2 * PCI Backend Operations - respond to PCI requests from Frontend
4 * Author: Ryan Wilson <hap9@epoch.ncsc.mil>
6 #include <linux/module.h>
7 #include <linux/wait.h>
8 #include <linux/bitops.h>
9 #include <xen/events.h>
10 #include <linux/sched.h>
11 #include <linux/ratelimit.h>
15 module_param(verbose_request, int, 0644);
17 static irqreturn_t xen_pcibk_guest_interrupt(int irq, void *dev_id);
19 /* Ensure a device is has the fake IRQ handler "turned on/off" and is
20 * ready to be exported. This MUST be run after xen_pcibk_reset_device
21 * which does the actual PCI device enable/disable.
23 static void xen_pcibk_control_isr(struct pci_dev *dev, int reset)
25 struct xen_pcibk_dev_data *dev_data;
29 dev_data = pci_get_drvdata(dev);
33 /* We don't deal with bridges */
34 if (dev->hdr_type != PCI_HEADER_TYPE_NORMAL)
38 dev_data->enable_intx = 0;
39 dev_data->ack_intr = 0;
41 enable = dev_data->enable_intx;
43 /* Asked to disable, but ISR isn't runnig */
44 if (!enable && !dev_data->isr_on)
47 /* Squirrel away the IRQs in the dev_data. We need this
48 * b/c when device transitions to MSI, the dev->irq is
49 * overwritten with the MSI vector.
52 dev_data->irq = dev->irq;
55 * SR-IOV devices in all use MSI-X and have no legacy
56 * interrupts, so inhibit creating a fake IRQ handler for them.
58 if (dev_data->irq == 0)
61 dev_dbg(&dev->dev, "%s: #%d %s %s%s %s-> %s\n",
64 pci_is_enabled(dev) ? "on" : "off",
65 dev->msi_enabled ? "MSI" : "",
66 dev->msix_enabled ? "MSI/X" : "",
67 dev_data->isr_on ? "enable" : "disable",
68 enable ? "enable" : "disable");
72 * The MSI or MSI-X should not have an IRQ handler. Otherwise
73 * if the guest terminates we BUG_ON in free_msi_irqs.
75 if (dev->msi_enabled || dev->msix_enabled)
78 rc = request_irq(dev_data->irq,
79 xen_pcibk_guest_interrupt, IRQF_SHARED,
80 dev_data->irq_name, dev);
82 dev_err(&dev->dev, "%s: failed to install fake IRQ " \
83 "handler for IRQ %d! (rc:%d)\n",
84 dev_data->irq_name, dev_data->irq, rc);
88 free_irq(dev_data->irq, dev);
91 dev_data->isr_on = enable;
92 dev_data->ack_intr = enable;
94 dev_dbg(&dev->dev, "%s: #%d %s %s%s %s\n",
97 pci_is_enabled(dev) ? "on" : "off",
98 dev->msi_enabled ? "MSI" : "",
99 dev->msix_enabled ? "MSI/X" : "",
100 enable ? (dev_data->isr_on ? "enabled" : "failed to enable") :
101 (dev_data->isr_on ? "failed to disable" : "disabled"));
104 /* Ensure a device is "turned off" and ready to be exported.
105 * (Also see xen_pcibk_config_reset to ensure virtual configuration space is
106 * ready to be re-exported)
108 void xen_pcibk_reset_device(struct pci_dev *dev)
112 xen_pcibk_control_isr(dev, 1 /* reset device */);
114 /* Disable devices (but not bridges) */
115 if (dev->hdr_type == PCI_HEADER_TYPE_NORMAL) {
116 #ifdef CONFIG_PCI_MSI
117 /* The guest could have been abruptly killed without
118 * disabling MSI/MSI-X interrupts.*/
119 if (dev->msix_enabled)
120 pci_disable_msix(dev);
121 if (dev->msi_enabled)
122 pci_disable_msi(dev);
124 if (pci_is_enabled(dev))
125 pci_disable_device(dev);
127 pci_write_config_word(dev, PCI_COMMAND, 0);
129 dev->is_busmaster = 0;
131 pci_read_config_word(dev, PCI_COMMAND, &cmd);
132 if (cmd & (PCI_COMMAND_INVALIDATE)) {
