2 * Copyright (c) 2007, Intel Corporation.
4 * This work is licensed under the terms of the GNU GPL, version 2. See
5 * the COPYING file in the top-level directory.
7 * Jiang Yunhong <yunhong.jiang@intel.com>
9 * This file implements direct PCI assignment to a HVM guest
14 #include "hw/xen/xen_backend.h"
16 #include "hw/i386/apic-msidef.h"
19 #define XEN_PT_AUTO_ASSIGN -1
21 /* shift count for gflags */
22 #define XEN_PT_GFLAGS_SHIFT_DEST_ID 0
23 #define XEN_PT_GFLAGS_SHIFT_RH 8
24 #define XEN_PT_GFLAGS_SHIFT_DM 9
25 #define XEN_PT_GFLAGSSHIFT_DELIV_MODE 12
26 #define XEN_PT_GFLAGSSHIFT_TRG_MODE 15
33 static inline uint8_t msi_vector(uint32_t data
)
35 return (data
& MSI_DATA_VECTOR_MASK
) >> MSI_DATA_VECTOR_SHIFT
;
38 static inline uint8_t msi_dest_id(uint32_t addr
)
40 return (addr
& MSI_ADDR_DEST_ID_MASK
) >> MSI_ADDR_DEST_ID_SHIFT
;
43 static inline uint32_t msi_ext_dest_id(uint32_t addr_hi
)
45 return addr_hi
& 0xffffff00;
48 static uint32_t msi_gflags(uint32_t data
, uint64_t addr
)
51 int rh
, dm
, dest_id
, deliv_mode
, trig_mode
;
53 rh
= (addr
>> MSI_ADDR_REDIRECTION_SHIFT
) & 0x1;
54 dm
= (addr
>> MSI_ADDR_DEST_MODE_SHIFT
) & 0x1;
55 dest_id
= msi_dest_id(addr
);
56 deliv_mode
= (data
>> MSI_DATA_DELIVERY_MODE_SHIFT
) & 0x7;
57 trig_mode
= (data
>> MSI_DATA_TRIGGER_SHIFT
) & 0x1;
59 result
= dest_id
| (rh
<< XEN_PT_GFLAGS_SHIFT_RH
)
60 | (dm
<< XEN_PT_GFLAGS_SHIFT_DM
)
61 | (deliv_mode
<< XEN_PT_GFLAGSSHIFT_DELIV_MODE
)
62 | (trig_mode
<< XEN_PT_GFLAGSSHIFT_TRG_MODE
);
67 static inline uint64_t msi_addr64(XenPTMSI
*msi
)
69 return (uint64_t)msi
->addr_hi
<< 32 | msi
->addr_lo
;
72 static int msi_msix_enable(XenPCIPassthroughState
*s
,
84 rc
= xen_host_pci_get_word(&s
->real_device
, address
, &val
);
86 XEN_PT_ERR(&s
->dev
, "Failed to read MSI/MSI-X register (0x%x), rc:%d\n",
95 rc
= xen_host_pci_set_word(&s
->real_device
, address
, val
);
97 XEN_PT_ERR(&s
->dev
, "Failed to write MSI/MSI-X register (0x%x), rc:%d\n",
103 static int msi_msix_setup(XenPCIPassthroughState
*s
,
111 uint8_t gvec
= msi_vector(data
);
114 assert((!is_msix
&& msix_entry
== 0) || is_msix
);
117 /* if gvec is 0, the guest is asking for a particular pirq that
118 * is passed as dest_id */
119 *ppirq
= msi_ext_dest_id(addr
>> 32) | msi_dest_id(addr
);
121 /* this probably identifies an misconfiguration of the guest,
122 * try the emulated path */
123 *ppirq
= XEN_PT_UNASSIGNED_PIRQ
;
125 XEN_PT_LOG(&s
->dev
, "requested pirq %d for MSI%s"
126 " (vec: %#x, entry: %#x)\n",
127 *ppirq
, is_msix
? "-X" : "", gvec
, msix_entry
);
132 uint64_t table_base
= 0;
135 table_base
= s
->msix
