hw/msix.c

   1 /*
   2  * MSI-X device support
   3  *
   4  * This module includes support for MSI-X in pci devices.
   5  *
   6  * Author: Michael S. Tsirkin <mst@redhat.com>
   7  *
   8  *  Copyright (c) 2009, Red Hat Inc, Michael S. Tsirkin (mst@redhat.com)
   9  *
  10  * This work is licensed under the terms of the GNU GPL, version 2.  See
  11  * the COPYING file in the top-level directory.
  12  */
  13
  14 #include "hw.h"
  15 #include "msix.h"
  16 #include "pci.h"
  17 #include "range.h"
  18
  19 /* MSI-X capability structure */
  20 #define MSIX_TABLE_OFFSET 4
  21 #define MSIX_PBA_OFFSET 8
  22 #define MSIX_CAP_LENGTH 12
  23
  24 /* MSI enable bit and maskall bit are in byte 1 in FLAGS register */
  25 #define MSIX_CONTROL_OFFSET (PCI_MSIX_FLAGS + 1)
  26 #define MSIX_ENABLE_MASK (PCI_MSIX_FLAGS_ENABLE >> 8)
  27 #define MSIX_MASKALL_MASK (PCI_MSIX_FLAGS_MASKALL >> 8)
  28
  29 /* MSI-X table format */
  30 #define MSIX_MSG_ADDR 0
  31 #define MSIX_MSG_UPPER_ADDR 4
  32 #define MSIX_MSG_DATA 8
  33 #define MSIX_VECTOR_CTRL 12
  34 #define MSIX_ENTRY_SIZE 16
  35 #define MSIX_VECTOR_MASK 0x1
  36
  37 /* How much space does an MSIX table need. */
  38 /* The spec requires giving the table structure
  39  * a 4K aligned region all by itself. */
  40 #define MSIX_PAGE_SIZE 0x1000
  41 /* Reserve second half of the page for pending bits */
  42 #define MSIX_PAGE_PENDING (MSIX_PAGE_SIZE / 2)
  43 #define MSIX_MAX_ENTRIES 32
  44
  45
  46 /* Flag for interrupt controller to declare MSI-X support */
  47 int msix_supported;
  48
  49 /* Add MSI-X capability to the config space for the device. */
  50 /* Given a bar and its size, add MSI-X table on top of it
  51  * and fill MSI-X capability in the config space.
  52  * Original bar size must be a power of 2 or 0.
  53  * New bar size is returned. */
  54 static int msix_add_config(struct PCIDevice *pdev, unsigned short nentries,
  55                            unsigned bar_nr, unsigned bar_size)
  56 {
  57     int config_offset;
  58     uint8_t *config;
  59     uint32_t new_size;
  60
  61     if (nentries < 1 || nentries > PCI_MSIX_FLAGS_QSIZE + 1)
  62         return -EINVAL;
  63     if (bar_size > 0x80000000)
  64         return -ENOSPC;
  65
  66     /* Add space for MSI-X structures */
  67     if (!bar_size) {
  68         new_size = MSIX_PAGE_SIZE;
  69     } else if (bar_size < MSIX_PAGE_SIZE) {
  70         bar_size = MSIX_PAGE_SIZE;
  71         new_size = MSIX_PAGE_SIZE * 2;
  72     } else {
  73         new_size = bar_size * 2;
  74     }
  75
  76     pdev->msix_bar_size = new_size;
  77     config_offset = pci_add_capability(pdev, PCI_CAP_ID_MSIX,
  78                                        0, MSIX_CAP_LENGTH);
  79     if (config_offset < 0)
  80         return config_offset;
  81     config = pdev->config + config_offset;
  82
  83     pci_set_word(config + PCI_MSIX_FLAGS, nentries - 1);
  84     /* Table on top of BAR */
  85     pci_set_long(config + MSIX_TABLE_OFFSET, bar_size | bar_nr);
  86     /* Pending bits on top of that */
  87     pci_set_long(config + MSIX_PBA_OFFSET, (bar_size + MSIX_PAGE_PENDING) |
  88                  bar_nr);
  89     pdev->msix_cap = config_offset;
  90     /* Make flags bit writeable. */
  91     pdev->wmask[config_offset + MSIX_CONTROL_OFFSET] |= MSIX_ENABLE_MASK |
  92             MSIX_MASKALL_MASK;
  93     return 0;
  94 }
  95
  96 static uint32_t msix_mmio_readl(void *opaque, target_phys_addr_t addr)
  97 {
  98     PCIDevice *dev = opaque;
  99     unsigned int offset = addr & (MSIX_PAGE_SIZE - 1) & ~0x3;
 100     void *page = dev->msix_table_page;
