drivers/xen/xen-pciback/pciback_ops.c

   1 /*
   2  * PCI Backend Operations - respond to PCI requests from Frontend
   3  *
   4  *   Author: Ryan Wilson <hap9@epoch.ncsc.mil>
   5  */
   6
   7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
   8
   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>
  14 #include "pciback.h"
  15
  16 int verbose_request;
  17 module_param(verbose_request, int, 0644);
  18
  19 static irqreturn_t xen_pcibk_guest_interrupt(int irq, void *dev_id);
  20
  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.
  24  */
  25 static void xen_pcibk_control_isr(struct pci_dev *dev, int reset)
  26 {
  27         struct xen_pcibk_dev_data *dev_data;
  28         int rc;
  29         int enable = 0;
  30
  31         dev_data = pci_get_drvdata(dev);
  32         if (!dev_data)
  33                 return;
  34
  35         /* We don't deal with bridges */
  36         if (dev->hdr_type != PCI_HEADER_TYPE_NORMAL)
  37                 return;
  38
  39         if (reset) {
  40                 dev_data->enable_intx = 0;
  41                 dev_data->ack_intr = 0;
  42         }
  43         enable =  dev_data->enable_intx;
  44
  45         /* Asked to disable, but ISR isn't runnig */
  46         if (!enable && !dev_data->isr_on)
  47                 return;
  48
  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.
  52          */
  53         if (enable)
  54                 dev_data->irq = dev->irq;
  55
  56         /*
  57          * SR-IOV devices in all use MSI-X and have no legacy
  58          * interrupts, so inhibit creating a fake IRQ handler for them.
  59          */
  60         if (dev_data->irq == 0)
  61                 goto out;
  62
  63         dev_dbg(&dev->dev, "%s: #%d %s %s%s %s-> %s\n",
  64                 dev_data->irq_name,
  65                 dev_data->irq,
  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");
  71
  72         if (enable) {
  73                 /*
  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.
  76                  */
  77                 if (dev->msi_enabled || dev->msix_enabled)
  78                         goto out;
  79
  80                 rc = request_irq(dev_data->irq,
  81                                 xen_pcibk_guest_interrupt, IRQF_SHARED,
  82                                 dev_data->irq_name, dev);
  83                 if (rc) {
  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);
  87                         goto out;
  88                 }
  89         } else {
  90                 free_irq(dev_data->irq, dev);
  91                 dev_data->irq = 0;
  92         }
  93         dev_data->isr_on = enable;
  94         dev_data->ack_intr = enable;
  95 out:
  96         dev_dbg(&dev->dev, "%s: #%d %s %s%s %s\n",
  97                 dev_data->irq_name,
  98                 dev_data->irq,
  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"));
 104 }
 105
 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)
 109  */
 110 void xen_pcibk_reset_device(struct pci_dev *dev)
 111 {
 112         u16 cmd;
 113
 114         xen_pcibk_control_isr(dev, 1 /* reset device */);
 115
 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);
 125 #endif
 126                 if (pci_is_enabled(dev))
 127                         pci_disable_device(dev);
 128
 129                 pci_write_config_word(dev, PCI_COMMAND, 0);
 130
 131                 dev->is_busmaster = 0;
 132         } else {
 133                 pci_read_config_word(dev, PCI_COMMAND, &cmd);
 134                 if (cmd & (PCI_COMMAND_INVALIDATE)) {
 135                         cmd &= ~(PCI_COMMAND_INVALIDATE);
 136                         pci_write_config_word(dev, PCI_COMMAND, cmd);
 137
 138                         dev->is_busmaster = 0;
 139                 }
 140         }
 141 }
 142
 143 #ifdef CONFIG_PCI_MSI
 144 static
 145 int xen_pcibk_enable_msi(struct xen_pcibk_device *pdev,
 146                          struct pci_dev *dev, struct xen_pci_op *op)
 147 {
 148         struct xen_pcibk_dev_data *dev_data;
 149         int status;
 150
 151         if (unlikely(verbose_request))
 152                 printk(KERN_DEBUG DRV_NAME ": %s: enable MSI\n", pci_name(dev));
 153
 154         if (dev->msi_enabled)
 155                 status = -EALREADY;
 156         else if (dev->msix_enabled)
 157                 status = -ENXIO;
 158         else
 159                 status = pci_enable_msi(dev);
 160
 161         if (status) {
 162                 pr_warn_ratelimited("%s: error enabling MSI for guest %u: err %d\n",