133 cmd &= ~(PCI_COMMAND_INVALIDATE);
134 pci_write_config_word(dev, PCI_COMMAND, cmd);
136 dev->is_busmaster = 0;
141 #ifdef CONFIG_PCI_MSI
143 int xen_pcibk_enable_msi(struct xen_pcibk_device *pdev,
144 struct pci_dev *dev, struct xen_pci_op *op)
146 struct xen_pcibk_dev_data *dev_data;
149 if (unlikely(verbose_request))
150 printk(KERN_DEBUG DRV_NAME ": %s: enable MSI\n", pci_name(dev));
152 if (dev->msi_enabled)
154 else if (dev->msix_enabled)
157 status = pci_enable_msi(dev);
160 pr_warn_ratelimited(DRV_NAME ": %s: error enabling MSI for guest %u: err %d\n",
161 pci_name(dev), pdev->xdev->otherend_id,
164 return XEN_PCI_ERR_op_failed;
167 /* The value the guest needs is actually the IDT vector, not the
168 * the local domain's IRQ number. */
170 op->value = dev->irq ? xen_pirq_from_irq(dev->irq) : 0;
171 if (unlikely(verbose_request))
172 printk(KERN_DEBUG DRV_NAME ": %s: MSI: %d\n", pci_name(dev),
175 dev_data = pci_get_drvdata(dev);
177 dev_data->ack_intr = 0;
183 int xen_pcibk_disable_msi(struct xen_pcibk_device *pdev,
184 struct pci_dev *dev, struct xen_pci_op *op)
186 if (unlikely(verbose_request))
187 printk(KERN_DEBUG DRV_NAME ": %s: disable MSI\n",
190 if (dev->msi_enabled) {
191 struct xen_pcibk_dev_data *dev_data;
193 pci_disable_msi(dev);
195 dev_data = pci_get_drvdata(dev);
197 dev_data->ack_intr = 1;
199 op->value = dev->irq ? xen_pirq_from_irq(dev->irq) : 0;
200 if (unlikely(verbose_request))
201 printk(KERN_DEBUG DRV_NAME ": %s: MSI: %d\n", pci_name(dev),
207 int xen_pcibk_enable_msix(struct xen_pcibk_device *pdev,
208 struct pci_dev *dev, struct xen_pci_op *op)
210 struct xen_pcibk_dev_data *dev_data;
212 struct msix_entry *entries;
215 if (unlikely(verbose_request))
216 printk(KERN_DEBUG DRV_NAME ": %s: enable MSI-X\n",
219 if (op->value > SH_INFO_MAX_VEC)
222 if (dev->msix_enabled)
226 * PCI_COMMAND_MEMORY must be enabled, otherwise we may not be able
227 * to access the BARs where the MSI-X entries reside.
228 * But VF devices are unique in which the PF needs to be checked.
230 pci_read_config_word(pci_physfn(dev), PCI_COMMAND, &cmd);
231 if (dev->msi_enabled || !(cmd & PCI_COMMAND_MEMORY))
234 entries = kmalloc(op->value * sizeof(*entries), GFP_KERNEL);
238 for (i = 0; i < op->value; i++) {
239 entries[i].entry = op->msix_entries[i].entry;
240 entries[i].vector = op->msix_entries[i].vector;
243 result = pci_enable_msix(dev, entries, op->value);
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: " \
255 op->msix_entries[i].vector);
258 pr_warn_ratelimited(DRV_NAME ": %s: error enabling MSI-X for guest %u: err %d!\n",
259 pci_name(dev), pdev->xdev->otherend_id,
264 dev_data = pci_get_drvdata(dev);
266 dev_data->ack_intr = 0;
272 int xen_pcibk_disable_msix(struct xen_pcibk_device *pdev,
273 struct pci_dev *dev, struct xen_pci_op *op)
275 if (unlikely(verbose_request))
276 printk(KERN_DEBUG DRV_NAME ": %s: disable MSI-X\n",
279 if (dev->msix_enabled) {
280 struct xen_pcibk_dev_data *dev_data;
282 pci_disable_msix(dev);
284 dev_data = pci_get_drvdata(dev);
286 dev_data->ack_intr = 1;
289 * SR-IOV devices (which don't have any legacy IRQ) have
290 * an undefined IRQ value of zero.
