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/module.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");
73 rc = request_irq(dev_data->irq,
74 xen_pcibk_guest_interrupt, IRQF_SHARED,
75 dev_data->irq_name, dev);
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);
83 free_irq(dev_data->irq, dev);
86 dev_data->isr_on = enable;
87 dev_data->ack_intr = enable;
89 dev_dbg(&dev->dev, "%s: #%d %s %s%s %s\n",
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"));
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)
103 void xen_pcibk_reset_device(struct pci_dev *dev)
107 xen_pcibk_control_isr(dev, 1 /* reset device */);
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);
119 if (pci_is_enabled(dev))
120 pci_disable_device(dev);
122 pci_write_config_word(dev, PCI_COMMAND, 0);
124 dev->is_busmaster = 0;
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);
131 dev->is_busmaster = 0;
136 #ifdef CONFIG_PCI_MSI
138 int xen_pcibk_enable_msi(struct xen_pcibk_device *pdev,
139 struct pci_dev *dev, struct xen_pci_op *op)
141 struct xen_pcibk_dev_data *dev_data;
144 if (unlikely(verbose_request))
145 printk(KERN_DEBUG DRV_NAME ": %s: enable MSI\n", pci_name(dev));
147 if (dev->msi_enabled)
149 else if (dev->msix_enabled)
152 status = pci_enable_msi(dev);
155 pr_warn_ratelimited("%s: error enabling MSI for guest %u: err %d\n",
156 pci_name(dev), pdev->xdev->otherend_id,
159 return XEN_PCI_ERR_op_failed;
162 /* The value the guest needs is actually the IDT vector, not the
163 * the local domain's IRQ number. */
165 op->value = dev->irq ? xen_pirq_from_irq(dev->irq) : 0;
166 if (unlikely(verbose_request))
167 printk(KERN_DEBUG DRV_NAME ": %s: MSI: %d\n", pci_name(dev),
170 dev_data = pci_get_drvdata(dev);
172 dev_data->ack_intr = 0;
178 int xen_pcibk_disable_msi(struct xen_pcibk_device *pdev,
179 struct pci_dev *dev, struct xen_pci_op *op)
181 struct xen_pcibk_dev_data *dev_data;
183 if (unlikely(verbose_request))
184 printk(KERN_DEBUG DRV_NAME ": %s: disable MSI\n",
186 pci_disable_msi(dev);
188 op->value = dev->irq ? xen_pirq_from_irq(dev->irq) : 0;
189 if (unlikely(verbose_request))
190 printk(KERN_DEBUG DRV_NAME ": %s: MSI: %d\n", pci_name(dev),
192 dev_data = pci_get_drvdata(dev);
194 dev_data->ack_intr = 1;
199 int xen_pcibk_enable_msix(struct xen_pcibk_device *pdev,
200 struct pci_dev *dev, struct xen_pci_op *op)
202 struct xen_pcibk_dev_data *dev_data;
204 struct msix_entry *entries;
206 if (unlikely(verbose_request))
207 printk(KERN_DEBUG DRV_NAME ": %s: enable MSI-X\n",
209 if (op->value > SH_INFO_MAX_VEC)
212 entries = kmalloc(op->value * sizeof(*entries), GFP_KERNEL);
216 for (i = 0; i < op->value; i++) {
217 entries[i].entry = op->msix_entries[i].entry;
218 entries[i].vector = op->msix_entries[i].vector;
221 result = pci_enable_msix_exact(dev, entries, op->value);
223 for (i = 0; i < op->value; i++) {
224 op->msix_entries[i].entry = entries[i].entry;
225 if (entries[i].vector) {
226 op->msix_entries[i].vector =
227 xen_pirq_from_irq(entries[i].vector);
228 if (unlikely(verbose_request))
229 printk(KERN_DEBUG DRV_NAME ": %s: " \
232 op->msix_entries[i].vector);
236 pr_warn_ratelimited("%s: error enabling MSI-X for guest %u: err %d!\n",
237 pci_name(dev), pdev->xdev->otherend_id,
242 dev_data = pci_get_drvdata(dev);
244 dev_data->ack_intr = 0;
246 return result > 0 ? 0 : result;
250 int xen_pcibk_disable_msix(struct xen_pcibk_device *pdev,
251 struct pci_dev *dev, struct xen_pci_op *op)
253 struct xen_pcibk_dev_data *dev_data;
254 if (unlikely(verbose_request))
255 printk(KERN_DEBUG DRV_NAME ": %s: disable MSI-X\n",
257 pci_disable_msix(dev);
260 * SR-IOV devices (which don't have any legacy IRQ) have
261 * an undefined IRQ value of zero.
