parallels: Switch to byte-based calls
[qemu/ar7.git] / hw / ppc / spapr_pci.c
blob497b896c7d2460cce8357c087b4e9667396d81bb
1 /*
2 * QEMU sPAPR PCI host originated from Uninorth PCI host
4 * Copyright (c) 2011 Alexey Kardashevskiy, IBM Corporation.
5 * Copyright (C) 2011 David Gibson, IBM Corporation.
7 * Permission is hereby granted, free of charge, to any person obtaining a copy
8 * of this software and associated documentation files (the "Software"), to deal
9 * in the Software without restriction, including without limitation the rights
10 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
11 * copies of the Software, and to permit persons to whom the Software is
12 * furnished to do so, subject to the following conditions:
14 * The above copyright notice and this permission notice shall be included in
15 * all copies or substantial portions of the Software.
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
22 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
23 * THE SOFTWARE.
25 #include "qemu/osdep.h"
26 #include "qapi/error.h"
27 #include "qemu-common.h"
28 #include "cpu.h"
29 #include "hw/hw.h"
30 #include "hw/sysbus.h"
31 #include "hw/pci/pci.h"
32 #include "hw/pci/msi.h"
33 #include "hw/pci/msix.h"
34 #include "hw/pci/pci_host.h"
35 #include "hw/ppc/spapr.h"
36 #include "hw/pci-host/spapr.h"
37 #include "exec/address-spaces.h"
38 #include "exec/ram_addr.h"
39 #include <libfdt.h>
40 #include "trace.h"
41 #include "qemu/error-report.h"
42 #include "qapi/qmp/qerror.h"
43 #include "hw/ppc/fdt.h"
44 #include "hw/pci/pci_bridge.h"
45 #include "hw/pci/pci_bus.h"
46 #include "hw/pci/pci_ids.h"
47 #include "hw/ppc/spapr_drc.h"
48 #include "sysemu/device_tree.h"
49 #include "sysemu/kvm.h"
50 #include "sysemu/hostmem.h"
51 #include "sysemu/numa.h"
53 /* Copied from the kernel arch/powerpc/platforms/pseries/msi.c */
54 #define RTAS_QUERY_FN 0
55 #define RTAS_CHANGE_FN 1
56 #define RTAS_RESET_FN 2
57 #define RTAS_CHANGE_MSI_FN 3
58 #define RTAS_CHANGE_MSIX_FN 4
60 /* Interrupt types to return on RTAS_CHANGE_* */
61 #define RTAS_TYPE_MSI 1
62 #define RTAS_TYPE_MSIX 2
64 sPAPRPHBState *spapr_pci_find_phb(sPAPRMachineState *spapr, uint64_t buid)
66 sPAPRPHBState *sphb;
68 QLIST_FOREACH(sphb, &spapr->phbs, list) {
69 if (sphb->buid != buid) {
70 continue;
72 return sphb;
75 return NULL;
78 PCIDevice *spapr_pci_find_dev(sPAPRMachineState *spapr, uint64_t buid,
79 uint32_t config_addr)
81 sPAPRPHBState *sphb = spapr_pci_find_phb(spapr, buid);
82 PCIHostState *phb = PCI_HOST_BRIDGE(sphb);
83 int bus_num = (config_addr >> 16) & 0xFF;
84 int devfn = (config_addr >> 8) & 0xFF;
86 if (!phb) {
87 return NULL;
90 return pci_find_device(phb->bus, bus_num, devfn);
93 static uint32_t rtas_pci_cfgaddr(uint32_t arg)
95 /* This handles the encoding of extended config space addresses */
96 return ((arg >> 20) & 0xf00) | (arg & 0xff);
99 static void finish_read_pci_config(sPAPRMachineState *spapr, uint64_t buid,
100 uint32_t addr, uint32_t size,
101 target_ulong rets)
103 PCIDevice *pci_dev;
104 uint32_t val;
106 if ((size != 1) && (size != 2) && (size != 4)) {
107 /* access must be 1, 2 or 4 bytes */
108 rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
109 return;
112 pci_dev = spapr_pci_find_dev(spapr, buid, addr);
113 addr = rtas_pci_cfgaddr(addr);
115 if (!pci_dev || (addr % size) || (addr >= pci_config_size(pci_dev))) {
116 /* Access must be to a valid device, within bounds and
117 * naturally aligned */
118 rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
119 return;
122 val = pci_host_config_read_common(pci_dev, addr,
123 pci_config_size(pci_dev), size);
125 rtas_st(rets, 0, RTAS_OUT_SUCCESS);
126 rtas_st(rets, 1, val);
129 static void rtas_ibm_read_pci_config(PowerPCCPU *cpu, sPAPRMachineState *spapr,
130 uint32_t token, uint32_t nargs,
131 target_ulong args,
132 uint32_t nret, target_ulong rets)
134 uint64_t buid;
135 uint32_t size, addr;
137 if ((nargs != 4) || (nret != 2)) {
138 rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
139 return;
142 buid = rtas_ldq(args, 1);
143 size = rtas_ld(args, 3);
144 addr = rtas_ld(args, 0);
146 finish_read_pci_config(spapr, buid, addr, size, rets);
149 static void rtas_read_pci_config(PowerPCCPU *cpu, sPAPRMachineState *spapr,
150 uint32_t token, uint32_t nargs,
151 target_ulong args,
152 uint32_t nret, target_ulong rets)
154 uint32_t size, addr;
156 if ((nargs != 2) || (nret != 2)) {
157 rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
158 return;
161 size = rtas_ld(args, 1);
162 addr = rtas_ld(args, 0);
164 finish_read_pci_config(spapr, 0, addr, size, rets);
167 static void finish_write_pci_config(sPAPRMachineState *spapr, uint64_t buid,
168 uint32_t addr, uint32_t size,
169 uint32_t val, target_ulong rets)
171 PCIDevice *pci_dev;
173 if ((size != 1) && (size != 2) && (size != 4)) {
174 /* access must be 1, 2 or 4 bytes */
175 rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
176 return;
179 pci_dev = spapr_pci_find_dev(spapr, buid, addr);
180 addr = rtas_pci_cfgaddr(addr);
182 if (!pci_dev || (addr % size) || (addr >= pci_config_size(pci_dev))) {
183 /* Access must be to a valid device, within bounds and
184 * naturally aligned */
185 rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
186 return;
189 pci_host_config_write_common(pci_dev, addr, pci_config_size(pci_dev),
190 val, size);
192 rtas_st(rets, 0, RTAS_OUT_SUCCESS);
195 static void rtas_ibm_write_pci_config(PowerPCCPU *cpu, sPAPRMachineState *spapr,
196 uint32_t token, uint32_t nargs,
197 target_ulong args,
198 uint32_t nret, target_ulong rets)
200 uint64_t buid;
201 uint32_t val, size, addr;
203 if ((nargs != 5) || (nret != 1)) {
204 rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
205 return;
208 buid = rtas_ldq(args, 1);
209 val = rtas_ld(args, 4);
210 size = rtas_ld(args, 3);
211 addr = rtas_ld(args, 0);
213 finish_write_pci_config(spapr, buid, addr, size, val, rets);
216 static void rtas_write_pci_config(PowerPCCPU *cpu, sPAPRMachineState *spapr,
217 uint32_t token, uint32_t nargs,
218 target_ulong args,
219 uint32_t nret, target_ulong rets)
221 uint32_t val, size, addr;
223 if ((nargs != 3) || (nret != 1)) {
224 rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
225 return;
229 val = rtas_ld(args, 2);
230 size = rtas_ld(args, 1);
231 addr = rtas_ld(args, 0);
233 finish_write_pci_config(spapr, 0, addr, size, val, rets);
237 * Set MSI/MSIX message data.
238 * This is required for msi_notify()/msix_notify() which
239 * will write at the addresses via spapr_msi_write().
241 * If hwaddr == 0, all entries will have .data == first_irq i.e.
242 * table will be reset.
244 static void spapr_msi_setmsg(PCIDevice *pdev, hwaddr addr, bool msix,
245 unsigned first_irq, unsigned req_num)
247 unsigned i;
248 MSIMessage msg = { .address = addr, .data = first_irq };
250 if (!msix) {
251 msi_set_message(pdev, msg);
252 trace_spapr_pci_msi_setup(pdev->name, 0, msg.address);
253 return;
256 for (i = 0; i < req_num; ++i) {
257 msix_set_message(pdev, i, msg);
258 trace_spapr_pci_msi_setup(pdev->name, i, msg.address);
259 if (addr) {
260 ++msg.data;
265 static void rtas_ibm_change_msi(PowerPCCPU *cpu, sPAPRMachineState *spapr,
266 uint32_t token, uint32_t nargs,
267 target_ulong args, uint32_t nret,
268 target_ulong rets)
270 uint32_t config_addr = rtas_ld(args, 0);
271 uint64_t buid = rtas_ldq(args, 1);
272 unsigned int func = rtas_ld(args, 3);
273 unsigned int req_num = rtas_ld(args, 4); /* 0 == remove all */
274 unsigned int seq_num = rtas_ld(args, 5);
275 unsigned int ret_intr_type;
276 unsigned int irq, max_irqs = 0;
277 sPAPRPHBState *phb = NULL;
278 PCIDevice *pdev = NULL;
279 spapr_pci_msi *msi;
280 int *config_addr_key;
281 Error *err = NULL;
282 int i;
284 /* Fins sPAPRPHBState */
285 phb = spapr_pci_find_phb(spapr, buid);
286 if (phb) {
287 pdev = spapr_pci_find_dev(spapr, buid, config_addr);
289 if (!phb || !pdev) {
290 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
291 return;
294 switch (func) {
295 case RTAS_CHANGE_FN:
296 if (msi_present(pdev)) {
297 ret_intr_type = RTAS_TYPE_MSI;
298 } else if (msix_present(pdev)) {
299 ret_intr_type = RTAS_TYPE_MSIX;
300 } else {
301 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
302 return;
304 break;
305 case RTAS_CHANGE_MSI_FN:
306 if (msi_present(pdev)) {
307 ret_intr_type = RTAS_TYPE_MSI;
308 } else {
309 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
310 return;
312 break;
313 case RTAS_CHANGE_MSIX_FN:
314 if (msix_present(pdev)) {
315 ret_intr_type = RTAS_TYPE_MSIX;
316 } else {
317 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
318 return;
320 break;
321 default:
322 error_report("rtas_ibm_change_msi(%u) is not implemented", func);
323 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
324 return;
327 msi = (spapr_pci_msi *) g_hash_table_lookup(phb->msi, &config_addr);
329 /* Releasing MSIs */
330 if (!req_num) {
331 if (!msi) {
332 trace_spapr_pci_msi("Releasing wrong config", config_addr);
333 rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
334 return;
337 spapr_irq_free(spapr, msi->first_irq, msi->num);
338 if (msi_present(pdev)) {
339 spapr_msi_setmsg(pdev, 0, false, 0, 0);
341 if (msix_present(pdev)) {
342 spapr_msi_setmsg(pdev, 0, true, 0, 0);
