bcs->Residue);
1350 if (rc != cmd->len - cmd->act_len) {
1351 /*
1352 * It is all right to transfer less, the caller has
1353 * to check. But it's not all right if the device
1354 * counts disagree with our counts.
1355 */
1356 /* P3 */ printk("%s: resid %d len %d act %d\n",
1357 sc->name, rc, cmd->len, cmd->act_len);
1358 goto Bad_End;
1359 }
1360
1258 switch (bcs->Status) { 1361 switch (bcs->Status) {
1259 case US_BULK_STAT_OK: 1362 case US_BULK_STAT_OK:
1260 break; 1363 break;
@@ -1272,6 +1375,10 @@ static void ub_scsi_urb_compl(struct ub_dev *sc, struct ub_scsi_cmd *cmd)
1272 } 1375 }
1273 1376
1274 /* Not zeroing error to preserve a babble indicator */ 1377 /* Not zeroing error to preserve a babble indicator */
1378 if (cmd->error != 0) {
1379 ub_state_sense(sc, cmd);
1380 return;
1381 }
1275 cmd->state = UB_CMDST_DONE; 1382 cmd->state = UB_CMDST_DONE;
1276 ub_cmdtr_state(sc, cmd); 1383 ub_cmdtr_state(sc, cmd);
1277 ub_cmdq_pop(sc); 1384 ub_cmdq_pop(sc);
@@ -1310,7 +1417,7 @@ static void ub_state_done(struct ub_dev *sc, struct ub_scsi_cmd *cmd, int rc)
1310 * Factorization helper for the command state machine: 1417 * Factorization helper for the command state machine:
1311 * Submit a CSW read. 1418 * Submit a CSW read.
1312 */ 1419 */
1313static void __ub_state_stat(struct ub_dev *sc, struct ub_scsi_cmd *cmd) 1420static int __ub_state_stat(struct ub_dev *sc, struct ub_scsi_cmd *cmd)
1314{ 1421{
1315 int rc; 1422 int rc;
1316 1423
@@ -1328,11 +1435,12 @@ static void __ub_state_stat(struct ub_dev *sc, struct ub_scsi_cmd *cmd)
1328 /* XXX Clear stalls */ 1435 /* XXX Clear stalls */
1329 ub_complete(&sc->work_done); 1436 ub_complete(&sc->work_done);
1330 ub_state_done(sc, cmd, rc); 1437 ub_state_done(sc, cmd, rc);
1331 return; 1438 return -1;
1332 } 1439 }
1333 1440
1334 sc->work_timer.expires = jiffies + UB_STAT_TIMEOUT; 1441 sc->work_timer.expires = jiffies + UB_STAT_TIMEOUT;
1335 add_timer(&sc->work_timer); 1442 add_timer(&sc->work_timer);
1443 return 0;
1336} 1444}
1337 1445
1338/* 1446/*
@@ -1341,7 +1449,9 @@ static void __ub_state_stat(struct ub_dev *sc, struct ub_scsi_cmd *cmd)
1341 */ 1449 */
1342static void ub_state_stat(struct ub_dev *sc, struct ub_scsi_cmd *cmd) 1450static void ub_state_stat(struct ub_dev *sc, struct ub_scsi_cmd *cmd)
1343{ 1451{
1344 __ub_state_stat(sc, cmd); 1452
1453 if (__ub_state_stat(sc, cmd) != 0)
1454 return;
1345 1455
1346 cmd->stat_count = 0; 1456 cmd->stat_count = 0;
1347 cmd->state = UB_CMDST_STAT; 1457 cmd->state = UB_CMDST_STAT;
@@ -1350,6 +1460,25 @@ static void ub_state_stat(struct ub_dev *sc, struct ub_scsi_cmd *cmd)
1350 1460
1351/* 1461/*
1352 * Factorization helper for the command state machine: 1462 * Factorization helper for the command state machine:
1463 * Submit a CSW read and go to STAT state with counter (along [C] path).
1464 */
1465static void ub_state_stat_counted(struct ub_dev *sc, struct ub_scsi_cmd *cmd)
1466{
1467
1468 if (++cmd->stat_count >= 4) {
1469 ub_state_sense(sc, cmd);
1470 return;
1471 }
1472
1473 if (__ub_state_stat(sc, cmd) != 0)
1474 return;
1475
1476 cmd->state = UB_CMDST_STAT;
1477 ub_cmdtr_state(sc, cmd);
1478}
1479
1480/*
1481 * Factorization helper for the command state machine:
1353 * Submit a REQUEST SENSE and go to SENSE state. 1482 * Submit a REQUEST SENSE and go to SENSE state.
1354 */ 1483 */
1355static void ub_state_sense(struct ub_dev *sc, struct ub_scsi_cmd *cmd) 1484static void ub_state_sense(struct ub_dev *sc, struct ub_scsi_cmd *cmd)
generated by cgit v1.2.2 (git 2.25.0) at 2024-11-23 03:08:50 -0500