spapr: introduce a fixed IRQ number space
[qemu/ar7.git] / hw / ppc / spapr_pci.c
blob3791ced6c536c0be7179f68f881caf15ba6a8407
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 if (!SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
338 spapr_irq_msi_free(spapr, msi->first_irq, msi->num);
340 spapr_irq_free(spapr, msi->first_irq, msi->num);
341 if (msi_present(pdev)) {
342 spapr_msi_setmsg(pdev, 0, false, 0, 0);
344 if (msix_present(pdev)) {
345 spapr_msi_setmsg(pdev, 0, true, 0, 0);
347 g_hash_table_remove(phb->msi, &config_addr);
349 trace_spapr_pci_msi("Released MSIs", config_addr);
350 rtas_st(rets, 0, RTAS_OUT_SUCCESS);
351 rtas_st(rets, 1, 0);
352 return;
355 /* Enabling MSI */
357 /* Check if the device supports as many IRQs as requested */
358 if (ret_intr_type == RTAS_TYPE_MSI) {
359 max_irqs = msi_nr_vectors_allocated(pdev);
360 } else if (ret_intr_type == RTAS_TYPE_MSIX) {
361 max_irqs = pdev->msix_entries_nr;
363 if (!max_irqs) {
364 error_report("Requested interrupt type %d is not enabled for device %x",
365 ret_intr_type, config_addr);
366 rtas_st(rets, 0, -1); /* Hardware error */
367 return;
369 /* Correct the number if the guest asked for too many */
370 if (req_num > max_irqs) {
371 trace_spapr_pci_msi_retry(config_addr, req_num, max_irqs);
372 req_num = max_irqs;
373 irq = 0; /* to avoid misleading trace */
374 goto out;
377 /* Allocate MSIs */
378 if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
379 irq = spapr_irq_find(spapr, req_num, ret_intr_type == RTAS_TYPE_MSI,
380 &err);
381 } else {
382 irq = spapr_irq_msi_alloc(spapr, req_num,
383 ret_intr_type == RTAS_TYPE_MSI, &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;
392 for (i = 0; i < req_num; i++) {
393 spapr_irq_claim(spapr, irq + i, false, &err);
394 if (err) {
395 error_reportf_err(err, "Can't allocate MSIs for device %x: ",
396 config_addr);
397 rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
398 return;
402 /* Release previous MSIs */
403 if (msi) {
404 if (!SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
405 spapr_irq_msi_free(spapr, msi->first_irq, msi->num);
407 spapr_irq_free(spapr, msi->first_irq, msi->num);
408 g_hash_table_remove(phb->msi, &config_addr);
411 /* Setup MSI/MSIX vectors in the device (via cfgspace or MSIX BAR) */
412 spapr_msi_setmsg(pdev, SPAPR_PCI_MSI_WINDOW, ret_intr_type == RTAS_TYPE_MSIX,
413 irq, req_num);
415 /* Add MSI device to cache */
416 msi = g_new(spapr_pci_msi, 1);
417 msi->first_irq = irq;
418 msi->num = req_num;
419 config_addr_key = g_new(int, 1);
420 *config_addr_key = config_addr;
421 g_hash_table_insert(phb->msi, config_addr_key, msi);
423 out:
424 rtas_st(rets, 0, RTAS_OUT_SUCCESS);
425 rtas_st(rets, 1, req_num);
426 rtas_st(rets, 2, ++seq_num);
427 if (nret > 3) {
428 rtas_st(rets, 3, ret_intr_type);
431 trace_spapr_pci_rtas_ibm_change_msi(config_addr, func, req_num, irq);
434 static void rtas_ibm_query_interrupt_source_number(PowerPCCPU *cpu,
435 sPAPRMachineState *spapr,
436 uint32_t token,
437 uint32_t nargs,
438 target_ulong args,
439 uint32_t nret,
440 target_ulong rets)
442 uint32_t config_addr = rtas_ld(args, 0);
443 uint64_t buid = rtas_ldq(args, 1);
444 unsigned int intr_src_num = -1, ioa_intr_num = rtas_ld(args, 3);
445 sPAPRPHBState *phb = NULL;
446 PCIDevice *pdev = NULL;
447 spapr_pci_msi *msi;
449 /* Find sPAPRPHBState */
450 phb = spapr_pci_find_phb(spapr, buid);
451 if (phb) {
452 pdev = spapr_pci_find_dev(spapr, buid, config_addr);
454 if (!phb || !pdev) {
455 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
456 return;
459 /* Find device descriptor and start IRQ */
460 msi = (spapr_pci_msi *) g_hash_table_lookup(phb->msi, &config_addr);
461 if (!msi || !msi->first_irq || !msi->num || (ioa_intr_num >= msi->num)) {
462 trace_spapr_pci_msi("Failed to return vector", config_addr);
463 rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
464 return;
466 intr_src_num = msi->first_irq + ioa_intr_num;
467 trace_spapr_pci_rtas_ibm_query_interrupt_source_number(ioa_intr_num,
468 intr_src_num);
470 rtas_st(rets, 0, RTAS_OUT_SUCCESS);
471 rtas_st(rets, 1, intr_src_num);
472 rtas_st(rets, 2, 1);/* 0 == level; 1 == edge */
475 static void rtas_ibm_set_eeh_option(PowerPCCPU *cpu,
476 sPAPRMachineState *spapr,
477 uint32_t token, uint32_t nargs,
478 target_ulong args, uint32_t nret,
479 target_ulong rets)
481 sPAPRPHBState *sphb;
482 uint32_t addr, option;
483 uint64_t buid;
484 int ret;
486 if ((nargs != 4) || (nret != 1)) {
487 goto param_error_exit;
490 buid = rtas_ldq(args, 1);
491 addr = rtas_ld(args, 0);
492 option = rtas_ld(args, 3);
494 sphb = spapr_pci_find_phb(spapr, buid);
495 if (!sphb) {
496 goto param_error_exit;
499 if (!spapr_phb_eeh_available(sphb)) {
500 goto param_error_exit;
503 ret = spapr_phb_vfio_eeh_set_option(sphb, addr, option);
504 rtas_st(rets, 0, ret);
505 return;
507 param_error_exit:
508 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
511 static void rtas_ibm_get_config_addr_info2(PowerPCCPU *cpu,
512 sPAPRMachineState *spapr,
513 uint32_t token, uint32_t nargs,
514 target_ulong args, uint32_t nret,
515 target_ulong rets)
517 sPAPRPHBState *sphb;
518 PCIDevice *pdev;
519 uint32_t addr, option;
520 uint64_t buid;
522 if ((nargs != 4) || (nret != 2)) {
523 goto param_error_exit;
526 buid = rtas_ldq(args, 1);
527 sphb = spapr_pci_find_phb(spapr, buid);
528 if (!sphb) {
529 goto param_error_exit;
532 if (!spapr_phb_eeh_available(sphb)) {
533 goto param_error_exit;
537 * We always have PE address of form "00BB0001". "BB"
538 * represents the bus number of PE's primary bus.
