4 * This module includes support for MSI-X in pci devices.
6 * Author: Michael S. Tsirkin <mst@redhat.com>
8 * Copyright (c) 2009, Red Hat Inc, Michael S. Tsirkin (mst@redhat.com)
10 * This work is licensed under the terms of the GNU GPL, version 2. See
11 * the COPYING file in the top-level directory.
18 /* Declaration from linux/pci_regs.h */
19 #define PCI_CAP_ID_MSIX 0x11 /* MSI-X */
20 #define PCI_MSIX_FLAGS 2 /* Table at lower 11 bits */
21 #define PCI_MSIX_FLAGS_QSIZE 0x7FF
22 #define PCI_MSIX_FLAGS_ENABLE (1 << 15)
23 #define PCI_MSIX_FLAGS_BIRMASK (7 << 0)
25 /* MSI-X capability structure */
26 #define MSIX_TABLE_OFFSET 4
27 #define MSIX_PBA_OFFSET 8
28 #define MSIX_CAP_LENGTH 12
30 /* MSI enable bit is in byte 1 in FLAGS register */
31 #define MSIX_ENABLE_OFFSET (PCI_MSIX_FLAGS + 1)
32 #define MSIX_ENABLE_MASK (PCI_MSIX_FLAGS_ENABLE >> 8)
34 /* MSI-X table format */
35 #define MSIX_MSG_ADDR 0
36 #define MSIX_MSG_UPPER_ADDR 4
37 #define MSIX_MSG_DATA 8
38 #define MSIX_VECTOR_CTRL 12
39 #define MSIX_ENTRY_SIZE 16
40 #define MSIX_VECTOR_MASK 0x1
42 /* How much space does an MSIX table need. */
43 /* The spec requires giving the table structure
44 * a 4K aligned region all by itself. Align it to
45 * target pages so that drivers can do passthrough
46 * on the rest of the region. */
47 #define MSIX_PAGE_SIZE TARGET_PAGE_ALIGN(0x1000)
48 /* Reserve second half of the page for pending bits */
49 #define MSIX_PAGE_PENDING (MSIX_PAGE_SIZE / 2)
50 #define MSIX_MAX_ENTRIES 32
54 #define DEBUG(fmt, ...) \
56 fprintf(stderr, "%s: " fmt, __func__ , __VA_ARGS__); \
59 #define DEBUG(fmt, ...) do { } while(0)
62 /* Flag for interrupt controller to declare MSI-X support */
65 /* Add MSI-X capability to the config space for the device. */
66 /* Given a bar and its size, add MSI-X table on top of it
67 * and fill MSI-X capability in the config space.
68 * Original bar size must be a power of 2 or 0.
69 * New bar size is returned. */
70 static int msix_add_config(struct PCIDevice *pdev, unsigned short nentries,
71 unsigned bar_nr, unsigned bar_size)
77 if (nentries < 1 || nentries > PCI_MSIX_FLAGS_QSIZE + 1)
79 if (bar_size > 0x80000000)
82 /* Add space for MSI-X structures */
84 new_size = MSIX_PAGE_SIZE;
85 else if (bar_size < MSIX_PAGE_SIZE) {
86 bar_size = MSIX_PAGE_SIZE;
87 new_size = MSIX_PAGE_SIZE * 2;
89 new_size = bar_size * 2;
91 pdev->msix_bar_size = new_size;
92 config_offset = pci_add_capability(pdev, PCI_CAP_ID_MSIX, MSIX_CAP_LENGTH);
93 if (config_offset < 0)
95 config = pdev->config + config_offset;
97 pci_set_word(config + PCI_MSIX_FLAGS, nentries - 1);
98 /* Table on top of BAR */
99 pci_set_long(config + MSIX_TABLE_OFFSET, bar_size | bar_nr);
100 /* Pending bits on top of that */
101 pci_set_long(config + MSIX_PBA_OFFSET, (bar_size + MSIX_PAGE_PENDING) |
103 pdev->msix_cap = config_offset;
104 /* Make flags bit writeable. */
105 pdev->wmask[config_offset + MSIX_ENABLE_OFFSET] |= MSIX_ENABLE_MASK;
109 static void msix_free_irq_entries(PCIDevice *dev)
113 for (vector = 0; vector < dev->msix_entries_nr; ++vector)
114 dev->msix_entry_used[vector] = 0;
117 /* Handle MSI-X capability config write. */
118 void msix_write_config(PCIDevice *dev, uint32_t addr,
119 uint32_t val, int len)
121 unsigned enable_pos = dev->msix_cap + MSIX_ENABLE_OFFSET;
122 if (addr + len <= enable_pos || addr > enable_pos)
125 if (msix_enabled(dev))
