xref: /linux/arch/powerpc/kernel/eeh_cache.c (revision 148f9bb87745ed45f7a11b2cbd3bc0f017d5d257)
1 /*
2  * PCI address cache; allows the lookup of PCI devices based on I/O address
3  *
4  * Copyright IBM Corporation 2004
5  * Copyright Linas Vepstas <linas@austin.ibm.com> 2004
6  *
7  * This program is free software; you can redistribute it and/or modify
8  * it under the terms of the GNU General Public License as published by
9  * the Free Software Foundation; either version 2 of the License, or
10  * (at your option) any later version.
11  *
12  * This program is distributed in the hope that it will be useful,
13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
15  * GNU General Public License for more details.
16  *
17  * You should have received a copy of the GNU General Public License
18  * along with this program; if not, write to the Free Software
19  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307 USA
20  */
21 
22 #include <linux/list.h>
23 #include <linux/pci.h>
24 #include <linux/rbtree.h>
25 #include <linux/slab.h>
26 #include <linux/spinlock.h>
27 #include <linux/atomic.h>
28 #include <asm/pci-bridge.h>
29 #include <asm/ppc-pci.h>
30 
31 
32 /**
33  * The pci address cache subsystem.  This subsystem places
34  * PCI device address resources into a red-black tree, sorted
35  * according to the address range, so that given only an i/o
36  * address, the corresponding PCI device can be **quickly**
37  * found. It is safe to perform an address lookup in an interrupt
38  * context; this ability is an important feature.
39  *
40  * Currently, the only customer of this code is the EEH subsystem;
41  * thus, this code has been somewhat tailored to suit EEH better.
42  * In particular, the cache does *not* hold the addresses of devices
43  * for which EEH is not enabled.
44  *
45  * (Implementation Note: The RB tree seems to be better/faster
46  * than any hash algo I could think of for this problem, even
47  * with the penalty of slow pointer chases for d-cache misses).
48  */
49 struct pci_io_addr_range {
50 	struct rb_node rb_node;
51 	unsigned long addr_lo;
52 	unsigned long addr_hi;
53 	struct eeh_dev *edev;
54 	struct pci_dev *pcidev;
55 	unsigned int flags;
56 };
57 
58 static struct pci_io_addr_cache {
59 	struct rb_root rb_root;
60 	spinlock_t piar_lock;
61 } pci_io_addr_cache_root;
62 
63 static inline struct eeh_dev *__eeh_addr_cache_get_device(unsigned long addr)
64 {
65 	struct rb_node *n = pci_io_addr_cache_root.rb_root.rb_node;
66 
67 	while (n) {
68 		struct pci_io_addr_range *piar;
69 		piar = rb_entry(n, struct pci_io_addr_range, rb_node);
70 
71 		if (addr < piar->addr_lo) {
72 			n = n->rb_left;
73 		} else {
74 			if (addr > piar->addr_hi) {
75 				n = n->rb_right;
76 			} else {
77 				pci_dev_get(piar->pcidev);
78 				return piar->edev;
79 			}
80 		}
81 	}
82 
83 	return NULL;
84 }
85 
86 /**
87  * eeh_addr_cache_get_dev - Get device, given only address
88  * @addr: mmio (PIO) phys address or i/o port number
89  *
90  * Given an mmio phys address, or a port number, find a pci device
91  * that implements this address.  Be sure to pci_dev_put the device
92  * when finished.  I/O port numbers are assumed to be offset
93  * from zero (that is, they do *not* have pci_io_addr added in).
94  * It is safe to call this function within an interrupt.
95  */
96 struct eeh_dev *eeh_addr_cache_get_dev(unsigned long addr)
97 {
98 	struct eeh_dev *edev;
99 	unsigned long flags;
100 
101 	spin_lock_irqsave(&pci_io_addr_cache_root.piar_lock, flags);
102 	edev = __eeh_addr_cache_get_device(addr);
103 	spin_unlock_irqrestore(&pci_io_addr_cache_root.piar_lock, flags);
104 	return edev;
105 }
106 
107 #ifdef DEBUG
108 /*
109  * Handy-dandy debug print routine, does nothing more
110  * than print out the contents of our addr cache.