540 option = rtas_ld(args, 3);
541 switch (option) {
542 case RTAS_GET_PE_ADDR:
543 addr = rtas_ld(args, 0);
544 pdev = spapr_pci_find_dev(spapr, buid, addr);
545 if (!pdev) {
546 goto param_error_exit;
549 rtas_st(rets, 1, (pci_bus_num(pci_get_bus(pdev)) << 16) + 1);
550 break;
551 case RTAS_GET_PE_MODE:
552 rtas_st(rets, 1, RTAS_PE_MODE_SHARED);
553 break;
554 default:
555 goto param_error_exit;
558 rtas_st(rets, 0, RTAS_OUT_SUCCESS);
559 return;
561 param_error_exit:
562 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
565 static void rtas_ibm_read_slot_reset_state2(PowerPCCPU *cpu,
566 sPAPRMachineState *spapr,
567 uint32_t token, uint32_t nargs,
568 target_ulong args, uint32_t nret,
569 target_ulong rets)
571 sPAPRPHBState *sphb;
572 uint64_t buid;
573 int state, ret;
575 if ((nargs != 3) || (nret != 4 && nret != 5)) {
576 goto param_error_exit;
579 buid = rtas_ldq(args, 1);
580 sphb = spapr_pci_find_phb(spapr, buid);
581 if (!sphb) {
582 goto param_error_exit;
585 if (!spapr_phb_eeh_available(sphb)) {
586 goto param_error_exit;
589 ret = spapr_phb_vfio_eeh_get_state(sphb, &state);
590 rtas_st(rets, 0, ret);
591 if (ret != RTAS_OUT_SUCCESS) {
592 return;
595 rtas_st(rets, 1, state);
596 rtas_st(rets, 2, RTAS_EEH_SUPPORT);
597 rtas_st(rets, 3, RTAS_EEH_PE_UNAVAIL_INFO);
598 if (nret >= 5) {
599 rtas_st(rets, 4, RTAS_EEH_PE_RECOVER_INFO);
601 return;
603 param_error_exit:
604 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
607 static void rtas_ibm_set_slot_reset(PowerPCCPU *cpu,
608 sPAPRMachineState *spapr,
609 uint32_t token, uint32_t nargs,
610 target_ulong args, uint32_t nret,
611 target_ulong rets)
613 sPAPRPHBState *sphb;
614 uint32_t option;
615 uint64_t buid;
616 int ret;
618 if ((nargs != 4) || (nret != 1)) {
619 goto param_error_exit;
622 buid = rtas_ldq(args, 1);
623 option = rtas_ld(args, 3);
624 sphb = spapr_pci_find_phb(spapr, buid);
625 if (!sphb) {
626 goto param_error_exit;
629 if (!spapr_phb_eeh_available(sphb)) {
630 goto param_error_exit;
633 ret = spapr_phb_vfio_eeh_reset(sphb, option);
634 rtas_st(rets, 0, ret);
635 return;
637 param_error_exit:
638 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
641 static void rtas_ibm_configure_pe(PowerPCCPU *cpu,
642 sPAPRMachineState *spapr,
643 uint32_t token, uint32_t nargs,
644 target_ulong args, uint32_t nret,
645 target_ulong rets)
647 sPAPRPHBState *sphb;
648 uint64_t buid;
649 int ret;
651 if ((nargs != 3) || (nret != 1)) {
652 goto param_error_exit;
655 buid = rtas_ldq(args, 1);
656 sphb = spapr_pci_find_phb(spapr, buid);
657 if (!sphb) {
658 goto param_error_exit;
661 if (!spapr_phb_eeh_available(sphb)) {
662 goto param_error_exit;
665 ret = spapr_phb_vfio_eeh_configure(sphb);
666 rtas_st(rets, 0, ret);
667 return;
669 param_error_exit:
670 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
673 /* To support it later */
674 static void rtas_ibm_slot_error_detail(PowerPCCPU *cpu,
675 sPAPRMachineState *spapr,
676 uint32_t token, uint32_t nargs,
677 target_ulong args, uint32_t nret,
678 target_ulong rets)
680 sPAPRPHBState *sphb;
681 int option;
682 uint64_t buid;
684 if ((nargs != 8) || (nret != 1)) {
685 goto param_error_exit;
688 buid = rtas_ldq(args, 1);
689 sphb = spapr_pci_find_phb(spapr, buid);
690 if (!sphb) {
691 goto param_error_exit;
694 if (!spapr_phb_eeh_available(sphb)) {
695 goto param_error_exit;
698 option = rtas_ld(args, 7);
699 switch (option) {
700 case RTAS_SLOT_TEMP_ERR_LOG:
701 case RTAS_SLOT_PERM_ERR_LOG:
702 break;
703 default:
704 goto param_error_exit;
707 /* We don't have error log yet */
708 rtas_st(rets, 0, RTAS_OUT_NO_ERRORS_FOUND);
709 return;
711 param_error_exit:
712 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
715 static int pci_spapr_swizzle(int slot, int pin)
717 return (slot + pin) % PCI_NUM_PINS;
720 static int pci_spapr_map_irq(PCIDevice *pci_dev, int irq_num)
723 * Here we need to convert pci_dev + irq_num to some unique value
724 * which is less than number of IRQs on the specific bus (4). We
725 * use standard PCI swizzling, that is (slot number + pin number)
726 * % 4.
728 return pci_spapr_swizzle(PCI_SLOT(pci_dev->devfn), irq_num);
731 static void pci_spapr_set_irq(void *opaque, int irq_num, int level)
734 * Here we use the number returned by pci_spapr_map_irq to find a
735 * corresponding qemu_irq.
737 sPAPRPHBState *phb = opaque;
739 trace_spapr_pci_lsi_set(phb->dtbusname, irq_num, phb->lsi_table[irq_num].irq);
740 qemu_set_irq(spapr_phb_lsi_qirq(phb, irq_num), level);
743 static PCIINTxRoute spapr_route_intx_pin_to_irq(void *opaque, int pin)
745 sPAPRPHBState *sphb = SPAPR_PCI_HOST_BRIDGE(opaque);
746 PCIINTxRoute route;
748 route.mode = PCI_INTX_ENABLED;
749 route.irq = sphb->lsi_table[pin].irq;
751 return route;
755 * MSI/MSIX memory region implementation.
756 * The handler handles both MSI and MSIX.
757 * The vector number is encoded in least bits in data.
759 static void spapr_msi_write(void *opaque, hwaddr addr,
760 uint64_t data, unsigned size)
762 sPAPRMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
763 uint32_t irq = data;
765 trace_spapr_pci_msi_write(addr, data, irq);
767 qemu_irq_pulse(spapr_qirq(spapr, irq));
770 static const MemoryRegionOps spapr_msi_ops = {
771 /* There is no .read as the read result is undefined by PCI spec */
772 .read = NULL,
773 .write = spapr_msi_write,
774 .endianness = DEVICE_LITTLE_ENDIAN
778 * PHB PCI device
780 static AddressSpace *spapr_pci_dma_iommu(PCIBus *bus, void *opaque, int devfn)
782 sPAPRPHBState *phb = opaque;
784 return &phb->iommu_as;
787 static char *spapr_phb_vfio_get_loc_code(sPAPRPHBState *sphb, PCIDevice *pdev)
789 char *path = NULL, *buf = NULL, *host = NULL;
791 /* Get the PCI VFIO host id */
792 host = object_property_get_str(OBJECT(pdev), "host", NULL);
793 if (!host) {
794 goto err_out;
797 /* Construct the path of the file that will give us the DT location */
798 path = g_strdup_printf("/sys/bus/pci/devices/%s/devspec", host);
799 g_free(host);
800 if (!g_file_get_contents(path, &buf, NULL, NULL)) {
801 goto err_out;
803 g_free(path);
805 /* Construct and read from host device tree the loc-code */
806 path = g_strdup_printf("/proc/device-tree%s/ibm,loc-code", buf);
807 g_free(buf);
808 if (!g_file_get_contents(path, &buf, NULL, NULL)) {
809 goto err_out;
811 return buf;
813 err_out:
814 g_free(path);
815 return NULL;
818 static char *spapr_phb_get_loc_code(sPAPRPHBState *sphb, PCIDevice *pdev)
820 char *buf;
821 const char *devtype = "qemu";
822 uint32_t busnr = pci_bus_num(PCI_BUS(qdev_get_parent_bus(DEVICE(pdev))));
824 if (object_dynamic_cast(OBJECT(pdev), "vfio-pci")) {
825 buf = spapr_phb_vfio_get_loc_code(sphb, pdev);
826 if (buf) {
827 return buf;
829 devtype = "vfio";
832 * For emulated devices and VFIO-failure case, make up
833 * the loc-code.
835 buf = g_strdup_printf("%s_%s:%04x:%02x:%02x.%x",
836 devtype, pdev->name, sphb->index, busnr,
837 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
838 return buf;
841 /* Macros to operate with address in OF binding to PCI */
842 #define b_x(x, p, l) (((x) & ((1<<(l))-1)) << (p))
843 #define b_n(x) b_x((x), 31, 1) /* 0 if relocatable */
844 #define b_p(x) b_x((x), 30, 1) /* 1 if prefetchable */
845 #define b_t(x) b_x((x), 29, 1) /* 1 if the address is aliased */
846 #define b_ss(x) b_x((x), 24, 2) /* the space code */
847 #define b_bbbbbbbb(x) b_x((x), 16, 8) /* bus number */
848 #define b_ddddd(x) b_x((x), 11, 5) /* device number */
849 #define b_fff(x) b_x((x), 8, 3) /* function number */
850 #define b_rrrrrrrr(x) b_x((x), 0, 8) /* register number */
852 /* for 'reg'/'assigned-addresses' OF properties */
853 #define RESOURCE_CELLS_SIZE 2
854 #define RESOURCE_CELLS_ADDRESS 3
856 typedef struct ResourceFields {
857 uint32_t phys_hi;
858 uint32_t phys_mid;
859 uint32_t phys_lo;
860 uint32_t size_hi;
861 uint32_t size_lo;
862 } QEMU_PACKED ResourceFields;
864 typedef struct ResourceProps {
865 ResourceFields reg[8];
866 ResourceFields assigned[7];
867 uint32_t reg_len;
868 uint32_t assigned_len;
869 } ResourceProps;
871 /* fill in the 'reg'/'assigned-resources' OF properties for
872 * a PCI device. 'reg' describes resource requirements for a
873 * device's IO/MEM regions, 'assigned-addresses' describes the
874 * actual resource assignments.
876 * the properties are arrays of ('phys-addr', 'size') pairs describing
877 * the addressable regions of the PCI device, where 'phys-addr' is a
878 * RESOURCE_CELLS_ADDRESS-tuple of 32-bit integers corresponding to
879 * (phys.hi, phys.mid, phys.lo), and 'size' is a
880 * RESOURCE_CELLS_SIZE-tuple corresponding to (size.hi, size.lo).
