1 /* 2 * PCI Backend Operations - respond to PCI requests from Frontend 3 * 4 * Author: Ryan Wilson <hap9@epoch.ncsc.mil> 5 */ 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 "pciback.h" 12 13 int verbose_request; 14 module_param(verbose_request, int, 0644); 15 16 static irqreturn_t xen_pcibk_guest_interrupt(int irq, void *dev_id); 17 18 /* Ensure a device is has the fake IRQ handler "turned on/off" and is 19 * ready to be exported. This MUST be run after xen_pcibk_reset_device 20 * which does the actual PCI device enable/disable. 21 */ 22 static void xen_pcibk_control_isr(struct pci_dev *dev, int reset) 23 { 24 struct xen_pcibk_dev_data *dev_data; 25 int rc; 26 int enable = 0; 27 28 dev_data = pci_get_drvdata(dev); 29 if (!dev_data) 30 return; 31 32 /* We don't deal with bridges */ 33 if (dev->hdr_type != PCI_HEADER_TYPE_NORMAL) 34 return; 35 36 if (reset) { 37 dev_data->enable_intx = 0; 38 dev_data->ack_intr = 0; 39 } 40 enable = dev_data->enable_intx; 41 42 /* Asked to disable, but ISR isn't runnig */ 43 if (!enable && !dev_data->isr_on) 44 return; 45 46 /* Squirrel away the IRQs in the dev_data. We need this 47 * b/c when device transitions to MSI, the dev->irq is 48 * overwritten with the MSI vector. 49 */ 50 if (enable) 51 dev_data->irq = dev->irq; 52 53 /* 54 * SR-IOV devices in all use MSI-X and have no legacy 55 * interrupts, so inhibit creating a fake IRQ handler for them. 56 */ 57 if (dev_data->irq == 0) 58 goto out; 59 60 dev_dbg(&dev->dev, "%s: #%d %s %s%s %s-> %s\n", 61 dev_data->irq_name, 62 dev_data->irq, 63 pci_is_enabled(dev) ? "on" : "off", 64 dev->msi_enabled ? "MSI" : "", 65 dev->msix_enabled ? "MSI/X" : "", 66 dev_data->isr_on ? "enable" : "disable", 67 enable ? "enable" : "disable"); 68 69 if (enable) { 70 rc = request_irq(dev_data->irq, 71 xen_pcibk_guest_interrupt, IRQF_SHARED, 72 dev_data->irq_name, dev); 73 if (rc) { 74 dev_err(&dev->dev, "%s: failed to install fake IRQ " \ 75 "handler for IRQ %d! (rc:%d)\n", 76 dev_data->irq_name, dev_data->irq, rc); 77 goto out; 78 } 79 } else { 80 free_irq(dev_data->irq, dev); 81 dev_data->irq = 0; 82 } 83 dev_data->isr_on = enable; 84 dev_data->ack_intr = enable; 85 out: 86 dev_dbg(&dev->dev, "%s: #%d %s %s%s %s\n", 87 dev_data->irq_name, 88 dev_data->irq, 89 pci_is_enabled(dev) ? "on" : "off", 90 dev->msi_enabled ? "MSI" : "", 91 dev->msix_enabled ? "MSI/X" : "", 92 enable ? (dev_data->isr_on ? "enabled" : "failed to enable") : 93 (dev_data->isr_on ? "failed to disable" : "disabled")); 94 } 95 96 /* Ensure a device is "turned off" and ready to be exported. 97 * (Also see xen_pcibk_config_reset to ensure virtual configuration space is 98 * ready to be re-exported) 99 */ 100 void xen_pcibk_reset_device(struct pci_dev *dev) 101 { 102 u16 cmd; 103 104 xen_pcibk_control_isr(dev, 1 /* reset device */); 105 106 /* Disable devices (but not bridges) */ 107 if (dev->hdr_type == PCI_HEADER_TYPE_NORMAL) { 108 #ifdef CONFIG_PCI_MSI 109 /* The guest could have been abruptly killed without 110 * disabling MSI/MSI-X interrupts.*/ 111 if (dev->msix_enabled) 112 pci_disable_msix(dev); 113 if (dev->msi_enabled) 114 pci_disable_msi(dev); 115 #endif 116 if (pci_is_enabled(dev)) 117 pci_disable_device(dev); 118 119 pci_write_config_word(dev, PCI_COMMAND, 0); 120 121 dev->is_busmaster = 0; 122 } else { 123 pci_read_config_word(dev, PCI_COMMAND, &cmd); 124 if (cmd & (PCI_COMMAND_INVALIDATE)) { 125 cmd &= ~(PCI_COMMAND_INVALIDATE); 126 pci_write_config_word(dev, PCI_COMMAND, cmd); 127 128 dev->is_busmaster = 0; 129 } 130 } 131 } 132 133 #ifdef CONFIG_PCI_MSI 134 static 135 int xen_pcibk_enable_msi(struct xen_pcibk_device *pdev, 136 struct pci_dev *dev, struct xen_pci_op *op) 137 { 138 struct xen_pcibk_dev_data *dev_data; 139 int status; 140 141 if (unlikely(verbose_request)) 142 printk(KERN_DEBUG