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 struct xen_pcibk_dev_data *dev_data;
188 if (unlikely(verbose_request))
189 printk(KERN_DEBUG DRV_NAME ": %s: disable MSI\n",
191 pci_disable_msi(dev);
193 op->value = dev->irq ? xen_pirq_from_irq(dev->irq) : 0;
194 if (unlikely(verbose_request))
195 printk(KERN_DEBUG DRV_NAME ": %s: MSI: %d\n", pci_name(dev),
197 dev_data = pci_get_drvdata(dev);
199 dev_data->ack_intr = 1;
204 int xen_pcibk_enable_msix(struct xen_pcibk_device *pdev,
205 struct pci_dev *dev, struct xen_pci_op *op)
207 struct xen_pcibk_dev_data *dev_data;
209 struct msix_entry *entries;
211 if (unlikely(verbose_request))
212 printk(KERN_DEBUG DRV_NAME ": %s: enable MSI-X\n",
215 if (op->value > SH_INFO_MAX_VEC)
218 if (dev->msix_enabled)
221 if (dev->msi_enabled)
224 entries = kmalloc(op->value * sizeof(*entries), GFP_KERNEL);
228 for (i = 0; i < op->value; i++) {
229 entries[i].entry = op->msix_entries[i].entry;
230 entries[i].vector = op->msix_entries[i].vector;
233 result = pci_enable_msix(dev, entries, op->value);
236 for (i = 0; i < op->value; i++) {
237 op->msix_entries[i].entry = entries[i].entry;
238 if (entries[i].vector)
239 op->msix_entries[i].vector =
240 xen_pirq_from_irq(entries[i].vector);
241 if (unlikely(verbose_request))
242 printk(KERN_DEBUG DRV_NAME ": %s: " \
245 op->msix_entries[i].vector);
248 pr_warn_ratelimited(DRV_NAME ": %s: error enabling MSI-X for guest %u: err %d!\n",
249 pci_name(dev), pdev->xdev->otherend_id,
254 dev_data = pci_get_drvdata(dev);
256 dev_data->ack_intr = 0;
262 int xen_pcibk_disable_msix(struct xen_pcibk_device *pdev,
263 struct pci_dev *dev, struct xen_pci_op *op)
265 struct xen_pcibk_dev_data *dev_data;
266 if (unlikely(verbose_request))
267 printk(KERN_DEBUG DRV_NAME ": %s: disable MSI-X\n",
269 pci_disable_msix(dev);
272 * SR-IOV devices (which don't have any legacy IRQ) have
273 * an undefined IRQ value of zero.
275 op->value = dev->irq ? xen_pirq_from_irq(dev->irq) : 0;
276 if (unlikely(verbose_request))
277 printk(KERN_DEBUG DRV_NAME ": %s: MSI-X: %d\n", pci_name(dev),
279 dev_data = pci_get_drvdata(dev);
281 dev_data->ack_intr = 1;
286 * Now the same evtchn is used for both pcifront conf_read_write request
287 * as well as pcie aer front end ack. We use a new work_queue to schedule
288 * xen_pcibk conf_read_write service for avoiding confict with aer_core
289 * do_recovery job which also use the system default work_queue
291 void xen_pcibk_test_and_schedule_op(struct xen_pcibk_device *pdev)
293 /* Check that frontend is requesting an operation and that we are not
294 * already processing a request */
295 if (test_bit(_XEN_PCIF_active, (unsigned long *)&pdev->sh_info->flags)
296 && !test_and_set_bit(_PDEVF_op_active, &pdev->flags)) {
297 queue_work(xen_pcibk_wq, &pdev->op_work);
299 /*_XEN_PCIB_active should have been cleared by pcifront. And also make
300 sure xen_pcibk is waiting for ack by checking _PCIB_op_pending*/
301 if (!test_bit(_XEN_PCIB_active, (unsigned long *)&pdev->sh_info->flags)
302 && test_bit(_PCIB_op_pending, &pdev->flags)) {
