2 * Copyright (C) 2010 Citrix Ltd.
4 * This work is licensed under the terms of the GNU GPL, version 2. See
5 * the COPYING file in the top-level directory.
12 #include "hw/xen_common.h"
13 #include "hw/xen_backend.h"
15 #include "xen-mapcache.h"
18 #include <xen/hvm/ioreq.h>
19 #include <xen/hvm/params.h>
24 #define DPRINTF(fmt, ...) \
25 do { fprintf(stderr, "xen: " fmt, ## __VA_ARGS__); } while (0)
27 #define DPRINTF(fmt, ...) \
31 /* Compatibility with older version */
32 #if __XEN_LATEST_INTERFACE_VERSION__ < 0x0003020a
33 static inline uint32_t xen_vcpu_eport(shared_iopage_t
*shared_page
, int i
)
35 return shared_page
->vcpu_iodata
[i
].vp_eport
;
37 static inline ioreq_t
*xen_vcpu_ioreq(shared_iopage_t
*shared_page
, int vcpu
)
39 return &shared_page
->vcpu_iodata
[vcpu
].vp_ioreq
;
41 # define FMT_ioreq_size PRIx64
43 static inline uint32_t xen_vcpu_eport(shared_iopage_t
*shared_page
, int i
)
45 return shared_page
->vcpu_ioreq
[i
].vp_eport
;
47 static inline ioreq_t
*xen_vcpu_ioreq(shared_iopage_t
*shared_page
, int vcpu
)
49 return &shared_page
->vcpu_ioreq
[vcpu
];
51 # define FMT_ioreq_size "u"
54 #define BUFFER_IO_MAX_DELAY 100
56 typedef struct XenIOState
{
57 shared_iopage_t
*shared_page
;
58 buffered_iopage_t
*buffered_io_page
;
59 QEMUTimer
*buffered_io_timer
;
60 /* the evtchn port for polling the notification, */
61 evtchn_port_t
*ioreq_local_port
;
62 /* the evtchn fd for polling */
64 /* which vcpu we are serving */
70 /* Xen specific function for piix pci */
72 int xen_pci_slot_get_pirq(PCIDevice
*pci_dev
, int irq_num
)
74 return irq_num
+ ((pci_dev
->devfn
>> 3) << 2);
77 void xen_piix3_set_irq(void *opaque
, int irq_num
, int level
)
79 xc_hvm_set_pci_intx_level(xen_xc
, xen_domid
, 0, 0, irq_num
>> 2,
83 void xen_piix_pci_write_config_client(uint32_t address
, uint32_t val
, int len
)
87 /* Scan for updates to PCI link routes (0x60-0x63). */
88 for (i
= 0; i
< len
; i
++) {
89 uint8_t v
= (val
>> (8 * i
)) & 0xff;
94 if (((address
+ i
) >= 0x60) && ((address
+ i
) <= 0x63)) {
95 xc_hvm_set_pci_link_route(xen_xc
, xen_domid
, address
+ i
- 0x60, v
);
100 /* Xen Interrupt Controller */
102 static void xen_set_irq(void *opaque
, int irq
, int level
)
104 xc_hvm_set_isa_irq_level(xen_xc
, xen_domid
, irq
, level
);
107 qemu_irq
*xen_interrupt_controller_init(void)
109 return qemu_allocate_irqs(xen_set_irq
, NULL
, 16);
114 static void xen_ram_init(ram_addr_t ram_size
)
117 ram_addr_t below_4g_mem_size
, above_4g_mem_size
= 0;
119 new_block
= qemu_mallocz(sizeof (*new_block
));
120 pstrcpy(new_block
->idstr
, sizeof (new_block
->idstr
), "xen.ram");
121 new_block
->host
= NULL
;
122 new_block
->offset
= 0;
123 new_block
->length
= ram_size
;
125 QLIST_INSERT_HEAD(&ram_list
.blocks
, new_block
, next
);
127 ram_list
.phys_dirty
= qemu_realloc(ram_list
.phys_dirty
,
128 new_block
->length
>> TARGET_PAGE_BITS
);
129 memset(ram_list
.phys_dirty
+ (new_block
->offset
>> TARGET_PAGE_BITS
),
130 0xff, new_block
->length
>> TARGET_PAGE_BITS
);
132 if (ram_size
>= 0xe0000000 ) {
133 above_4g_mem_size
= ram_size
- 0xe0000000;
134 below_4g_mem_size
= 0xe0000000;
136 below_4g_mem_size
= ram_size
;
139 cpu_register_physical_memory(0, below_4g_mem_size
, new_block
->offset
);
140 #if TARGET_PHYS_ADDR_BITS > 32
141 if (above_4g_mem_size
> 0) {
