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