882 * phys.hi = 0xYYXXXXZZ, where:
883 * 0xYY = npt000ss
884 * ||| |
885 * ||| +-- space code
886 * ||| |
887 * ||| + 00 if configuration space
888 * ||| + 01 if IO region,
889 * ||| + 10 if 32-bit MEM region
890 * ||| + 11 if 64-bit MEM region
891 * |||
892 * ||+------ for non-relocatable IO: 1 if aliased
893 * || for relocatable IO: 1 if below 64KB
894 * || for MEM: 1 if below 1MB
895 * |+------- 1 if region is prefetchable
896 * +-------- 1 if region is non-relocatable
897 * 0xXXXX = bbbbbbbb dddddfff, encoding bus, slot, and function
898 * bits respectively
899 * 0xZZ = rrrrrrrr, the register number of the BAR corresponding
900 * to the region
902 * phys.mid and phys.lo correspond respectively to the hi/lo portions
903 * of the actual address of the region.
905 * how the phys-addr/size values are used differ slightly between
906 * 'reg' and 'assigned-addresses' properties. namely, 'reg' has
907 * an additional description for the config space region of the
908 * device, and in the case of QEMU has n=0 and phys.mid=phys.lo=0
909 * to describe the region as relocatable, with an address-mapping
910 * that corresponds directly to the PHB's address space for the
911 * resource. 'assigned-addresses' always has n=1 set with an absolute
912 * address assigned for the resource. in general, 'assigned-addresses'
913 * won't be populated, since addresses for PCI devices are generally
914 * unmapped initially and left to the guest to assign.
916 * note also that addresses defined in these properties are, at least
917 * for PAPR guests, relative to the PHBs IO/MEM windows, and
918 * correspond directly to the addresses in the BARs.
920 * in accordance with PCI Bus Binding to Open Firmware,
921 * IEEE Std 1275-1994, section 4.1.1, as implemented by PAPR+ v2.7,
922 * Appendix C.
924 static void populate_resource_props(PCIDevice *d, ResourceProps *rp)
926 int bus_num = pci_bus_num(PCI_BUS(qdev_get_parent_bus(DEVICE(d))));
927 uint32_t dev_id = (b_bbbbbbbb(bus_num) |
928 b_ddddd(PCI_SLOT(d->devfn)) |
929 b_fff(PCI_FUNC(d->devfn)));
930 ResourceFields *reg, *assigned;
931 int i, reg_idx = 0, assigned_idx = 0;
933 /* config space region */
934 reg = &rp->reg[reg_idx++];
935 reg->phys_hi = cpu_to_be32(dev_id);
936 reg->phys_mid = 0;
937 reg->phys_lo = 0;
938 reg->size_hi = 0;
939 reg->size_lo = 0;
941 for (i = 0; i < PCI_NUM_REGIONS; i++) {
942 if (!d->io_regions[i].size) {
943 continue;
946 reg = &rp->reg[reg_idx++];
948 reg->phys_hi = cpu_to_be32(dev_id | b_rrrrrrrr(pci_bar(d, i)));
949 if (d->io_regions[i].type & PCI_BASE_ADDRESS_SPACE_IO) {
950 reg->phys_hi |= cpu_to_be32(b_ss(1));
951 } else if (d->io_regions[i].type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
952 reg->phys_hi |= cpu_to_be32(b_ss(3));
953 } else {
954 reg->phys_hi |= cpu_to_be32(b_ss(2));
956 reg->phys_mid = 0;
957 reg->phys_lo = 0;
958 reg->size_hi = cpu_to_be32(d->io_regions[i].size >> 32);
959 reg->size_lo = cpu_to_be32(d->io_regions[i].size);
961 if (d->io_regions[i].addr == PCI_BAR_UNMAPPED) {
962 continue;
965 assigned = &rp->assigned[assigned_idx++];
966 assigned->phys_hi = cpu_to_be32(reg->phys_hi | b_n(1));
967 assigned->phys_mid = cpu_to_be32(d->io_regions[i].addr >> 32);
968 assigned->phys_lo = cpu_to_be32(d->io_regions[i].addr);
969 assigned->size_hi = reg->size_hi;
970 assigned->size_lo = reg->size_lo;
973 rp->reg_len = reg_idx * sizeof(ResourceFields);
974 rp->assigned_len = assigned_idx * sizeof(ResourceFields);
977 typedef struct PCIClass PCIClass;
978 typedef struct PCISubClass PCISubClass;
979 typedef struct PCIIFace PCIIFace;
981 struct PCIIFace {
982 int iface;
983 const char *name;
986 struct PCISubClass {
987 int subclass;
988 const char *name;
989 const PCIIFace *iface;
992 struct PCIClass {
993 const char *name;
994 const PCISubClass *subc;
997 static const PCISubClass undef_subclass[] = {
998 { PCI_CLASS_NOT_DEFINED_VGA, "display", NULL },
999 { 0xFF, NULL, NULL },
1002 static const PCISubClass mass_subclass[] = {
1003 { PCI_CLASS_STORAGE_SCSI, "scsi", NULL },
1004 { PCI_CLASS_STORAGE_IDE, "ide", NULL },
1005 { PCI_CLASS_STORAGE_FLOPPY, "fdc", NULL },
1006 { PCI_CLASS_STORAGE_IPI, "ipi", NULL },
1007 { PCI_CLASS_STORAGE_RAID, "raid", NULL },
1008 { PCI_CLASS_STORAGE_ATA, "ata", NULL },
1009 { PCI_CLASS_STORAGE_SATA, "sata", NULL },
1010 { PCI_CLASS_STORAGE_SAS, "sas", NULL },
1011 { 0xFF, NULL, NULL },
1014 static const PCISubClass net_subclass[] = {
1015 { PCI_CLASS_NETWORK_ETHERNET, "ethernet", NULL },
1016 { PCI_CLASS_NETWORK_TOKEN_RING, "token-ring", NULL },
1017 { PCI_CLASS_NETWORK_FDDI, "fddi", NULL },
1018 { PCI_CLASS_NETWORK_ATM, "atm", NULL },
1019 { PCI_CLASS_NETWORK_ISDN, "isdn", NULL },
1020 { PCI_CLASS_NETWORK_WORLDFIP, "worldfip", NULL },
1021 { PCI_CLASS_NETWORK_PICMG214, "picmg", NULL },
1022 { 0xFF, NULL, NULL },
1025 static const PCISubClass displ_subclass[] = {
1026 { PCI_CLASS_DISPLAY_VGA, "vga", NULL },
1027 { PCI_CLASS_DISPLAY_XGA, "xga", NULL },
1028 { PCI_CLASS_DISPLAY_3D, "3d-controller", NULL },
1029 { 0xFF, NULL, NULL },
1032 static const PCISubClass media_subclass[] = {
1033 { PCI_CLASS_MULTIMEDIA_VIDEO, "video", NULL },
1034 { PCI_CLASS_MULTIMEDIA_AUDIO, "sound", NULL },
1035 { PCI_CLASS_MULTIMEDIA_PHONE, "telephony", NULL },
1036 { 0xFF, NULL, NULL },
1039 static const PCISubClass mem_subclass[] = {
1040 { PCI_CLASS_MEMORY_RAM, "memory", NULL },
1041 { PCI_CLASS_MEMORY_FLASH, "flash", NULL },
1042 { 0xFF, NULL, NULL },
1045 static const PCISubClass bridg_subclass[] = {
1046 { PCI_CLASS_BRIDGE_HOST, "host", NULL },
1047 { PCI_CLASS_BRIDGE_ISA, "isa", NULL },
1048 { PCI_CLASS_BRIDGE_EISA, "eisa", NULL },
1049 { PCI_CLASS_BRIDGE_MC, "mca", NULL },
1050 { PCI_CLASS_BRIDGE_PCI, "pci", NULL },
1051 { PCI_CLASS_BRIDGE_PCMCIA, "pcmcia", NULL },
1052 { PCI_CLASS_BRIDGE_NUBUS, "nubus", NULL },
1053 { PCI_CLASS_BRIDGE_CARDBUS, "cardbus", NULL },
1054 { PCI_CLASS_BRIDGE_RACEWAY, "raceway", NULL },
1055 { PCI_CLASS_BRIDGE_PCI_SEMITP, "semi-transparent-pci", NULL },
1056 { PCI_CLASS_BRIDGE_IB_PCI, "infiniband", NULL },
1057 { 0xFF, NULL, NULL },
1060 static const PCISubClass comm_subclass[] = {
1061 { PCI_CLASS_COMMUNICATION_SERIAL, "serial", NULL },
1062 { PCI_CLASS_COMMUNICATION_PARALLEL, "parallel", NULL },
1063 { PCI_CLASS_COMMUNICATION_MULTISERIAL, "multiport-serial", NULL },
1064 { PCI_CLASS_COMMUNICATION_MODEM, "modem", NULL },
1065 { PCI_CLASS_COMMUNICATION_GPIB, "gpib", NULL },
