2 * PCI Backend Operations - respond to PCI requests from Frontend
4 * Author: Ryan Wilson <hap9@epoch.ncsc.mil>
7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
9 #include <linux/moduleparam.h>
10 #include <linux/wait.h>
11 #include <linux/bitops.h>
12 #include <xen/events.h>
13 #include <linux/sched.h>
17 module_param(verbose_request, int, 0644);
19 static irqreturn_t xen_pcibk_guest_interrupt(int irq, void *dev_id);
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.
25 static void xen_pcibk_control_isr(struct pci_dev *dev, int reset)
27 struct xen_pcibk_dev_data *dev_data;
31 dev_data = pci_get_drvdata(dev);
35 /* We don't deal with bridges */
36 if (dev->hdr_type != PCI_HEADER_TYPE_NORMAL)
40 dev_data->enable_intx = 0;
41 dev_data->ack_intr = 0;
43 enable = dev_data->enable_intx;
45 /* Asked to disable, but ISR isn't runnig */
46 if (!enable && !dev_data->isr_on)
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.
54 dev_data->irq = dev->irq;
57 * SR-IOV devices in all use MSI-X and have no legacy
58 * interrupts, so inhibit creating a fake IRQ handler for them.
60 if (dev_data->irq == 0)
63 dev_dbg(&dev->dev, "%s: #%d %s %s%s %s-> %s\n",
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");
74 * The MSI or MSI-X should not have an IRQ handler. Otherwise
75 * if the guest terminates we BUG_ON in free_msi_irqs.
77 if (dev->msi_enabled || dev->msix_enabled)
80 rc = request_irq(dev_data->irq,
81 xen_pcibk_guest_interrupt, IRQF_SHARED,
82 dev_data->irq_name, dev);
84 dev_err(&dev->dev, "%s: failed to install fake IRQ " \
85 "handler for IRQ %d! (rc:%d)\n",
86 dev_data->irq_name, dev_data->irq, rc);
90 free_irq(dev_data->irq, dev);
93 dev_data->isr_on = enable;
94 dev_data->ack_intr = enable;
96 dev_dbg(&dev->dev, "%s: #%d %s %s%s %s\n",
99 pci_is_enabled(dev) ? "on" : "off",
100 dev->msi_enabled ? "MSI" : "",
101 dev->msix_enabled ? "MSI/X" : "",
102 enable ? (dev_data->isr_on ? "enabled" : "failed to enable") :
103 (dev_data->isr_on ? "failed to disable" : "disabled"));
106 /* Ensure a device is "turned off" and ready to be exported.
107 * (Also see xen_pcibk_config_reset to ensure virtual configuration space is
108 * ready to be re-exported)
110 void xen_pcibk_reset_device(struct pci_dev *dev)
114 xen_pcibk_control_isr(dev, 1 /* reset device */);
116 /* Disable devices (but not bridges) */
117 if (dev->hdr_type == PCI_HEADER_TYPE_NORMAL) {
118 #ifdef CONFIG_PCI_MSI
119 /* The guest could have been abruptly killed without
120 * disabling MSI/MSI-X interrupts.*/
121 if (dev->msix_enabled)
122 pci_disable_msix(dev);
123 if (dev->msi_enabled)
124 pci_disable_msi(dev);
126 if (pci_is_enabled(dev))
127 pci_disable_device(dev);
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("%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_exact(dev, entries, op->value);
245 for (i = 0; i < op->value; i++) {
246 op->msix_entries[i].entry = entries[i].entry;
247 if (entries[i].vector) {
248 op->msix_entries[i].vector =
249 xen_pirq_from_irq(entries[i].vector);
250 if (unlikely(verbose_request))
251 printk(KERN_DEBUG DRV_NAME ": %s: " \
254 op->msix_entries[i].vector);
258 pr_warn_ratelimited("%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;
268 return result > 0 ? 0 : result;
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 static inline bool xen_pcibk_test_op_pending(struct xen_pcibk_device *pdev)
302 return test_bit(_XEN_PCIF_active,
303 (unsigned long *)&pdev->sh_info->flags) &&
304 !test_and_set_bit(_PDEVF_op_active, &pdev->flags);
308 * Now the same evtchn is used for both pcifront conf_read_write request
309 * as well as pcie aer front end ack. We use a new work_queue to schedule
310 * xen_pcibk conf_read_write service for avoiding confict with aer_core
311 * do_recovery job which also use the system default work_queue
313 static void xen_pcibk_test_and_schedule_op(struct xen_pcibk_device *pdev)