->table_base
;
138 rc
= xc_physdev_map_pirq_msi(xen_xc
, xen_domid
, XEN_PT_AUTO_ASSIGN
,
139 ppirq
, PCI_DEVFN(s
->real_device
.dev
,
140 s
->real_device
.func
),
142 msix_entry
, table_base
);
145 "Mapping of MSI%s (err: %i, vec: %#x, entry %#x)\n",
146 is_msix
? "-X" : "", errno
, gvec
, msix_entry
);
153 static int msi_msix_update(XenPCIPassthroughState
*s
,
161 PCIDevice
*d
= &s
->dev
;
162 uint8_t gvec
= msi_vector(data
);
163 uint32_t gflags
= msi_gflags(data
, addr
);
165 uint64_t table_addr
= 0;
167 XEN_PT_LOG(d
, "Updating MSI%s with pirq %d gvec %#x gflags %#x"
169 is_msix
? "-X" : "", pirq
, gvec
, gflags
, msix_entry
);
172 table_addr
= s
->msix
->mmio_base_addr
;
175 rc
= xc_domain_update_msi_irq(xen_xc
, xen_domid
, gvec
,
176 pirq
, gflags
, table_addr
);
179 XEN_PT_ERR(d
, "Updating of MSI%s failed. (err: %d)\n",
180 is_msix
? "-X" : "", errno
);
182 if (xc_physdev_unmap_pirq(xen_xc
, xen_domid
, *old_pirq
)) {
183 XEN_PT_ERR(d
, "Unmapping of MSI%s pirq %d failed. (err: %d)\n",
184 is_msix
? "-X" : "", *old_pirq
, errno
);
186 *old_pirq
= XEN_PT_UNASSIGNED_PIRQ
;
191 static int msi_msix_disable(XenPCIPassthroughState
*s
,
198 PCIDevice
*d
= &s
->dev
;
199 uint8_t gvec
= msi_vector(data
);
200 uint32_t gflags
= msi_gflags(data
, addr
);
203 if (pirq
== XEN_PT_UNASSIGNED_PIRQ
) {
208 XEN_PT_LOG(d
, "Unbind MSI%s with pirq %d, gvec %#x\n",
209 is_msix
? "-X" : "", pirq
, gvec
);
210 rc
= xc_domain_unbind_msi_irq(xen_xc
, xen_domid
, gvec
, pirq
, gflags
);
212 XEN_PT_ERR(d
, "Unbinding of MSI%s failed. (err: %d, pirq: %d, gvec: %#x)\n",
213 is_msix
? "-X" : "", errno
, pirq
, gvec
);
218 XEN_PT_LOG(d
, "Unmap MSI%s pirq %d\n", is_msix
? "-X" : "", pirq
);
219 rc
= xc_physdev_unmap_pirq(xen_xc
, xen_domid
, pirq
);
221 XEN_PT_ERR(d
, "Unmapping of MSI%s pirq %d failed. (err: %i)\n",
222 is_msix
? "-X" : "", pirq
, errno
);
230 * MSI virtualization functions
233 static int xen_pt_msi_set_enable(XenPCIPassthroughState
*s
, bool enable
)
235 XEN_PT_LOG(&s
->dev
, "%s MSI.\n", enable
? "enabling" : "disabling");
241 return msi_msix_enable(s
, s
->msi
->ctrl_offset
, PCI_MSI_FLAGS_ENABLE
,
245 /* setup physical msi, but don't enable it */
246 int xen_pt_msi_setup(XenPCIPassthroughState
*s
)
248 int pirq
= XEN_PT_UNASSIGNED_PIRQ
;
250 XenPTMSI
*msi
= s
->msi
;
252 if (msi
->initialized
) {
254 "Setup physical MSI when it has been properly initialized.\n");
258 rc
= msi_msix_setup(s
, msi_addr64(msi
), msi
->data
, &pirq
, false, 0, true);
264 XEN_PT_ERR(&s
->dev
, "Invalid pirq number: %d.\n", pirq
);
269 XEN_PT_LOG(&s
->dev
, "MSI mapped with pirq %d.\n", pirq