303 wake_up(&xen_pcibk_aer_wait_queue);
307 /* Performing the configuration space reads/writes must not be done in atomic
308 * context because some of the pci_* functions can sleep (mostly due to ACPI
309 * use of semaphores). This function is intended to be called from a work
310 * queue in process context taking a struct xen_pcibk_device as a parameter */
312 void xen_pcibk_do_op(struct work_struct *data)
314 struct xen_pcibk_device *pdev =
315 container_of(data, struct xen_pcibk_device, op_work);
317 struct xen_pcibk_dev_data *dev_data = NULL;
318 struct xen_pci_op *op = &pdev->op;
321 *op = pdev->sh_info->op;
323 dev = xen_pcibk_get_pci_dev(pdev, op->domain, op->bus, op->devfn);
326 op->err = XEN_PCI_ERR_dev_not_found;
328 dev_data = pci_get_drvdata(dev);
330 test_intx = dev_data->enable_intx;
332 case XEN_PCI_OP_conf_read:
333 op->err = xen_pcibk_config_read(dev,
334 op->offset, op->size, &op->value);
336 case XEN_PCI_OP_conf_write:
337 op->err = xen_pcibk_config_write(dev,
338 op->offset, op->size, op->value);
340 #ifdef CONFIG_PCI_MSI
341 case XEN_PCI_OP_enable_msi:
342 op->err = xen_pcibk_enable_msi(pdev, dev, op);
344 case XEN_PCI_OP_disable_msi:
345 op->err = xen_pcibk_disable_msi(pdev, dev, op);
347 case XEN_PCI_OP_enable_msix:
348 op->err = xen_pcibk_enable_msix(pdev, dev, op);
350 case XEN_PCI_OP_disable_msix:
351 op->err = xen_pcibk_disable_msix(pdev, dev, op);
355 op->err = XEN_PCI_ERR_not_implemented;
359 if (!op->err && dev && dev_data) {
360 /* Transition detected */
361 if ((dev_data->enable_intx != test_intx))
362 xen_pcibk_control_isr(dev, 0 /* no reset */);
364 pdev->sh_info->op.err = op->err;
365 pdev->sh_info->op.value = op->value;
366 #ifdef CONFIG_PCI_MSI
367 if (op->cmd == XEN_PCI_OP_enable_msix && op->err == 0) {
370 for (i = 0; i < op->value; i++)
371 pdev->sh_info->op.msix_entries[i].vector =
372 op->msix_entries[i].vector;
375 /* Tell the driver domain that we're done. */
377 clear_bit(_XEN_PCIF_active, (unsigned long *)&pdev->sh_info->flags);
378 notify_remote_via_irq(pdev->evtchn_irq);
380 /* Mark that we're done. */
381 smp_mb__before_clear_bit(); /* /after/ clearing PCIF_active */
382 clear_bit(_PDEVF_op_active, &pdev->flags);
383 smp_mb__after_clear_bit(); /* /before/ final check for work */
385 /* Check to see if the driver domain tried to start another request in
386 * between clearing _XEN_PCIF_active and clearing _PDEVF_op_active.
388 xen_pcibk_test_and_schedule_op(pdev);
391 irqreturn_t xen_pcibk_handle_event(int irq, void *dev_id)
393 struct xen_pcibk_device *pdev = dev_id;
395 xen_pcibk_test_and_schedule_op(pdev);
399 static irqreturn_t xen_pcibk_guest_interrupt(int irq, void *dev_id)
401 struct pci_dev *dev = (struct pci_dev *)dev_id;
402 struct xen_pcibk_dev_data *dev_data = pci_get_drvdata(dev);
404 if (dev_data->isr_on && dev_data->ack_intr) {
406 if ((dev_data->handled % 1000) == 0) {
407 if (xen_test_irq_shared(irq)) {
408 printk(KERN_INFO "%s IRQ line is not shared "
409 "with other domains. Turning ISR off\n",
411 dev_data->ack_intr = 0;