126 qemu_set_irq(dev->irq[0], 0);
129 static uint32_t msix_mmio_readl(void *opaque, target_phys_addr_t addr)
131 PCIDevice *dev = opaque;
132 unsigned int offset = addr & (MSIX_PAGE_SIZE - 1);
133 void *page = dev->msix_table_page;
136 memcpy(&val, (void *)((char *)page + offset), 4);
141 static uint32_t msix_mmio_read_unallowed(void *opaque, target_phys_addr_t addr)
143 fprintf(stderr, "MSI-X: only dword read is allowed!\n");
147 static uint8_t msix_pending_mask(int vector)
149 return 1 << (vector % 8);
152 static uint8_t *msix_pending_byte(PCIDevice *dev, int vector)
154 return dev->msix_table_page + MSIX_PAGE_PENDING + vector / 8;
157 static int msix_is_pending(PCIDevice *dev, int vector)
159 return *msix_pending_byte(dev, vector) & msix_pending_mask(vector);
162 static void msix_set_pending(PCIDevice *dev, int vector)
164 *msix_pending_byte(dev, vector) |= msix_pending_mask(vector);
167 static void msix_clr_pending(PCIDevice *dev, int vector)
169 *msix_pending_byte(dev, vector) &= ~msix_pending_mask(vector);
172 static int msix_is_masked(PCIDevice *dev, int vector)
174 unsigned offset = vector * MSIX_ENTRY_SIZE + MSIX_VECTOR_CTRL;
175 return dev->msix_table_page[offset] & MSIX_VECTOR_MASK;
178 static void msix_mmio_writel(void *opaque, target_phys_addr_t addr,
181 PCIDevice *dev = opaque;
182 unsigned int offset = addr & (MSIX_PAGE_SIZE - 1);
183 int vector = offset / MSIX_ENTRY_SIZE;
184 memcpy(dev->msix_table_page + offset, &val, 4);
185 if (!msix_is_masked(dev, vector) && msix_is_pending(dev, vector)) {
186 msix_clr_pending(dev, vector);
187 msix_notify(dev, vector);
191 static void msix_mmio_write_unallowed(void *opaque, target_phys_addr_t addr,
194 fprintf(stderr, "MSI-X: only dword write is allowed!\n");
197 static CPUWriteMemoryFunc *msix_mmio_write[] = {
198 msix_mmio_write_unallowed, msix_mmio_write_unallowed, msix_mmio_writel
201 static CPUReadMemoryFunc *msix_mmio_read[] = {
202 msix_mmio_read_unallowed, msix_mmio_read_unallowed, msix_mmio_readl
205 /* Should be called from device's map method. */
206 void msix_mmio_map(PCIDevice *d, int region_num,
207 uint32_t addr, uint32_t size, int type)
209 uint8_t *config = d->config + d->msix_cap;
210 uint32_t table = pci_get_long(config + MSIX_TABLE_OFFSET);
211 uint32_t offset = table & ~(MSIX_PAGE_SIZE - 1);
212 /* TODO: for assigned devices, we'll want to make it possible to map
213 * pending bits separately in case they are in a separate bar. */
214 int table_bir = table & PCI_MSIX_FLAGS_BIRMASK;
216 if (table_bir != region_num)
220 cpu_register_physical_memory(addr + offset, size - offset,
224 /* Initialize the MSI-X structures. Note: if MSI-X is supported, BAR size is
225 * modified, it should be retrieved with msix_bar_size. */
226 int msix_init(struct PCIDevice *dev, unsigned short nentries,
227 unsigned bar_nr, unsigned bar_size)
230 /* Nothing to do if MSI is not supported by interrupt controller */
234 if (nentries > MSIX_MAX_ENTRIES)
237 dev->msix_entry_used = qemu_mallocz(MSIX_MAX_ENTRIES *
238 sizeof *dev->msix_entry_used);
240 dev->msix_table_page = qemu_mallocz(MSIX_PAGE_SIZE);
242 dev->msix_mmio_index = cpu_register_io_memory(msix_mmio_read,
243 msix_mmio_write, dev);
244 if (dev->msix_mmio_index == -1) {
249 dev->msix_entries_nr = nentries;
250 ret = msix_add_config(dev, nentries, bar_nr, bar_size);
254 dev->cap_present |= QEMU_PCI_CAP_MSIX;