);
274 int xen_pt_msi_update(XenPCIPassthroughState
*s
)
276 XenPTMSI
*msi
= s
->msi
;
277 return msi_msix_update(s
, msi_addr64(msi
), msi
->data
, msi
->pirq
,
278 false, 0, &msi
->pirq
);
281 void xen_pt_msi_disable(XenPCIPassthroughState
*s
)
283 XenPTMSI
*msi
= s
->msi
;
289 (void)xen_pt_msi_set_enable(s
, false);
291 msi_msix_disable(s
, msi_addr64(msi
), msi
->data
, msi
->pirq
, false,
295 msi
->flags
&= ~PCI_MSI_FLAGS_ENABLE
;
296 msi
->initialized
= false;
298 msi
->pirq
= XEN_PT_UNASSIGNED_PIRQ
;
302 * MSI-X virtualization functions
305 static int msix_set_enable(XenPCIPassthroughState
*s
, bool enabled
)
307 XEN_PT_LOG(&s
->dev
, "%s MSI-X.\n", enabled
? "enabling" : "disabling");
313 return msi_msix_enable(s
, s
->msix
->ctrl_offset
, PCI_MSIX_FLAGS_ENABLE
,
317 static int xen_pt_msix_update_one(XenPCIPassthroughState
*s
, int entry_nr
)
319 XenPTMSIXEntry
*entry
= NULL
;
323 if (entry_nr
< 0 || entry_nr
>= s
->msix
->total_entries
) {
327 entry
= &s
->msix
->msix_entry
[entry_nr
];
329 if (!entry
->updated
) {
335 rc
= msi_msix_setup(s
, entry
->addr
, entry
->data
, &pirq
, true, entry_nr
,
336 entry
->pirq
== XEN_PT_UNASSIGNED_PIRQ
);
340 if (entry
->pirq
== XEN_PT_UNASSIGNED_PIRQ
) {
344 rc
= msi_msix_update(s
, entry
->addr
, entry
->data
, pirq
, true,
345 entry_nr
, &entry
->pirq
);
348 entry
->updated
= false;
354 int xen_pt_msix_update(XenPCIPassthroughState
*s
)
356 XenPTMSIX
*msix
= s
->msix
;
359 for (i
= 0; i
< msix
->total_entries
; i
++) {
360 xen_pt_msix_update_one(s
, i
);
366 void xen_pt_msix_disable(XenPCIPassthroughState
*s
)
370 msix_set_enable(s
, false);
372 for (i
= 0; i
< s
->msix
->total_entries
; i
++) {
373 XenPTMSIXEntry
*entry
= &s
->msix
->msix_entry
[i
];
375 msi_msix_disable(s
, entry
->addr
, entry
->data
, entry
->pirq
, true, true);
377 /* clear MSI-X info */
378 entry
->pirq
= XEN_PT_UNASSIGNED_PIRQ
;
379 entry
->updated
= false;
383 int xen_pt_msix_update_remap(XenPCIPassthroughState
*s
, int bar_index
)
385 XenPTMSIXEntry
*entry
;
388 if (!(s
->msix
&& s
->msix
->bar_index
== bar_index
)) {
392 for (i
= 0; i
< s
->msix
->total_entries
; i
++) {
393 entry
= &s
->msix
->msix_entry
[i
];
394 if (entry
->pirq
!= XEN_PT_UNASSIGNED_PIRQ
) {
395 ret
= xc_domain_unbind_pt_irq(xen_xc
, xen_domid
, entry
->pirq
,
396 PT_IRQ_TYPE_MSI
, 0, 0, 0, 0);
398 XEN_PT_ERR(&s
->dev
, "unbind MSI-X entry %d failed (err: %d)\n",
401 entry
->updated
= true;
404 return xen_pt_msix_update(s
);
407 static uint32_t get_entry_value(XenPTMSIXEntry
*e
, int offset
)
410 case PCI_MSIX_ENTRY_LOWER_ADDR
:
411 return e
->addr
& UINT32_MAX