 163                                     pci_name(dev), pdev->xdev->otherend_id,
 164                                     status);
 165                 op->value = 0;
 166                 return XEN_PCI_ERR_op_failed;
 167         }
 168
 169         /* The value the guest needs is actually the IDT vector, not the
 170          * the local domain's IRQ number. */
 171
 172         op->value = dev->irq ? xen_pirq_from_irq(dev->irq) : 0;
 173         if (unlikely(verbose_request))
 174                 printk(KERN_DEBUG DRV_NAME ": %s: MSI: %d\n", pci_name(dev),
 175                         op->value);
 176
 177         dev_data = pci_get_drvdata(dev);
 178         if (dev_data)
 179                 dev_data->ack_intr = 0;
 180
 181         return 0;
 182 }
 183
 184 static
 185 int xen_pcibk_disable_msi(struct xen_pcibk_device *pdev,
 186                           struct pci_dev *dev, struct xen_pci_op *op)
 187 {
 188         struct xen_pcibk_dev_data *dev_data;
 189
 190         if (unlikely(verbose_request))
 191                 printk(KERN_DEBUG DRV_NAME ": %s: disable MSI\n",
 192                        pci_name(dev));
 193         pci_disable_msi(dev);
 194
 195         op->value = dev->irq ? xen_pirq_from_irq(dev->irq) : 0;
 196         if (unlikely(verbose_request))
 197                 printk(KERN_DEBUG DRV_NAME ": %s: MSI: %d\n", pci_name(dev),
 198                         op->value);
 199         dev_data = pci_get_drvdata(dev);
 200         if (dev_data)
 201                 dev_data->ack_intr = 1;
 202         return 0;
 203 }
 204
 205 static
 206 int xen_pcibk_enable_msix(struct xen_pcibk_device *pdev,
 207                           struct pci_dev *dev, struct xen_pci_op *op)
 208 {
 209         struct xen_pcibk_dev_data *dev_data;
 210         int i, result;
 211         struct msix_entry *entries;
 212
 213         if (unlikely(verbose_request))
 214                 printk(KERN_DEBUG DRV_NAME ": %s: enable MSI-X\n",
 215                        pci_name(dev));
 216
 217         if (op->value > SH_INFO_MAX_VEC)
 218                 return -EINVAL;
 219
 220         if (dev->msix_enabled)
 221                 return -EALREADY;
 222
 223         if (dev->msi_enabled)
 224                 return -ENXIO;
 225
 226         entries = kmalloc(op->value * sizeof(*entries), GFP_KERNEL);
 227         if (entries == NULL)
 228                 return -ENOMEM;
 229
 230         for (i = 0; i < op->value; i++) {
 231                 entries[i].entry = op->msix_entries[i].entry;
 232                 entries[i].vector = op->msix_entries[i].vector;
 233         }
 234
 235         result = pci_enable_msix_exact(dev, entries, op->value);
 236         if (result == 0) {
 237                 for (i = 0; i < op->value; i++) {
 238                         op->msix_entries[i].entry = entries[i].entry;
 239                         if (entries[i].vector) {
 240                                 op->msix_entries[i].vector =
 241                                         xen_pirq_from_irq(entries[i].vector);
 242                                 if (unlikely(verbose_request))
 243                                         printk(KERN_DEBUG DRV_NAME ": %s: " \
 244                                                 "MSI-X[%d]: %d\n",
 245                                                 pci_name(dev), i,
 246                                                 op->msix_entries[i].vector);
 247                         }
 248                 }
 249         } else
 250                 pr_warn_ratelimited("%s: error enabling MSI-X for guest %u: err %d!\n",
 251                                     pci_name(dev), pdev->xdev->otherend_id,
 252                                     result);
 253         kfree(entries);
 254
 255         op->value = result;
 256         dev_data = pci_get_drvdata(dev);
 257         if (dev_data)
 258                 dev_data->ack_intr = 0;
 259
 260         return result > 0 ? 0 : result;
 261 }
 262
 263 static
 264 int xen_pcibk_disable_msix(struct xen_pcibk_device *pdev,
 265                            struct pci_dev *dev, struct xen_pci_op *op)
 266 {
 267         struct xen_pcibk_dev_data *dev_data;
 268         if (unlikely(verbose_request))
 269                 printk(KERN_DEBUG DRV_NAME ": %s: disable MSI-X\n",
 270                         pci_name(dev));
 271         pci_disable_msix(dev);
 272
 273         /*
 274          * SR-IOV devices (which don't have any legacy IRQ) have
 275          * an undefined IRQ value of zero.
