1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * PCI Backend Operations - respond to PCI requests from Frontend
4 *
5 * Author: Ryan Wilson <hap9@epoch.ncsc.mil>
6 */
7
8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
9
10 #include <linux/moduleparam.h>
11 #include <linux/wait.h>
12 #include <linux/bitops.h>
13 #include <xen/events.h>
14 #include <linux/sched.h>
15 #include "pciback.h"
16
17 int verbose_request;
18 module_param(verbose_request, int, 0644);
19
20 static irqreturn_t xen_pcibk_guest_interrupt(int irq, void *dev_id);
21
22 /* Ensure a device is has the fake IRQ handler "turned on/off" and is
23 * ready to be exported. This MUST be run after xen_pcibk_reset_device
24 * which does the actual PCI device enable/disable.
25 */
xen_pcibk_control_isr(struct pci_dev * dev,int reset)26 static void xen_pcibk_control_isr(struct pci_dev *dev, int reset)
27 {
28 struct xen_pcibk_dev_data *dev_data;
29 int rc;
30 int enable = 0;
31
32 dev_data = pci_get_drvdata(dev);
33 if (!dev_data)
34 return;
35
36 /* We don't deal with bridges */
37 if (dev->hdr_type != PCI_HEADER_TYPE_NORMAL)
38 return;
39
40 if (reset) {
41 dev_data->enable_intx = 0;
42 dev_data->ack_intr = 0;
43 }
44 enable = dev_data->enable_intx;
45
46 /* Asked to disable, but ISR isn't runnig */
47 if (!enable && !dev_data->isr_on)
48 return;
49
50 /* Squirrel away the IRQs in the dev_data. We need this
51 * b/c when device transitions to MSI, the dev->irq is
52 * overwritten with the MSI vector.
53 */
54 if (enable)
55 dev_data->irq = dev->irq;
56
57 /*
58 * SR-IOV devices in all use MSI-X and have no legacy
59 * interrupts, so inhibit creating a fake IRQ handler for them.
60 */
61 if (dev_data->irq == 0)
62 goto out;
63
64 dev_dbg(&dev->dev, "%s: #%d %s %s%s %s-> %s\n",
65 dev_data->irq_name,
66 dev_data->irq,
67 pci_is_enabled(dev) ? "on" : "off",
68 dev->msi_enabled ? "MSI" : "",
69 dev->msix_enabled ? "MSI/X" : "",
70 dev_data->isr_on ? "enable" : "disable",
71 enable ? "enable" : "disable");
72
73 if (enable) {
74 /*
75 * The MSI or MSI-X should not have an IRQ handler. Otherwise
76 * if the guest terminates we BUG_ON in free_msi_irqs.
77 */
78 if (dev->msi_enabled || dev->msix_enabled)
79 goto out;
80
81 rc = request_irq(dev_data->irq,
82 xen_pcibk_guest_interrupt, IRQF_SHARED,
83 dev_data->irq_name, dev);
84 if (rc) {
85 dev_err(&dev->dev, "%s: failed to install fake IRQ " \
86 "handler for IRQ %d! (rc:%d)\n",
87 dev_data->irq_name, dev_data->irq, rc);
88 goto out;
89 }
90 } else {
91 free_irq(dev_data->irq, dev);
92 dev_data->irq = 0;
93 }
94 dev_data->isr_on = enable;
95 dev_data->ack_intr = enable;
96 out:
97 dev_dbg(&dev->dev, "%s: #%d %s %s%s %s\n",
98 dev_data->irq_name,
99 dev_data->irq,
100 pci_is_enabled(dev) ? "on" : "off",
101 dev->msi_enabled ? "MSI" : "",
102 dev->msix_enabled ? "MSI/X" : "",
103 enable ? (dev_data->isr_on ? "enabled" : "failed to enable") :
104 (dev_data->isr_on ? "failed to disable" : "disabled"));
105 }
106
107 /* Ensure a device is "turned off" and ready to be exported.