1066 { PCI_CLASS_COMMUNICATION_SC, "smart-card", NULL },
1067 { 0xFF, NULL, NULL, },
1070 static const PCIIFace pic_iface[] = {
1071 { PCI_CLASS_SYSTEM_PIC_IOAPIC, "io-apic" },
1072 { PCI_CLASS_SYSTEM_PIC_IOXAPIC, "io-xapic" },
1073 { 0xFF, NULL },
1076 static const PCISubClass sys_subclass[] = {
1077 { PCI_CLASS_SYSTEM_PIC, "interrupt-controller", pic_iface },
1078 { PCI_CLASS_SYSTEM_DMA, "dma-controller", NULL },
1079 { PCI_CLASS_SYSTEM_TIMER, "timer", NULL },
1080 { PCI_CLASS_SYSTEM_RTC, "rtc", NULL },
1081 { PCI_CLASS_SYSTEM_PCI_HOTPLUG, "hot-plug-controller", NULL },
1082 { PCI_CLASS_SYSTEM_SDHCI, "sd-host-controller", NULL },
1083 { 0xFF, NULL, NULL },
1086 static const PCISubClass inp_subclass[] = {
1087 { PCI_CLASS_INPUT_KEYBOARD, "keyboard", NULL },
1088 { PCI_CLASS_INPUT_PEN, "pen", NULL },
1089 { PCI_CLASS_INPUT_MOUSE, "mouse", NULL },
1090 { PCI_CLASS_INPUT_SCANNER, "scanner", NULL },
1091 { PCI_CLASS_INPUT_GAMEPORT, "gameport", NULL },
1092 { 0xFF, NULL, NULL },
1095 static const PCISubClass dock_subclass[] = {
1096 { PCI_CLASS_DOCKING_GENERIC, "dock", NULL },
1097 { 0xFF, NULL, NULL },
1100 static const PCISubClass cpu_subclass[] = {
1101 { PCI_CLASS_PROCESSOR_PENTIUM, "pentium", NULL },
1102 { PCI_CLASS_PROCESSOR_POWERPC, "powerpc", NULL },
1103 { PCI_CLASS_PROCESSOR_MIPS, "mips", NULL },
1104 { PCI_CLASS_PROCESSOR_CO, "co-processor", NULL },
1105 { 0xFF, NULL, NULL },
1108 static const PCIIFace usb_iface[] = {
1109 { PCI_CLASS_SERIAL_USB_UHCI, "usb-uhci" },
1110 { PCI_CLASS_SERIAL_USB_OHCI, "usb-ohci", },
1111 { PCI_CLASS_SERIAL_USB_EHCI, "usb-ehci" },
1112 { PCI_CLASS_SERIAL_USB_XHCI, "usb-xhci" },
1113 { PCI_CLASS_SERIAL_USB_UNKNOWN, "usb-unknown" },
1114 { PCI_CLASS_SERIAL_USB_DEVICE, "usb-device" },
1115 { 0xFF, NULL },
1118 static const PCISubClass ser_subclass[] = {
1119 { PCI_CLASS_SERIAL_FIREWIRE, "firewire", NULL },
1120 { PCI_CLASS_SERIAL_ACCESS, "access-bus", NULL },
1121 { PCI_CLASS_SERIAL_SSA, "ssa", NULL },
1122 { PCI_CLASS_SERIAL_USB, "usb", usb_iface },
1123 { PCI_CLASS_SERIAL_FIBER, "fibre-channel", NULL },
1124 { PCI_CLASS_SERIAL_SMBUS, "smb", NULL },
1125 { PCI_CLASS_SERIAL_IB, "infiniband", NULL },
1126 { PCI_CLASS_SERIAL_IPMI, "ipmi", NULL },
1127 { PCI_CLASS_SERIAL_SERCOS, "sercos", NULL },
1128 { PCI_CLASS_SERIAL_CANBUS, "canbus", NULL },
1129 { 0xFF, NULL, NULL },
1132 static const PCISubClass wrl_subclass[] = {
1133 { PCI_CLASS_WIRELESS_IRDA, "irda", NULL },
1134 { PCI_CLASS_WIRELESS_CIR, "consumer-ir", NULL },
1135 { PCI_CLASS_WIRELESS_RF_CONTROLLER, "rf-controller", NULL },
1136 { PCI_CLASS_WIRELESS_BLUETOOTH, "bluetooth", NULL },
1137 { PCI_CLASS_WIRELESS_BROADBAND, "broadband", NULL },
1138 { 0xFF, NULL, NULL },
1141 static const PCISubClass sat_subclass[] = {
1142 { PCI_CLASS_SATELLITE_TV, "satellite-tv", NULL },
1143 { PCI_CLASS_SATELLITE_AUDIO, "satellite-audio", NULL },
1144 { PCI_CLASS_SATELLITE_VOICE, "satellite-voice", NULL },
1145 { PCI_CLASS_SATELLITE_DATA, "satellite-data", NULL },
1146 { 0xFF, NULL, NULL },
1149 static const PCISubClass crypt_subclass[] = {
1150 { PCI_CLASS_CRYPT_NETWORK, "network-encryption", NULL },
1151 { PCI_CLASS_CRYPT_ENTERTAINMENT,
1152 "entertainment-encryption", NULL },
1153 { 0xFF, NULL, NULL },
1156 static const PCISubClass spc_subclass[] = {
1157 { PCI_CLASS_SP_DPIO, "dpio", NULL },
1158 { PCI_CLASS_SP_PERF, "counter", NULL },
1159 { PCI_CLASS_SP_SYNCH, "measurement", NULL },
1160 { PCI_CLASS_SP_MANAGEMENT, "management-card", NULL },
1161 { 0xFF, NULL, NULL },
1164 static const PCIClass pci_classes[] = {
1165 { "legacy-device", undef_subclass },
1166 { "mass-storage", mass_subclass },
1167 { "network", net_subclass },
1168 { "display", displ_subclass, },
1169 { "multimedia-device", media_subclass },
1170 { "memory-controller", mem_subclass },
1171 { "unknown-bridge", bridg_subclass },
1172 { "communication-controller", comm_subclass},
1173 { "system-peripheral", sys_subclass },
1174 { "input-controller", inp_subclass },
1175 { "docking-station", dock_subclass },
1176 { "cpu", cpu_subclass },
1177 { "serial-bus", ser_subclass },
1178 { "wireless-controller", wrl_subclass },
1179 { "intelligent-io", NULL },
1180 { "satellite-device", sat_subclass },
1181 { "encryption", crypt_subclass },
1182 { "data-processing-controller", spc_subclass },
1185 static const char *pci_find_device_name(uint8_t class, uint8_t subclass,
1186 uint8_t iface)
1188 const PCIClass *pclass;
1189 const PCISubClass *psubclass;
1190 const PCIIFace *piface;
1191 const char *name;
1193 if (class >= ARRAY_SIZE(pci_classes)) {
1194 return "pci";
1197 pclass = pci_classes + class;
1198 name = pclass->name;
1200 if (pclass->subc == NULL) {
1201 return name;
1204 psubclass = pclass->subc;
1205 while ((psubclass->subclass & 0xff) != 0xff) {
1206 if ((psubclass->subclass & 0xff) == subclass) {
1207 name = psubclass->name;
1208 break;
1210 psubclass++;
1213 piface = psubclass->iface;
1214 if (piface == NULL) {
1215 return name;
1217 while ((piface->iface & 0xff) != 0xff) {
1218 if ((piface->iface & 0xff) == iface) {
1219 name = piface->name;
1220 break;
1222 piface++;
1225 return name;
1228 static gchar *pci_get_node_name(PCIDevice *dev)
1230 int slot = PCI_SLOT(dev->devfn);
1231 int func = PCI_FUNC(dev->devfn);
1232 uint32_t ccode = pci_default_read_config(dev, PCI_CLASS_PROG, 3);
1233 const char *name;
1235 name = pci_find_device_name((ccode >> 16) & 0xff, (ccode >> 8) & 0xff,
1236 ccode & 0xff);
1238 if (func != 0) {
1239 return g_strdup_printf("%s@%x,%x", name, slot, func);
1240 } else {
1241 return g_strdup_printf("%s@%x", name, slot);
1245 static uint32_t spapr_phb_get_pci_drc_index(sPAPRPHBState *phb,
1246 PCIDevice *pdev);
1248 static void spapr_populate_pci_child_dt(PCIDevice *dev, void *fdt, int offset,
1249 sPAPRPHBState *sphb)
1251 ResourceProps rp;
1252 bool is_bridge = false;
1253 int pci_status;
1254 char *buf = NULL;
1255 uint32_t drc_index = spapr_phb_get_pci_drc_index(sphb, dev);
1256 uint32_t ccode = pci_default_read_config(dev, PCI_CLASS_PROG, 3);
1257 uint32_t max_msi, max_msix;
1259 if (pci_default_read_config(dev, PCI_HEADER_TYPE, 1) ==
1260 PCI_HEADER_TYPE_BRIDGE) {
1261 is_bridge = true;
1264 /* in accordance with PAPR+ v2.7 13.6.3, Table 181 */
1265 _FDT(fdt_setprop_cell(fdt, offset, "vendor-id",
1266 pci_default_read_config(dev, PCI_VENDOR_ID, 2)));
1267 _FDT(fdt_setprop_cell(fdt, offset, "device-id",
1268 pci_default_read_config(dev, PCI_DEVICE_ID, 2)));