317 /* Check that frontend is requesting an operation and that we are not
318 * already processing a request */
319 if (xen_pcibk_test_op_pending(pdev)) {
320 schedule_work(&pdev->op_work);
323 /*_XEN_PCIB_active should have been cleared by pcifront. And also make
324 sure xen_pcibk is waiting for ack by checking _PCIB_op_pending*/
325 if (!test_bit(_XEN_PCIB_active, (unsigned long *)&pdev->sh_info->flags)
326 && test_bit(_PCIB_op_pending, &pdev->flags)) {
327 wake_up(&xen_pcibk_aer_wait_queue);
331 /* EOI if there was nothing to do. */
333 xen_pcibk_lateeoi(pdev, XEN_EOI_FLAG_SPURIOUS);
336 /* Performing the configuration space reads/writes must not be done in atomic
337 * context because some of the pci_* functions can sleep (mostly due to ACPI
338 * use of semaphores). This function is intended to be called from a work
339 * queue in process context taking a struct xen_pcibk_device as a parameter */
341 static void xen_pcibk_do_one_op(struct xen_pcibk_device *pdev)
344 struct xen_pcibk_dev_data *dev_data = NULL;
345 struct xen_pci_op *op = &pdev->op;
347 #ifdef CONFIG_PCI_MSI
351 *op = pdev->sh_info->op;
353 dev = xen_pcibk_get_pci_dev(pdev, op->domain, op->bus, op->devfn);
356 op->err = XEN_PCI_ERR_dev_not_found;
358 dev_data = pci_get_drvdata(dev);
360 test_intx = dev_data->enable_intx;
362 case XEN_PCI_OP_conf_read:
363 op->err = xen_pcibk_config_read(dev,
364 op->offset, op->size, &op->value);
366 case XEN_PCI_OP_conf_write:
367 op->err = xen_pcibk_config_write(dev,
368 op->offset, op->size, op->value);
370 #ifdef CONFIG_PCI_MSI
371 case XEN_PCI_OP_enable_msi:
372 op->err = xen_pcibk_enable_msi(pdev, dev, op);
374 case XEN_PCI_OP_disable_msi:
375 op->err = xen_pcibk_disable_msi(pdev, dev, op);
377 case XEN_PCI_OP_enable_msix:
379 op->err = xen_pcibk_enable_msix(pdev, dev, op);
381 case XEN_PCI_OP_disable_msix:
382 op->err = xen_pcibk_disable_msix(pdev, dev, op);
386 op->err = XEN_PCI_ERR_not_implemented;
390 if (!op->err && dev && dev_data) {
391 /* Transition detected */
392 if ((dev_data->enable_intx != test_intx))
393 xen_pcibk_control_isr(dev, 0 /* no reset */);
395 pdev->sh_info->op.err = op->err;
396 pdev->sh_info->op.value = op->value;
397 #ifdef CONFIG_PCI_MSI
398 if (op->cmd == XEN_PCI_OP_enable_msix && op->err == 0) {
401 for (i = 0; i < nr; i++)
402 pdev->sh_info->op.msix_entries[i].vector =
403 op->msix_entries[i].vector;
406 /* Tell the driver domain that we're done. */
408 clear_bit(_XEN_PCIF_active, (unsigned long *)&pdev->sh_info->flags);
409 notify_remote_via_irq(pdev->evtchn_irq);
411 /* Mark that we're done. */
412 smp_mb__before_atomic(); /* /after/ clearing PCIF_active */
413 clear_bit(_PDEVF_op_active, &pdev->flags);
414 smp_mb__after_atomic(); /* /before/ final check for work */
417 void xen_pcibk_do_op(struct work_struct *data)
419 struct xen_pcibk_device *pdev =
420 container_of(data, struct xen_pcibk_device, op_work);
423 xen_pcibk_do_one_op(pdev);
424 } while (xen_pcibk_test_op_pending(pdev));
426 xen_pcibk_lateeoi(pdev, 0);
429 irqreturn_t xen_pcibk_handle_event(int irq, void *dev_id)
431 struct xen_pcibk_device *pdev = dev_id;
434 /* IRQs might come in before pdev->evtchn_irq is written. */
435 if (unlikely(pdev->evtchn_irq != irq))
436 pdev->evtchn_irq = irq;
438 eoi = test_and_set_bit(_EOI_pending, &pdev->flags);
439 WARN(eoi, "IRQ while EOI pending\n");
441 xen_pcibk_test_and_schedule_op(pdev);
445 static irqreturn_t xen_pcibk_guest_interrupt(int irq, void *dev_id)
447 struct pci_dev *dev = (struct pci_dev *)dev_id;
448 struct xen_pcibk_dev_data *dev_data = pci_get_drvdata(dev);
450 if (dev_data->isr_on && dev_data->ack_intr) {
452 if ((dev_data->handled % 1000) == 0) {
453 if (xen_test_irq_shared(irq)) {
454 pr_info("%s IRQ line is not shared "
455 "with other domains. Turning ISR off\n",
457 dev_data->ack_intr = 0;