344 g_hash_table_remove(phb->msi, &config_addr);
346 trace_spapr_pci_msi("Released MSIs", config_addr);
347 rtas_st(rets, 0, RTAS_OUT_SUCCESS);
348 rtas_st(rets, 1, 0);
349 return;
352 /* Enabling MSI */
354 /* Check if the device supports as many IRQs as requested */
355 if (ret_intr_type == RTAS_TYPE_MSI) {
356 max_irqs = msi_nr_vectors_allocated(pdev);
357 } else if (ret_intr_type == RTAS_TYPE_MSIX) {
358 max_irqs = pdev->msix_entries_nr;
360 if (!max_irqs) {
361 error_report("Requested interrupt type %d is not enabled for device %x",
362 ret_intr_type, config_addr);
363 rtas_st(rets, 0, -1); /* Hardware error */
364 return;
366 /* Correct the number if the guest asked for too many */
367 if (req_num > max_irqs) {
368 trace_spapr_pci_msi_retry(config_addr, req_num, max_irqs);
369 req_num = max_irqs;
370 irq = 0; /* to avoid misleading trace */
371 goto out;
374 /* Allocate MSIs */
375 irq = spapr_irq_find(spapr, req_num, ret_intr_type == RTAS_TYPE_MSI, &err);
376 if (err) {
377 error_reportf_err(err, "Can't allocate MSIs for device %x: ",
378 config_addr);
379 rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
380 return;
383 for (i = 0; i < req_num; i++) {
384 spapr_irq_claim(spapr, irq + i, false, &err);
385 if (err) {
386 error_reportf_err(err, "Can't allocate MSIs for device %x: ",
387 config_addr);
388 rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
389 return;
393 /* Release previous MSIs */
394 if (msi) {
395 spapr_irq_free(spapr, msi->first_irq, msi->num);
396 g_hash_table_remove(phb->msi, &config_addr);
399 /* Setup MSI/MSIX vectors in the device (via cfgspace or MSIX BAR) */
400 spapr_msi_setmsg(pdev, SPAPR_PCI_MSI_WINDOW, ret_intr_type == RTAS_TYPE_MSIX,
401 irq, req_num);
403 /* Add MSI device to cache */
404 msi = g_new(spapr_pci_msi, 1);
405 msi->first_irq = irq;
406 msi->num = req_num;
407 config_addr_key = g_new(int, 1);
408 *config_addr_key = config_addr;
409 g_hash_table_insert(phb->msi, config_addr_key, msi);
411 out:
412 rtas_st(rets, 0, RTAS_OUT_SUCCESS);
413 rtas_st(rets, 1, req_num);
414 rtas_st(rets, 2, ++seq_num);
415 if (nret > 3) {
416 rtas_st(rets, 3, ret_intr_type);
419 trace_spapr_pci_rtas_ibm_change_msi(config_addr, func, req_num, irq);
422 static void rtas_ibm_query_interrupt_source_number(PowerPCCPU *cpu,
423 sPAPRMachineState *spapr,
424 uint32_t token,
425 uint32_t nargs,
426 target_ulong args,
427 uint32_t nret,
428 target_ulong rets)
430 uint32_t config_addr = rtas_ld(args, 0);
431 uint64_t buid = rtas_ldq(args, 1);
432 unsigned int intr_src_num = -1, ioa_intr_num = rtas_ld(args, 3);
433 sPAPRPHBState *phb = NULL;
434 PCIDevice *pdev = NULL;
435 spapr_pci_msi *msi;
437 /* Find sPAPRPHBState */
438 phb = spapr_pci_find_phb(spapr, buid);
439 if (phb) {
440 pdev = spapr_pci_find_dev(spapr, buid, config_addr);
442 if (!phb || !pdev) {
443 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
444 return;
447 /* Find device descriptor and start IRQ */
448 msi = (spapr_pci_msi *) g_hash_table_lookup(phb->msi, &config_addr);
449 if (!msi || !msi->first_irq || !msi->num || (ioa_intr_num >= msi->num)) {
450 trace_spapr_pci_msi("Failed to return vector", config_addr);
451 rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
452 return;
454 intr_src_num = msi->first_irq + ioa_intr_num;
455 trace_spapr_pci_rtas_ibm_query_interrupt_source_number(ioa_intr_num,
456 intr_src_num);
458 rtas_st(rets, 0, RTAS_OUT_SUCCESS);
459 rtas_st(rets, 1, intr_src_num);
460 rtas_st(rets, 2, 1);/* 0 == level; 1 == edge */
463 static void rtas_ibm_set_eeh_option(PowerPCCPU *cpu,
464 sPAPRMachineState *spapr,
465 uint32_t token, uint32_t nargs,
466 target_ulong args, uint32_t nret,
467 target_ulong rets)
469 sPAPRPHBState *sphb;
470 uint32_t addr, option;
471 uint64_t buid;
472 int ret;
474 if ((nargs != 4) || (nret != 1)) {
475 goto param_error_exit;
478 buid = rtas_ldq(args, 1);
479 addr = rtas_ld(args, 0);
480 option = rtas_ld(args, 3);
482 sphb = spapr_pci_find_phb(spapr, buid);
483 if (!sphb) {
484 goto param_error_exit;
487 if (!spapr_phb_eeh_available(sphb)) {
488 goto param_error_exit;
491 ret = spapr_phb_vfio_eeh_set_option(sphb, addr, option);
492 rtas_st(rets, 0, ret);
493 return;
495 param_error_exit:
496 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
499 static void rtas_ibm_get_config_addr_info2(PowerPCCPU *cpu,
500 sPAPRMachineState *spapr,
501 uint32_t token, uint32_t nargs,
502 target_ulong args, uint32_t nret,
503 target_ulong rets)
505 sPAPRPHBState *sphb;
506 PCIDevice *pdev;
507 uint32_t addr, option;
508 uint64_t buid;
510 if ((nargs != 4) || (nret != 2)) {
511 goto param_error_exit;
514 buid = rtas_ldq(args, 1);
515 sphb = spapr_pci_find_phb(spapr, buid);
516 if (!sphb) {
517 goto param_error_exit;
520 if (!spapr_phb_eeh_available(sphb)) {
521 goto param_error_exit;
525 * We always have PE address of form "00BB0001". "BB"
526 * represents the bus number of PE's primary bus.
528 option = rtas_ld(args, 3);
529 switch (option) {
530 case RTAS_GET_PE_ADDR:
531 addr = rtas_ld(args, 0);
532 pdev = spapr_pci_find_dev(spapr, buid, addr);
533 if (!pdev) {
534 goto param_error_exit;
537 rtas_st(rets, 1, (pci_bus_num(pci_get_bus(pdev)) << 16) + 1);
538 break;
539 case RTAS_GET_PE_MODE:
540 rtas_st(rets, 1, RTAS_PE_MODE_SHARED);
541 break;
542 default:
543 goto param_error_exit;
546 rtas_st(rets, 0, RTAS_OUT_SUCCESS);
547 return;
549 param_error_exit:
550 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
553 static void rtas_ibm_read_slot_reset_state2(PowerPCCPU *cpu,
554 sPAPRMachineState *spapr,
555 uint32_t token, uint32_t nargs,
556 target_ulong args, uint32_t nret,
557 target_ulong rets)
559 sPAPRPHBState *sphb;
560 uint64_t buid;
561 int state, ret;
563 if ((nargs != 3) || (nret != 4 && nret != 5)) {
564 goto param_error_exit;
567 buid = rtas_ldq(args, 1);
568 sphb = spapr_pci_find_phb(spapr, buid);
569 if (!sphb) {
570 goto param_error_exit;
573 if (!spapr_phb_eeh_available(sphb)) {
574 goto param_error_exit;
577 ret = spapr_phb_vfio_eeh_get_state(sphb, &state);
578 rtas_st(rets, 0, ret);
579 if (ret != RTAS_OUT_SUCCESS) {
580 return;
583 rtas_st(rets, 1, state);
584 rtas_st(rets, 2, RTAS_EEH_SUPPORT);
585 rtas_st(rets, 3, RTAS_EEH_PE_UNAVAIL_INFO);
586 if (nret >= 5) {
587 rtas_st(rets, 4, RTAS_EEH_PE_RECOVER_INFO);
589 return;
591 param_error_exit:
592 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
595 static void rtas_ibm_set_slot_reset(PowerPCCPU *cpu,
596 sPAPRMachineState *spapr,
597 uint32_t token, uint32_t nargs,
598 target_ulong args, uint32_t nret,
599 target_ulong rets)
601 sPAPRPHBState *sphb;
602 uint32_t option;
603 uint64_t buid;
604 int ret;
606 if ((nargs != 4) || (nret != 1)) {
607 goto param_error_exit;
610 buid = rtas_ldq(args, 1);
611 option = rtas_ld(args, 3);
612 sphb = spapr_pci_find_phb(spapr, buid);
613 if (!sphb) {
614 goto param_error_exit;
617 if (!spapr_phb_eeh_available(sphb)) {
618 goto param_error_exit;
621 ret = spapr_phb_vfio_eeh_reset(sphb, option);
622 rtas_st(rets, 0, ret);
623 return;
625 param_error_exit:
626 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
629 static void rtas_ibm_configure_pe(PowerPCCPU *cpu,
630 sPAPRMachineState *spapr,
631 uint32_t token, uint32_t nargs,
632 target_ulong args, uint32_t nret,
633 target_ulong rets)
635 sPAPRPHBState *sphb;
636 uint64_t buid;
637 int ret;
639 if ((nargs != 3) || (nret != 1)) {
640 goto param_error_exit;
643 buid = rtas_ldq(args, 1);
644 sphb = spapr_pci_find_phb(spapr, buid);
645 if (!sphb) {
646 goto param_error_exit;
649 if (!spapr_phb_eeh_available(sphb)) {
650 goto param_error_exit;
653 ret = spapr_phb_vfio_eeh_configure(sphb);
654 rtas_st(rets, 0, ret);
655 return;
657 param_error_exit:
658 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
661 /* To support it later */
662 static void rtas_ibm_slot_error_detail(PowerPCCPU *cpu,
663 sPAPRMachineState *spapr,
664 uint32_t token, uint32_t nargs,
665 target_ulong args, uint32_t nret,
666 target_ulong rets)
668 sPAPRPHBState *sphb;
669 int option;
670 uint64_t buid;
672 if ((nargs != 8) || (nret != 1)) {
673 goto param_error_exit;
676 buid = rtas_ldq(args, 1);
677 sphb = spapr_pci_find_phb(spapr, buid);
678 if (!sphb) {
679 goto param_error_exit;
682 if (!spapr_phb_eeh_available(sphb)) {
683 goto param_error_exit;
686 option = rtas_ld(args, 7);
687 switch (option) {
688 case RTAS_SLOT_TEMP_ERR_LOG:
689 case RTAS_SLOT_PERM_ERR_LOG:
690 break;
691 default:
692 goto param_error_exit;
695 /* We don't have error log yet */
696 rtas_st(rets, 0, RTAS_OUT_NO_ERRORS_FOUND);
697 return;
699 param_error_exit:
700 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
703 static int pci_spapr_swizzle(int slot, int pin)
705 return (slot + pin) % PCI_NUM_PINS;
708 static int pci_spapr_map_irq(PCIDevice *pci_dev, int irq_num)
711 * Here we need to convert pci_dev + irq_num to some unique value
712 * which is less than number of IRQs on the specific bus (4). We
713 * use standard PCI swizzling, that is (slot number + pin number)
714 * % 4.
716 return pci_spapr_swizzle(PCI_SLOT(pci_dev->devfn), irq_num);
719 static void pci_spapr_set_irq(void *opaque, int irq_num, int level)
722 * Here we use the number returned by pci_spapr_map_irq to find a
723 * corresponding qemu_irq.