108 * (Also see xen_pcibk_config_reset to ensure virtual configuration space is
109 * ready to be re-exported)
110 */
xen_pcibk_reset_device(struct pci_dev * dev)111 void xen_pcibk_reset_device(struct pci_dev *dev)
112 {
113 u16 cmd;
114
115 xen_pcibk_control_isr(dev, 1 /* reset device */);
116
117 /* Disable devices (but not bridges) */
118 if (dev->hdr_type == PCI_HEADER_TYPE_NORMAL) {
119 #ifdef CONFIG_PCI_MSI
120 /* The guest could have been abruptly killed without
121 * disabling MSI/MSI-X interrupts.*/
122 if (dev->msix_enabled)
123 pci_disable_msix(dev);
124 if (dev->msi_enabled)
125 pci_disable_msi(dev);
126 #endif
127 if (pci_is_enabled(dev))
128 pci_disable_device(dev);
129
130 dev->is_busmaster = 0;
131 } else {
132 pci_read_config_word(dev, PCI_COMMAND, &cmd);
133 if (cmd & (PCI_COMMAND_INVALIDATE)) {
134 cmd &= ~(PCI_COMMAND_INVALIDATE);
135 pci_write_config_word(dev, PCI_COMMAND, cmd);
136
137 dev->is_busmaster = 0;
138 }
139 }
140 }
141
142 #ifdef CONFIG_PCI_MSI
143 static
xen_pcibk_enable_msi(struct xen_pcibk_device * pdev,struct pci_dev * dev,struct xen_pci_op * op)144 int xen_pcibk_enable_msi(struct xen_pcibk_device *pdev,
145 struct pci_dev *dev, struct xen_pci_op *op)
146 {
147 struct xen_pcibk_dev_data *dev_data;
148 int status;
149
150 if (unlikely(verbose_request))
151 printk(KERN_DEBUG DRV_NAME ": %s: enable MSI\n", pci_name(dev));
152
153 if (dev->msi_enabled)
154 status = -EALREADY;
155 else if (dev->msix_enabled)
156 status = -ENXIO;
157 else
158 status = pci_enable_msi(dev);
159
160 if (status) {
161 pr_warn_ratelimited("%s: error enabling MSI for guest %u: err %d\n",
162 pci_name(dev), pdev->xdev->otherend_id,
163 status);
164 op->value = 0;
165 return XEN_PCI_ERR_op_failed;
166 }
167
168 /* The value the guest needs is actually the IDT vector, not the
169 * the local domain's IRQ number. */
170
171 op->value = dev->irq ? xen_pirq_from_irq(dev->irq) : 0;
172 if (unlikely(verbose_request))
173 printk(KERN_DEBUG DRV_NAME ": %s: MSI: %d\n", pci_name(dev),
174 op->value);
175
176 dev_data = pci_get_drvdata(dev);
177 if (dev_data)
178 dev_data->ack_intr = 0;
179
180 return 0;
181 }
182
183 static
xen_pcibk_disable_msi(struct xen_pcibk_device * pdev,struct pci_dev * dev,struct xen_pci_op * op)184 int xen_pcibk_disable_msi(struct xen_pcibk_device *pdev,
185 struct pci_dev *dev, struct xen_pci_op *op)
186 {
187 if (unlikely(verbose_request))
188 printk(KERN_DEBUG DRV_NAME ": %s: disable MSI\n",
189 pci_name(dev));
190
191 if (dev->msi_enabled) {
192 struct xen_pcibk_dev_data *dev_data;
193
194 pci_disable_msi(dev);
195
196 dev_data = pci_get_drvdata(dev);
197 if (dev_data)
198 dev_data->ack_intr = 1;
199 }
200 op->value = dev->irq ? xen_pirq_from_irq(dev->irq) : 0;
201 if (unlikely(verbose_request))
202 printk(KERN_DEBUG DRV_NAME ": %s: MSI: %d\n", pci_name(dev),
203 op->value);
204 return 0;
205 }
206
207 static
xen_pcibk_enable_msix(struct xen_pcibk_device * pdev,struct pci_dev * dev,struct xen_pci_op * op)208 int xen_pcibk_enable_msix(struct xen_pcibk_device *pdev,
209 struct pci_dev *dev, struct xen_pci_op *op)
210 {
211 struct xen_pcibk_dev_data *dev_data;
212 int i, result;
213 struct msix_entry *entries;
214 u16 cmd;
215
216 if (unlikely(verbose_request))
217 printk(KERN_DEBUG DRV_NAME ": %s: enable MSI-X\n",
218 pci_name(dev));
219
220 if (op->value > SH_INFO_MAX_VEC)
221 return -EINVAL;
222
223 if (dev->msix_enabled)
224 return -EALREADY;
225
226 /*
227 * PCI_COMMAND_MEMORY must be enabled, otherwise we may not be able
228 * to access the BARs where the MSI-X entries reside.
229 * But VF devices are unique in which the PF needs to be checked.
