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");
71 rc = request_irq(dev_data->irq,
72 xen_pcibk_guest_interrupt, IRQF_SHARED,
73 dev_data->irq_name, dev);
75 dev_err(&dev->dev, "%s: failed to install fake IRQ " \
76 "handler for IRQ %d! (rc:%d)\n",
77 dev_data->irq_name, dev_data->irq, rc);
81 free_irq(dev_data->irq, dev);
84 dev_data->isr_on = enable;
85 dev_data->ack_intr = enable;
87 dev_dbg(&dev->dev, "%s: #%d %s %s%s %s\n",
90 pci_is_enabled(dev) ? "on" : "off",
91 dev->msi_enabled ? "MSI" : "",
92 dev->msix_enabled ? "MSI/X" : "",
93 enable ? (dev_data->isr_on ? "enabled" : "failed to enable") :
94 (dev_data->isr_on ? "failed to disable" : "disabled"));
97 /* Ensure a device is "turned off" and ready to be exported.
98 * (Also see xen_pcibk_config_reset to ensure virtual configuration space is
99 * ready to be re-exported)
101 void xen_pcibk_reset_device(struct pci_dev *dev)
105 xen_pcibk_control_isr(dev, 1 /* reset device */);
107 /* Disable devices (but not bridges) */
108 if (dev->hdr_type == PCI_HEADER_TYPE_NORMAL) {
109 #ifdef CONFIG_PCI_MSI
110 /* The guest could have been abruptly killed without
111 * disabling MSI/MSI-X interrupts.*/
112 if (dev->msix_enabled)
113 pci_disable_msix(dev);
114 if (dev->msi_enabled)
115 pci_disable_msi(dev);
117 if (pci_is_enabled(dev))
118 pci_disable_device(dev);
120 pci_write_config_word(dev, PCI_COMMAND, 0);
122 dev->is_busmaster = 0;
124 pci_read_config_word(dev, PCI_COMMAND, &cmd);
125 if (cmd & (PCI_COMMAND_INVALIDATE)) {
126 cmd &= ~(PCI_COMMAND_INVALIDATE);
127 pci_write_config_word(dev, PCI_COMMAND, cmd);
129 dev->is_busmaster = 0;
134 #ifdef CONFIG_PCI_MSI
136 int xen_pcibk_enable_msi(struct xen_pcibk_device *pdev,
137 struct pci_dev *dev, struct xen_pci_op *op)
139 struct xen_pcibk_dev_data *dev_data;
142 if (unlikely(verbose_request))
143 printk(KERN_DEBUG DRV_NAME ": %s: enable MSI\n", pci_name(dev));
145 status = pci_enable_msi(dev);
148 pr_warn_ratelimited(DRV_NAME ": %s: error enabling MSI for guest %u: err %d\n",
149 pci_name(dev), pdev->xdev->otherend_id,
152 return XEN_PCI_ERR_op_failed;
155 /* The value the guest needs is actually the IDT vector, not the
156 * the local domain's IRQ number. */
158 op->value = dev->irq ? xen_pirq_from_irq(dev->irq) : 0;
159 if (unlikely(verbose_request))
160 printk(KERN_DEBUG DRV_NAME ": %s: MSI: %d\n", pci_name(dev),
163 dev_data = pci_get_drvdata(dev);
165 dev_data->ack_intr = 0;
171 int xen_pcibk_disable_msi(struct xen_pcibk_device *pdev,
172 struct pci_dev *dev, struct xen_pci_op *op)
174 struct xen_pcibk_dev_data *dev_data;
176 if (unlikely(verbose_request))
177 printk(KERN_DEBUG DRV_NAME ": %s: disable MSI\n",
179 pci_disable_msi(dev);
181 op->value = dev->irq ? xen_pirq_from_irq(dev->irq) : 0;
182 if (unlikely(verbose_request))
183 printk(KERN_DEBUG DRV_NAME ": %s: MSI: %d\n", pci_name(dev),
185 dev_data = pci_get_drvdata(dev);
187 dev_data->ack_intr = 1;
192 int xen_pcibk_enable_msix(struct xen_pcibk_device *pdev,
193 struct pci_dev *dev, struct xen_pci_op *op)
195 struct xen_pcibk_dev_data *dev_data;
197 struct msix_entry *entries;
199 if (unlikely(verbose_request))
200 printk(KERN_DEBUG DRV_NAME ": %s: enable MSI-X\n",
202 if (op->value > SH_INFO_MAX_VEC)
205 entries = kmalloc(op->value * sizeof(*entries), GFP_KERNEL);
209 for (i = 0; i < op->value; i++) {
210 entries[i].entry = op->msix_entries[i].entry;
211 entries[i].vector = op->msix_entries[i].vector;
214 result = pci_enable_msix(dev, entries, op->value);
217 for (i = 0; i < op->value; i++) {
218 op->msix_entries[i].entry = entries[i].entry;
219 if (entries[i].vector)
220 op->msix_entries[i].vector =
221 xen_pirq_from_irq(entries[i].vector);
222 if (unlikely(verbose_request))
223 printk(KERN_DEBUG DRV_NAME ": %s: " \
226 op->msix_entries[i].vector);
229 pr_warn_ratelimited(DRV_NAME ": %s: error enabling MSI-X for guest %u: err %d!\n",
230 pci_name(dev), pdev->xdev->otherend_id,
235 dev_data = pci_get_drvdata(dev);
237 dev_data->ack_intr = 0;
243 int xen_pcibk_disable_msix(struct xen_pcibk_device *pdev,
244 struct pci_dev *dev, struct xen_pci_op *op)
246 struct xen_pcibk_dev_data *dev_data;
247 if (unlikely(verbose_request))
248 printk(KERN_DEBUG DRV_NAME ": %s: disable MSI-X\n",
250 pci_disable_msix(dev);
253 * SR-IOV devices (which don't have any legacy IRQ) have
254 * an undefined IRQ value of zero.