725 sPAPRPHBState *phb = opaque;
727 trace_spapr_pci_lsi_set(phb->dtbusname, irq_num, phb->lsi_table[irq_num].irq);
728 qemu_set_irq(spapr_phb_lsi_qirq(phb, irq_num), level);
731 static PCIINTxRoute spapr_route_intx_pin_to_irq(void *opaque, int pin)
733 sPAPRPHBState *sphb = SPAPR_PCI_HOST_BRIDGE(opaque);
734 PCIINTxRoute route;
736 route.mode = PCI_INTX_ENABLED;
737 route.irq = sphb->lsi_table[pin].irq;
739 return route;
743 * MSI/MSIX memory region implementation.
744 * The handler handles both MSI and MSIX.
745 * The vector number is encoded in least bits in data.
747 static void spapr_msi_write(void *opaque, hwaddr addr,
748 uint64_t data, unsigned size)
750 sPAPRMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
751 uint32_t irq = data;
753 trace_spapr_pci_msi_write(addr, data, irq);
755 qemu_irq_pulse(spapr_qirq(spapr, irq));
758 static const MemoryRegionOps spapr_msi_ops = {
759 /* There is no .read as the read result is undefined by PCI spec */
760 .read = NULL,
761 .write = spapr_msi_write,
762 .endianness = DEVICE_LITTLE_ENDIAN
766 * PHB PCI device
768 static AddressSpace *spapr_pci_dma_iommu(PCIBus *bus, void *opaque, int devfn)
770 sPAPRPHBState *phb = opaque;
772 return &phb->iommu_as;
775 static char *spapr_phb_vfio_get_loc_code(sPAPRPHBState *sphb, PCIDevice *pdev)
777 char *path = NULL, *buf = NULL, *host = NULL;
779 /* Get the PCI VFIO host id */
780 host = object_property_get_str(OBJECT(pdev), "host", NULL);
781 if (!host) {
782 goto err_out;
785 /* Construct the path of the file that will give us the DT location */
786 path = g_strdup_printf("/sys/bus/pci/devices/%s/devspec", host);
787 g_free(host);
788 if (!g_file_get_contents(path, &buf, NULL, NULL)) {
789 goto err_out;
791 g_free(path);
793 /* Construct and read from host device tree the loc-code */
794 path = g_strdup_printf("/proc/device-tree%s/ibm,loc-code", buf);
795 g_free(buf);
796 if (!g_file_get_contents(path, &buf, NULL, NULL)) {
797 goto err_out;
799 return buf;
801 err_out:
802 g_free(path);
803 return NULL;
806 static char *spapr_phb_get_loc_code(sPAPRPHBState *sphb, PCIDevice *pdev)
808 char *buf;
809 const char *devtype = "qemu";
810 uint32_t busnr = pci_bus_num(PCI_BUS(qdev_get_parent_bus(DEVICE(pdev))));
812 if (object_dynamic_cast(OBJECT(pdev), "vfio-pci")) {
813 buf = spapr_phb_vfio_get_loc_code(sphb, pdev);
814 if (buf) {
815 return buf;
817 devtype = "vfio";
820 * For emulated devices and VFIO-failure case, make up
821 * the loc-code.
823 buf = g_strdup_printf("%s_%s:%04x:%02x:%02x.%x",
824 devtype, pdev->name, sphb->index, busnr,
825 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
826 return buf;
829 /* Macros to operate with address in OF binding to PCI */
830 #define b_x(x, p, l) (((x) & ((1<<(l))-1)) << (p))
831 #define b_n(x) b_x((x), 31, 1) /* 0 if relocatable */
832 #define b_p(x) b_x((x), 30, 1) /* 1 if prefetchable */
833 #define b_t(x) b_x((x), 29, 1) /* 1 if the address is aliased */
834 #define b_ss(x) b_x((x), 24, 2) /* the space code */
835 #define b_bbbbbbbb(x) b_x((x), 16, 8) /* bus number */
836 #define b_ddddd(x) b_x((x), 11, 5) /* device number */
837 #define b_fff(x) b_x((x), 8, 3) /* function number */
838 #define b_rrrrrrrr(x) b_x((x), 0, 8) /* register number */
840 /* for 'reg'/'assigned-addresses' OF properties */
841 #define RESOURCE_CELLS_SIZE 2
842 #define RESOURCE_CELLS_ADDRESS 3
844 typedef struct ResourceFields {
845 uint32_t phys_hi;
846 uint32_t phys_mid;
847 uint32_t phys_lo;
848 uint32_t size_hi;
849 uint32_t size_lo;
850 } QEMU_PACKED ResourceFields;
852 typedef struct ResourceProps {
853 ResourceFields reg[8];
854 ResourceFields assigned[7];
855 uint32_t reg_len;
856 uint32_t assigned_len;
857 } ResourceProps;
859 /* fill in the 'reg'/'assigned-resources' OF properties for
860 * a PCI device. 'reg' describes resource requirements for a
861 * device's IO/MEM regions, 'assigned-addresses' describes the
862 * actual resource assignments.
864 * the properties are arrays of ('phys-addr', 'size') pairs describing
865 * the addressable regions of the PCI device, where 'phys-addr' is a
866 * RESOURCE_CELLS_ADDRESS-tuple of 32-bit integers corresponding to
867 * (phys.hi, phys.mid, phys.lo), and 'size' is a
868 * RESOURCE_CELLS_SIZE-tuple corresponding to (size.hi, size.lo).
870 * phys.hi = 0xYYXXXXZZ, where:
871 * 0xYY = npt000ss
872 * ||| |
873 * ||| +-- space code
874 * ||| |
875 * ||| + 00 if configuration space
876 * ||| + 01 if IO region,
877 * ||| + 10 if 32-bit MEM region
878 * ||| + 11 if 64-bit MEM region
879 * |||
880 * ||+------ for non-relocatable IO: 1 if aliased
881 * || for relocatable IO: 1 if below 64KB
882 * || for MEM: 1 if below 1MB
883 * |+------- 1 if region is prefetchable
884 * +-------- 1 if region is non-relocatable
885 * 0xXXXX = bbbbbbbb dddddfff, encoding bus, slot, and function
886 * bits respectively
887 * 0xZZ = rrrrrrrr, the register number of the BAR corresponding
888 * to the region
890 * phys.mid and phys.lo correspond respectively to the hi/lo portions
891 * of the actual address of the region.
893 * how the phys-addr/size values are used differ slightly between
894 * 'reg' and 'assigned-addresses' properties. namely, 'reg' has
895 * an additional description for the config space region of the
896 * device, and in the case of QEMU has n=0 and phys.mid=phys.lo=0
897 * to describe the region as relocatable, with an address-mapping
898 * that corresponds directly to the PHB's address space for the
899 * resource. 'assigned-addresses' always has n=1 set with an absolute
900 * address assigned for the resource. in general, 'assigned-addresses'
901 * won't be populated, since addresses for PCI devices are generally
902 * unmapped initially and left to the guest to assign.
904 * note also that addresses defined in these properties are, at least
905 * for PAPR guests, relative to the PHBs IO/MEM windows, and
906 * correspond directly to the addresses in the BARs.
908 * in accordance with PCI Bus Binding to Open Firmware,
909 * IEEE Std 1275-1994, section 4.1.1, as implemented by PAPR+ v2.7,
910 * Appendix C.
912 static void populate_resource_props(PCIDevice *d, ResourceProps *rp)
914 int bus_num = pci_bus_num(PCI_BUS(qdev_get_parent_bus(DEVICE(d))));
915 uint32_t dev_id = (b_bbbbbbbb(bus_num) |
916 b_ddddd(PCI_SLOT(d->devfn)) |
917 b_fff(PCI_FUNC(d->devfn)));
918 ResourceFields *reg, *assigned;
919 int i, reg_idx = 0, assigned_idx = 0;
921 /* config space region */
922 reg = &rp->reg[reg_idx++];
923 reg->phys_hi = cpu_to_be32(dev_id);
924 reg->phys_mid = 0;
925 reg->phys_lo = 0;
926 reg->size_hi = 0;
927 reg->size_lo = 0;
929 for (i = 0; i < PCI_NUM_REGIONS; i++) {
930 if (!d->io_regions[i].size) {
931 continue;
934 reg = &rp->reg[reg_idx++];
936 reg->phys_hi = cpu_to_be32(dev_id | b_rrrrrrrr(pci_bar(d, i)));
937 if (d->io_regions[i].type & PCI_BASE_ADDRESS_SPACE_IO) {
938 reg->phys_hi |= cpu_to_be32(b_ss(1));
939 } else if (d->io_regions[i].type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
940 reg->phys_hi |= cpu_to_be32(b_ss(3));
941 } else {
942 reg->phys_hi |= cpu_to_be32(b_ss(2));
944 reg->phys_mid = 0;
945 reg->phys_lo = 0;
946 reg->size_hi = cpu_to_be32(d->io_regions[i].size >> 32);
947 reg->size_lo = cpu_to_be32(d->io_regions[i].size);
949 if (d->io_regions[i].addr == PCI_BAR_UNMAPPED) {
950 continue;
953 assigned = &rp->assigned[assigned_idx++];
954 assigned->phys_hi = cpu_to_be32(reg->phys_hi | b_n(1));
955 assigned->phys_mid = cpu_to_be32(d->io_regions[i].addr >> 32);
956 assigned->phys_lo = cpu_to_be32(d->io_regions[i].addr);
957 assigned->size_hi = reg->size_hi;
958 assigned->size_lo = reg->size_lo;
961 rp->reg_len = reg_idx * sizeof(ResourceFields);
962 rp->assigned_len = assigned_idx * sizeof(ResourceFields);
965 typedef struct PCIClass PCIClass;
966 typedef struct PCISubClass PCISubClass;
967 typedef struct PCIIFace PCIIFace;
969 struct PCIIFace {
970 int iface;
971 const char *name;
974 struct PCISubClass {
975 int subclass;
976 const char *name;
977 const PCIIFace *iface;
980 struct PCIClass {
981 const char *name;
982 const PCISubClass *subc;
985 static const PCISubClass undef_subclass[] = {
986 { PCI_CLASS_NOT_DEFINED_VGA, "display", NULL },
987 { 0xFF, NULL, NULL },
990 static const PCISubClass mass_subclass[] = {
991 { PCI_CLASS_STORAGE_SCSI, "scsi", NULL },
992 { PCI_CLASS_STORAGE_IDE, "ide", NULL },
993 { PCI_CLASS_STORAGE_FLOPPY, "fdc", NULL },
994 { PCI_CLASS_STORAGE_IPI, "ipi", NULL },
995 { PCI_CLASS_STORAGE_RAID, "raid", NULL },
996 { PCI_CLASS_STORAGE_ATA, "ata", NULL },
997 { PCI_CLASS_STORAGE_SATA, "sata", NULL },
998 { PCI_CLASS_STORAGE_SAS, "sas", NULL },
999 { 0xFF, NULL, NULL },
1002 static const PCISubClass net_subclass[] = {
1003 { PCI_CLASS_NETWORK_ETHERNET, "ethernet", NULL },