1269 _FDT(fdt_setprop_cell(fdt, offset, "revision-id",
1270 pci_default_read_config(dev, PCI_REVISION_ID, 1)));
1271 _FDT(fdt_setprop_cell(fdt, offset, "class-code", ccode));
1272 if (pci_default_read_config(dev, PCI_INTERRUPT_PIN, 1)) {
1273 _FDT(fdt_setprop_cell(fdt, offset, "interrupts",
1274 pci_default_read_config(dev, PCI_INTERRUPT_PIN, 1)));
1277 if (!is_bridge) {
1278 _FDT(fdt_setprop_cell(fdt, offset, "min-grant",
1279 pci_default_read_config(dev, PCI_MIN_GNT, 1)));
1280 _FDT(fdt_setprop_cell(fdt, offset, "max-latency",
1281 pci_default_read_config(dev, PCI_MAX_LAT, 1)));
1284 if (pci_default_read_config(dev, PCI_SUBSYSTEM_ID, 2)) {
1285 _FDT(fdt_setprop_cell(fdt, offset, "subsystem-id",
1286 pci_default_read_config(dev, PCI_SUBSYSTEM_ID, 2)));
1289 if (pci_default_read_config(dev, PCI_SUBSYSTEM_VENDOR_ID, 2)) {
1290 _FDT(fdt_setprop_cell(fdt, offset, "subsystem-vendor-id",
1291 pci_default_read_config(dev, PCI_SUBSYSTEM_VENDOR_ID, 2)));
1294 _FDT(fdt_setprop_cell(fdt, offset, "cache-line-size",
1295 pci_default_read_config(dev, PCI_CACHE_LINE_SIZE, 1)));
1297 /* the following fdt cells are masked off the pci status register */
1298 pci_status = pci_default_read_config(dev, PCI_STATUS, 2);
1299 _FDT(fdt_setprop_cell(fdt, offset, "devsel-speed",
1300 PCI_STATUS_DEVSEL_MASK & pci_status));
1302 if (pci_status & PCI_STATUS_FAST_BACK) {
1303 _FDT(fdt_setprop(fdt, offset, "fast-back-to-back", NULL, 0));
1305 if (pci_status & PCI_STATUS_66MHZ) {
1306 _FDT(fdt_setprop(fdt, offset, "66mhz-capable", NULL, 0));
1308 if (pci_status & PCI_STATUS_UDF) {
1309 _FDT(fdt_setprop(fdt, offset, "udf-supported", NULL, 0));
1312 _FDT(fdt_setprop_string(fdt, offset, "name",
1313 pci_find_device_name((ccode >> 16) & 0xff,
1314 (ccode >> 8) & 0xff,
1315 ccode & 0xff)));
1317 buf = spapr_phb_get_loc_code(sphb, dev);
1318 _FDT(fdt_setprop_string(fdt, offset, "ibm,loc-code", buf));
1319 g_free(buf);
1321 if (drc_index) {
1322 _FDT(fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index));
1325 _FDT(fdt_setprop_cell(fdt, offset, "#address-cells",
1326 RESOURCE_CELLS_ADDRESS));
1327 _FDT(fdt_setprop_cell(fdt, offset, "#size-cells",
1328 RESOURCE_CELLS_SIZE));
1330 if (msi_present(dev)) {
1331 max_msi = msi_nr_vectors_allocated(dev);
1332 if (max_msi) {
1333 _FDT(fdt_setprop_cell(fdt, offset, "ibm,req#msi", max_msi));
1336 if (msix_present(dev)) {
1337 max_msix = dev->msix_entries_nr;
1338 if (max_msix) {
1339 _FDT(fdt_setprop_cell(fdt, offset, "ibm,req#msi-x", max_msix));
1343 populate_resource_props(dev, &rp);
1344 _FDT(fdt_setprop(fdt, offset, "reg", (uint8_t *)rp.reg, rp.reg_len));
1345 _FDT(fdt_setprop(fdt, offset, "assigned-addresses",
1346 (uint8_t *)rp.assigned, rp.assigned_len));
1348 if (sphb->pcie_ecs && pci_is_express(dev)) {
1349 _FDT(fdt_setprop_cell(fdt, offset, "ibm,pci-config-space-type", 0x1));
1353 /* create OF node for pci device and required OF DT properties */
1354 static int spapr_create_pci_child_dt(sPAPRPHBState *phb, PCIDevice *dev,
1355 void *fdt, int node_offset)
1357 int offset;
1358 gchar *nodename;
1360 nodename = pci_get_node_name(dev);
1361 _FDT(offset = fdt_add_subnode(fdt, node_offset, nodename));
1362 g_free(nodename);
1364 spapr_populate_pci_child_dt(dev, fdt, offset, phb);
1366 return offset;
1369 /* Callback to be called during DRC release. */
1370 void spapr_phb_remove_pci_device_cb(DeviceState *dev)
1372 /* some version guests do not wait for completion of a device
1373 * cleanup (generally done asynchronously by the kernel) before
1374 * signaling to QEMU that the device is safe, but instead sleep
1375 * for some 'safe' period of time. unfortunately on a busy host
1376 * this sleep isn't guaranteed to be long enough, resulting in
1377 * bad things like IRQ lines being left asserted during final
1378 * device removal. to deal with this we call reset just prior
1379 * to finalizing the device, which will put the device back into
1380 * an 'idle' state, as the device cleanup code expects.
1382 pci_device_reset(PCI_DEVICE(dev));
1383 object_unparent(OBJECT(dev));
1386 static sPAPRDRConnector *spapr_phb_get_pci_func_drc(sPAPRPHBState *phb,
1387 uint32_t busnr,
1388 int32_t devfn)
1390 return spapr_drc_by_id(TYPE_SPAPR_DRC_PCI,
1391 (phb->index << 16) | (busnr << 8) | devfn);
1394 static sPAPRDRConnector *spapr_phb_get_pci_drc(sPAPRPHBState *phb,
1395 PCIDevice *pdev)
1397 uint32_t busnr = pci_bus_num(PCI_BUS(qdev_get_parent_bus(DEVICE(pdev))));
1398 return spapr_phb_get_pci_func_drc(phb, busnr, pdev->devfn);
1401 static uint32_t spapr_phb_get_pci_drc_index(sPAPRPHBState *phb,
1402 PCIDevice *pdev)
1404 sPAPRDRConnector *drc = spapr_phb_get_pci_drc(phb, pdev);
1406 if (!drc) {
1407 return 0;
1410 return spapr_drc_index(drc);
1413 static void spapr_pci_plug(HotplugHandler *plug_handler,
1414 DeviceState *plugged_dev, Error **errp)
1416 sPAPRPHBState *phb = SPAPR_PCI_HOST_BRIDGE(DEVICE(plug_handler));
1417 PCIDevice *pdev = PCI_DEVICE(plugged_dev);
1418 sPAPRDRConnector *drc = spapr_phb_get_pci_drc(phb, pdev);
1419 Error *local_err = NULL;
1420 PCIBus *bus = PCI_BUS(qdev_get_parent_bus(DEVICE(pdev)));
1421 uint32_t slotnr = PCI_SLOT(pdev->devfn);
1422 void *fdt = NULL;
1423 int fdt_start_offset, fdt_size;
1425 /* if DR is disabled we don't need to do anything in the case of
1426 * hotplug or coldplug callbacks
1428 if (!phb->dr_enabled) {
1429 /* if this is a hotplug operation initiated by the user
1430 * we need to let them know it's not enabled
1432 if (plugged_dev->hotplugged) {
1433 error_setg(&local_err, QERR_BUS_NO_HOTPLUG,
1434 object_get_typename(OBJECT(phb)));
1436 goto out;
1439 g_assert(drc);
1441 /* Following the QEMU convention used for PCIe multifunction
1442 * hotplug, we do not allow functions to be hotplugged to a
1443 * slot that already has function 0 present
1445 if (plugged_dev->hotplugged && bus->devices[PCI_DEVFN(slotnr, 0)] &&
1446 PCI_FUNC(pdev->devfn) != 0) {
1447 error_setg(&local_err, "PCI: slot %d function 0 already ocuppied by %s,"
1448 " additional functions can no longer be exposed to guest.",
1449 slotnr, bus->devices[PCI_DEVFN(slotnr, 0)]->name);
1450 goto out;
1453 fdt = create_device_tree(&fdt_size);
1454 fdt_start_offset = spapr_create_pci_child_dt(phb, pdev, fdt, 0);
1456 spapr_drc_attach(drc, DEVICE(pdev), fdt, fdt_start_offset, &local_err);
1457 if (local_err) {
1458 goto out;
1461 /* If this is function 0, signal hotplug for all the device functions.