111  */
112 static void eeh_addr_cache_print(struct pci_io_addr_cache *cache)
113 {
114 	struct rb_node *n;
115 	int cnt = 0;
116 
117 	n = rb_first(&cache->rb_root);
118 	while (n) {
119 		struct pci_io_addr_range *piar;
120 		piar = rb_entry(n, struct pci_io_addr_range, rb_node);
121 		pr_debug("PCI: %s addr range %d [%lx-%lx]: %s\n",
122 		       (piar->flags & IORESOURCE_IO) ? "i/o" : "mem", cnt,
123 		       piar->addr_lo, piar->addr_hi, pci_name(piar->pcidev));
124 		cnt++;
125 		n = rb_next(n);
126 	}
127 }
128 #endif
129 
130 /* Insert address range into the rb tree. */
131 static struct pci_io_addr_range *
132 eeh_addr_cache_insert(struct pci_dev *dev, unsigned long alo,
133 		      unsigned long ahi, unsigned int flags)
134 {
135 	struct rb_node **p = &pci_io_addr_cache_root.rb_root.rb_node;
136 	struct rb_node *parent = NULL;
137 	struct pci_io_addr_range *piar;
138 
139 	/* Walk tree, find a place to insert into tree */
140 	while (*p) {
141 		parent = *p;
142 		piar = rb_entry(parent, struct pci_io_addr_range, rb_node);
143 		if (ahi < piar->addr_lo) {
144 			p = &parent->rb_left;
145 		} else if (alo > piar->addr_hi) {
146 			p = &parent->rb_right;
147 		} else {
148 			if (dev != piar->pcidev ||
149 			    alo != piar->addr_lo || ahi != piar->addr_hi) {
150 				pr_warning("PIAR: overlapping address range\n");
151 			}
152 			return piar;
153 		}
154 	}
155 	piar = kzalloc(sizeof(struct pci_io_addr_range), GFP_ATOMIC);
156 	if (!piar)
157 		return NULL;
158 
159 	pci_dev_get(dev);
160 	piar->addr_lo = alo;
161 	piar->addr_hi = ahi;
162 	piar->edev = pci_dev_to_eeh_dev(dev);
163 	piar->pcidev = dev;
164 	piar->flags = flags;
165 
166 #ifdef DEBUG
167 	pr_debug("PIAR: insert range=[%lx:%lx] dev=%s\n",
168 	                  alo, ahi, pci_name(dev));
169 #endif
170 
171 	rb_link_node(&piar->rb_node, parent, p);
172 	rb_insert_color(&piar->rb_node, &pci_io_addr_cache_root.rb_root);
173 
174 	return piar;
175 }
176 
177 static void __eeh_addr_cache_insert_dev(struct pci_dev *dev)
178 {
179 	struct device_node *dn;
180 	struct eeh_dev *edev;
181 	int i;
182 
183 	dn = pci_device_to_OF_node(dev);
184 	if (!dn) {
185 		pr_warning("PCI: no pci dn found for dev=%s\n", pci_name(dev));
186 		return;
187 	}
188 
189 	edev = of_node_to_eeh_dev(dn);
190 	if (!edev) {
191 		pr_warning("PCI: no EEH dev found for dn=%s\n",
192 			dn->full_name);
193 		return;
194 	}
195 
196 	/* Skip any devices for which EEH is not enabled. */
197 	if (!eeh_probe_mode_dev() && !edev->pe) {
198 #ifdef DEBUG
199 		pr_info("PCI: skip building address cache for=%s - %s\n",
200 			pci_name(dev), dn->full_name);
201 #endif
202 		return;
203 	}
204 
205 	/* Walk resources on this device, poke them into the tree */
206 	for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
207 		unsigned long start = pci_resource_start(dev,i);
208 		unsigned long end = pci_resource_end(dev,i);
209 		unsigned int flags = pci_resource_flags(dev,i);
210 
211 		/* We are interested only bus addresses, not dma or other stuff */
212 		if (0 == (flags & (IORESOURCE_IO | IORESOURCE_MEM)))
213 			continue;
214 		if (start == 0 || ~start == 0 || end == 0 || ~end == 0)
215 			 continue;
216 		eeh_addr_cache_insert(dev, start, end, flags);
217 	}
218 }
219 
220 /**
221  * eeh_addr_cache_insert_dev - Add a device to the address cache
222  * @dev: PCI device whose I/O addresses we are interested in.