292 op->value = dev->irq ? xen_pirq_from_irq(dev->irq) : 0;
293 if (unlikely(verbose_request))
294 printk(KERN_DEBUG DRV_NAME ": %s: MSI-X: %d\n",
295 pci_name(dev), op->value);
300 * Now the same evtchn is used for both pcifront conf_read_write request
301 * as well as pcie aer front end ack. We use a new work_queue to schedule
302 * xen_pcibk conf_read_write service for avoiding confict with aer_core
303 * do_recovery job which also use the system default work_queue
305 void xen_pcibk_test_and_schedule_op(struct xen_pcibk_device *pdev)
307 /* Check that frontend is requesting an operation and that we are not
308 * already processing a request */
309 if (test_bit(_XEN_PCIF_active, (unsigned long *)&pdev->sh_info->flags)
310 && !test_and_set_bit(_PDEVF_op_active, &pdev->flags)) {
311 queue_work(xen_pcibk_wq, &pdev->op_work);
313 /*_XEN_PCIB_active should have been cleared by pcifront. And also make
314 sure xen_pcibk is waiting for ack by checking _PCIB_op_pending*/
315 if (!test_bit(_XEN_PCIB_active, (unsigned long *)&pdev->sh_info->flags)
316 && test_bit(_PCIB_op_pending, &pdev->flags)) {
317 wake_up(&xen_pcibk_aer_wait_queue);
321 /* Performing the configuration space reads/writes must not be done in atomic
322 * context because some of the pci_* functions can sleep (mostly due to ACPI
323 * use of semaphores). This function is intended to be called from a work
324 * queue in process context taking a struct xen_pcibk_device as a parameter */
326 void xen_pcibk_do_op(struct work_struct *data)
328 struct xen_pcibk_device *pdev =
329 container_of(data, struct xen_pcibk_device, op_work);
331 struct xen_pcibk_dev_data *dev_data = NULL;
332 struct xen_pci_op *op = &pdev->op;
335 *op = pdev->sh_info->op;
337 dev = xen_pcibk_get_pci_dev(pdev, op->domain, op->bus, op->devfn);
340 op->err = XEN_PCI_ERR_dev_not_found;
342 dev_data = pci_get_drvdata(dev);
344 test_intx = dev_data->enable_intx;
346 case XEN_PCI_OP_conf_read:
347 op->err = xen_pcibk_config_read(dev,
348 op->offset, op->size, &op->value);
350 case XEN_PCI_OP_conf_write:
351 op->err = xen_pcibk_config_write(dev,
352 op->offset, op->size, op->value);
354 #ifdef CONFIG_PCI_MSI
355 case XEN_PCI_OP_enable_msi:
356 op->err = xen_pcibk_enable_msi(pdev, dev, op);
358 case XEN_PCI_OP_disable_msi:
359 op->err = xen_pcibk_disable_msi(pdev, dev, op);
361 case XEN_PCI_OP_enable_msix:
362 op->err = xen_pcibk_enable_msix(pdev, dev, op);
364 case XEN_PCI_OP_disable_msix:
365 op->err = xen_pcibk_disable_msix(pdev, dev, op);
369 op->err = XEN_PCI_ERR_not_implemented;
373 if (!op->err && dev && dev_data) {
374 /* Transition detected */
375 if ((dev_data->enable_intx != test_intx))
376 xen_pcibk_control_isr(dev, 0 /* no reset */);
378 pdev->sh_info->op.err = op->err;
379 pdev->sh_info->op.value = op->value;
380 #ifdef CONFIG_PCI_MSI
381 if (op->cmd == XEN_PCI_OP_enable_msix && op->err == 0) {
384 for (i = 0; i < op->value; i++)
385 pdev->sh_info->op.msix_entries[i].vector =
386 op->msix_entries[i].vector;
389 /* Tell the driver domain that we're done. */
391 clear_bit(_XEN_PCIF_active, (unsigned long *)&pdev->sh_info->flags);
392 notify_remote_via_irq(pdev->evtchn_irq);
394 /* Mark that we're done. */
395 smp_mb__before_clear_bit(); /* /after/ clearing PCIF_active */
396 clear_bit(_PDEVF_op_active, &pdev->flags);
397 smp_mb__after_clear_bit(); /* /before/ final check for work */
399 /* Check to see if the driver domain tried to start another request in
400 * between clearing _XEN_PCIF_active and clearing _PDEVF_op_active.
402 xen_pcibk_test_and_schedule_op(pdev);
405 irqreturn_t xen_pcibk_handle_event(int irq, void *dev_id)
407 struct xen_pcibk_device *pdev = dev_id;
409 xen_pcibk_test_and_schedule_op(pdev);
413 static irqreturn_t xen_pcibk_guest_interrupt(int irq, void *dev_id)
415 struct pci_dev *dev = (struct pci_dev *)dev_id;
416 struct xen_pcibk_dev_data *dev_data = pci_get_drvdata(dev);
418 if (dev_data->isr_on && dev_data->ack_intr) {
420 if ((dev_data->handled % 1000) == 0) {
421 if (xen_test_irq_shared(irq)) {
422 printk(KERN_INFO "%s IRQ line is not shared "
423 "with other domains. Turning ISR off\n",
425 dev_data->ack_intr = 0;