142 cpu_register_physical_memory(0x100000000ULL
, above_4g_mem_size
,
143 new_block
->offset
+ below_4g_mem_size
);
148 void xen_ram_alloc(ram_addr_t ram_addr
, ram_addr_t size
)
150 unsigned long nr_pfn
;
154 trace_xen_ram_alloc(ram_addr
, size
);
156 nr_pfn
= size
>> TARGET_PAGE_BITS
;
157 pfn_list
= qemu_malloc(sizeof (*pfn_list
) * nr_pfn
);
159 for (i
= 0; i
< nr_pfn
; i
++) {
160 pfn_list
[i
] = (ram_addr
>> TARGET_PAGE_BITS
) + i
;
163 if (xc_domain_populate_physmap_exact(xen_xc
, xen_domid
, nr_pfn
, 0, 0, pfn_list
)) {
164 hw_error("xen: failed to populate ram at %lx", ram_addr
);
171 /* VCPU Operations, MMIO, IO ring ... */
173 static void xen_reset_vcpu(void *opaque
)
175 CPUState
*env
= opaque
;
180 void xen_vcpu_init(void)
184 if ((first_cpu
= qemu_get_cpu(0))) {
185 qemu_register_reset(xen_reset_vcpu
, first_cpu
);
186 xen_reset_vcpu(first_cpu
);
190 /* get the ioreq packets from share mem */
191 static ioreq_t
*cpu_get_ioreq_from_shared_memory(XenIOState
*state
, int vcpu
)
193 ioreq_t
*req
= xen_vcpu_ioreq(state
->shared_page
, vcpu
);
195 if (req
->state
!= STATE_IOREQ_READY
) {
196 DPRINTF("I/O request not ready: "
197 "%x, ptr: %x, port: %"PRIx64
", "
198 "data: %"PRIx64
", count: %" FMT_ioreq_size
", size: %" FMT_ioreq_size
"\n",
199 req
->state
, req
->data_is_ptr
, req
->addr
,
200 req
->data
, req
->count
, req
->size
);
204 xen_rmb(); /* see IOREQ_READY /then/ read contents of ioreq */
206 req
->state
= STATE_IOREQ_INPROCESS
;
210 /* use poll to get the port notification */
211 /* ioreq_vec--out,the */
212 /* retval--the number of ioreq packet */
213 static ioreq_t
*cpu_get_ioreq(XenIOState
*state
)
218 port
= xc_evtchn_pending(state
->xce_handle
);
220 for (i
= 0; i
< smp_cpus
; i
++) {
221 if (state
->ioreq_local_port
[i
] == port
) {
227 hw_error("Fatal error while trying to get io event!\n");
230 /* unmask the wanted port again */
231 xc_evtchn_unmask(state
->xce_handle
, port
);
233 /* get the io packet from shared memory */
234 state
->send_vcpu
= i
;
235 return cpu_get_ioreq_from_shared_memory(state
, i
);
238 /* read error or read nothing */
242 static uint32_t do_inp(pio_addr_t addr
, unsigned long size
)
246 return cpu_inb(addr
);
248 return cpu_inw(addr
);
250 return cpu_inl(addr
);
252 hw_error("inp: bad size: %04"FMT_pioaddr
" %lx", addr
, size
);
256 static void do_outp(pio_addr_t addr
,
257 unsigned long size
, uint32_t val
)
261 return cpu_outb(addr
, val
);
263 return cpu_outw(addr
, val
);
265 return cpu_outl(addr
, val
);
267 hw_error("outp: bad size: %04"FMT_pioaddr
" %lx", addr
, size
);
271 static void cpu_ioreq_pio(ioreq_t
*req
)
275 sign
= req
->df
? -1 : 1;
277 if (req
->dir
== IOREQ_READ
) {
278 if (!req
->data_is_ptr
) {
279 req
->data
= do_inp(req
->addr
, req
->size
);
283 for (i
= 0; i
< req
->count
; i
++) {
284 tmp
= do_inp(req
->addr
, req
->size
);
285 cpu_physical_memory_write(req
->data
+ (sign
* i
* req
->size
),
286 (uint8_t *) &tmp
, req
->size
);
289 } else if (req
->dir
== IOREQ_WRITE
) {
290 if (!req
->data_is_ptr
) {
291 do_outp(req
->addr
, req
->size
, req
->data
);
293 for (i
= 0; i
< req
->count
; i
++) {
296 cpu_physical_memory_read(req
->data
+ (sign
* i
* req
->size
),
297 (uint8_t*) &tmp
, req
->size
);
298 do_outp(req
->addr
, req
->size
, tmp
);
304 static void cpu_ioreq_move(ioreq_t