 101
 102     return pci_get_long(page + offset);
 103 }
 104
 105 static uint32_t msix_mmio_read_unallowed(void *opaque, target_phys_addr_t addr)
 106 {
 107     fprintf(stderr, "MSI-X: only dword read is allowed!\n");
 108     return 0;
 109 }
 110
 111 static uint8_t msix_pending_mask(int vector)
 112 {
 113     return 1 << (vector % 8);
 114 }
 115
 116 static uint8_t *msix_pending_byte(PCIDevice *dev, int vector)
 117 {
 118     return dev->msix_table_page + MSIX_PAGE_PENDING + vector / 8;
 119 }
 120
 121 static int msix_is_pending(PCIDevice *dev, int vector)
 122 {
 123     return *msix_pending_byte(dev, vector) & msix_pending_mask(vector);
 124 }
 125
 126 static void msix_set_pending(PCIDevice *dev, int vector)
 127 {
 128     *msix_pending_byte(dev, vector) |= msix_pending_mask(vector);
 129 }
 130
 131 static void msix_clr_pending(PCIDevice *dev, int vector)
 132 {
 133     *msix_pending_byte(dev, vector) &= ~msix_pending_mask(vector);
 134 }
 135
 136 static int msix_function_masked(PCIDevice *dev)
 137 {
 138     return dev->config[dev->msix_cap + MSIX_CONTROL_OFFSET] & MSIX_MASKALL_MASK;
 139 }
 140
 141 static int msix_is_masked(PCIDevice *dev, int vector)
 142 {
 143     unsigned offset = vector * MSIX_ENTRY_SIZE + MSIX_VECTOR_CTRL;
 144     return msix_function_masked(dev) ||
 145            dev->msix_table_page[offset] & MSIX_VECTOR_MASK;
 146 }
 147
 148 static void msix_handle_mask_update(PCIDevice *dev, int vector)
 149 {
 150     if (!msix_is_masked(dev, vector) && msix_is_pending(dev, vector)) {
 151         msix_clr_pending(dev, vector);
 152         msix_notify(dev, vector);
 153     }
 154 }
 155
 156 /* Handle MSI-X capability config write. */
 157 void msix_write_config(PCIDevice *dev, uint32_t addr,
 158                        uint32_t val, int len)
 159 {
 160     unsigned enable_pos = dev->msix_cap + MSIX_CONTROL_OFFSET;
 161     int vector;
 162     int i;
 163
 164     if (!range_covers_byte(addr, len, enable_pos)) {
 165         return;
 166     }
 167
 168     if (!msix_enabled(dev)) {
 169         return;
 170     }
 171
 172     for (i = 0; i < PCI_NUM_PINS; ++i) {
 173         qemu_set_irq(dev->irq[i], 0);
 174     }
 175
 176     if (msix_function_masked(dev)) {
 177         return;
 178     }
 179
 180     for (vector = 0; vector < dev->msix_entries_nr; ++vector) {
 181         msix_handle_mask_update(dev, vector);
 182     }
 183 }
 184
 185 static void msix_mmio_writel(void *opaque, target_phys_addr_t addr,
 186                              uint32_t val)
 187 {
 188     PCIDevice *dev = opaque;
 189     unsigned int offset = addr & (MSIX_PAGE_SIZE - 1) & ~0x3;
 190     int vector = offset / MSIX_ENTRY_SIZE;
 191     pci_set_long(dev->msix_table_page + offset, val);
 192     msix_handle_mask_update(dev, vector);
 193 }
 194
 195 static void msix_mmio_write_unallowed(void *opaque, target_phys_addr_t addr,
 196                                       uint32_t val)
 197 {
 198     fprintf(stderr, "MSI-X: only dword write is allowed!\n");
 199 }
 200
 201 static CPUWriteMemoryFunc * const msix_mmio_write[] = {
 202     msix_mmio_write_unallowed, msix_mmio_write_unallowed, msix_mmio_writel
 203 };
 204
 205 static CPUReadMemoryFunc * const msix_mmio_read[] = {
 206     msix_mmio_read_unallowed, msix_mmio_read_unallowed, msix_mmio_readl
 207 };
 208
 209 /* Should be called from device's map method. */
 210 void msix_mmio_map(PCIDevice *d, int region_num,
 211                    pcibus_t addr, pcibus_t size, int type)
 212 {
 213     uint8_t *config = d->config + d->msix_cap;