258 cpu_unregister_io_memory(dev->msix_mmio_index);
260 qemu_free(dev->msix_table_page);
261 dev->msix_table_page = NULL;
262 qemu_free(dev->msix_entry_used);
263 dev->msix_entry_used = NULL;
267 /* Clean up resources for the device. */
268 int msix_uninit(PCIDevice *dev)
270 if (!(dev->cap_present & QEMU_PCI_CAP_MSIX))
272 pci_del_capability(dev, PCI_CAP_ID_MSIX, MSIX_CAP_LENGTH);
274 msix_free_irq_entries(dev);
275 dev->msix_entries_nr = 0;
276 cpu_unregister_io_memory(dev->msix_mmio_index);
277 qemu_free(dev->msix_table_page);
278 dev->msix_table_page = NULL;
279 qemu_free(dev->msix_entry_used);
280 dev->msix_entry_used = NULL;
281 dev->cap_present &= ~QEMU_PCI_CAP_MSIX;
285 void msix_save(PCIDevice *dev, QEMUFile *f)
287 unsigned n = dev->msix_entries_nr;
289 if (!(dev->cap_present & QEMU_PCI_CAP_MSIX)) {
293 qemu_put_buffer(f, dev->msix_table_page, n * MSIX_ENTRY_SIZE);
294 qemu_put_buffer(f, dev->msix_table_page + MSIX_PAGE_PENDING, (n + 7) / 8);
297 /* Should be called after restoring the config space. */
298 void msix_load(PCIDevice *dev, QEMUFile *f)
300 unsigned n = dev->msix_entries_nr;
302 if (!(dev->cap_present & QEMU_PCI_CAP_MSIX)) {
306 msix_free_irq_entries(dev);
307 qemu_get_buffer(f, dev->msix_table_page, n * MSIX_ENTRY_SIZE);
308 qemu_get_buffer(f, dev->msix_table_page + MSIX_PAGE_PENDING, (n + 7) / 8);
311 /* Does device support MSI-X? */
312 int msix_present(PCIDevice *dev)
314 return dev->cap_present & QEMU_PCI_CAP_MSIX;
317 /* Is MSI-X enabled? */
318 int msix_enabled(PCIDevice *dev)
320 return (dev->cap_present & QEMU_PCI_CAP_MSIX) &&
321 (dev->config[dev->msix_cap + MSIX_ENABLE_OFFSET] &
325 /* Size of bar where MSI-X table resides, or 0 if MSI-X not supported. */
326 uint32_t msix_bar_size(PCIDevice *dev)
328 return (dev->cap_present & QEMU_PCI_CAP_MSIX) ?
329 dev->msix_bar_size : 0;
332 /* Send an MSI-X message */
333 void msix_notify(PCIDevice *dev, unsigned vector)
335 uint8_t *table_entry = dev->msix_table_page + vector * MSIX_ENTRY_SIZE;
339 if (vector >= dev->msix_entries_nr || !dev->msix_entry_used[vector])
341 if (msix_is_masked(dev, vector)) {
342 msix_set_pending(dev, vector);
346 address = pci_get_long(table_entry + MSIX_MSG_UPPER_ADDR);
347 address = (address << 32) | pci_get_long(table_entry + MSIX_MSG_ADDR);
348 data = pci_get_long(table_entry + MSIX_MSG_DATA);
349 stl_phys(address, data);
352 void msix_reset(PCIDevice *dev)
354 if (!(dev->cap_present & QEMU_PCI_CAP_MSIX))
356 msix_free_irq_entries(dev);
357 dev->config[dev->msix_cap + MSIX_ENABLE_OFFSET] &= MSIX_ENABLE_MASK;
358 memset(dev->msix_table_page, 0, MSIX_PAGE_SIZE);
361 /* PCI spec suggests that devices make it possible for software to configure
362 * less vectors than supported by the device, but does not specify a standard
363 * mechanism for devices to do so.
365 * We support this by asking devices to declare vectors software is going to
366 * actually use, and checking this on the notification path. Devices that
367 * don't want to follow the spec suggestion can declare all vectors as used. */
369 /* Mark vector as used. */
370 int msix_vector_use(PCIDevice *dev, unsigned vector)
372 if (vector >= dev->msix_entries_nr)
374 dev->msix_entry_used[vector]++;
378 /* Mark vector as unused. */
379 void msix_vector_unuse(PCIDevice *dev, unsigned vector)
381 if (vector < dev->msix_entries_nr && dev->msix_entry_used[vector])
382 --dev->msix_entry_used[vector];