230 */
231 pci_read_config_word(pci_physfn(dev), PCI_COMMAND, &cmd);
232 if (dev->msi_enabled || !(cmd & PCI_COMMAND_MEMORY))
233 return -ENXIO;
234
235 entries = kmalloc_array(op->value, sizeof(*entries), GFP_KERNEL);
236 if (entries == NULL)
237 return -ENOMEM;
238
239 for (i = 0; i < op->value; i++) {
240 entries[i].entry = op->msix_entries[i].entry;
241 entries[i].vector = op->msix_entries[i].vector;
242 }
243
244 result = pci_enable_msix_exact(dev, entries, op->value);
245 if (result == 0) {
246 for (i = 0; i < op->value; i++) {
247 op->msix_entries[i].entry = entries[i].entry;
248 if (entries[i].vector) {
249 op->msix_entries[i].vector =
250 xen_pirq_from_irq(entries[i].vector);
251 if (unlikely(verbose_request))
252 printk(KERN_DEBUG DRV_NAME ": %s: " \
253 "MSI-X[%d]: %d\n",
254 pci_name(dev), i,
255 op->msix_entries[i].vector);
256 }
257 }
258 } else
259 pr_warn_ratelimited("%s: error enabling MSI-X for guest %u: err %d!\n",
260 pci_name(dev), pdev->xdev->otherend_id,
261 result);
262 kfree(entries);
263
264 op->value = result;
265 dev_data = pci_get_drvdata(dev);
266 if (dev_data)
267 dev_data->ack_intr = 0;
268
269 return result > 0 ? 0 : result;
270 }
271
272 static
xen_pcibk_disable_msix(struct xen_pcibk_device * pdev,struct pci_dev * dev,struct xen_pci_op * op)273 int xen_pcibk_disable_msix(struct xen_pcibk_device *pdev,
274 struct pci_dev *dev, struct xen_pci_op *op)
275 {
276 if (unlikely(verbose_request))
277 printk(KERN_DEBUG DRV_NAME ": %s: disable MSI-X\n",
278 pci_name(dev));
279
280 if (dev->msix_enabled) {
281 struct xen_pcibk_dev_data *dev_data;
282
283 pci_disable_msix(dev);
284
285 dev_data = pci_get_drvdata(dev);
286 if (dev_data)
287 dev_data->ack_intr = 1;
288 }
289 /*
290 * SR-IOV devices (which don't have any legacy IRQ) have
291 * an undefined IRQ value of zero.
292 */
293 op->value = dev->irq ? xen_pirq_from_irq(dev->irq) : 0;
294 if (unlikely(verbose_request))
295 printk(KERN_DEBUG DRV_NAME ": %s: MSI-X: %d\n",
296 pci_name(dev), op->value);
297 return 0;
298 }
299 #endif
300
xen_pcibk_test_op_pending(struct xen_pcibk_device * pdev)301 static inline bool xen_pcibk_test_op_pending(struct xen_pcibk_device *pdev)
302 {
303 return test_bit(_XEN_PCIF_active,
304 (unsigned long *)&pdev->sh_info->flags) &&
305 !test_and_set_bit(_PDEVF_op_active, &pdev->flags);
306 }
307
308 /*
309 * Now the same evtchn is used for both pcifront conf_read_write request
310 * as well as pcie aer front end ack. We use a new work_queue to schedule
311 * xen_pcibk conf_read_write service for avoiding confict with aer_core
312 * do_recovery job which also use the system default work_queue
313 */
xen_pcibk_test_and_schedule_op(struct xen_pcibk_device * pdev)314 static void xen_pcibk_test_and_schedule_op(struct xen_pcibk_device *pdev)
315 {
316 bool eoi = true;
317
318 /* Check that frontend is requesting an operation and that we are not
319 * already processing a request */
320 if (xen_pcibk_test_op_pending(pdev)) {
321 schedule_work(&pdev->op_work);
322 eoi = false;
323 }
324 /*_XEN_PCIB_active should have been cleared by pcifront. And also make
325 sure xen_pcibk is waiting for ack by checking _PCIB_op_pending*/
326 if (!test_bit(_XEN_PCIB_active, (unsigned long *)&pdev->sh_info->flags)
327 && test_bit(_PCIB_op_pending, &pdev->flags)) {
328 wake_up(&xen_pcibk_aer_wait_queue);
329 eoi = false;
330 }
331
332 /* EOI if there was nothing to do. */
333 if (eoi)
334 xen_pcibk_lateeoi(pdev, XEN_EOI_FLAG_SPURIOUS);
335 }
336
337 /* Performing the configuration space reads/writes must not be done in atomic
338 * context because some of the pci_* functions can sleep (mostly due to ACPI
339 * use of semaphores). This function is intended to be called from a work