 214     uint32_t table = pci_get_long(config + MSIX_TABLE_OFFSET);
 215     uint32_t offset = table & ~(MSIX_PAGE_SIZE - 1);
 216     /* TODO: for assigned devices, we'll want to make it possible to map
 217      * pending bits separately in case they are in a separate bar. */
 218     int table_bir = table & PCI_MSIX_FLAGS_BIRMASK;
 219
 220     if (table_bir != region_num)
 221         return;
 222     if (size <= offset)
 223         return;
 224     cpu_register_physical_memory(addr + offset, size - offset,
 225                                  d->msix_mmio_index);
 226 }
 227
 228 static void msix_mask_all(struct PCIDevice *dev, unsigned nentries)
 229 {
 230     int vector;
 231     for (vector = 0; vector < nentries; ++vector) {
 232         unsigned offset = vector * MSIX_ENTRY_SIZE + MSIX_VECTOR_CTRL;
 233         dev->msix_table_page[offset] |= MSIX_VECTOR_MASK;
 234     }
 235 }
 236
 237 /* Initialize the MSI-X structures. Note: if MSI-X is supported, BAR size is
 238  * modified, it should be retrieved with msix_bar_size. */
 239 int msix_init(struct PCIDevice *dev, unsigned short nentries,
 240               unsigned bar_nr, unsigned bar_size)
 241 {
 242     int ret;
 243     /* Nothing to do if MSI is not supported by interrupt controller */
 244     if (!msix_supported)
 245         return -ENOTSUP;
 246
 247     if (nentries > MSIX_MAX_ENTRIES)
 248         return -EINVAL;
 249
 250     dev->msix_entry_used = qemu_mallocz(MSIX_MAX_ENTRIES *
 251                                         sizeof *dev->msix_entry_used);
 252
 253     dev->msix_table_page = qemu_mallocz(MSIX_PAGE_SIZE);
 254     msix_mask_all(dev, nentries);
 255
 256     dev->msix_mmio_index = cpu_register_io_memory(msix_mmio_read,
 257                                                   msix_mmio_write, dev,
 258                                                   DEVICE_NATIVE_ENDIAN);
 259     if (dev->msix_mmio_index == -1) {
 260         ret = -EBUSY;
 261         goto err_index;
 262     }
 263
 264     dev->msix_entries_nr = nentries;
 265     ret = msix_add_config(dev, nentries, bar_nr, bar_size);
 266     if (ret)
 267         goto err_config;
 268
 269     dev->cap_present |= QEMU_PCI_CAP_MSIX;
 270     return 0;
 271
 272 err_config:
 273     dev->msix_entries_nr = 0;
 274     cpu_unregister_io_memory(dev->msix_mmio_index);
 275 err_index:
 276     qemu_free(dev->msix_table_page);
 277     dev->msix_table_page = NULL;
 278     qemu_free(dev->msix_entry_used);
 279     dev->msix_entry_used = NULL;
 280     return ret;
 281 }
 282
 283 static void msix_free_irq_entries(PCIDevice *dev)
 284 {
 285     int vector;
 286
 287     for (vector = 0; vector < dev->msix_entries_nr; ++vector) {
 288         dev->msix_entry_used[vector] = 0;
 289         msix_clr_pending(dev, vector);
 290     }
 291 }
 292
 293 /* Clean up resources for the device. */
 294 int msix_uninit(PCIDevice *dev)
 295 {
 296     if (!(dev->cap_present & QEMU_PCI_CAP_MSIX))
 297         return 0;
 298     pci_del_capability(dev, PCI_CAP_ID_MSIX, MSIX_CAP_LENGTH);
 299     dev->msix_cap = 0;
 300     msix_free_irq_entries(dev);
 301     dev->msix_entries_nr = 0;
 302     cpu_unregister_io_memory(dev->msix_mmio_index);
 303     qemu_free(dev->msix_table_page);
 304     dev->msix_table_page = NULL;
 305     qemu_free(dev->msix_entry_used);
 306     dev->msix_entry_used = NULL;
 307     dev->cap_present &= ~QEMU_PCI_CAP_MSIX;
 308     return 0;
 309 }
 310
 311 void msix_save(PCIDevice *dev, QEMUFile *f)
 312 {
 313     unsigned n = dev->msix_entries_nr;
 314
 315     if (!(dev->cap_present & QEMU_PCI_CAP_MSIX)) {
 316         return;