263 op->value = dev->irq ? xen_pirq_from_irq(dev->irq) : 0;
264 if (unlikely(verbose_request))
265 printk(KERN_DEBUG DRV_NAME ": %s: MSI-X: %d\n", pci_name(dev),
267 dev_data = pci_get_drvdata(dev);
269 dev_data->ack_intr = 1;
274 * Now the same evtchn is used for both pcifront conf_read_write request
275 * as well as pcie aer front end ack. We use a new work_queue to schedule
276 * xen_pcibk conf_read_write service for avoiding confict with aer_core
277 * do_recovery job which also use the system default work_queue
279 void xen_pcibk_test_and_schedule_op(struct xen_pcibk_device *pdev)
281 /* Check that frontend is requesting an operation and that we are not
282 * already processing a request */
283 if (test_bit(_XEN_PCIF_active, (unsigned long *)&pdev->sh_info->flags)
284 && !test_and_set_bit(_PDEVF_op_active, &pdev->flags)) {
285 queue_work(xen_pcibk_wq, &pdev->op_work);
287 /*_XEN_PCIB_active should have been cleared by pcifront. And also make
288 sure xen_pcibk is waiting for ack by checking _PCIB_op_pending*/
289 if (!test_bit(_XEN_PCIB_active, (unsigned long *)&pdev->sh_info->flags)
290 && test_bit(_PCIB_op_pending, &pdev->flags)) {
291 wake_up(&xen_pcibk_aer_wait_queue);
295 /* Performing the configuration space reads/writes must not be done in atomic
296 * context because some of the pci_* functions can sleep (mostly due to ACPI
297 * use of semaphores). This function is intended to be called from a work
298 * queue in process context taking a struct xen_pcibk_device as a parameter */
300 void xen_pcibk_do_op(struct work_struct *data)
302 struct xen_pcibk_device *pdev =
303 container_of(data, struct xen_pcibk_device, op_work);
305 struct xen_pcibk_dev_data *dev_data = NULL;
306 struct xen_pci_op *op = &pdev->op;
309 *op = pdev->sh_info->op;
311 dev = xen_pcibk_get_pci_dev(pdev, op->domain, op->bus, op->devfn);
314 op->err = XEN_PCI_ERR_dev_not_found;
316 dev_data = pci_get_drvdata(dev);
318 test_intx = dev_data->enable_intx;
320 case XEN_PCI_OP_conf_read:
321 op->err = xen_pcibk_config_read(dev,
322 op->offset, op->size, &op->value);
324 case XEN_PCI_OP_conf_write:
325 op->err = xen_pcibk_config_write(dev,
326 op->offset, op->size, op->value);
328 #ifdef CONFIG_PCI_MSI
329 case XEN_PCI_OP_enable_msi:
330 op->err = xen_pcibk_enable_msi(pdev, dev, op);
332 case XEN_PCI_OP_disable_msi:
333 op->err = xen_pcibk_disable_msi(pdev, dev, op);
335 case XEN_PCI_OP_enable_msix:
336 op->err = xen_pcibk_enable_msix(pdev, dev, op);
338 case XEN_PCI_OP_disable_msix:
339 op->err = xen_pcibk_disable_msix(pdev, dev, op);
343 op->err = XEN_PCI_ERR_not_implemented;
347 if (!op->err && dev && dev_data) {
348 /* Transition detected */
349 if ((dev_data->enable_intx != test_intx))
350 xen_pcibk_control_isr(dev, 0 /* no reset */);
352 pdev->sh_info->op.err = op->err;
353 pdev->sh_info->op.value = op->value;
354 #ifdef CONFIG_PCI_MSI
355 if (op->cmd == XEN_PCI_OP_enable_msix && op->err == 0) {
358 for (i = 0; i < op->value; i++)
359 pdev->sh_info->op.msix_entries[i].vector =
360 op->msix_entries[i].vector;
363 /* Tell the driver domain that we're done. */
365 clear_bit(_XEN_PCIF_active, (unsigned long *)&pdev->sh_info->flags);
366 notify_remote_via_irq(pdev->evtchn_irq);
368 /* Mark that we're done. */
369 smp_mb__before_atomic(); /* /after/ clearing PCIF_active */
370 clear_bit(_PDEVF_op_active, &pdev->flags);
371 smp_mb__after_atomic(); /* /before/ final check for work */
373 /* Check to see if the driver domain tried to start another request in
374 * between clearing _XEN_PCIF_active and clearing _PDEVF_op_active.
376 xen_pcibk_test_and_schedule_op(pdev);
379 irqreturn_t xen_pcibk_handle_event(int irq, void *dev_id)
381 struct xen_pcibk_device *pdev = dev_id;
383 xen_pcibk_test_and_schedule_op(pdev);
387 static irqreturn_t xen_pcibk_guest_interrupt(int irq, void *dev_id)
389 struct pci_dev *dev = (struct pci_dev *)dev_id;
390 struct xen_pcibk_dev_data *dev_data = pci_get_drvdata(dev);
392 if (dev_data->isr_on && dev_data->ack_intr) {
394 if ((dev_data->handled % 1000) == 0) {
395 if (xen_test_irq_shared(irq)) {
396 pr_info("%s IRQ line is not shared "
397 "with other domains. Turning ISR off\n",
399 dev_data->ack_intr = 0;