1462 * Otherwise defer sending the hotplug event.
1464 if (!spapr_drc_hotplugged(plugged_dev)) {
1465 spapr_drc_reset(drc);
1466 } else if (PCI_FUNC(pdev->devfn) == 0) {
1467 int i;
1469 for (i = 0; i < 8; i++) {
1470 sPAPRDRConnector *func_drc;
1471 sPAPRDRConnectorClass *func_drck;
1472 sPAPRDREntitySense state;
1474 func_drc = spapr_phb_get_pci_func_drc(phb, pci_bus_num(bus),
1475 PCI_DEVFN(slotnr, i));
1476 func_drck = SPAPR_DR_CONNECTOR_GET_CLASS(func_drc);
1477 state = func_drck->dr_entity_sense(func_drc);
1479 if (state == SPAPR_DR_ENTITY_SENSE_PRESENT) {
1480 spapr_hotplug_req_add_by_index(func_drc);
1485 out:
1486 if (local_err) {
1487 error_propagate(errp, local_err);
1488 g_free(fdt);
1492 static void spapr_pci_unplug_request(HotplugHandler *plug_handler,
1493 DeviceState *plugged_dev, Error **errp)
1495 sPAPRPHBState *phb = SPAPR_PCI_HOST_BRIDGE(DEVICE(plug_handler));
1496 PCIDevice *pdev = PCI_DEVICE(plugged_dev);
1497 sPAPRDRConnector *drc = spapr_phb_get_pci_drc(phb, pdev);
1499 if (!phb->dr_enabled) {
1500 error_setg(errp, QERR_BUS_NO_HOTPLUG,
1501 object_get_typename(OBJECT(phb)));
1502 return;
1505 g_assert(drc);
1506 g_assert(drc->dev == plugged_dev);
1508 if (!spapr_drc_unplug_requested(drc)) {
1509 PCIBus *bus = PCI_BUS(qdev_get_parent_bus(DEVICE(pdev)));
1510 uint32_t slotnr = PCI_SLOT(pdev->devfn);
1511 sPAPRDRConnector *func_drc;
1512 sPAPRDRConnectorClass *func_drck;
1513 sPAPRDREntitySense state;
1514 int i;
1516 /* ensure any other present functions are pending unplug */
1517 if (PCI_FUNC(pdev->devfn) == 0) {
1518 for (i = 1; i < 8; i++) {
1519 func_drc = spapr_phb_get_pci_func_drc(phb, pci_bus_num(bus),
1520 PCI_DEVFN(slotnr, i));
1521 func_drck = SPAPR_DR_CONNECTOR_GET_CLASS(func_drc);
1522 state = func_drck->dr_entity_sense(func_drc);
1523 if (state == SPAPR_DR_ENTITY_SENSE_PRESENT
1524 && !spapr_drc_unplug_requested(func_drc)) {
1525 error_setg(errp,
1526 "PCI: slot %d, function %d still present. "
1527 "Must unplug all non-0 functions first.",
1528 slotnr, i);
1529 return;
1534 spapr_drc_detach(drc);
1536 /* if this isn't func 0, defer unplug event. otherwise signal removal
1537 * for all present functions
1539 if (PCI_FUNC(pdev->devfn) == 0) {
1540 for (i = 7; i >= 0; i--) {
1541 func_drc = spapr_phb_get_pci_func_drc(phb, pci_bus_num(bus),
1542 PCI_DEVFN(slotnr, i));
1543 func_drck = SPAPR_DR_CONNECTOR_GET_CLASS(func_drc);
1544 state = func_drck->dr_entity_sense(func_drc);
1545 if (state == SPAPR_DR_ENTITY_SENSE_PRESENT) {
1546 spapr_hotplug_req_remove_by_index(func_drc);
1553 static void spapr_phb_realize(DeviceState *dev, Error **errp)
1555 /* We don't use SPAPR_MACHINE() in order to exit gracefully if the user
1556 * tries to add a sPAPR PHB to a non-pseries machine.
1558 sPAPRMachineState *spapr =
1559 (sPAPRMachineState *) object_dynamic_cast(qdev_get_machine(),
1560 TYPE_SPAPR_MACHINE);
1561 SysBusDevice *s = SYS_BUS_DEVICE(dev);
1562 sPAPRPHBState *sphb = SPAPR_PCI_HOST_BRIDGE(s);
1563 PCIHostState *phb = PCI_HOST_BRIDGE(s);
1564 char *namebuf;
1565 int i;
1566 PCIBus *bus;
1567 uint64_t msi_window_size = 4096;
1568 sPAPRTCETable *tcet;
1569 const unsigned windows_supported =
1570 sphb->ddw_enabled ? SPAPR_PCI_DMA_MAX_WINDOWS : 1;
1572 if (!spapr) {
1573 error_setg(errp, TYPE_SPAPR_PCI_HOST_BRIDGE " needs a pseries machine");
1574 return;
1577 if (sphb->index != (uint32_t)-1) {
1578 sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
1579 Error *local_err = NULL;
1581 smc->phb_placement(spapr, sphb->index,
1582 &sphb->buid, &sphb->io_win_addr,
1583 &sphb->mem_win_addr, &sphb->mem64_win_addr,
1584 windows_supported, sphb->dma_liobn, &local_err);
1585 if (local_err) {
1586 error_propagate(errp, local_err);
1587 return;
1589 } else {
1590 error_setg(errp, "\"index\" for PAPR PHB is mandatory");
1591 return;
1594 if (sphb->mem64_win_size != 0) {
1595 if (sphb->mem_win_size > SPAPR_PCI_MEM32_WIN_SIZE) {
1596 error_setg(errp, "32-bit memory window of size 0x%"HWADDR_PRIx
1597 " (max 2 GiB)", sphb->mem_win_size);
1598 return;
1601 /* 64-bit window defaults to identity mapping */
1602 sphb->mem64_win_pciaddr = sphb->mem64_win_addr;
1603 } else if (sphb->mem_win_size > SPAPR_PCI_MEM32_WIN_SIZE) {
1605 * For compatibility with old configuration, if no 64-bit MMIO
1606 * window is specified, but the ordinary (32-bit) memory
1607 * window is specified as > 2GiB, we treat it as a 2GiB 32-bit
1608 * window, with a 64-bit MMIO window following on immediately
1609 * afterwards
1611 sphb->mem64_win_size = sphb->mem_win_size - SPAPR_PCI_MEM32_WIN_SIZE;
1612 sphb->mem64_win_addr = sphb->mem_win_addr + SPAPR_PCI_MEM32_WIN_SIZE;
1613 sphb->mem64_win_pciaddr =
1614 SPAPR_PCI_MEM_WIN_BUS_OFFSET + SPAPR_PCI_MEM32_WIN_SIZE;
1615 sphb->mem_win_size = SPAPR_PCI_MEM32_WIN_SIZE;
1618 if (spapr_pci_find_phb(spapr, sphb->buid)) {
1619 error_setg(errp, "PCI host bridges must have unique BUIDs");
1620 return;
1623 if (sphb->numa_node != -1 &&
1624 (sphb->numa_node >= MAX_NODES || !numa_info[sphb->numa_node].present)) {
1625 error_setg(errp, "Invalid NUMA node ID for PCI host bridge");
1626 return;
1629 sphb->dtbusname = g_strdup_printf("pci@%" PRIx64, sphb->buid);
1631 /* Initialize memory regions */
1632 namebuf = g_strdup_printf("%s.mmio", sphb->dtbusname);
1633 memory_region_init(&sphb->memspace, OBJECT(sphb), namebuf, UINT64_MAX);
1634 g_free(namebuf);
1636 namebuf = g_strdup_printf("%s.mmio32-alias", sphb->dtbusname);
1637 memory_region_init_alias(&sphb->mem32window, OBJECT(sphb),
1638 namebuf, &sphb->memspace,
1639 SPAPR_PCI_MEM_WIN_BUS_OFFSET, sphb->mem_win_size);
1640 g_free(namebuf);
1641 memory_region_add_subregion(get_system_memory(), sphb->mem_win_addr,
1642 &sphb->mem32window);
1644 if (sphb->mem64_win_size != 0) {
1645 namebuf = g_strdup_printf("%s.mmio64-alias", sphb->dtbusname);
1646 memory_region_init_alias(&sphb->mem64window, OBJECT(sphb),
1647 namebuf, &sphb->memspace,
1648 sphb->mem64_win_pciaddr, sphb->mem64_win_size);
1649 g_free(namebuf);
1651 memory_region_add_subregion(get_system_memory(),
1652 sphb->mem64_win_addr,
1653 &sphb->mem64window);
1656 /* Initialize IO regions */
1657 namebuf = g_strdup_printf("%s.io", sphb->dtbusname);
1658 memory_region_init(&sphb->iospace, OBJECT(sphb),
1659 namebuf, SPAPR_PCI_IO_WIN_SIZE);
1660 g_free(namebuf);
1662 namebuf = g_strdup_printf("%s.io-alias", sphb->dtbusname);
1663 memory_region_init_alias(&sphb->iowindow, OBJECT(sphb), namebuf,
1664 &sphb->iospace, 0, SPAPR_PCI_IO_WIN_SIZE);
1665 g_free(namebuf);
1666 memory_region_add_subregion(get_system_memory(), sphb->io_win_addr,
1667 &sphb->iowindow);
1669 bus = pci_register_root_bus(dev, NULL,
1670 pci_spapr_set_irq, pci_spapr_map_irq, sphb,
1671 &sphb->memspace, &sphb->iospace,
1672 PCI_DEVFN(0, 0), PCI_NUM_PINS, TYPE_PCI_BUS);
1673 phb->bus = bus;
1674 qbus_set_hotplug_handler(BUS(phb->bus), DEVICE(sphb), NULL);
1677 * Initialize PHB address space.
