| 1 | |
| 2 | |
| 3 | * PCI address cache; allows the lookup of PCI devices based on I/O address
|
| 4 | *
|
| 5 | * Copyright IBM Corporation 2004
|
| 6 | * Copyright Linas Vepstas <[email protected]> 2004
|
| 7 |
|
| 8 |
|
| 9 | #include <linux/list.h>
|
| 10 | #include <linux/pci.h>
|
| 11 | #include <linux/rbtree.h>
|
| 12 | #include <linux/slab.h>
|
| 13 | #include <linux/spinlock.h>
|
| 14 | #include <linux/atomic.h>
|
| 15 | #include <linux/debugfs.h>
|
| 16 | #include <asm/pci-bridge.h>
|
| 17 | #include <asm/ppc-pci.h>
|
| 18 |
|
| 19 |
|
| 20 | |
| 21 | * DOC: Overview
|
| 22 | *
|
| 23 | * The pci address cache subsystem. This subsystem places
|
| 24 | * PCI device address resources into a red-black tree, sorted
|
| 25 | * according to the address range, so that given only an i/o
|
| 26 | * address, the corresponding PCI device can be **quickly**
|
| 27 | * found. It is safe to perform an address lookup in an interrupt
|
| 28 | * context; this ability is an important feature.
|
| 29 | *
|
| 30 | * Currently, the only customer of this code is the EEH subsystem;
|
| 31 | * thus, this code has been somewhat tailored to suit EEH better.
|
| 32 | * In particular, the cache does *not* hold the addresses of devices
|
| 33 | * for which EEH is not enabled.
|
| 34 | *
|
| 35 | * (Implementation Note: The RB tree seems to be better/faster
|
| 36 | * than any hash algo I could think of for this problem, even
|
| 37 | * with the penalty of slow pointer chases for d-cache misses).
|
| 38 |
|
| 39 |
|
| 40 | struct pci_io_addr_range {
|
| 41 | struct rb_node rb_node;
|
| 42 | resource_size_t addr_lo;
|
| 43 | resource_size_t addr_hi;
|
| 44 | struct eeh_dev *edev;
|
| 45 | struct pci_dev *pcidev;
|
| 46 | unsigned long flags;
|
| 47 | };
|
| 48 |
|
| 49 | static struct pci_io_addr_cache {
|
| 50 | struct rb_root rb_root;
|
| 51 | spinlock_t piar_lock;
|
| 52 | } pci_io_addr_cache_root;
|
| 53 |
|
| 54 | static inline struct eeh_dev *__eeh_addr_cache_get_device(unsigned long addr)
|
| 55 | {
|
| 56 | struct rb_node *n = pci_io_addr_cache_root.rb_root.rb_node;
|
| 57 |
|
| 58 | while (n) {
|
| 59 | struct pci_io_addr_range *piar;
|
| 60 | piar = rb_entry(n, struct pci_io_addr_range, rb_node);
|
| 61 |
|
| 62 | if (addr < piar->addr_lo)
|
| 63 | n = n->rb_left;
|
| 64 | else if (addr > piar->addr_hi)
|
| 65 | n = n->rb_right;
|
| 66 | else
|
| 67 | return piar->edev;
|
| 68 | }
|
| 69 |
|
| 70 | return NULL;
|
| 71 | }
|
| 72 |
|
| 73 | |
| 74 | * eeh_addr_cache_get_dev - Get device, given only address
|
| 75 | * @addr: mmio (PIO) phys address or i/o port number
|
| 76 | *
|
| 77 | * Given an mmio phys address, or a port number, find a pci device
|
| 78 | * that implements this address. I/O port numbers are assumed to be offset
|
| 79 | * from zero (that is, they do *not* have pci_io_addr added in).
|
| 80 | * It is safe to call this function within an interrupt.
|
| 81 |
|
| 82 | struct eeh_dev *eeh_addr_cache_get_dev(unsigned long addr)
|
| 83 | {
|
| 84 | struct eeh_dev *edev;
|
| 85 | unsigned long flags;
|
| 86 |
|
| 87 | spin_lock_irqsave(&pci_io_addr_cache_root.piar_lock, flags);
|
| 88 | edev = __eeh_addr_cache_get_device(addr);
|
| 89 | spin_unlock_irqrestore(&pci_io_addr_cache_root.piar_lock, flags);
|
| 90 | return edev;
|
| 91 | }
|
| 92 |
|
| 93 | #ifdef DEBUG
|
| 94 | |
| 95 | * Handy-dandy debug print routine, does nothing more
|
| 96 | * than print out the contents of our addr cache.