1004 { PCI_CLASS_NETWORK_TOKEN_RING, "token-ring", NULL },
1005 { PCI_CLASS_NETWORK_FDDI, "fddi", NULL },
1006 { PCI_CLASS_NETWORK_ATM, "atm", NULL },
1007 { PCI_CLASS_NETWORK_ISDN, "isdn", NULL },
1008 { PCI_CLASS_NETWORK_WORLDFIP, "worldfip", NULL },
1009 { PCI_CLASS_NETWORK_PICMG214, "picmg", NULL },
1010 { 0xFF, NULL, NULL },
1013 static const PCISubClass displ_subclass[] = {
1014 { PCI_CLASS_DISPLAY_VGA, "vga", NULL },
1015 { PCI_CLASS_DISPLAY_XGA, "xga", NULL },
1016 { PCI_CLASS_DISPLAY_3D, "3d-controller", NULL },
1017 { 0xFF, NULL, NULL },
1020 static const PCISubClass media_subclass[] = {
1021 { PCI_CLASS_MULTIMEDIA_VIDEO, "video", NULL },
1022 { PCI_CLASS_MULTIMEDIA_AUDIO, "sound", NULL },
1023 { PCI_CLASS_MULTIMEDIA_PHONE, "telephony", NULL },
1024 { 0xFF, NULL, NULL },
1027 static const PCISubClass mem_subclass[] = {
1028 { PCI_CLASS_MEMORY_RAM, "memory", NULL },
1029 { PCI_CLASS_MEMORY_FLASH, "flash", NULL },
1030 { 0xFF, NULL, NULL },
1033 static const PCISubClass bridg_subclass[] = {
1034 { PCI_CLASS_BRIDGE_HOST, "host", NULL },
1035 { PCI_CLASS_BRIDGE_ISA, "isa", NULL },
1036 { PCI_CLASS_BRIDGE_EISA, "eisa", NULL },
1037 { PCI_CLASS_BRIDGE_MC, "mca", NULL },
1038 { PCI_CLASS_BRIDGE_PCI, "pci", NULL },
1039 { PCI_CLASS_BRIDGE_PCMCIA, "pcmcia", NULL },
1040 { PCI_CLASS_BRIDGE_NUBUS, "nubus", NULL },
1041 { PCI_CLASS_BRIDGE_CARDBUS, "cardbus", NULL },
1042 { PCI_CLASS_BRIDGE_RACEWAY, "raceway", NULL },
1043 { PCI_CLASS_BRIDGE_PCI_SEMITP, "semi-transparent-pci", NULL },
1044 { PCI_CLASS_BRIDGE_IB_PCI, "infiniband", NULL },
1045 { 0xFF, NULL, NULL },
1048 static const PCISubClass comm_subclass[] = {
1049 { PCI_CLASS_COMMUNICATION_SERIAL, "serial", NULL },
1050 { PCI_CLASS_COMMUNICATION_PARALLEL, "parallel", NULL },
1051 { PCI_CLASS_COMMUNICATION_MULTISERIAL, "multiport-serial", NULL },
1052 { PCI_CLASS_COMMUNICATION_MODEM, "modem", NULL },
1053 { PCI_CLASS_COMMUNICATION_GPIB, "gpib", NULL },
1054 { PCI_CLASS_COMMUNICATION_SC, "smart-card", NULL },
1055 { 0xFF, NULL, NULL, },
1058 static const PCIIFace pic_iface[] = {
1059 { PCI_CLASS_SYSTEM_PIC_IOAPIC, "io-apic" },
1060 { PCI_CLASS_SYSTEM_PIC_IOXAPIC, "io-xapic" },
1061 { 0xFF, NULL },
1064 static const PCISubClass sys_subclass[] = {
1065 { PCI_CLASS_SYSTEM_PIC, "interrupt-controller", pic_iface },
1066 { PCI_CLASS_SYSTEM_DMA, "dma-controller", NULL },
1067 { PCI_CLASS_SYSTEM_TIMER, "timer", NULL },
1068 { PCI_CLASS_SYSTEM_RTC, "rtc", NULL },
1069 { PCI_CLASS_SYSTEM_PCI_HOTPLUG, "hot-plug-controller", NULL },
1070 { PCI_CLASS_SYSTEM_SDHCI, "sd-host-controller", NULL },
1071 { 0xFF, NULL, NULL },
1074 static const PCISubClass inp_subclass[] = {
1075 { PCI_CLASS_INPUT_KEYBOARD, "keyboard", NULL },
1076 { PCI_CLASS_INPUT_PEN, "pen", NULL },
1077 { PCI_CLASS_INPUT_MOUSE, "mouse", NULL },
1078 { PCI_CLASS_INPUT_SCANNER, "scanner", NULL },
1079 { PCI_CLASS_INPUT_GAMEPORT, "gameport", NULL },
1080 { 0xFF, NULL, NULL },
1083 static const PCISubClass dock_subclass[] = {
1084 { PCI_CLASS_DOCKING_GENERIC, "dock", NULL },
1085 { 0xFF, NULL, NULL },
1088 static const PCISubClass cpu_subclass[] = {
1089 { PCI_CLASS_PROCESSOR_PENTIUM, "pentium", NULL },
1090 { PCI_CLASS_PROCESSOR_POWERPC, "powerpc", NULL },
1091 { PCI_CLASS_PROCESSOR_MIPS, "mips", NULL },
1092 { PCI_CLASS_PROCESSOR_CO, "co-processor", NULL },
1093 { 0xFF, NULL, NULL },
1096 static const PCIIFace usb_iface[] = {
1097 { PCI_CLASS_SERIAL_USB_UHCI, "usb-uhci" },
1098 { PCI_CLASS_SERIAL_USB_OHCI, "usb-ohci", },
1099 { PCI_CLASS_SERIAL_USB_EHCI, "usb-ehci" },
1100 { PCI_CLASS_SERIAL_USB_XHCI, "usb-xhci" },
1101 { PCI_CLASS_SERIAL_USB_UNKNOWN, "usb-unknown" },
1102 { PCI_CLASS_SERIAL_USB_DEVICE, "usb-device" },
1103 { 0xFF, NULL },
1106 static const PCISubClass ser_subclass[] = {
1107 { PCI_CLASS_SERIAL_FIREWIRE, "firewire", NULL },
1108 { PCI_CLASS_SERIAL_ACCESS, "access-bus", NULL },
1109 { PCI_CLASS_SERIAL_SSA, "ssa", NULL },
1110 { PCI_CLASS_SERIAL_USB, "usb", usb_iface },
1111 { PCI_CLASS_SERIAL_FIBER, "fibre-channel", NULL },
1112 { PCI_CLASS_SERIAL_SMBUS, "smb", NULL },
1113 { PCI_CLASS_SERIAL_IB, "infiniband", NULL },
1114 { PCI_CLASS_SERIAL_IPMI, "ipmi", NULL },
1115 { PCI_CLASS_SERIAL_SERCOS, "sercos", NULL },
1116 { PCI_CLASS_SERIAL_CANBUS, "canbus", NULL },
1117 { 0xFF, NULL, NULL },
1120 static const PCISubClass wrl_subclass[] = {
1121 { PCI_CLASS_WIRELESS_IRDA, "irda", NULL },
1122 { PCI_CLASS_WIRELESS_CIR, "consumer-ir", NULL },
1123 { PCI_CLASS_WIRELESS_RF_CONTROLLER, "rf-controller", NULL },
1124 { PCI_CLASS_WIRELESS_BLUETOOTH, "bluetooth", NULL },
1125 { PCI_CLASS_WIRELESS_BROADBAND, "broadband", NULL },
1126 { 0xFF, NULL, NULL },
1129 static const PCISubClass sat_subclass[] = {
1130 { PCI_CLASS_SATELLITE_TV, "satellite-tv", NULL },
1131 { PCI_CLASS_SATELLITE_AUDIO, "satellite-audio", NULL },
1132 { PCI_CLASS_SATELLITE_VOICE, "satellite-voice", NULL },
1133 { PCI_CLASS_SATELLITE_DATA, "satellite-data", NULL },
1134 { 0xFF, NULL, NULL },
1137 static const PCISubClass crypt_subclass[] = {
1138 { PCI_CLASS_CRYPT_NETWORK, "network-encryption", NULL },
1139 { PCI_CLASS_CRYPT_ENTERTAINMENT,
1140 "entertainment-encryption", NULL },
1141 { 0xFF, NULL, NULL },
1144 static const PCISubClass spc_subclass[] = {
1145 { PCI_CLASS_SP_DPIO, "dpio", NULL },
1146 { PCI_CLASS_SP_PERF, "counter", NULL },
1147 { PCI_CLASS_SP_SYNCH, "measurement", NULL },
1148 { PCI_CLASS_SP_MANAGEMENT, "management-card", NULL },
1149 { 0xFF, NULL, NULL },
1152 static const PCIClass pci_classes[] = {
1153 { "legacy-device", undef_subclass },
1154 { "mass-storage", mass_subclass },
1155 { "network", net_subclass },
1156 { "display", displ_subclass, },
1157 { "multimedia-device", media_subclass },
1158 { "memory-controller", mem_subclass },
1159 { "unknown-bridge", bridg_subclass },
1160 { "communication-controller", comm_subclass},
1161 { "system-peripheral", sys_subclass },
1162 { "input-controller", inp_subclass },
1163 { "docking-station", dock_subclass },
1164 { "cpu", cpu_subclass },
1165 { "serial-bus", ser_subclass },
1166 { "wireless-controller", wrl_subclass },
1167 { "intelligent-io", NULL },
1168 { "satellite-device", sat_subclass },
1169 { "encryption", crypt_subclass },
1170 { "data-processing-controller", spc_subclass },
1173 static const char *pci_find_device_name(uint8_t class, uint8_t subclass,
1174 uint8_t iface)
1176 const PCIClass *pclass;
1177 const PCISubClass *psubclass;
1178 const PCIIFace *piface;
1179 const char *name;
1181 if (class >= ARRAY_SIZE(pci_classes)) {
1182 return "pci";
1185 pclass = pci_classes + class;
1186 name = pclass->name;
1188 if (pclass->subc == NULL) {
1189 return name;
1192 psubclass = pclass->subc;
1193 while ((psubclass->subclass & 0xff) != 0xff) {
1194 if ((psubclass->subclass & 0xff) == subclass) {
1195 name = psubclass->name;
1196 break;
1198 psubclass++;
1201 piface = psubclass->iface;
1202 if (piface == NULL) {
1203 return name;
1205 while ((piface->iface & 0xff) != 0xff) {
1206 if ((piface->iface & 0xff) == iface) {
1207 name = piface->name;
1208 break;
1210 piface++;
1213 return name;
1216 static gchar *pci_get_node_name(PCIDevice *dev)
1218 int slot = PCI_SLOT(dev->devfn);
1219 int func = PCI_FUNC(dev->devfn);
1220 uint32_t ccode = pci_default_read_config(dev, PCI_CLASS_PROG, 3);
1221 const char *name;
1223 name = pci_find_device_name((ccode >> 16) & 0xff, (ccode >> 8) & 0xff,
1224 ccode & 0xff);
1226 if (func != 0) {
1227 return g_strdup_printf("%s@%x,%x", name, slot, func);
1228 } else {
1229 return g_strdup_printf("%s@%x", name, slot);
1233 static uint32_t spapr_phb_get_pci_drc_index(sPAPRPHBState *phb,
1234 PCIDevice *pdev);
1236 static void spapr_populate_pci_child_dt(PCIDevice *dev, void *fdt, int offset,
1237 sPAPRPHBState *sphb)
1239 ResourceProps rp;
1240 bool is_bridge = false;
1241 int pci_status;
1242 char *buf = NULL;
1243 uint32_t drc_index = spapr_phb_get_pci_drc_index(sphb, dev);
1244 uint32_t ccode = pci_default_read_config(dev, PCI_CLASS_PROG, 3);
1245 uint32_t max_msi, max_msix;
1247 if (pci_default_read_config(dev, PCI_HEADER_TYPE, 1) ==
1248 PCI_HEADER_TYPE_BRIDGE) {
1249 is_bridge = true;
1252 /* in accordance with PAPR+ v2.7 13.6.3, Table 181 */
1253 _FDT(fdt_setprop_cell(fdt, offset, "vendor-id",
1254 pci_default_read_config(dev, PCI_VENDOR_ID, 2)));
1255 _FDT(fdt_setprop_cell(fdt, offset, "device-id",
1256 pci_default_read_config(dev, PCI_DEVICE_ID, 2)));
1257 _FDT(fdt_setprop_cell(fdt, offset, "revision-id",
1258 pci_default_read_config(dev, PCI_REVISION_ID, 1)));
1259 _FDT(fdt_setprop_cell(fdt, offset, "class-code", ccode));
1260 if (pci_default_read_config(dev, PCI_INTERRUPT_PIN, 1)) {
1261 _FDT(fdt_setprop_cell(fdt, offset, "interrupts",
1262 pci_default_read_config(dev, PCI_INTERRUPT_PIN, 1)));
1265 if (!is_bridge) {
1266 _FDT(fdt_setprop_cell(fdt, offset, "min-grant",