1678 * By default there will be at least one subregion for default
1679 * 32bit DMA window.
1680 * Later the guest might want to create another DMA window
1681 * which will become another memory subregion.
1683 namebuf = g_strdup_printf("%s.iommu-root", sphb->dtbusname);
1684 memory_region_init(&sphb->iommu_root, OBJECT(sphb),
1685 namebuf, UINT64_MAX);
1686 g_free(namebuf);
1687 address_space_init(&sphb->iommu_as, &sphb->iommu_root,
1688 sphb->dtbusname);
1691 * As MSI/MSIX interrupts trigger by writing at MSI/MSIX vectors,
1692 * we need to allocate some memory to catch those writes coming
1693 * from msi_notify()/msix_notify().
1694 * As MSIMessage:addr is going to be the same and MSIMessage:data
1695 * is going to be a VIRQ number, 4 bytes of the MSI MR will only
1696 * be used.
1698 * For KVM we want to ensure that this memory is a full page so that
1699 * our memory slot is of page size granularity.
1701 #ifdef CONFIG_KVM
1702 if (kvm_enabled()) {
1703 msi_window_size = getpagesize();
1705 #endif
1707 memory_region_init_io(&sphb->msiwindow, OBJECT(sphb), &spapr_msi_ops, spapr,
1708 "msi", msi_window_size);
1709 memory_region_add_subregion(&sphb->iommu_root, SPAPR_PCI_MSI_WINDOW,
1710 &sphb->msiwindow);
1712 pci_setup_iommu(bus, spapr_pci_dma_iommu, sphb);
1714 pci_bus_set_route_irq_fn(bus, spapr_route_intx_pin_to_irq);
1716 QLIST_INSERT_HEAD(&spapr->phbs, sphb, list);
1718 /* Initialize the LSI table */
1719 for (i = 0; i < PCI_NUM_PINS; i++) {
1720 uint32_t irq = SPAPR_IRQ_PCI_LSI + sphb->index * PCI_NUM_PINS + i;
1721 Error *local_err = NULL;
1723 if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
1724 irq = spapr_irq_findone(spapr, &local_err);
1725 if (local_err) {
1726 error_propagate(errp, local_err);
1727 error_prepend(errp, "can't allocate LSIs: ");
1728 return;
1732 spapr_irq_claim(spapr, irq, true, &local_err);
1733 if (local_err) {
1734 error_propagate(errp, local_err);
1735 error_prepend(errp, "can't allocate LSIs: ");
1736 return;
1739 sphb->lsi_table[i].irq = irq;
1742 /* allocate connectors for child PCI devices */
1743 if (sphb->dr_enabled) {
1744 for (i = 0; i < PCI_SLOT_MAX * 8; i++) {
1745 spapr_dr_connector_new(OBJECT(phb), TYPE_SPAPR_DRC_PCI,
1746 (sphb->index << 16) | i);
1750 /* DMA setup */
1751 for (i = 0; i < windows_supported; ++i) {
1752 tcet = spapr_tce_new_table(DEVICE(sphb), sphb->dma_liobn[i]);
1753 if (!tcet) {
1754 error_setg(errp, "Creating window#%d failed for %s",
1755 i, sphb->dtbusname);
1756 return;
1758 memory_region_add_subregion(&sphb->iommu_root, 0,
1759 spapr_tce_get_iommu(tcet));
1762 sphb->msi = g_hash_table_new_full(g_int_hash, g_int_equal, g_free, g_free);
1765 static int spapr_phb_children_reset(Object *child, void *opaque)
1767 DeviceState *dev = (DeviceState *) object_dynamic_cast(child, TYPE_DEVICE);
1769 if (dev) {
1770 device_reset(dev);
1773 return 0;
1776 void spapr_phb_dma_reset(sPAPRPHBState *sphb)
1778 int i;
1779 sPAPRTCETable *tcet;
1781 for (i = 0; i < SPAPR_PCI_DMA_MAX_WINDOWS; ++i) {
1782 tcet = spapr_tce_find_by_liobn(sphb->dma_liobn[i]);
1784 if (tcet && tcet->nb_table) {
1785 spapr_tce_table_disable(tcet);
1789 /* Register default 32bit DMA window */
1790 tcet = spapr_tce_find_by_liobn(sphb->dma_liobn[0]);
1791 spapr_tce_table_enable(tcet, SPAPR_TCE_PAGE_SHIFT, sphb->dma_win_addr,
1792 sphb->dma_win_size >> SPAPR_TCE_PAGE_SHIFT);
1795 static void spapr_phb_reset(DeviceState *qdev)
1797 sPAPRPHBState *sphb = SPAPR_PCI_HOST_BRIDGE(qdev);
1799 spapr_phb_dma_reset(sphb);
1801 /* Reset the IOMMU state */
1802 object_child_foreach(OBJECT(qdev), spapr_phb_children_reset, NULL);
1804 if (spapr_phb_eeh_available(SPAPR_PCI_HOST_BRIDGE(qdev))) {
1805 spapr_phb_vfio_reset(qdev);
1809 static Property spapr_phb_properties[] = {
1810 DEFINE_PROP_UINT32("index", sPAPRPHBState, index, -1),
1811 DEFINE_PROP_UINT64("mem_win_size", sPAPRPHBState, mem_win_size,
1812 SPAPR_PCI_MEM32_WIN_SIZE),
1813 DEFINE_PROP_UINT64("mem64_win_size", sPAPRPHBState, mem64_win_size,
1814 SPAPR_PCI_MEM64_WIN_SIZE),
1815 DEFINE_PROP_UINT64("io_win_size", sPAPRPHBState, io_win_size,
1816 SPAPR_PCI_IO_WIN_SIZE),
1817 DEFINE_PROP_BOOL("dynamic-reconfiguration", sPAPRPHBState, dr_enabled,
1818 true),
1819 /* Default DMA window is 0..1GB */
1820 DEFINE_PROP_UINT64("dma_win_addr", sPAPRPHBState, dma_win_addr, 0),
1821 DEFINE_PROP_UINT64("dma_win_size", sPAPRPHBState, dma_win_size, 0x40000000),
1822 DEFINE_PROP_UINT64("dma64_win_addr", sPAPRPHBState, dma64_win_addr,
1823 0x800000000000000ULL),
1824 DEFINE_PROP_BOOL("ddw", sPAPRPHBState, ddw_enabled, true),
1825 DEFINE_PROP_UINT64("pgsz", sPAPRPHBState, page_size_mask,
1826 (1ULL << 12) | (1ULL << 16)),
1827 DEFINE_PROP_UINT32("numa_node", sPAPRPHBState, numa_node, -1),
1828 DEFINE_PROP_BOOL("pre-2.8-migration", sPAPRPHBState,
1829 pre_2_8_migration, false),
1830 DEFINE_PROP_BOOL("pcie-extended-configuration-space", sPAPRPHBState,
1831 pcie_ecs, true),
1832 DEFINE_PROP_END_OF_LIST(),
1835 static const VMStateDescription vmstate_spapr_pci_lsi = {
1836 .name = "spapr_pci/lsi",
1837 .version_id = 1,
1838 .minimum_version_id = 1,
1839 .fields = (VMStateField[]) {
1840 VMSTATE_UINT32_EQUAL(irq, struct spapr_pci_lsi, NULL),
1842 VMSTATE_END_OF_LIST()
1846 static const VMStateDescription vmstate_spapr_pci_msi = {
1847 .name = "spapr_pci/msi",
1848 .version_id = 1,
1849 .minimum_version_id = 1,
1850 .fields = (VMStateField []) {
1851 VMSTATE_UINT32(key, spapr_pci_msi_mig),
1852 VMSTATE_UINT32(value.first_irq, spapr_pci_msi_mig),
1853 VMSTATE_UINT32(value.num, spapr_pci_msi_mig),
1854 VMSTATE_END_OF_LIST()
1858 static int spapr_pci_pre_save(void *opaque)
1860 sPAPRPHBState *sphb = opaque;
1861 GHashTableIter iter;
1862 gpointer key, value;
1863 int i;
1865 if (sphb->pre_2_8_migration) {
1866 sphb->mig_liobn = sphb->dma_liobn[0];
1867 sphb->mig_mem_win_addr = sphb->mem_win_addr;