|
| 97 |
|
| 98 | static void eeh_addr_cache_print(struct pci_io_addr_cache *cache)
|
| 99 | {
|
| 100 | struct rb_node *n;
|
| 101 | int cnt = 0;
|
| 102 |
|
| 103 | n = rb_first(&cache->rb_root);
|
| 104 | while (n) {
|
| 105 | struct pci_io_addr_range *piar;
|
| 106 | piar = rb_entry(n, struct pci_io_addr_range, rb_node);
|
| 107 | pr_info("PCI: %s addr range %d [%pap-%pap]: %s\n",
|
| 108 | (piar->flags & IORESOURCE_IO) ? "i/o" : "mem", cnt,
|
| 109 | &piar->addr_lo, &piar->addr_hi, pci_name(piar->pcidev));
|
| 110 | cnt++;
|
| 111 | n = rb_next(n);
|
| 112 | }
|
| 113 | }
|
| 114 | #endif
|
| 115 |
|
| 116 |
|
| 117 | static struct pci_io_addr_range *
|
| 118 | eeh_addr_cache_insert(struct pci_dev *dev, resource_size_t alo,
|
| 119 | resource_size_t ahi, unsigned long flags)
|
| 120 | {
|
| 121 | struct rb_node **p = &pci_io_addr_cache_root.rb_root.rb_node;
|
| 122 | struct rb_node *parent = NULL;
|
| 123 | struct pci_io_addr_range *piar;
|
| 124 |
|
| 125 |
|
| 126 | while (*p) {
|
| 127 | parent = *p;
|
| 128 | piar = rb_entry(parent, struct pci_io_addr_range, rb_node);
|
| 129 | if (ahi < piar->addr_lo) {
|
| 130 | p = &parent->rb_left;
|
| 131 | } else if (alo > piar->addr_hi) {
|
| 132 | p = &parent->rb_right;
|
| 133 | } else {
|
| 134 | if (dev != piar->pcidev ||
|
| 135 | alo != piar->addr_lo || ahi != piar->addr_hi) {
|
| 136 | pr_warn("PIAR: overlapping address range\n");
|
| 137 | }
|
| 138 | return piar;
|
| 139 | }
|
| 140 | }
|
| 141 | piar = kzalloc_obj(struct pci_io_addr_range, GFP_ATOMIC);
|
| 142 | if (!piar)
|
| 143 | return NULL;
|
| 144 |
|
| 145 | piar->addr_lo = alo;
|
| 146 | piar->addr_hi = ahi;
|
| 147 | piar->edev = pci_dev_to_eeh_dev(dev);
|
| 148 | piar->pcidev = dev;
|
| 149 | piar->flags = flags;
|
| 150 |
|
| 151 | eeh_edev_dbg(piar->edev, "PIAR: insert range=[%pap:%pap]\n",
|
| 152 | &alo, &ahi);
|
| 153 |
|
| 154 | rb_link_node(&piar->rb_node, parent, p);
|
| 155 | rb_insert_color(&piar->rb_node, &pci_io_addr_cache_root.rb_root);
|
| 156 |
|
| 157 | return piar;
|
| 158 | }
|
| 159 |
|
| 160 | static void __eeh_addr_cache_insert_dev(struct pci_dev *dev)
|
| 161 | {
|
| 162 | struct eeh_dev *edev;
|
| 163 | int i;
|
| 164 |
|
| 165 | edev = pci_dev_to_eeh_dev(dev);
|
| 166 | if (!edev) {
|
| 167 | pr_warn("PCI: no EEH dev found for %s\n",
|
| 168 | pci_name(dev));
|
| 169 | return;
|
| 170 | }
|
| 171 |
|
| 172 |
|
| 173 | if (!edev->pe) {
|
| 174 | dev_dbg(&dev->dev, "EEH: Skip building address cache\n");
|
| 175 | return;
|
| 176 | }
|
| 177 |
|
| 178 | |
| 179 | * Walk resources on this device, poke the first 7 (6 normal BAR and 1
|
| 180 | * ROM BAR) into the tree.
|
| 181 |
|
| 182 | for (i = 0; i <= PCI_ROM_RESOURCE; i++) {
|
| 183 | resource_size_t start = pci_resource_start(dev,i);
|
| 184 | resource_size_t end = pci_resource_end(dev,i);
|
| 185 | unsigned long flags = pci_resource_flags(dev,i);
|
| 186 |
|
| 187 |
|
| 188 | if (0 == (flags & (IORESOURCE_IO | IORESOURCE_MEM)))
|
| 189 | continue;
|
| 190 | if (start == 0 || ~start == 0 || end == 0 || ~end == 0)
|
| 191 | continue;
|
| 192 | eeh_addr_cache_insert(dev, start, end, flags);
|
| 193 | }
|
| 194 | }
|
| 195 |
|
| 196 | |
| 197 | * eeh_addr_cache_insert_dev - Add a device to the address cache
|
| 198 | * @dev: PCI device whose I/O addresses we are interested in.