*req
)
308 sign
= req
->df
? -1 : 1;
310 if (!req
->data_is_ptr
) {
311 if (req
->dir
== IOREQ_READ
) {
312 for (i
= 0; i
< req
->count
; i
++) {
313 cpu_physical_memory_read(req
->addr
+ (sign
* i
* req
->size
),
314 (uint8_t *) &req
->data
, req
->size
);
316 } else if (req
->dir
== IOREQ_WRITE
) {
317 for (i
= 0; i
< req
->count
; i
++) {
318 cpu_physical_memory_write(req
->addr
+ (sign
* i
* req
->size
),
319 (uint8_t *) &req
->data
, req
->size
);
325 if (req
->dir
== IOREQ_READ
) {
326 for (i
= 0; i
< req
->count
; i
++) {
327 cpu_physical_memory_read(req
->addr
+ (sign
* i
* req
->size
),
328 (uint8_t*) &tmp
, req
->size
);
329 cpu_physical_memory_write(req
->data
+ (sign
* i
* req
->size
),
330 (uint8_t*) &tmp
, req
->size
);
332 } else if (req
->dir
== IOREQ_WRITE
) {
333 for (i
= 0; i
< req
->count
; i
++) {
334 cpu_physical_memory_read(req
->data
+ (sign
* i
* req
->size
),
335 (uint8_t*) &tmp
, req
->size
);
336 cpu_physical_memory_write(req
->addr
+ (sign
* i
* req
->size
),
337 (uint8_t*) &tmp
, req
->size
);
343 static void handle_ioreq(ioreq_t
*req
)
345 if (!req
->data_is_ptr
&& (req
->dir
== IOREQ_WRITE
) &&
346 (req
->size
< sizeof (target_ulong
))) {
347 req
->data
&= ((target_ulong
) 1 << (8 * req
->size
)) - 1;
354 case IOREQ_TYPE_COPY
:
357 case IOREQ_TYPE_TIMEOFFSET
:
359 case IOREQ_TYPE_INVALIDATE
:
360 qemu_invalidate_map_cache();
363 hw_error("Invalid ioreq type 0x%x\n", req
->type
);
367 static void handle_buffered_iopage(XenIOState
*state
)
369 buf_ioreq_t
*buf_req
= NULL
;
373 if (!state
->buffered_io_page
) {
377 while (state
->buffered_io_page
->read_pointer
!= state
->buffered_io_page
->write_pointer
) {
378 buf_req
= &state
->buffered_io_page
->buf_ioreq
[
379 state
->buffered_io_page
->read_pointer
% IOREQ_BUFFER_SLOT_NUM
];
380 req
.size
= 1UL << buf_req
->size
;
382 req
.addr
= buf_req
->addr
;
383 req
.data
= buf_req
->data
;
384 req
.state
= STATE_IOREQ_READY
;
385 req
.dir
= buf_req
->dir
;
387 req
.type
= buf_req
->type
;
389 qw
= (req
.size
== 8);
391 buf_req
= &state
->buffered_io_page
->buf_ioreq
[
392 (state
->buffered_io_page
->read_pointer
+ 1) % IOREQ_BUFFER_SLOT_NUM
];
393 req
.data
|= ((uint64_t)buf_req
->data
) << 32;
399 state
->buffered_io_page
->read_pointer
+= qw
? 2 : 1;
403 static void handle_buffered_io(void *opaque
)
405 XenIOState
*state
= opaque
;
407 handle_buffered_iopage(state
);
408 qemu_mod_timer(state
->buffered_io_timer
,
409 BUFFER_IO_MAX_DELAY
+ qemu_get_clock_ms(rt_clock
));
412 static void cpu_handle_ioreq(void *opaque
)
414 XenIOState
*state
= opaque
;
415 ioreq_t
*req
= cpu_get_ioreq(state
);
417 handle_buffered_iopage(state
);
421 if (req
->state
!= STATE_IOREQ_INPROCESS
) {
422 fprintf(stderr
, "Badness in I/O request ... not in service?!: "
423 "%x, ptr: %x, port: %"PRIx64
", "
424 "data: %"PRIx64
", count: %" FMT_ioreq_size
", size: %" FMT_ioreq_size
"\n",
425 req
->state
, req
->data_is_ptr
, req
->addr
,
426 req
->data
, req
->count
, req
->size
);
427 destroy_hvm_domain();
431 xen_wmb(); /* Update ioreq contents /then/ update state. */
434 * We do this before we send the response so that the tools
435 * have the opportunity to pick up on the reset before the
436 * guest resumes and does a hlt with interrupts disabled which
437 * causes Xen to powerdown the domain.