;
412 case PCI_MSIX_ENTRY_UPPER_ADDR
:
413 return e
->addr
>> 32;
414 case PCI_MSIX_ENTRY_DATA
:
416 case PCI_MSIX_ENTRY_VECTOR_CTRL
:
417 return e
->vector_ctrl
;
423 static void set_entry_value(XenPTMSIXEntry
*e
, int offset
, uint32_t val
)
426 case PCI_MSIX_ENTRY_LOWER_ADDR
:
427 e
->addr
= (e
->addr
& ((uint64_t)UINT32_MAX
<< 32)) | val
;
429 case PCI_MSIX_ENTRY_UPPER_ADDR
:
430 e
->addr
= (uint64_t)val
<< 32 | (e
->addr
& UINT32_MAX
);
432 case PCI_MSIX_ENTRY_DATA
:
435 case PCI_MSIX_ENTRY_VECTOR_CTRL
:
436 e
->vector_ctrl
= val
;
441 static void pci_msix_write(void *opaque
, hwaddr addr
,
442 uint64_t val
, unsigned size
)
444 XenPCIPassthroughState
*s
= opaque
;
445 XenPTMSIX
*msix
= s
->msix
;
446 XenPTMSIXEntry
*entry
;
447 unsigned int entry_nr
, offset
;
449 entry_nr
= addr
/ PCI_MSIX_ENTRY_SIZE
;
450 if (entry_nr
>= msix
->total_entries
) {
453 entry
= &msix
->msix_entry
[entry_nr
];
454 offset
= addr
% PCI_MSIX_ENTRY_SIZE
;
456 if (offset
!= PCI_MSIX_ENTRY_VECTOR_CTRL
) {
457 const volatile uint32_t *vec_ctrl
;
459 if (get_entry_value(entry
, offset
) == val
460 && entry
->pirq
!= XEN_PT_UNASSIGNED_PIRQ
) {
465 * If Xen intercepts the mask bit access, entry->vec_ctrl may not be
466 * up-to-date. Read from hardware directly.
468 vec_ctrl
= s
->msix
->phys_iomem_base
+ entry_nr
* PCI_MSIX_ENTRY_SIZE
469 + PCI_MSIX_ENTRY_VECTOR_CTRL
;
471 if (msix
->enabled
&& !(*vec_ctrl
& PCI_MSIX_ENTRY_CTRL_MASKBIT
)) {
472 if (!entry
->warned
) {
473 entry
->warned
= true;
474 XEN_PT_ERR(&s
->dev
, "Can't update msix entry %d since MSI-X is"
475 " already enabled.\n", entry_nr
);
480 entry
->updated
= true;
483 set_entry_value(entry
, offset
, val
);
485 if (offset
== PCI_MSIX_ENTRY_VECTOR_CTRL
) {
486 if (msix
->enabled
&& !(val
& PCI_MSIX_ENTRY_CTRL_MASKBIT
)) {
487 xen_pt_msix_update_one(s
, entry_nr
);
492 static uint64_t pci_msix_read(void *opaque
, hwaddr addr
,
495 XenPCIPassthroughState
*s
= opaque
;
496 XenPTMSIX
*msix
= s
->msix
;
497 int entry_nr
, offset
;
499 entry_nr
= addr
/ PCI_MSIX_ENTRY_SIZE
;
501 XEN_PT_ERR(&s
->dev
, "asked MSI-X entry '%i' invalid!\n", entry_nr
);
505 offset
= addr
% PCI_MSIX_ENTRY_SIZE
;
507 if (addr
< msix
->total_entries
* PCI_MSIX_ENTRY_SIZE
) {
508 return get_entry_value(&msix
->msix_entry
[entry_nr
], offset
);
510 /* Pending Bit Array (PBA) */
511 return *(uint32_t *)(msix
->phys_iomem_base
+ addr
);
515 static const MemoryRegionOps pci_msix_ops
= {
516 .read
= pci_msix_read
,
517 .write
= pci_msix_write
,
518 .endianness
= DEVICE_NATIVE_ENDIAN
,
520 .min_access_size
= 4,
521 .max_access_size
= 4,