|
| 199 | *
|
| 200 | * In order to support the fast lookup of devices based on addresses,
|
| 201 | * we maintain a cache of devices that can be quickly searched.
|
| 202 | * This routine adds a device to that cache.
|
| 203 |
|
| 204 | void eeh_addr_cache_insert_dev(struct pci_dev *dev)
|
| 205 | {
|
| 206 | unsigned long flags;
|
| 207 |
|
| 208 | spin_lock_irqsave(&pci_io_addr_cache_root.piar_lock, flags);
|
| 209 | __eeh_addr_cache_insert_dev(dev);
|
| 210 | spin_unlock_irqrestore(&pci_io_addr_cache_root.piar_lock, flags);
|
| 211 | }
|
| 212 |
|
| 213 | static inline void __eeh_addr_cache_rmv_dev(struct pci_dev *dev)
|
| 214 | {
|
| 215 | struct rb_node *n;
|
| 216 |
|
| 217 | restart:
|
| 218 | n = rb_first(&pci_io_addr_cache_root.rb_root);
|
| 219 | while (n) {
|
| 220 | struct pci_io_addr_range *piar;
|
| 221 | piar = rb_entry(n, struct pci_io_addr_range, rb_node);
|
| 222 |
|
| 223 | if (piar->pcidev == dev) {
|
| 224 | eeh_edev_dbg(piar->edev, "PIAR: remove range=[%pap:%pap]\n",
|
| 225 | &piar->addr_lo, &piar->addr_hi);
|
| 226 | rb_erase(n, &pci_io_addr_cache_root.rb_root);
|
| 227 | kfree(piar);
|
| 228 | goto restart;
|
| 229 | }
|
| 230 | n = rb_next(n);
|
| 231 | }
|
| 232 | }
|
| 233 |
|
| 234 | |
| 235 | * eeh_addr_cache_rmv_dev - remove pci device from addr cache
|
| 236 | * @dev: device to remove
|
| 237 | *
|
| 238 | * Remove a device from the addr-cache tree.
|
| 239 | * This is potentially expensive, since it will walk
|
| 240 | * the tree multiple times (once per resource).
|
| 241 | * But so what; device removal doesn't need to be that fast.
|
| 242 |
|
| 243 | void eeh_addr_cache_rmv_dev(struct pci_dev *dev)
|
| 244 | {
|
| 245 | unsigned long flags;
|
| 246 |
|
| 247 | spin_lock_irqsave(&pci_io_addr_cache_root.piar_lock, flags);
|
| 248 | __eeh_addr_cache_rmv_dev(dev);
|
| 249 | spin_unlock_irqrestore(&pci_io_addr_cache_root.piar_lock, flags);
|
| 250 | }
|
| 251 |
|
| 252 | |
| 253 | * eeh_addr_cache_init - Initialize a cache of I/O addresses
|
| 254 | *
|
| 255 | * Initialize a cache of pci i/o addresses. This cache will be used to
|
| 256 | * find the pci device that corresponds to a given address.
|
| 257 |
|
| 258 | void eeh_addr_cache_init(void)
|
| 259 | {
|
| 260 | spin_lock_init(&pci_io_addr_cache_root.piar_lock);
|
| 261 | }
|
| 262 |
|
| 263 | static int eeh_addr_cache_show(struct seq_file *s, void *v)
|
| 264 | {
|
| 265 | struct pci_io_addr_range *piar;
|
| 266 | struct rb_node *n;
|
| 267 | unsigned long flags;
|
| 268 |
|
| 269 | spin_lock_irqsave(&pci_io_addr_cache_root.piar_lock, flags);
|
| 270 | for (n = rb_first(&pci_io_addr_cache_root.rb_root); n; n = rb_next(n)) {
|
| 271 | piar = rb_entry(n, struct pci_io_addr_range, rb_node);
|
| 272 |
|
| 273 | seq_printf(s, "%s addr range [%pap-%pap]: %s\n",
|
| 274 | (piar->flags & IORESOURCE_IO) ? "i/o" : "mem",
|
| 275 | &piar->addr_lo, &piar->addr_hi, pci_name(piar->pcidev));
|
| 276 | }
|
| 277 | spin_unlock_irqrestore(&pci_io_addr_cache_root.piar_lock, flags);
|
| 278 |
|
| 279 | return 0;
|
| 280 | }
|
| 281 | DEFINE_SHOW_ATTRIBUTE(eeh_addr_cache);
|
| 282 |
|
| 283 | void __init eeh_cache_debugfs_init(void)
|
| 284 | {
|
| 285 | debugfs_create_file_unsafe("eeh_address_cache", 0400,
|
| 286 | arch_debugfs_dir, NULL,
|
| 287 | &eeh_addr_cache_fops);
|
| 288 | }
|