1267 pci_default_read_config(dev, PCI_MIN_GNT, 1)));
1268 _FDT(fdt_setprop_cell(fdt, offset, "max-latency",
1269 pci_default_read_config(dev, PCI_MAX_LAT, 1)));
1272 if (pci_default_read_config(dev, PCI_SUBSYSTEM_ID, 2)) {
1273 _FDT(fdt_setprop_cell(fdt, offset, "subsystem-id",
1274 pci_default_read_config(dev, PCI_SUBSYSTEM_ID, 2)));
1277 if (pci_default_read_config(dev, PCI_SUBSYSTEM_VENDOR_ID, 2)) {
1278 _FDT(fdt_setprop_cell(fdt, offset, "subsystem-vendor-id",
1279 pci_default_read_config(dev, PCI_SUBSYSTEM_VENDOR_ID, 2)));
1282 _FDT(fdt_setprop_cell(fdt, offset, "cache-line-size",
1283 pci_default_read_config(dev, PCI_CACHE_LINE_SIZE, 1)));
1285 /* the following fdt cells are masked off the pci status register */
1286 pci_status = pci_default_read_config(dev, PCI_STATUS, 2);
1287 _FDT(fdt_setprop_cell(fdt, offset, "devsel-speed",
1288 PCI_STATUS_DEVSEL_MASK & pci_status));
1290 if (pci_status & PCI_STATUS_FAST_BACK) {
1291 _FDT(fdt_setprop(fdt, offset, "fast-back-to-back", NULL, 0));
1293 if (pci_status & PCI_STATUS_66MHZ) {
1294 _FDT(fdt_setprop(fdt, offset, "66mhz-capable", NULL, 0));
1296 if (pci_status & PCI_STATUS_UDF) {
1297 _FDT(fdt_setprop(fdt, offset, "udf-supported", NULL, 0));
1300 _FDT(fdt_setprop_string(fdt, offset, "name",
1301 pci_find_device_name((ccode >> 16) & 0xff,
1302 (ccode >> 8) & 0xff,
1303 ccode & 0xff)));
1305 buf = spapr_phb_get_loc_code(sphb, dev);
1306 _FDT(fdt_setprop_string(fdt, offset, "ibm,loc-code", buf));
1307 g_free(buf);
1309 if (drc_index) {
1310 _FDT(fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index));
1313 _FDT(fdt_setprop_cell(fdt, offset, "#address-cells",
1314 RESOURCE_CELLS_ADDRESS));
1315 _FDT(fdt_setprop_cell(fdt, offset, "#size-cells",
1316 RESOURCE_CELLS_SIZE));
1318 if (msi_present(dev)) {
1319 max_msi = msi_nr_vectors_allocated(dev);
1320 if (max_msi) {
1321 _FDT(fdt_setprop_cell(fdt, offset, "ibm,req#msi", max_msi));
1324 if (msix_present(dev)) {
1325 max_msix = dev->msix_entries_nr;
1326 if (max_msix) {
1327 _FDT(fdt_setprop_cell(fdt, offset, "ibm,req#msi-x", max_msix));
1331 populate_resource_props(dev, &rp);
1332 _FDT(fdt_setprop(fdt, offset, "reg", (uint8_t *)rp.reg, rp.reg_len));
1333 _FDT(fdt_setprop(fdt, offset, "assigned-addresses",
1334 (uint8_t *)rp.assigned, rp.assigned_len));
1336 if (sphb->pcie_ecs && pci_is_express(dev)) {
1337 _FDT(fdt_setprop_cell(fdt, offset, "ibm,pci-config-space-type", 0x1));
1341 /* create OF node for pci device and required OF DT properties */
1342 static int spapr_create_pci_child_dt(sPAPRPHBState *phb, PCIDevice *dev,
1343 void *fdt, int node_offset)
1345 int offset;
1346 gchar *nodename;
1348 nodename = pci_get_node_name(dev);
1349 _FDT(offset = fdt_add_subnode(fdt, node_offset, nodename));
1350 g_free(nodename);
1352 spapr_populate_pci_child_dt(dev, fdt, offset, phb);
1354 return offset;
1357 /* Callback to be called during DRC release. */
1358 void spapr_phb_remove_pci_device_cb(DeviceState *dev)
1360 /* some version guests do not wait for completion of a device
1361 * cleanup (generally done asynchronously by the kernel) before
1362 * signaling to QEMU that the device is safe, but instead sleep
1363 * for some 'safe' period of time. unfortunately on a busy host
1364 * this sleep isn't guaranteed to be long enough, resulting in
1365 * bad things like IRQ lines being left asserted during final
1366 * device removal. to deal with this we call reset just prior
1367 * to finalizing the device, which will put the device back into
1368 * an 'idle' state, as the device cleanup code expects.
1370 pci_device_reset(PCI_DEVICE(dev));
1371 object_unparent(OBJECT(dev));
1374 static sPAPRDRConnector *spapr_phb_get_pci_func_drc(sPAPRPHBState *phb,
1375 uint32_t busnr,
1376 int32_t devfn)
1378 return spapr_drc_by_id(TYPE_SPAPR_DRC_PCI,
1379 (phb->index << 16) | (busnr << 8) | devfn);
1382 static sPAPRDRConnector *spapr_phb_get_pci_drc(sPAPRPHBState *phb,
1383 PCIDevice *pdev)
1385 uint32_t busnr = pci_bus_num(PCI_BUS(qdev_get_parent_bus(DEVICE(pdev))));
1386 return spapr_phb_get_pci_func_drc(phb, busnr, pdev->devfn);
1389 static uint32_t spapr_phb_get_pci_drc_index(sPAPRPHBState *phb,
1390 PCIDevice *pdev)
1392 sPAPRDRConnector *drc = spapr_phb_get_pci_drc(phb, pdev);
1394 if (!drc) {
1395 return 0;
1398 return spapr_drc_index(drc);
1401 static void spapr_pci_plug(HotplugHandler *plug_handler,
1402 DeviceState *plugged_dev, Error **errp)
1404 sPAPRPHBState *phb = SPAPR_PCI_HOST_BRIDGE(DEVICE(plug_handler));
1405 PCIDevice *pdev = PCI_DEVICE(plugged_dev);
1406 sPAPRDRConnector *drc = spapr_phb_get_pci_drc(phb, pdev);
1407 Error *local_err = NULL;
1408 PCIBus *bus = PCI_BUS(qdev_get_parent_bus(DEVICE(pdev)));
1409 uint32_t slotnr = PCI_SLOT(pdev->devfn);
1410 void *fdt = NULL;
1411 int fdt_start_offset, fdt_size;
1413 /* if DR is disabled we don't need to do anything in the case of
1414 * hotplug or coldplug callbacks
1416 if (!phb->dr_enabled) {
1417 /* if this is a hotplug operation initiated by the user
1418 * we need to let them know it's not enabled
1420 if (plugged_dev->hotplugged) {
1421 error_setg(&local_err, QERR_BUS_NO_HOTPLUG,
1422 object_get_typename(OBJECT(phb)));
1424 goto out;
1427 g_assert(drc);
1429 /* Following the QEMU convention used for PCIe multifunction
1430 * hotplug, we do not allow functions to be hotplugged to a
1431 * slot that already has function 0 present
1433 if (plugged_dev->hotplugged && bus->devices[PCI_DEVFN(slotnr, 0)] &&
1434 PCI_FUNC(pdev->devfn) != 0) {
1435 error_setg(&local_err, "PCI: slot %d function 0 already ocuppied by %s,"
1436 " additional functions can no longer be exposed to guest.",
1437 slotnr, bus->devices[PCI_DEVFN(slotnr, 0)]->name);
1438 goto out;
1441 fdt = create_device_tree(&fdt_size);
1442 fdt_start_offset = spapr_create_pci_child_dt(phb, pdev, fdt, 0);
1444 spapr_drc_attach(drc, DEVICE(pdev), fdt, fdt_start_offset, &local_err);
1445 if (local_err) {
1446 goto out;
1449 /* If this is function 0, signal hotplug for all the device functions.
1450 * Otherwise defer sending the hotplug event.
1452 if (!spapr_drc_hotplugged(plugged_dev)) {
1453 spapr_drc_reset(drc);
1454 } else if (PCI_FUNC(pdev->devfn) == 0) {
1455 int i;
1457 for (i = 0; i < 8; i++) {
1458 sPAPRDRConnector *func_drc;
1459 sPAPRDRConnectorClass *func_drck;
1460 sPAPRDREntitySense state;
1462 func_drc = spapr_phb_get_pci_func_drc(phb, pci_bus_num(bus),
1463 PCI_DEVFN(slotnr, i));
1464 func_drck = SPAPR_DR_CONNECTOR_GET_CLASS(func_drc);
1465 state = func_drck->dr_entity_sense(func_drc);
1467 if (state == SPAPR_DR_ENTITY_SENSE_PRESENT) {
1468 spapr_hotplug_req_add_by_index(func_drc);
1473 out:
1474 if (local_err) {
1475 error_propagate(errp, local_err);
1476 g_free(fdt);
1480 static void spapr_pci_unplug_request(HotplugHandler *plug_handler,
1481 DeviceState *plugged_dev, Error **errp)
1483 sPAPRPHBState *phb = SPAPR_PCI_HOST_BRIDGE(DEVICE(plug_handler));
1484 PCIDevice *pdev = PCI_DEVICE(plugged_dev);
1485 sPAPRDRConnector *drc = spapr_phb_get_pci_drc(phb, pdev);
1487 if (!phb->dr_enabled) {
1488 error_setg(errp, QERR_BUS_NO_HOTPLUG,
1489 object_get_typename(OBJECT(phb)));
1490 return;
1493 g_assert(drc);
1494 g_assert(drc->dev == plugged_dev);
1496 if (!spapr_drc_unplug_requested(drc)) {
1497 PCIBus *bus = PCI_BUS(qdev_get_parent_bus(DEVICE(pdev)));
1498 uint32_t slotnr = PCI_SLOT(pdev->devfn);
1499 sPAPRDRConnector *func_drc;
1500 sPAPRDRConnectorClass *func_drck;
1501 sPAPRDREntitySense state;
1502 int i;
1504 /* ensure any other present functions are pending unplug */
1505 if (PCI_FUNC(pdev->devfn) == 0) {
1506 for (i = 1; i < 8; i++) {
1507 func_drc = spapr_phb_get_pci_func_drc(phb, pci_bus_num(bus),
1508 PCI_DEVFN(slotnr, i));
1509 func_drck = SPAPR_DR_CONNECTOR_GET_CLASS(func_drc);
1510 state = func_drck->dr_entity_sense(func_drc);
1511 if (state == SPAPR_DR_ENTITY_SENSE_PRESENT
1512 && !spapr_drc_unplug_requested(func_drc)) {
1513 error_setg(errp,
1514 "PCI: slot %d, function %d still present. "
1515 "Must unplug all non-0 functions first.",
1516 slotnr, i);
1517 return;
1522 spapr_drc_detach(drc);
1524 /* if this isn't func 0, defer unplug event. otherwise signal removal
1525 * for all present functions
1527 if (PCI_FUNC(pdev->devfn) == 0) {
1528 for (i = 7; i >= 0; i--) {
1529 func_drc = spapr_phb_get_pci_func_drc(phb, pci_bus_num(bus),
1530 PCI_DEVFN(slotnr, i));
1531 func_drck = SPAPR_DR_CONNECTOR_GET_CLASS(func_drc);
1532 state = func_drck->dr_entity_sense(func_drc);
1533 if (state == SPAPR_DR_ENTITY_SENSE_PRESENT) {
1534 spapr_hotplug_req_remove_by_index(func_drc);
1541 static void spapr_phb_realize(DeviceState *dev, Error **errp)
1543 /* We don't use SPAPR_MACHINE() in order to exit gracefully if the user
1544 * tries to add a sPAPR PHB to a non-pseries machine.