223  *
224  * In order to support the fast lookup of devices based on addresses,
225  * we maintain a cache of devices that can be quickly searched.
226  * This routine adds a device to that cache.
227  */
228 void eeh_addr_cache_insert_dev(struct pci_dev *dev)
229 {
230 	unsigned long flags;
231 
232 	/* Ignore PCI bridges */
233 	if ((dev->class >> 16) == PCI_BASE_CLASS_BRIDGE)
234 		return;
235 
236 	spin_lock_irqsave(&pci_io_addr_cache_root.piar_lock, flags);
237 	__eeh_addr_cache_insert_dev(dev);
238 	spin_unlock_irqrestore(&pci_io_addr_cache_root.piar_lock, flags);
239 }
240 
241 static inline void __eeh_addr_cache_rmv_dev(struct pci_dev *dev)
242 {
243 	struct rb_node *n;
244 
245 restart:
246 	n = rb_first(&pci_io_addr_cache_root.rb_root);
247 	while (n) {
248 		struct pci_io_addr_range *piar;
249 		piar = rb_entry(n, struct pci_io_addr_range, rb_node);
250 
251 		if (piar->pcidev == dev) {
252 			rb_erase(n, &pci_io_addr_cache_root.rb_root);
253 			pci_dev_put(piar->pcidev);
254 			kfree(piar);
255 			goto restart;
256 		}
257 		n = rb_next(n);
258 	}
259 }
260 
261 /**
262  * eeh_addr_cache_rmv_dev - remove pci device from addr cache
263  * @dev: device to remove
264  *
265  * Remove a device from the addr-cache tree.
266  * This is potentially expensive, since it will walk
267  * the tree multiple times (once per resource).
268  * But so what; device removal doesn't need to be that fast.
269  */
270 void eeh_addr_cache_rmv_dev(struct pci_dev *dev)
271 {
272 	unsigned long flags;
273 
274 	spin_lock_irqsave(&pci_io_addr_cache_root.piar_lock, flags);
275 	__eeh_addr_cache_rmv_dev(dev);
276 	spin_unlock_irqrestore(&pci_io_addr_cache_root.piar_lock, flags);
277 }
278 
279 /**
280  * eeh_addr_cache_build - Build a cache of I/O addresses
281  *
282  * Build a cache of pci i/o addresses.  This cache will be used to
283  * find the pci device that corresponds to a given address.
284  * This routine scans all pci busses to build the cache.
285  * Must be run late in boot process, after the pci controllers
286  * have been scanned for devices (after all device resources are known).
287  */
288 void eeh_addr_cache_build(void)
289 {
290 	struct device_node *dn;
291 	struct eeh_dev *edev;
292 	struct pci_dev *dev = NULL;
293 
294 	spin_lock_init(&pci_io_addr_cache_root.piar_lock);
295 
296 	for_each_pci_dev(dev) {
297 		dn = pci_device_to_OF_node(dev);
298 		if (!dn)
299 			continue;
300 
301 		edev = of_node_to_eeh_dev(dn);
302 		if (!edev)
303 			continue;
304 
305 		pci_dev_get(dev);  /* matching put is in eeh_remove_device() */
306 		dev->dev.archdata.edev = edev;
307 		edev->pdev = dev;
308 
309 		eeh_addr_cache_insert_dev(dev);
310 
311 		eeh_sysfs_add_device(dev);
312 	}
313 
314 #ifdef DEBUG
315 	/* Verify tree built up above, echo back the list of addrs. */
316 	eeh_addr_cache_print(&pci_io_addr_cache_root);
317 #endif
318 }
319