DRV_NAME ": %s: enable MSI\n", pci_name(dev)); 143 144 status = pci_enable_msi(dev); 145 146 if (status) { 147 pr_warn_ratelimited(DRV_NAME ": %s: error enabling MSI for guest %u: err %d\n", 148 pci_name(dev), pdev->xdev->otherend_id, 149 status); 150 op->value = 0; 151 return XEN_PCI_ERR_op_failed; 152 } 153 154 /* The value the guest needs is actually the IDT vector, not the 155 * the local domain's IRQ number. */ 156 157 op->value = dev->irq ? xen_pirq_from_irq(dev->irq) : 0; 158 if (unlikely(verbose_request)) 159 printk(KERN_DEBUG DRV_NAME ": %s: MSI: %d\n", pci_name(dev), 160 op->value); 161 162 dev_data = pci_get_drvdata(dev); 163 if (dev_data) 164 dev_data->ack_intr = 0; 165 166 return 0; 167 } 168 169 static 170 int xen_pcibk_disable_msi(struct xen_pcibk_device *pdev, 171 struct pci_dev *dev, struct xen_pci_op *op) 172 { 173 struct xen_pcibk_dev_data *dev_data; 174 175 if (unlikely(verbose_request)) 176 printk(KERN_DEBUG DRV_NAME ": %s: disable MSI\n", 177 pci_name(dev)); 178 pci_disable_msi(dev); 179 180 op->value = dev->irq ? xen_pirq_from_irq(dev->irq) : 0; 181 if (unlikely(verbose_request)) 182 printk(KERN_DEBUG DRV_NAME ": %s: MSI: %d\n", pci_name(dev), 183 op->value); 184 dev_data = pci_get_drvdata(dev); 185 if (dev_data) 186 dev_data->ack_intr = 1; 187 return 0; 188 } 189 190 static 191 int xen_pcibk_enable_msix(struct xen_pcibk_device *pdev, 192 struct pci_dev *dev, struct xen_pci_op *op) 193 { 194 struct xen_pcibk_dev_data *dev_data; 195 int i, result; 196 struct msix_entry *entries; 197 198 if (unlikely(verbose_request)) 199 printk(KERN_DEBUG DRV_NAME ": %s: enable MSI-X\n", 200 pci_name(dev)); 201 if (op->value > SH_INFO_MAX_VEC) 202 return -EINVAL; 203 204 entries = kmalloc(op->value * sizeof(*entries), GFP_KERNEL); 205 if (entries == NULL) 206 return -ENOMEM; 207 208 for (i = 0; i < op->value; i++) { 209 entries[i].entry = op->msix_entries[i].entry; 210 entries[i].vector = op->msix_entries[i].vector; 211 } 212 213 result = pci_enable_msix(dev, entries, op->value); 214 215 if (result == 0) { 216 for (i = 0; i < op->value; i++) { 217 op->msix_entries[i].entry = entries[i].entry; 218 if (entries[i].vector) 219 op->msix_entries[i].vector = 220 xen_pirq_from_irq(entries[i].vector); 221 if (unlikely(verbose_request)) 222 printk(KERN_DEBUG DRV_NAME ": %s: " \ 223 "MSI-X[%d]: %d\n", 224 pci_name(dev), i, 225 op->msix_entries[i].vector); 226 } 227 } else 228 pr_warn_ratelimited(DRV_NAME ": %s: error enabling MSI-X for guest %u: err %d!\n", 229 pci_name(dev), pdev->xdev->otherend_id, 230 result); 231 kfree(entries); 232 233 op->value = result; 234 dev_data = pci_get_drvdata(dev); 235 if (dev_data) 236 dev_data->ack_intr = 0; 237 238 return result > 0 ? 0 : result; 239 } 240 241 static 242 int xen_pcibk_disable_msix(struct xen_pcibk_device *pdev, 243 struct pci_dev *dev, struct xen_pci_op *op) 244 { 245 struct xen_pcibk_dev_data *dev_data; 246 if (unlikely(verbose_request)) 247 printk(KERN_DEBUG DRV_NAME ": %s: disable MSI-X\n", 248 pci_name(dev)); 249 pci_disable_msix(dev); 250 251 /* 252 * SR-IOV devices (which don't have any legacy IRQ) have 253 * an undefined IRQ value of zero. 254 */ 255 op->value = dev->irq ? xen_pirq_from_irq(dev->irq) : 0; 256 if (unlikely(verbose_request)) 257 printk(KERN_DEBUG DRV_NAME ": %s: MSI-X: %d\n", pci_name(dev), 258 op->value); 259 dev_data = pci_get_drvdata(dev); 260 if (dev_data) 261 dev_data->ack_intr = 1; 262 return 0; 263 } 264 #endif 265 /* 266 * Now the same evtchn is used for both pcifront conf_read_write request 267 * as well as pcie aer front end ack. We use a new work_queue to schedule 268 * xen_pcibk conf_read_write service for avoiding confict with aer_core 269 * do_recovery job which also use the system default work_queue 270 */ 271 void xen_pcibk_test_and_schedule_op(struct xen_pcibk_device *pdev) 272 { 273 /* Check that frontend is requesting an operation and that we are not 274 * already processing a request */ 275 if (test_bit(_XEN_PCIF_active, (unsigned long *)&pdev->sh_info->flags) 276 && !test_and_set_bit(_PDEVF_op_active, &pdev->flags)) { 277 queue_work(xen_pcibk_wq, &pdev->op_work); 278 } 279 /*_XEN_PCIB_active should have been cleared by pcifront. And also make 280 sure xen_pcibk is waiting for ack by checking _PCIB_op_pending*/ 281 if (!test_bit(_XEN_PCIB_active, (unsigned long *)&pdev->sh_info->flags) 282 && test_bit(_PCIB_op_pending, &pdev->flags)) { 283 wake_up(&xen_pcibk_aer_wait_queue); 284 } 285 } 286 287 /* Performing the configuration space reads/writes must not be done in atomic 288 * context because some of the pci_* functions can sleep (mostly due to ACPI 289 * use of semaphores). This function is intended to be called from a work 290 * queue in process context taking a struct xen_pcibk_device as a parameter */ 291 292 void xen_pcibk_do_op(struct work_struct *data) 293 { 294 struct xen_pcibk_device *pdev = 295 container_of(data, struct xen_pcibk_device, op_work); 296 struct pci_dev *dev; 297 struct xen_pcibk_dev_data *dev_data = NULL; 298 struct xen_pci_op *op = &pdev->sh_info->op; 299 int test_intx = 0; 300 301 dev = xen_pcibk_get_pci_dev(pdev, op->domain, op->bus, op->devfn); 302 303 if (dev == NULL) 304 op->err = XEN_PCI_ERR_dev_not_found; 305 else { 306 dev_data = pci_get_drvdata(dev); 307 if (dev_data) 308 test_intx = dev_data->enable_intx; 309 switch (op->cmd) { 310 case XEN_PCI_OP_conf_read: 311 op->err = xen_pcibk_config_read(dev, 312 op->offset, op->size, &op->value); 313 break; 314 case XEN_PCI_OP_conf_write: 315 op->err = xen_pcibk_config_write(dev, 316 op->offset, op->size, op->value); 317 break; 318 #ifdef CONFIG_PCI_MSI 319 case XEN_PCI_OP_enable_msi: 320 op->err = xen_pcibk_enable_msi(pdev, dev, op); 321 break; 322 case XEN_PCI_OP_disable_msi: 323 op->err = xen_pcibk_disable_msi(pdev, dev, op); 324 break; 325 case XEN_PCI_OP_enable_msix: 326 op->err = xen_pcibk_enable_msix(pdev, dev, op); 327 break; 328 case XEN_PCI_OP_disable_msix: 329 op->err = xen_pcibk_disable_msix(pdev, dev, op); 330 break; 331 #endif 332 default: 333 op->err = XEN_PCI_ERR_not_implemented; 334 break; 335 } 336 } 337 if (!op->err && dev && dev_data) { 338 /* Transition detected */ 339 if ((dev_data->enable_intx != test_intx)) 340 xen_pcibk_control_isr(dev, 0 /* no reset */); 341 } 342 /* Tell the driver domain that we're done. */ 343 wmb(); 344 clear_bit(_XEN_PCIF_active, (unsigned long *)&pdev->sh_info->flags); 345 notify_remote_via_irq(pdev->evtchn_irq); 346 347 /* Mark that we're done. */ 348 smp_mb__before_clear_bit(); /* /after/ clearing PCIF_active */ 349 clear_bit(_PDEVF_op_active, &pdev->flags); 350 smp_mb__after_clear_bit(); /* /before/ final check for work */ 351 352 /* Check to see if the driver domain tried to start another request in 353 * between clearing _XEN_PCIF_active and clearing _PDEVF_op_active. 354 */ 355 xen_pcibk_test_and_schedule_op(pdev); 356 } 357 358 irqreturn_t xen_pcibk_handle_event(int irq, void *dev_id) 359 { 360 struct xen_pcibk_device *pdev = dev_id; 361 362 xen_pcibk_test_and_schedule_op(pdev); 363 364 return IRQ_HANDLED; 365 } 366 static irqreturn_t xen_pcibk_guest_interrupt(int irq, void *dev_id) 367 { 368 struct pci_dev *dev = (struct pci_dev *)dev_id; 369 struct xen_pcibk_dev_data *dev_data = pci_get_drvdata(dev); 370 371 if (dev_data->isr_on && dev_data->ack_intr) { 372 dev_data->handled++; 373 if ((dev_data->handled % 1000) == 0) { 374 if (xen_test_irq_shared(irq)) { 375 printk(KERN_INFO "%s IRQ line is not shared " 376 "with other domains. Turning ISR off\n", 377 dev_data->irq_name); 378 dev_data->ack_intr = 0; 379 } 380 } 381 return IRQ_HANDLED; 382 } 383 return IRQ_NONE; 384 } 385