1546 sPAPRMachineState *spapr =
1547 (sPAPRMachineState *) object_dynamic_cast(qdev_get_machine(),
1548 TYPE_SPAPR_MACHINE);
1549 SysBusDevice *s = SYS_BUS_DEVICE(dev);
1550 sPAPRPHBState *sphb = SPAPR_PCI_HOST_BRIDGE(s);
1551 PCIHostState *phb = PCI_HOST_BRIDGE(s);
1552 char *namebuf;
1553 int i;
1554 PCIBus *bus;
1555 uint64_t msi_window_size = 4096;
1556 sPAPRTCETable *tcet;
1557 const unsigned windows_supported =
1558 sphb->ddw_enabled ? SPAPR_PCI_DMA_MAX_WINDOWS : 1;
1560 if (!spapr) {
1561 error_setg(errp, TYPE_SPAPR_PCI_HOST_BRIDGE " needs a pseries machine");
1562 return;
1565 if (sphb->index != (uint32_t)-1) {
1566 sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
1567 Error *local_err = NULL;
1569 smc->phb_placement(spapr, sphb->index,
1570 &sphb->buid, &sphb->io_win_addr,
1571 &sphb->mem_win_addr, &sphb->mem64_win_addr,
1572 windows_supported, sphb->dma_liobn, &local_err);
1573 if (local_err) {
1574 error_propagate(errp, local_err);
1575 return;
1577 } else {
1578 error_setg(errp, "\"index\" for PAPR PHB is mandatory");
1579 return;
1582 if (sphb->mem64_win_size != 0) {
1583 if (sphb->mem_win_size > SPAPR_PCI_MEM32_WIN_SIZE) {
1584 error_setg(errp, "32-bit memory window of size 0x%"HWADDR_PRIx
1585 " (max 2 GiB)", sphb->mem_win_size);
1586 return;
1589 /* 64-bit window defaults to identity mapping */
1590 sphb->mem64_win_pciaddr = sphb->mem64_win_addr;
1591 } else if (sphb->mem_win_size > SPAPR_PCI_MEM32_WIN_SIZE) {
1593 * For compatibility with old configuration, if no 64-bit MMIO
1594 * window is specified, but the ordinary (32-bit) memory
1595 * window is specified as > 2GiB, we treat it as a 2GiB 32-bit
1596 * window, with a 64-bit MMIO window following on immediately
1597 * afterwards
1599 sphb->mem64_win_size = sphb->mem_win_size - SPAPR_PCI_MEM32_WIN_SIZE;
1600 sphb->mem64_win_addr = sphb->mem_win_addr + SPAPR_PCI_MEM32_WIN_SIZE;
1601 sphb->mem64_win_pciaddr =
1602 SPAPR_PCI_MEM_WIN_BUS_OFFSET + SPAPR_PCI_MEM32_WIN_SIZE;
1603 sphb->mem_win_size = SPAPR_PCI_MEM32_WIN_SIZE;
1606 if (spapr_pci_find_phb(spapr, sphb->buid)) {
1607 error_setg(errp, "PCI host bridges must have unique BUIDs");
1608 return;
1611 if (sphb->numa_node != -1 &&
1612 (sphb->numa_node >= MAX_NODES || !numa_info[sphb->numa_node].present)) {
1613 error_setg(errp, "Invalid NUMA node ID for PCI host bridge");
1614 return;
1617 sphb->dtbusname = g_strdup_printf("pci@%" PRIx64, sphb->buid);
1619 /* Initialize memory regions */
1620 namebuf = g_strdup_printf("%s.mmio", sphb->dtbusname);
1621 memory_region_init(&sphb->memspace, OBJECT(sphb), namebuf, UINT64_MAX);
1622 g_free(namebuf);
1624 namebuf = g_strdup_printf("%s.mmio32-alias", sphb->dtbusname);
1625 memory_region_init_alias(&sphb->mem32window, OBJECT(sphb),
1626 namebuf, &sphb->memspace,
1627 SPAPR_PCI_MEM_WIN_BUS_OFFSET, sphb->mem_win_size);
1628 g_free(namebuf);
1629 memory_region_add_subregion(get_system_memory(), sphb->mem_win_addr,
1630 &sphb->mem32window);
1632 if (sphb->mem64_win_size != 0) {
1633 namebuf = g_strdup_printf("%s.mmio64-alias", sphb->dtbusname);
1634 memory_region_init_alias(&sphb->mem64window, OBJECT(sphb),
1635 namebuf, &sphb->memspace,
1636 sphb->mem64_win_pciaddr, sphb->mem64_win_size);
1637 g_free(namebuf);
1639 memory_region_add_subregion(get_system_memory(),
1640 sphb->mem64_win_addr,
1641 &sphb->mem64window);
1644 /* Initialize IO regions */
1645 namebuf = g_strdup_printf("%s.io", sphb->dtbusname);
1646 memory_region_init(&sphb->iospace, OBJECT(sphb),
1647 namebuf, SPAPR_PCI_IO_WIN_SIZE);
1648 g_free(namebuf);
1650 namebuf = g_strdup_printf("%s.io-alias", sphb->dtbusname);
1651 memory_region_init_alias(&sphb->iowindow, OBJECT(sphb), namebuf,
1652 &sphb->iospace, 0, SPAPR_PCI_IO_WIN_SIZE);
1653 g_free(namebuf);
1654 memory_region_add_subregion(get_system_memory(), sphb->io_win_addr,
1655 &sphb->iowindow);
1657 bus = pci_register_root_bus(dev, NULL,
1658 pci_spapr_set_irq, pci_spapr_map_irq, sphb,
1659 &sphb->memspace, &sphb->iospace,
1660 PCI_DEVFN(0, 0), PCI_NUM_PINS, TYPE_PCI_BUS);
1661 phb->bus = bus;
1662 qbus_set_hotplug_handler(BUS(phb->bus), DEVICE(sphb), NULL);
1665 * Initialize PHB address space.
1666 * By default there will be at least one subregion for default
1667 * 32bit DMA window.
1668 * Later the guest might want to create another DMA window
1669 * which will become another memory subregion.
1671 namebuf = g_strdup_printf("%s.iommu-root", sphb->dtbusname);
1672 memory_region_init(&sphb->iommu_root, OBJECT(sphb),
1673 namebuf, UINT64_MAX);
1674 g_free(namebuf);
1675 address_space_init(&sphb->iommu_as, &sphb->iommu_root,
1676 sphb->dtbusname);
1679 * As MSI/MSIX interrupts trigger by writing at MSI/MSIX vectors,
1680 * we need to allocate some memory to catch those writes coming
1681 * from msi_notify()/msix_notify().
1682 * As MSIMessage:addr is going to be the same and MSIMessage:data
1683 * is going to be a VIRQ number, 4 bytes of the MSI MR will only
1684 * be used.
1686 * For KVM we want to ensure that this memory is a full page so that
1687 * our memory slot is of page size granularity.