340 * queue in process context taking a struct xen_pcibk_device as a parameter */
341
xen_pcibk_do_one_op(struct xen_pcibk_device * pdev)342 static void xen_pcibk_do_one_op(struct xen_pcibk_device *pdev)
343 {
344 struct pci_dev *dev;
345 struct xen_pcibk_dev_data *dev_data = NULL;
346 struct xen_pci_op *op = &pdev->op;
347 int test_intx = 0;
348 #ifdef CONFIG_PCI_MSI
349 unsigned int nr = 0;
350 #endif
351
352 *op = pdev->sh_info->op;
353 barrier();
354 dev = xen_pcibk_get_pci_dev(pdev, op->domain, op->bus, op->devfn);
355
356 if (dev == NULL)
357 op->err = XEN_PCI_ERR_dev_not_found;
358 else {
359 dev_data = pci_get_drvdata(dev);
360 if (dev_data)
361 test_intx = dev_data->enable_intx;
362 switch (op->cmd) {
363 case XEN_PCI_OP_conf_read:
364 op->err = xen_pcibk_config_read(dev,
365 op->offset, op->size, &op->value);
366 break;
367 case XEN_PCI_OP_conf_write:
368 op->err = xen_pcibk_config_write(dev,
369 op->offset, op->size, op->value);
370 break;
371 #ifdef CONFIG_PCI_MSI
372 case XEN_PCI_OP_enable_msi:
373 op->err = xen_pcibk_enable_msi(pdev, dev, op);
374 break;
375 case XEN_PCI_OP_disable_msi:
376 op->err = xen_pcibk_disable_msi(pdev, dev, op);
377 break;
378 case XEN_PCI_OP_enable_msix:
379 nr = op->value;
380 op->err = xen_pcibk_enable_msix(pdev, dev, op);
381 break;
382 case XEN_PCI_OP_disable_msix:
383 op->err = xen_pcibk_disable_msix(pdev, dev, op);
384 break;
385 #endif
386 default:
387 op->err = XEN_PCI_ERR_not_implemented;
388 break;
389 }
390 }
391 if (!op->err && dev && dev_data) {
392 /* Transition detected */
393 if ((dev_data->enable_intx != test_intx))
394 xen_pcibk_control_isr(dev, 0 /* no reset */);
395 }
396 pdev->sh_info->op.err = op->err;
397 pdev->sh_info->op.value = op->value;
398 #ifdef CONFIG_PCI_MSI
399 if (op->cmd == XEN_PCI_OP_enable_msix && op->err == 0) {
400 unsigned int i;
401
402 for (i = 0; i < nr; i++)
403 pdev->sh_info->op.msix_entries[i].vector =
404 op->msix_entries[i].vector;
405 }
406 #endif
407 /* Tell the driver domain that we're done. */
408 wmb();
409 clear_bit(_XEN_PCIF_active, (unsigned long *)&pdev->sh_info->flags);
410 notify_remote_via_irq(pdev->evtchn_irq);
411
412 /* Mark that we're done. */
413 smp_mb__before_atomic(); /* /after/ clearing PCIF_active */
414 clear_bit(_PDEVF_op_active, &pdev->flags);
415 smp_mb__after_atomic(); /* /before/ final check for work */
416 }
417
xen_pcibk_do_op(struct work_struct * data)418 void xen_pcibk_do_op(struct work_struct *data)
419 {
420 struct xen_pcibk_device *pdev =
421 container_of(data, struct xen_pcibk_device, op_work);
422
423 do {
424 xen_pcibk_do_one_op(pdev);
425 } while (xen_pcibk_test_op_pending(pdev));
426
427 xen_pcibk_lateeoi(pdev, 0);
428 }
429
xen_pcibk_handle_event(int irq,void * dev_id)430 irqreturn_t xen_pcibk_handle_event(int irq, void *dev_id)
431 {
432 struct xen_pcibk_device *pdev = dev_id;
433 bool eoi;
434
435 /* IRQs might come in before pdev->evtchn_irq is written. */
436 if (unlikely(pdev->evtchn_irq != irq))
437 pdev->evtchn_irq = irq;
438
439 eoi = test_and_set_bit(_EOI_pending, &pdev->flags);
440 WARN(eoi, "IRQ while EOI pending\n");
441
442 xen_pcibk_test_and_schedule_op(pdev);
443
444 return IRQ_HANDLED;
445 }
xen_pcibk_guest_interrupt(int irq,void * dev_id)446 static irqreturn_t xen_pcibk_guest_interrupt(int irq, void *dev_id)
447 {
448 struct pci_dev *dev = (struct pci_dev *)dev_id;
449 struct xen_pcibk_dev_data *dev_data = pci_get_drvdata(dev);
450
451 if (dev_data->isr_on && dev_data->ack_intr) {
452 dev_data->handled++;
453 if ((dev_data->handled % 1000) == 0) {
454 if (xen_test_irq_shared(irq)) {
455 pr_info("%s IRQ line is not shared "
456 "with other domains. Turning ISR off\n",
457 dev_data->irq_name);
458 dev_data->ack_intr = 0;
459 }
460 }
461 return IRQ_HANDLED;
462 }
463 return IRQ_NONE;
464 }
465