440 if (qemu_shutdown_requested_get()) {
441 destroy_hvm_domain();
443 if (qemu_reset_requested_get()) {
448 req
->state
= STATE_IORESP_READY
;
449 xc_evtchn_notify(state
->xce_handle
, state
->ioreq_local_port
[state
->send_vcpu
]);
453 static void xen_main_loop_prepare(XenIOState
*state
)
457 if (state
->xce_handle
!= XC_HANDLER_INITIAL_VALUE
) {
458 evtchn_fd
= xc_evtchn_fd(state
->xce_handle
);
461 state
->buffered_io_timer
= qemu_new_timer_ms(rt_clock
, handle_buffered_io
,
463 qemu_mod_timer(state
->buffered_io_timer
, qemu_get_clock_ms(rt_clock
));
465 if (evtchn_fd
!= -1) {
466 qemu_set_fd_handler(evtchn_fd
, cpu_handle_ioreq
, NULL
, state
);
473 static void xen_vm_change_state_handler(void *opaque
, int running
, int reason
)
475 XenIOState
*state
= opaque
;
477 xen_main_loop_prepare(state
);
481 static void xen_exit_notifier(Notifier
*n
)
483 XenIOState
*state
= container_of(n
, XenIOState
, exit
);
485 xc_evtchn_close(state
->xce_handle
);
490 xen_xc
= xen_xc_interface_open(0, 0, 0);
491 if (xen_xc
== XC_HANDLER_INITIAL_VALUE
) {
492 xen_be_printf(NULL
, 0, "can't open xen interface\n");
499 int xen_hvm_init(void)
502 unsigned long ioreq_pfn
;
505 state
= qemu_mallocz(sizeof (XenIOState
));
507 state
->xce_handle
= xen_xc_evtchn_open(NULL
, 0);
508 if (state
->xce_handle
== XC_HANDLER_INITIAL_VALUE
) {
509 perror("xen: event channel open");
513 state
->exit
.notify
= xen_exit_notifier
;
514 qemu_add_exit_notifier(&state
->exit
);
516 xc_get_hvm_param(xen_xc
, xen_domid
, HVM_PARAM_IOREQ_PFN
, &ioreq_pfn
);
517 DPRINTF("shared page at pfn %lx\n", ioreq_pfn
);
518 state
->shared_page
= xc_map_foreign_range(xen_xc
, xen_domid
, XC_PAGE_SIZE
,
519 PROT_READ
|PROT_WRITE
, ioreq_pfn
);
520 if (state
->shared_page
== NULL
) {
521 hw_error("map shared IO page returned error %d handle=" XC_INTERFACE_FMT
,
525 xc_get_hvm_param(xen_xc
, xen_domid
, HVM_PARAM_BUFIOREQ_PFN
, &ioreq_pfn
);
526 DPRINTF("buffered io page at pfn %lx\n", ioreq_pfn
);
527 state
->buffered_io_page
= xc_map_foreign_range(xen_xc
, xen_domid
, XC_PAGE_SIZE
,
528 PROT_READ
|PROT_WRITE
, ioreq_pfn
);
529 if (state
->buffered_io_page
== NULL
) {
530 hw_error("map buffered IO page returned error %d", errno
);
533 state
->ioreq_local_port
= qemu_mallocz(smp_cpus
* sizeof (evtchn_port_t
));
535 /* FIXME: how about if we overflow the page here? */
536 for (i
= 0; i
< smp_cpus
; i
++) {
537 rc
= xc_evtchn_bind_interdomain(state
->xce_handle
, xen_domid
,
538 xen_vcpu_eport(state
->shared_page
, i
));
540 fprintf(stderr
, "bind interdomain ioctl error %d\n", errno
);
543 state
->ioreq_local_port
[i
] = rc
;
546 /* Init RAM management */
547 qemu_map_cache_init();
548 xen_ram_init(ram_size
);
550 qemu_add_vm_change_state_handler(xen_vm_change_state_handler
, state
);
555 void destroy_hvm_domain(void)
560 xc_handle
= xen_xc_interface_open(0, 0, 0);
561 if (xc_handle
== XC_HANDLER_INITIAL_VALUE
) {
562 fprintf(stderr
, "Cannot acquire xenctrl handle\n");
564 sts
= xc_domain_shutdown(xc_handle
, xen_domid
, SHUTDOWN_poweroff
);
566 fprintf(stderr
, "? xc_domain_shutdown failed to issue poweroff, "
567 "sts %d, %s\n", sts
, strerror(errno
));
569 fprintf(stderr
, "Issued domain %d poweroff\n", xen_domid
);
571 xc_interface_close(xc_handle
);