1689 #ifdef CONFIG_KVM
1690 if (kvm_enabled()) {
1691 msi_window_size = getpagesize();
1693 #endif
1695 memory_region_init_io(&sphb->msiwindow, OBJECT(sphb), &spapr_msi_ops, spapr,
1696 "msi", msi_window_size);
1697 memory_region_add_subregion(&sphb->iommu_root, SPAPR_PCI_MSI_WINDOW,
1698 &sphb->msiwindow);
1700 pci_setup_iommu(bus, spapr_pci_dma_iommu, sphb);
1702 pci_bus_set_route_irq_fn(bus, spapr_route_intx_pin_to_irq);
1704 QLIST_INSERT_HEAD(&spapr->phbs, sphb, list);
1706 /* Initialize the LSI table */
1707 for (i = 0; i < PCI_NUM_PINS; i++) {
1708 uint32_t irq;
1709 Error *local_err = NULL;
1711 irq = spapr_irq_findone(spapr, &local_err);
1712 if (local_err) {
1713 error_propagate(errp, local_err);
1714 error_prepend(errp, "can't allocate LSIs: ");
1715 return;
1718 spapr_irq_claim(spapr, irq, true, &local_err);
1719 if (local_err) {
1720 error_propagate(errp, local_err);
1721 error_prepend(errp, "can't allocate LSIs: ");
1722 return;
1725 sphb->lsi_table[i].irq = irq;
1728 /* allocate connectors for child PCI devices */
1729 if (sphb->dr_enabled) {
1730 for (i = 0; i < PCI_SLOT_MAX * 8; i++) {
1731 spapr_dr_connector_new(OBJECT(phb), TYPE_SPAPR_DRC_PCI,
1732 (sphb->index << 16) | i);
1736 /* DMA setup */
1737 for (i = 0; i < windows_supported; ++i) {
1738 tcet = spapr_tce_new_table(DEVICE(sphb), sphb->dma_liobn[i]);
1739 if (!tcet) {
1740 error_setg(errp, "Creating window#%d failed for %s",
1741 i, sphb->dtbusname);
1742 return;
1744 memory_region_add_subregion(&sphb->iommu_root, 0,
1745 spapr_tce_get_iommu(tcet));
1748 sphb->msi = g_hash_table_new_full(g_int_hash, g_int_equal, g_free, g_free);
1751 static int spapr_phb_children_reset(Object *child, void *opaque)
1753 DeviceState *dev = (DeviceState *) object_dynamic_cast(child, TYPE_DEVICE);
1755 if (dev) {
1756 device_reset(dev);
1759 return 0;
1762 void spapr_phb_dma_reset(sPAPRPHBState *sphb)
1764 int i;
1765 sPAPRTCETable *tcet;
1767 for (i = 0; i < SPAPR_PCI_DMA_MAX_WINDOWS; ++i) {
1768 tcet = spapr_tce_find_by_liobn(sphb->dma_liobn[i]);
1770 if (tcet && tcet->nb_table) {
1771 spapr_tce_table_disable(tcet);
1775 /* Register default 32bit DMA window */
1776 tcet = spapr_tce_find_by_liobn(sphb->dma_liobn[0]);
1777 spapr_tce_table_enable(tcet, SPAPR_TCE_PAGE_SHIFT, sphb->dma_win_addr,
1778 sphb->dma_win_size >> SPAPR_TCE_PAGE_SHIFT);
1781 static void spapr_phb_reset(DeviceState *qdev)
1783 sPAPRPHBState *sphb = SPAPR_PCI_HOST_BRIDGE(qdev);
1785 spapr_phb_dma_reset(sphb);
1787 /* Reset the IOMMU state */
1788 object_child_foreach(OBJECT(qdev), spapr_phb_children_reset, NULL);
1790 if (spapr_phb_eeh_available(SPAPR_PCI_HOST_BRIDGE(qdev))) {
1791 spapr_phb_vfio_reset(qdev);
1795 static Property spapr_phb_properties[] = {
1796 DEFINE_PROP_UINT32("index", sPAPRPHBState, index, -1),
1797 DEFINE_PROP_UINT64("mem_win_size", sPAPRPHBState, mem_win_size,
1798 SPAPR_PCI_MEM32_WIN_SIZE),
1799 DEFINE_PROP_UINT64("mem64_win_size", sPAPRPHBState, mem64_win_size,
1800 SPAPR_PCI_MEM64_WIN_SIZE),
1801 DEFINE_PROP_UINT64("io_win_size", sPAPRPHBState, io_win_size,
1802 SPAPR_PCI_IO_WIN_SIZE),
1803 DEFINE_PROP_BOOL("dynamic-reconfiguration", sPAPRPHBState, dr_enabled,
1804 true),
1805 /* Default DMA window is 0..1GB */
1806 DEFINE_PROP_UINT64("dma_win_addr", sPAPRPHBState, dma_win_addr, 0),
1807 DEFINE_PROP_UINT64("dma_win_size", sPAPRPHBState, dma_win_size, 0x40000000),
1808 DEFINE_PROP_UINT64("dma64_win_addr", sPAPRPHBState, dma64_win_addr,
1809 0x800000000000000ULL),
1810 DEFINE_PROP_BOOL("ddw", sPAPRPHBState, ddw_enabled, true),
1811 DEFINE_PROP_UINT64("pgsz", sPAPRPHBState, page_size_mask,
1812 (1ULL << 12) | (1ULL << 16)),
1813 DEFINE_PROP_UINT32("numa_node", sPAPRPHBState, numa_node, -1),
1814 DEFINE_PROP_BOOL("pre-2.8-migration", sPAPRPHBState,
1815 pre_2_8_migration, false),
1816 DEFINE_PROP_BOOL("pcie-extended-configuration-space", sPAPRPHBState,
1817 pcie_ecs, true),
1818 DEFINE_PROP_END_OF_LIST(),
1821 static const VMStateDescription vmstate_spapr_pci_lsi = {
1822 .name = "spapr_pci/lsi",
1823 .version_id = 1,
1824 .minimum_version_id = 1,
1825 .fields = (VMStateField[]) {
1826 VMSTATE_UINT32_EQUAL(irq, struct spapr_pci_lsi, NULL),
1828 VMSTATE_END_OF_LIST()
1832 static const VMStateDescription vmstate_spapr_pci_msi = {
1833 .name = "spapr_pci/msi",
1834 .version_id = 1,
1835 .minimum_version_id = 1,
1836 .fields = (VMStateField []) {
1837 VMSTATE_UINT32(key, spapr_pci_msi_mig),
1838 VMSTATE_UINT32(value.first_irq, spapr_pci_msi_mig),
1839 VMSTATE_UINT32(value.num, spapr_pci_msi_mig),
1840 VMSTATE_END_OF_LIST()
1844 static int spapr_pci_pre_save(void *opaque)
1846 sPAPRPHBState *sphb = opaque;
1847 GHashTableIter iter;
1848 gpointer key, value;
1849 int i;
1851 if (sphb->pre_2_8_migration) {
1852 sphb->mig_liobn = sphb->dma_liobn[0];
1853 sphb->mig_mem_win_addr = sphb->mem_win_addr;
1854 sphb->mig_mem_win_size = sphb->mem_win_size;
1855 sphb->mig_io_win_addr = sphb->io_win_addr;
1856 sphb->mig_io_win_size = sphb->io_win_size;
1858 if ((sphb->mem64_win_size != 0)
1859 && (sphb->mem64_win_addr
1860 == (sphb->mem_win_addr + sphb->mem_win_size))) {
1861 sphb->mig_mem_win_size += sphb->mem64_win_size;
1865 g_free(sphb->msi_devs);
1866 sphb->msi_devs = NULL;
1867 sphb->msi_devs_num = g_hash_table_size(sphb->msi);
1868 if (!sphb->msi_devs_num) {
1869 return 0;
1871 sphb->msi_devs = g_malloc(sphb->msi_devs_num * sizeof(spapr_pci_msi_mig));
1873 g_hash_table_iter_init(&iter, sphb->msi);
1874 for (i = 0; g_hash_table_iter_next(&iter, &key, &value); ++i) {
1875 sphb->msi_devs[i].key = *(uint32_t *) key;
1876 sphb->msi_devs[i].value = *(spapr_pci_msi *) value;
1879 return 0;
1882 static int spapr_pci_post_load(void *opaque, int version_id)
1884 sPAPRPHBState *sphb = opaque;
1885 gpointer key, value;
1886 int i;
1888 for (i = 0; i < sphb->msi_devs_num; ++i) {
1889 key = g_memdup(&sphb->msi_devs[i].key,
1890 sizeof(sphb->msi_devs[i].key));
1891 value = g_memdup(&sphb->msi_devs[i].value,
1892 sizeof(sphb->msi_devs[i].value));
1893 g_hash_table_insert(sphb->msi, key, value);
1895 g_free(sphb->msi_devs);
1896 sphb->msi_devs = NULL;
1897 sphb->msi_devs_num = 0;
1899 return 0;
1902 static bool pre_2_8_migration(void *opaque, int version_id)
1904 sPAPRPHBState *sphb = opaque;
1906 return sphb->pre_2_8_migration;
1909 static const VMStateDescription vmstate_spapr_pci = {
1910 .name = "spapr_pci",
1911 .version_id = 2,
1912 .minimum_version_id = 2,
1913 .pre_save = spapr_pci_pre_save,
1914 .post_load = spapr_pci_post_load,
1915 .fields = (VMStateField[]) {
1916 VMSTATE_UINT64_EQUAL(buid, sPAPRPHBState, NULL),
1917 VMSTATE_UINT32_TEST(mig_liobn, sPAPRPHBState, pre_2_8_migration),
1918 VMSTATE_UINT64_TEST(mig_mem_win_addr, sPAPRPHBState, pre_2_8_migration),
1919 VMSTATE_UINT64_TEST(mig_mem_win_size, sPAPRPHBState, pre_2_8_migration),
1920 VMSTATE_UINT64_TEST(mig_io_win_addr, sPAPRPHBState, pre_2_8_migration),
1921 VMSTATE_UINT64_TEST(mig_io_win_size, sPAPRPHBState, pre_2_8_migration),
1922 VMSTATE_STRUCT_ARRAY(lsi_table, sPAPRPHBState, PCI_NUM_PINS, 0,
1923 vmstate_spapr_pci_lsi, struct spapr_pci_lsi),
1924 VMSTATE_INT32(msi_devs_num, sPAPRPHBState),
1925 VMSTATE_STRUCT_VARRAY_ALLOC(msi_devs, sPAPRPHBState, msi_devs_num, 0,
1926 vmstate_spapr_pci_msi, spapr_pci_msi_mig),
1927 VMSTATE_END_OF_LIST()
1931 static const char *spapr_phb_root_bus_path(PCIHostState *host_bridge,
1932 PCIBus *rootbus)
1934 sPAPRPHBState *sphb = SPAPR_PCI_HOST_BRIDGE(host_bridge);
1936 return sphb->dtbusname;
1939 static void spapr_phb_class_init(ObjectClass *klass, void *data)
1941 PCIHostBridgeClass *hc = PCI_HOST_BRIDGE_CLASS(klass);
1942 DeviceClass *dc = DEVICE_CLASS(klass);
1943 HotplugHandlerClass *hp = HOTPLUG_HANDLER_CLASS(klass);
1945 hc->root_bus_path = spapr_phb_root_bus_path;
1946 dc->realize = spapr_phb_realize;
1947 dc->props = spapr_phb_properties;
1948 dc->reset = spapr_phb_reset;
1949 dc->vmsd = &vmstate_spapr_pci;
1950 /* Supported by TYPE_SPAPR_MACHINE */
1951 dc->user_creatable = true;
1952 set_bit(DEVICE_CATEGORY_BRIDGE, dc->categories);
1953 hp->plug = spapr_pci_plug;
1954 hp->unplug_request = spapr_pci_unplug_request;
1957 static const TypeInfo spapr_phb_info = {
1958 .name = TYPE_SPAPR_PCI_HOST_BRIDGE,
1959 .parent = TYPE_PCI_HOST_BRIDGE,
1960 .instance_size = sizeof(sPAPRPHBState),
1961 .class_init = spapr_phb_class_init,
1962 .interfaces = (InterfaceInfo[]) {
1963 { TYPE_HOTPLUG_HANDLER },
1968 PCIHostState *spapr_create_phb(sPAPRMachineState *spapr, int index)
1970 DeviceState *dev;
1972 dev = qdev_create(NULL, TYPE_SPAPR_PCI_HOST_BRIDGE);
1973 qdev_prop_set_uint32(dev, "index", index);
1974 qdev_init_nofail(dev);
1976 return PCI_HOST_BRIDGE(dev);
1979 typedef struct sPAPRFDT {
1980 void *fdt;
1981 int node_off;
1982 sPAPRPHBState *sphb;
1983 } sPAPRFDT;