 317     }
 318
 319     qemu_put_buffer(f, dev->msix_table_page, n * MSIX_ENTRY_SIZE);
 320     qemu_put_buffer(f, dev->msix_table_page + MSIX_PAGE_PENDING, (n + 7) / 8);
 321 }
 322
 323 /* Should be called after restoring the config space. */
 324 void msix_load(PCIDevice *dev, QEMUFile *f)
 325 {
 326     unsigned n = dev->msix_entries_nr;
 327
 328     if (!(dev->cap_present & QEMU_PCI_CAP_MSIX)) {
 329         return;
 330     }
 331
 332     msix_free_irq_entries(dev);
 333     qemu_get_buffer(f, dev->msix_table_page, n * MSIX_ENTRY_SIZE);
 334     qemu_get_buffer(f, dev->msix_table_page + MSIX_PAGE_PENDING, (n + 7) / 8);
 335 }
 336
 337 /* Does device support MSI-X? */
 338 int msix_present(PCIDevice *dev)
 339 {
 340     return dev->cap_present & QEMU_PCI_CAP_MSIX;
 341 }
 342
 343 /* Is MSI-X enabled? */
 344 int msix_enabled(PCIDevice *dev)
 345 {
 346     return (dev->cap_present & QEMU_PCI_CAP_MSIX) &&
 347         (dev->config[dev->msix_cap + MSIX_CONTROL_OFFSET] &
 348          MSIX_ENABLE_MASK);
 349 }
 350
 351 /* Size of bar where MSI-X table resides, or 0 if MSI-X not supported. */
 352 uint32_t msix_bar_size(PCIDevice *dev)
 353 {
 354     return (dev->cap_present & QEMU_PCI_CAP_MSIX) ?
 355         dev->msix_bar_size : 0;
 356 }
 357
 358 /* Send an MSI-X message */
 359 void msix_notify(PCIDevice *dev, unsigned vector)
 360 {
 361     uint8_t *table_entry = dev->msix_table_page + vector * MSIX_ENTRY_SIZE;
 362     uint64_t address;
 363     uint32_t data;
 364
 365     if (vector >= dev->msix_entries_nr || !dev->msix_entry_used[vector])
 366         return;
 367     if (msix_is_masked(dev, vector)) {
 368         msix_set_pending(dev, vector);
 369         return;
 370     }
 371
 372     address = pci_get_long(table_entry + MSIX_MSG_UPPER_ADDR);
 373     address = (address << 32) | pci_get_long(table_entry + MSIX_MSG_ADDR);
 374     data = pci_get_long(table_entry + MSIX_MSG_DATA);
 375     stl_phys(address, data);
 376 }
 377
 378 void msix_reset(PCIDevice *dev)
 379 {
 380     if (!(dev->cap_present & QEMU_PCI_CAP_MSIX))
 381         return;
 382     msix_free_irq_entries(dev);
 383     dev->config[dev->msix_cap + MSIX_CONTROL_OFFSET] &=
 384             ~dev->wmask[dev->msix_cap + MSIX_CONTROL_OFFSET];
 385     memset(dev->msix_table_page, 0, MSIX_PAGE_SIZE);
 386     msix_mask_all(dev, dev->msix_entries_nr);
 387 }
 388
 389 /* PCI spec suggests that devices make it possible for software to configure
 390  * less vectors than supported by the device, but does not specify a standard
 391  * mechanism for devices to do so.
 392  *
 393  * We support this by asking devices to declare vectors software is going to
 394  * actually use, and checking this on the notification path. Devices that
 395  * don't want to follow the spec suggestion can declare all vectors as used. */
 396
 397 /* Mark vector as used. */
 398 int msix_vector_use(PCIDevice *dev, unsigned vector)
 399 {
 400     if (vector >= dev->msix_entries_nr)
 401         return -EINVAL;
 402     dev->msix_entry_used[vector]++;
 403     return 0;
 404 }
 405
 406 /* Mark vector as unused. */
 407 void msix_vector_unuse(PCIDevice *dev, unsigned vector)
 408 {
 409     if (vector >= dev->msix_entries_nr || !dev->msix_entry_used[vector]) {
 410         return;
 411     }
 412     if (--dev->msix_entry_used[vector]) {
 413         return;
 414     }
 415     msix_clr_pending(dev, vector);
 416 }
 417
 418 void msix_unuse_all_vectors(PCIDevice *dev)
 419 {
 420     if (!(dev->cap_present & QEMU_PCI_CAP_MSIX))
 421         return;
 422     msix_free_irq_entries(dev);
 423 }