526 int xen_pt_msix_init(XenPCIPassthroughState
*s
, uint32_t base
)
529 uint16_t control
= 0;
530 uint32_t table_off
= 0;
531 int i
, total_entries
, bar_index
;
532 XenHostPCIDevice
*hd
= &s
->real_device
;
533 PCIDevice
*d
= &s
->dev
;
535 XenPTMSIX
*msix
= NULL
;
538 rc
= xen_host_pci_get_byte(hd
, base
+ PCI_CAP_LIST_ID
, &id
);
543 if (id
!= PCI_CAP_ID_MSIX
) {
544 XEN_PT_ERR(d
, "Invalid id %#x base %#x\n", id
, base
);
548 xen_host_pci_get_word(hd
, base
+ PCI_MSIX_FLAGS
, &control
);
549 total_entries
= control
& PCI_MSIX_FLAGS_QSIZE
;
552 s
->msix
= g_malloc0(sizeof (XenPTMSIX
)
553 + total_entries
* sizeof (XenPTMSIXEntry
));
556 msix
->total_entries
= total_entries
;
557 for (i
= 0; i
< total_entries
; i
++) {
558 msix
->msix_entry
[i
].pirq
= XEN_PT_UNASSIGNED_PIRQ
;
561 memory_region_init_io(&msix
->mmio
, OBJECT(s
), &pci_msix_ops
,
562 s
, "xen-pci-pt-msix",
563 (total_entries
* PCI_MSIX_ENTRY_SIZE
567 xen_host_pci_get_long(hd
, base
+ PCI_MSIX_TABLE
, &table_off
);
568 bar_index
= msix
->bar_index
= table_off
& PCI_MSIX_FLAGS_BIRMASK
;
569 table_off
= table_off
& ~PCI_MSIX_FLAGS_BIRMASK
;
570 msix
->table_base
= s
->real_device
.io_regions
[bar_index
].base_addr
;
571 XEN_PT_LOG(d
, "get MSI-X table BAR base 0x%"PRIx64
"\n", msix
->table_base
);
573 fd
= open("/dev/mem", O_RDWR
);
576 XEN_PT_ERR(d
, "Can't open /dev/mem: %s\n", strerror(errno
));
579 XEN_PT_LOG(d
, "table_off = %#x, total_entries = %d\n",
580 table_off
, total_entries
);
581 msix
->table_offset_adjust
= table_off
& 0x0fff;
582 msix
->phys_iomem_base
=
584 total_entries
* PCI_MSIX_ENTRY_SIZE
+ msix
->table_offset_adjust
,
586 MAP_SHARED
| MAP_LOCKED
,
588 msix
->table_base
+ table_off
- msix
->table_offset_adjust
);
590 if (msix
->phys_iomem_base
== MAP_FAILED
) {
592 XEN_PT_ERR(d
, "Can't map physical MSI-X table: %s\n", strerror(errno
));
595 msix
->phys_iomem_base
= (char *)msix
->phys_iomem_base
596 + msix
->table_offset_adjust
;
598 XEN_PT_LOG(d
, "mapping physical MSI-X table to %p\n",
599 msix
->phys_iomem_base
);
601 memory_region_add_subregion_overlap(&s
->bar
[bar_index
], table_off
,
603 2); /* Priority: pci default + 1 */
613 void xen_pt_msix_delete(XenPCIPassthroughState
*s
)
615 XenPTMSIX
*msix
= s
->msix
;
621 /* unmap the MSI-X memory mapped register area */
622 if (msix
->phys_iomem_base
) {
623 XEN_PT_LOG(&s
->dev
, "unmapping physical MSI-X table from %p\n",
624 msix
->phys_iomem_base
);
625 munmap(msix
->phys_iomem_base
, msix
->total_entries
* PCI_MSIX_ENTRY_SIZE
626 + msix
->table_offset_adjust
);
629 memory_region_del_subregion(&s
->bar
[msix
->bar_index
], &msix
->mmio
);