1985 static void spapr_populate_pci_devices_dt(PCIBus *bus, PCIDevice *pdev,
1986 void *opaque)
1988 PCIBus *sec_bus;
1989 sPAPRFDT *p = opaque;
1990 int offset;
1991 sPAPRFDT s_fdt;
1993 offset = spapr_create_pci_child_dt(p->sphb, pdev, p->fdt, p->node_off);
1994 if (!offset) {
1995 error_report("Failed to create pci child device tree node");
1996 return;
1999 if ((pci_default_read_config(pdev, PCI_HEADER_TYPE, 1) !=
2000 PCI_HEADER_TYPE_BRIDGE)) {
2001 return;
2004 sec_bus = pci_bridge_get_sec_bus(PCI_BRIDGE(pdev));
2005 if (!sec_bus) {
2006 return;
2009 s_fdt.fdt = p->fdt;
2010 s_fdt.node_off = offset;
2011 s_fdt.sphb = p->sphb;
2012 pci_for_each_device_reverse(sec_bus, pci_bus_num(sec_bus),
2013 spapr_populate_pci_devices_dt,
2014 &s_fdt);
2017 static void spapr_phb_pci_enumerate_bridge(PCIBus *bus, PCIDevice *pdev,
2018 void *opaque)
2020 unsigned int *bus_no = opaque;
2021 unsigned int primary = *bus_no;
2022 unsigned int subordinate = 0xff;
2023 PCIBus *sec_bus = NULL;
2025 if ((pci_default_read_config(pdev, PCI_HEADER_TYPE, 1) !=
2026 PCI_HEADER_TYPE_BRIDGE)) {
2027 return;
2030 (*bus_no)++;
2031 pci_default_write_config(pdev, PCI_PRIMARY_BUS, primary, 1);
2032 pci_default_write_config(pdev, PCI_SECONDARY_BUS, *bus_no, 1);
2033 pci_default_write_config(pdev, PCI_SUBORDINATE_BUS, *bus_no, 1);
2035 sec_bus = pci_bridge_get_sec_bus(PCI_BRIDGE(pdev));
2036 if (!sec_bus) {
2037 return;
2040 pci_default_write_config(pdev, PCI_SUBORDINATE_BUS, subordinate, 1);
2041 pci_for_each_device(sec_bus, pci_bus_num(sec_bus),
2042 spapr_phb_pci_enumerate_bridge, bus_no);
2043 pci_default_write_config(pdev, PCI_SUBORDINATE_BUS, *bus_no, 1);
2046 static void spapr_phb_pci_enumerate(sPAPRPHBState *phb)
2048 PCIBus *bus = PCI_HOST_BRIDGE(phb)->bus;
2049 unsigned int bus_no = 0;
2051 pci_for_each_device(bus, pci_bus_num(bus),
2052 spapr_phb_pci_enumerate_bridge,
2053 &bus_no);
2057 int spapr_populate_pci_dt(sPAPRPHBState *phb,
2058 uint32_t xics_phandle,
2059 void *fdt)
2061 int bus_off, i, j, ret;
2062 gchar *nodename;
2063 uint32_t bus_range[] = { cpu_to_be32(0), cpu_to_be32(0xff) };
2064 struct {
2065 uint32_t hi;
2066 uint64_t child;
2067 uint64_t parent;
2068 uint64_t size;
2069 } QEMU_PACKED ranges[] = {
2071 cpu_to_be32(b_ss(1)), cpu_to_be64(0),
2072 cpu_to_be64(phb->io_win_addr),
2073 cpu_to_be64(memory_region_size(&phb->iospace)),
2076 cpu_to_be32(b_ss(2)), cpu_to_be64(SPAPR_PCI_MEM_WIN_BUS_OFFSET),
2077 cpu_to_be64(phb->mem_win_addr),
2078 cpu_to_be64(phb->mem_win_size),
2081 cpu_to_be32(b_ss(3)), cpu_to_be64(phb->mem64_win_pciaddr),
2082 cpu_to_be64(phb->mem64_win_addr),
2083 cpu_to_be64(phb->mem64_win_size),
2086 const unsigned sizeof_ranges =
2087 (phb->mem64_win_size ? 3 : 2) * sizeof(ranges[0]);
2088 uint64_t bus_reg[] = { cpu_to_be64(phb->buid), 0 };
2089 uint32_t interrupt_map_mask[] = {
2090 cpu_to_be32(b_ddddd(-1)|b_fff(0)), 0x0, 0x0, cpu_to_be32(-1)};
2091 uint32_t interrupt_map[PCI_SLOT_MAX * PCI_NUM_PINS][7];
2092 uint32_t ddw_applicable[] = {
2093 cpu_to_be32(RTAS_IBM_QUERY_PE_DMA_WINDOW),
2094 cpu_to_be32(RTAS_IBM_CREATE_PE_DMA_WINDOW),
2095 cpu_to_be32(RTAS_IBM_REMOVE_PE_DMA_WINDOW)
2097 uint32_t ddw_extensions[] = {
2098 cpu_to_be32(1),
2099 cpu_to_be32(RTAS_IBM_RESET_PE_DMA_WINDOW)
2101 uint32_t associativity[] = {cpu_to_be32(0x4),
2102 cpu_to_be32(0x0),
2103 cpu_to_be32(0x0),
2104 cpu_to_be32(0x0),
2105 cpu_to_be32(phb->numa_node)};
2106 sPAPRTCETable *tcet;
2107 PCIBus *bus = PCI_HOST_BRIDGE(phb)->bus;
2108 sPAPRFDT s_fdt;
2110 /* Start populating the FDT */
2111 nodename = g_strdup_printf("pci@%" PRIx64, phb->buid);
2112 _FDT(bus_off = fdt_add_subnode(fdt, 0, nodename));
2113 g_free(nodename);
2115 /* Write PHB properties */
2116 _FDT(fdt_setprop_string(fdt, bus_off, "device_type", "pci"));
2117 _FDT(fdt_setprop_string(fdt, bus_off, "compatible", "IBM,Logical_PHB"));
2118 _FDT(fdt_setprop_cell(fdt, bus_off, "#address-cells", 0x3));
2119 _FDT(fdt_setprop_cell(fdt, bus_off, "#size-cells", 0x2));
2120 _FDT(fdt_setprop_cell(fdt, bus_off, "#interrupt-cells", 0x1));
2121 _FDT(fdt_setprop(fdt, bus_off, "used-by-rtas", NULL, 0));
2122 _FDT(fdt_setprop(fdt, bus_off, "bus-range", &bus_range, sizeof(bus_range)));
2123 _FDT(fdt_setprop(fdt, bus_off, "ranges", &ranges, sizeof_ranges));
2124 _FDT(fdt_setprop(fdt, bus_off, "reg", &bus_reg, sizeof(bus_reg)));
2125 _FDT(fdt_setprop_cell(fdt, bus_off, "ibm,pci-config-space-type", 0x1));
2126 _FDT(fdt_setprop_cell(fdt, bus_off, "ibm,pe-total-#msi", XICS_IRQS_SPAPR));
2128 /* Dynamic DMA window */
2129 if (phb->ddw_enabled) {
2130 _FDT(fdt_setprop(fdt, bus_off, "ibm,ddw-applicable", &ddw_applicable,
2131 sizeof(ddw_applicable)));
2132 _FDT(fdt_setprop(fdt, bus_off, "ibm,ddw-extensions",
2133 &ddw_extensions, sizeof(ddw_extensions)));
2136 /* Advertise NUMA via ibm,associativity */
2137 if (phb->numa_node != -1) {
2138 _FDT(fdt_setprop(fdt, bus_off, "ibm,associativity", associativity,
2139 sizeof(associativity)));
2142 /* Build the interrupt-map, this must matches what is done
2143 * in pci_spapr_map_irq
2145 _FDT(fdt_setprop(fdt, bus_off, "interrupt-map-mask",
2146 &interrupt_map_mask, sizeof(interrupt_map_mask)));
2147 for (i = 0; i < PCI_SLOT_MAX; i++) {
2148 for (j = 0; j < PCI_NUM_PINS; j++) {
2149 uint32_t *irqmap = interrupt_map[i*PCI_NUM_PINS + j];
2150 int lsi_num = pci_spapr_swizzle(i, j);
2152 irqmap[0] = cpu_to_be32(b_ddddd(i)|b_fff(0));
2153 irqmap[1] = 0;
2154 irqmap[2] = 0;
2155 irqmap[3] = cpu_to_be32(j+1);
2156 irqmap[4] = cpu_to_be32(xics_phandle);
2157 spapr_dt_xics_irq(&irqmap[5], phb->lsi_table[lsi_num].irq, true);
2160 /* Write interrupt map */
2161 _FDT(fdt_setprop(fdt, bus_off, "interrupt-map", &interrupt_map,
2162 sizeof(interrupt_map)));
2164 tcet = spapr_tce_find_by_liobn(phb->dma_liobn[0]);
2165 if (!tcet) {
2166 return -1;
2168 spapr_dma_dt(fdt, bus_off, "ibm,dma-window",
2169 tcet->liobn, tcet->bus_offset,
2170 tcet->nb_table << tcet->page_shift);
2172 /* Walk the bridges and program the bus numbers*/
2173 spapr_phb_pci_enumerate(phb);
2174 _FDT(fdt_setprop_cell(fdt, bus_off, "qemu,phb-enumerated", 0x1));
2176 /* Populate tree nodes with PCI devices attached */
2177 s_fdt.fdt = fdt;
2178 s_fdt.node_off = bus_off;
2179 s_fdt.sphb = phb;
2180 pci_for_each_device_reverse(bus, pci_bus_num(bus),
2181 spapr_populate_pci_devices_dt,
2182 &s_fdt);
2184 ret = spapr_drc_populate_dt(fdt, bus_off, OBJECT(phb),
2185 SPAPR_DR_CONNECTOR_TYPE_PCI);
2186 if (ret) {
2187 return ret;
2190 return 0;
2193 void spapr_pci_rtas_init(void)
2195 spapr_rtas_register(RTAS_READ_PCI_CONFIG, "read-pci-config",
2196 rtas_read_pci_config);
2197 spapr_rtas_register(RTAS_WRITE_PCI_CONFIG, "write-pci-config",
2198 rtas_write_pci_config);
2199 spapr_rtas_register(RTAS_IBM_READ_PCI_CONFIG, "ibm,read-pci-config",
2200 rtas_ibm_read_pci_config);
2201 spapr_rtas_register(RTAS_IBM_WRITE_PCI_CONFIG, "ibm,write-pci-config",
2202 rtas_ibm_write_pci_config);
2203 if (msi_nonbroken) {
2204 spapr_rtas_register(RTAS_IBM_QUERY_INTERRUPT_SOURCE_NUMBER,
2205 "ibm,query-interrupt-source-number",
2206 rtas_ibm_query_interrupt_source_number);
2207 spapr_rtas_register(RTAS_IBM_CHANGE_MSI, "ibm,change-msi",
2208 rtas_ibm_change_msi);
2211 spapr_rtas_register(RTAS_IBM_SET_EEH_OPTION,
2212 "ibm,set-eeh-option",
2213 rtas_ibm_set_eeh_option);
2214 spapr_rtas_register(RTAS_IBM_GET_CONFIG_ADDR_INFO2,
2215 "ibm,get-config-addr-info2",
2216 rtas_ibm_get_config_addr_info2);
2217 spapr_rtas_register(RTAS_IBM_READ_SLOT_RESET_STATE2,
2218 "ibm,read-slot-reset-state2",
2219 rtas_ibm_read_slot_reset_state2);
2220 spapr_rtas_register(RTAS_IBM_SET_SLOT_RESET,
2221 "ibm,set-slot-reset",
2222 rtas_ibm_set_slot_reset);
2223 spapr_rtas_register(RTAS_IBM_CONFIGURE_PE,
2224 "ibm,configure-pe",
2225 rtas_ibm_configure_pe);
2226 spapr_rtas_register(RTAS_IBM_SLOT_ERROR_DETAIL,
2227 "ibm,slot-error-detail",
2228 rtas_ibm_slot_error_detail);
2231 static void spapr_pci_register_types(void)
2233 type_register_static(&spapr_phb_info);
2236 type_init(spapr_pci_register_types)
2238 static int spapr_switch_one_vga(DeviceState *dev, void *opaque)
2240 bool be = *(bool *)opaque;
2242 if (object_dynamic_cast(OBJECT(dev), "VGA")
2243 || object_dynamic_cast(OBJECT(dev), "secondary-vga")) {
2244 object_property_set_bool(OBJECT(dev), be, "big-endian-framebuffer",
2245 &error_abort);
2247 return 0;
2250 void spapr_pci_switch_vga(bool big_endian)
2252 sPAPRMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
2253 sPAPRPHBState *sphb;
2256 * For backward compatibility with existing guests, we switch
2257 * the endianness of the VGA controller when changing the guest
2258 * interrupt mode
2260 QLIST_FOREACH(sphb, &spapr->phbs, list) {
2261 BusState *bus = &PCI_HOST_BRIDGE(sphb)->bus->qbus;
2262 qbus_walk_children(bus, spapr_switch_one_vga, NULL, NULL, NULL,
2263 &big_endian);