256 op->value = dev->irq ? xen_pirq_from_irq(dev->irq) : 0;
257 if (unlikely(verbose_request))
258 printk(KERN_DEBUG DRV_NAME ": %s: MSI-X: %d\n", pci_name(dev),
260 dev_data = pci_get_drvdata(dev);
262 dev_data->ack_intr = 1;
267 * Now the same evtchn is used for both pcifront conf_read_write request
268 * as well as pcie aer front end ack. We use a new work_queue to schedule
269 * xen_pcibk conf_read_write service for avoiding confict with aer_core
270 * do_recovery job which also use the system default work_queue
272 void xen_pcibk_test_and_schedule_op(struct xen_pcibk_device *pdev)
274 /* Check that frontend is requesting an operation and that we are not
275 * already processing a request */
276 if (test_bit(_XEN_PCIF_active, (unsigned long *)&pdev->sh_info->flags)
277 && !test_and_set_bit(_PDEVF_op_active, &pdev->flags)) {
278 queue_work(xen_pcibk_wq, &pdev->op_work);
280 /*_XEN_PCIB_active should have been cleared by pcifront. And also make
281 sure xen_pcibk is waiting for ack by checking _PCIB_op_pending*/
282 if (!test_bit(_XEN_PCIB_active, (unsigned long *)&pdev->sh_info->flags)
283 && test_bit(_PCIB_op_pending, &pdev->flags)) {
284 wake_up(&xen_pcibk_aer_wait_queue);
288 /* Performing the configuration space reads/writes must not be done in atomic
289 * context because some of the pci_* functions can sleep (mostly due to ACPI
290 * use of semaphores). This function is intended to be called from a work
291 * queue in process context taking a struct xen_pcibk_device as a parameter */
293 void xen_pcibk_do_op(struct work_struct *data)
295 struct xen_pcibk_device *pdev =
296 container_of(data, struct xen_pcibk_device, op_work);
298 struct xen_pcibk_dev_data *dev_data = NULL;
299 struct xen_pci_op *op = &pdev->op;
302 *op = pdev->sh_info->op;
304 dev = xen_pcibk_get_pci_dev(pdev, op->domain, op->bus, op->devfn);
307 op->err = XEN_PCI_ERR_dev_not_found;
309 dev_data = pci_get_drvdata(dev);
311 test_intx = dev_data->enable_intx;
313 case XEN_PCI_OP_conf_read:
314 op->err = xen_pcibk_config_read(dev,
315 op->offset, op->size, &op->value);
317 case XEN_PCI_OP_conf_write:
318 op->err = xen_pcibk_config_write(dev,
319 op->offset, op->size, op->value);
321 #ifdef CONFIG_PCI_MSI
322 case XEN_PCI_OP_enable_msi:
323 op->err = xen_pcibk_enable_msi(pdev, dev, op);
325 case XEN_PCI_OP_disable_msi:
326 op->err = xen_pcibk_disable_msi(pdev, dev, op);
328 case XEN_PCI_OP_enable_msix:
329 op->err = xen_pcibk_enable_msix(pdev, dev, op);
331 case XEN_PCI_OP_disable_msix:
332 op->err = xen_pcibk_disable_msix(pdev, dev, op);
336 op->err = XEN_PCI_ERR_not_implemented;
340 if (!op->err && dev && dev_data) {
341 /* Transition detected */
342 if ((dev_data->enable_intx != test_intx))
343 xen_pcibk_control_isr(dev, 0 /* no reset */);
345 pdev->sh_info->op.err = op->err;
346 pdev->sh_info->op.value = op->value;
347 #ifdef CONFIG_PCI_MSI
348 if (op->cmd == XEN_PCI_OP_enable_msix && op->err == 0) {
351 for (i = 0; i < op->value; i++)
352 pdev->sh_info->op.msix_entries[i].vector =
353 op->msix_entries[i].vector;
356 /* Tell the driver domain that we're done. */
358 clear_bit(_XEN_PCIF_active, (unsigned long *)&pdev->sh_info->flags);
359 notify_remote_via_irq(pdev->evtchn_irq);
361 /* Mark that we're done. */
362 smp_mb__before_clear_bit(); /* /after/ clearing PCIF_active */
363 clear_bit(_PDEVF_op_active, &pdev->flags);
364 smp_mb__after_clear_bit(); /* /before/ final check for work */
366 /* Check to see if the driver domain tried to start another request in
367 * between clearing _XEN_PCIF_active and clearing _PDEVF_op_active.
369 xen_pcibk_test_and_schedule_op(pdev);
372 irqreturn_t xen_pcibk_handle_event(int irq, void *dev_id)
374 struct xen_pcibk_device *pdev = dev_id;
376 xen_pcibk_test_and_schedule_op(pdev);
380 static irqreturn_t xen_pcibk_guest_interrupt(int irq, void *dev_id)
382 struct pci_dev *dev = (struct pci_dev *)dev_id;
383 struct xen_pcibk_dev_data *dev_data = pci_get_drvdata(dev);
385 if (dev_data->isr_on && dev_data->ack_intr) {
387 if ((dev_data->handled % 1000) == 0) {
388 if (xen_test_irq_shared(irq)) {
389 printk(KERN_INFO "%s IRQ line is not shared "
390 "with other domains. Turning ISR off\n",
392 dev_data->ack_intr = 0;