 276          */
 277         op->value = dev->irq ? xen_pirq_from_irq(dev->irq) : 0;
 278         if (unlikely(verbose_request))
 279                 printk(KERN_DEBUG DRV_NAME ": %s: MSI-X: %d\n", pci_name(dev),
 280                         op->value);
 281         dev_data = pci_get_drvdata(dev);
 282         if (dev_data)
 283                 dev_data->ack_intr = 1;
 284         return 0;
 285 }
 286 #endif
 287 /*
 288 * Now the same evtchn is used for both pcifront conf_read_write request
 289 * as well as pcie aer front end ack. We use a new work_queue to schedule
 290 * xen_pcibk conf_read_write service for avoiding confict with aer_core
 291 * do_recovery job which also use the system default work_queue
 292 */
 293 void xen_pcibk_test_and_schedule_op(struct xen_pcibk_device *pdev)
 294 {
 295         /* Check that frontend is requesting an operation and that we are not
 296          * already processing a request */
 297         if (test_bit(_XEN_PCIF_active, (unsigned long *)&pdev->sh_info->flags)
 298             && !test_and_set_bit(_PDEVF_op_active, &pdev->flags)) {
 299                 queue_work(xen_pcibk_wq, &pdev->op_work);
 300         }
 301         /*_XEN_PCIB_active should have been cleared by pcifront. And also make
 302         sure xen_pcibk is waiting for ack by checking _PCIB_op_pending*/
 303         if (!test_bit(_XEN_PCIB_active, (unsigned long *)&pdev->sh_info->flags)
 304             && test_bit(_PCIB_op_pending, &pdev->flags)) {
 305                 wake_up(&xen_pcibk_aer_wait_queue);
 306         }
 307 }
 308
 309 /* Performing the configuration space reads/writes must not be done in atomic
 310  * context because some of the pci_* functions can sleep (mostly due to ACPI
 311  * use of semaphores). This function is intended to be called from a work
 312  * queue in process context taking a struct xen_pcibk_device as a parameter */
 313
 314 void xen_pcibk_do_op(struct work_struct *data)
 315 {
 316         struct xen_pcibk_device *pdev =
 317                 container_of(data, struct xen_pcibk_device, op_work);
 318         struct pci_dev *dev;
 319         struct xen_pcibk_dev_data *dev_data = NULL;
 320         struct xen_pci_op *op = &pdev->op;
 321         int test_intx = 0;
 322
 323         *op = pdev->sh_info->op;
 324         barrier();
 325         dev = xen_pcibk_get_pci_dev(pdev, op->domain, op->bus, op->devfn);
 326
 327         if (dev == NULL)
 328                 op->err = XEN_PCI_ERR_dev_not_found;
 329         else {
 330                 dev_data = pci_get_drvdata(dev);
 331                 if (dev_data)
 332                         test_intx = dev_data->enable_intx;
 333                 switch (op->cmd) {
 334                 case XEN_PCI_OP_conf_read:
 335                         op->err = xen_pcibk_config_read(dev,
 336                                   op->offset, op->size, &op->value);
 337                         break;
 338                 case XEN_PCI_OP_conf_write:
 339                         op->err = xen_pcibk_config_write(dev,
 340                                   op->offset, op->size, op->value);
 341                         break;
 342 #ifdef CONFIG_PCI_MSI
 343                 case XEN_PCI_OP_enable_msi:
 344                         op->err = xen_pcibk_enable_msi(pdev, dev, op);
 345                         break;
 346                 case XEN_PCI_OP_disable_msi:
 347                         op->err = xen_pcibk_disable_msi(pdev, dev, op);
 348                         break;
 349                 case XEN_PCI_OP_enable_msix:
 350                         op->err = xen_pcibk_enable_msix(pdev, dev, op);
 351                         break;
 352                 case XEN_PCI_OP_disable_msix:
 353                         op->err = xen_pcibk_disable_msix(pdev, dev, op);
 354                         break;
 355 #endif
 356                 default:
 357                         op->err = XEN_PCI_ERR_not_implemented;
 358                         break;
 359                 }
 360         }
 361         if (!op->err && dev && dev_data) {
 362                 /* Transition detected */
 363                 if ((dev_data->enable_intx != test_intx))
 364                         xen_pcibk_control_isr(dev, 0 /* no reset */);
 365         }
 366         pdev->sh_info->op.err = op->err;
 367         pdev->sh_info->op.value = op->value;
 368 #ifdef CONFIG_PCI_MSI
 369         if (op->cmd == XEN_PCI_OP_enable_msix && op->err == 0) {
 370                 unsigned int i;
 371
 372                 for (i = 0; i < op->value; i++)
 373                         pdev->sh_info->op.msix_entries[i].vector =
 374                                 op->msix_entries[i].vector;
 375         }
 376 #endif
 377         /* Tell the driver domain that we're done. */
 378         wmb();
 379         clear_bit(_XEN_PCIF_active, (unsigned long *)&pdev->sh_info->flags);
 380         notify_remote_via_irq(pdev->evtchn_irq);
 381
 382         /* Mark that we're done. */
 383         smp_mb__before_atomic(); /* /after/ clearing PCIF_active */
 384         clear_bit(_PDEVF_op_active, &pdev->flags);
 385         smp_mb__after_atomic(); /* /before/ final check for work */
 386
 387         /* Check to see if the driver domain tried to start another request in
 388          * between clearing _XEN_PCIF_active and clearing _PDEVF_op_active.
 389         */
 390         xen_pcibk_test_and_schedule_op(pdev);
 391 }
 392
 393 irqreturn_t xen_pcibk_handle_event(int irq, void *dev_id)
 394 {
 395         struct xen_pcibk_device *pdev = dev_id;
 396
 397         xen_pcibk_test_and_schedule_op(pdev);
 398
 399         return IRQ_HANDLED;
 400 }
 401 static irqreturn_t xen_pcibk_guest_interrupt(int irq, void *dev_id)
 402 {
 403         struct pci_dev *dev = (struct pci_dev *)dev_id;
 404         struct xen_pcibk_dev_data *dev_data = pci_get_drvdata(dev);
 405
 406         if (dev_data->isr_on && dev_data->ack_intr) {
 407                 dev_data->handled++;
 408                 if ((dev_data->handled % 1000) == 0) {
 409                         if (xen_test_irq_shared(irq)) {
 410                                 pr_info("%s IRQ line is not shared "
 411                                         "with other domains. Turning ISR off\n",
 412                                          dev_data->irq_name);
 413                                 dev_data->ack_intr = 0;
 414                         }
 415                 }
 416                 return IRQ_HANDLED;
 417         }
 418         return IRQ_NONE;
 419 }