1868 sphb->mig_mem_win_size = sphb->mem_win_size;
1869 sphb->mig_io_win_addr = sphb->io_win_addr;
1870 sphb->mig_io_win_size = sphb->io_win_size;
1872 if ((sphb->mem64_win_size != 0)
1873 && (sphb->mem64_win_addr
1874 == (sphb->mem_win_addr + sphb->mem_win_size))) {
1875 sphb->mig_mem_win_size += sphb->mem64_win_size;
1879 g_free(sphb->msi_devs);
1880 sphb->msi_devs = NULL;
1881 sphb->msi_devs_num = g_hash_table_size(sphb->msi);
1882 if (!sphb->msi_devs_num) {
1883 return 0;
1885 sphb->msi_devs = g_malloc(sphb->msi_devs_num * sizeof(spapr_pci_msi_mig));
1887 g_hash_table_iter_init(&iter, sphb->msi);
1888 for (i = 0; g_hash_table_iter_next(&iter, &key, &value); ++i) {
1889 sphb->msi_devs[i].key = *(uint32_t *) key;
1890 sphb->msi_devs[i].value = *(spapr_pci_msi *) value;
1893 return 0;
1896 static int spapr_pci_post_load(void *opaque, int version_id)
1898 sPAPRPHBState *sphb = opaque;
1899 gpointer key, value;
1900 int i;
1902 for (i = 0; i < sphb->msi_devs_num; ++i) {
1903 key = g_memdup(&sphb->msi_devs[i].key,
1904 sizeof(sphb->msi_devs[i].key));
1905 value = g_memdup(&sphb->msi_devs[i].value,
1906 sizeof(sphb->msi_devs[i].value));
1907 g_hash_table_insert(sphb->msi, key, value);
1909 g_free(sphb->msi_devs);
1910 sphb->msi_devs = NULL;
1911 sphb->msi_devs_num = 0;
1913 return 0;
1916 static bool pre_2_8_migration(void *opaque, int version_id)
1918 sPAPRPHBState *sphb = opaque;
1920 return sphb->pre_2_8_migration;
1923 static const VMStateDescription vmstate_spapr_pci = {
1924 .name = "spapr_pci",
1925 .version_id = 2,
1926 .minimum_version_id = 2,
1927 .pre_save = spapr_pci_pre_save,
1928 .post_load = spapr_pci_post_load,
1929 .fields = (VMStateField[]) {
1930 VMSTATE_UINT64_EQUAL(buid, sPAPRPHBState, NULL),
1931 VMSTATE_UINT32_TEST(mig_liobn, sPAPRPHBState, pre_2_8_migration),
1932 VMSTATE_UINT64_TEST(mig_mem_win_addr, sPAPRPHBState, pre_2_8_migration),
1933 VMSTATE_UINT64_TEST(mig_mem_win_size, sPAPRPHBState, pre_2_8_migration),
1934 VMSTATE_UINT64_TEST(mig_io_win_addr, sPAPRPHBState, pre_2_8_migration),
1935 VMSTATE_UINT64_TEST(mig_io_win_size, sPAPRPHBState, pre_2_8_migration),
1936 VMSTATE_STRUCT_ARRAY(lsi_table, sPAPRPHBState, PCI_NUM_PINS, 0,
1937 vmstate_spapr_pci_lsi, struct spapr_pci_lsi),
1938 VMSTATE_INT32(msi_devs_num, sPAPRPHBState),
1939 VMSTATE_STRUCT_VARRAY_ALLOC(msi_devs, sPAPRPHBState, msi_devs_num, 0,
1940 vmstate_spapr_pci_msi, spapr_pci_msi_mig),
1941 VMSTATE_END_OF_LIST()
1945 static const char *spapr_phb_root_bus_path(PCIHostState *host_bridge,
1946 PCIBus *rootbus)
1948 sPAPRPHBState *sphb = SPAPR_PCI_HOST_BRIDGE(host_bridge);
1950 return sphb->dtbusname;
1953 static void spapr_phb_class_init(ObjectClass *klass, void *data)
1955 PCIHostBridgeClass *hc = PCI_HOST_BRIDGE_CLASS(klass);
1956 DeviceClass *dc = DEVICE_CLASS(klass);
1957 HotplugHandlerClass *hp = HOTPLUG_HANDLER_CLASS(klass);
1959 hc->root_bus_path = spapr_phb_root_bus_path;
1960 dc->realize = spapr_phb_realize;
1961 dc->props = spapr_phb_properties;
1962 dc->reset = spapr_phb_reset;
1963 dc->vmsd = &vmstate_spapr_pci;
1964 /* Supported by TYPE_SPAPR_MACHINE */
1965 dc->user_creatable = true;
1966 set_bit(DEVICE_CATEGORY_BRIDGE, dc->categories);
1967 hp->plug = spapr_pci_plug;
1968 hp->unplug_request = spapr_pci_unplug_request;
1971 static const TypeInfo spapr_phb_info = {
1972 .name = TYPE_SPAPR_PCI_HOST_BRIDGE,
1973 .parent = TYPE_PCI_HOST_BRIDGE,
1974 .instance_size = sizeof(sPAPRPHBState),
1975 .class_init = spapr_phb_class_init,
1976 .interfaces = (InterfaceInfo[]) {
1977 { TYPE_HOTPLUG_HANDLER },
1982 PCIHostState *spapr_create_phb(sPAPRMachineState *spapr, int index)
1984 DeviceState *dev;
1986 dev = qdev_create(NULL, TYPE_SPAPR_PCI_HOST_BRIDGE);
1987 qdev_prop_set_uint32(dev, "index", index);
1988 qdev_init_nofail(dev);
1990 return PCI_HOST_BRIDGE(dev);
1993 typedef struct sPAPRFDT {
1994 void *fdt;
1995 int node_off;
1996 sPAPRPHBState *sphb;
1997 } sPAPRFDT;
1999 static void spapr_populate_pci_devices_dt(PCIBus *bus, PCIDevice *pdev,
2000 void *opaque)
2002 PCIBus *sec_bus;
2003 sPAPRFDT *p = opaque;
2004 int offset;
2005 sPAPRFDT s_fdt;
2007 offset = spapr_create_pci_child_dt(p->sphb, pdev, p->fdt, p->node_off);
2008 if (!offset) {
2009 error_report("Failed to create pci child device tree node");
2010 return;
2013 if ((pci_default_read_config(pdev, PCI_HEADER_TYPE, 1) !=
2014 PCI_HEADER_TYPE_BRIDGE)) {
2015 return;
2018 sec_bus = pci_bridge_get_sec_bus(PCI_BRIDGE(pdev));
2019 if (!sec_bus) {
2020 return;
2023 s_fdt.fdt = p->fdt;
2024 s_fdt.node_off = offset;
2025 s_fdt.sphb = p->sphb;
2026 pci_for_each_device_reverse(sec_bus, pci_bus_num(sec_bus),
2027 spapr_populate_pci_devices_dt,
2028 &s_fdt);
2031 static void spapr_phb_pci_enumerate_bridge(PCIBus *bus, PCIDevice *pdev,
2032 void *opaque)
2034 unsigned int *bus_no = opaque;
2035 unsigned int primary = *bus_no;
2036 unsigned int subordinate = 0xff;
2037 PCIBus *sec_bus = NULL;
2039 if ((pci_default_read_config(pdev, PCI_HEADER_TYPE, 1) !=
2040 PCI_HEADER_TYPE_BRIDGE)) {
2041 return;
2044 (*bus_no)++;
2045 pci_default_write_config(pdev, PCI_PRIMARY_BUS, primary, 1);
2046 pci_default_write_config(pdev, PCI_SECONDARY_BUS, *bus_no, 1);
2047 pci_default_write_config(pdev, PCI_SUBORDINATE_BUS, *bus_no, 1);
2049 sec_bus = pci_bridge_get_sec_bus(PCI_BRIDGE(pdev));
2050 if (!sec_bus) {
2051 return;
2054 pci_default_write_config(pdev, PCI_SUBORDINATE_BUS, subordinate, 1);
2055 pci_for_each_device(sec_bus, pci_bus_num(sec_bus),
2056 spapr_phb_pci_enumerate_bridge, bus_no);
2057 pci_default_write_config(pdev, PCI_SUBORDINATE_BUS, *bus_no, 1);
2060 static void spapr_phb_pci_enumerate(sPAPRPHBState *phb)
2062 PCIBus *bus = PCI_HOST_BRIDGE(phb)->bus;
2063 unsigned int bus_no = 0;
2065 pci_for_each_device(bus, pci_bus_num(bus),
2066 spapr_phb_pci_enumerate_bridge,
2067 &bus_no);
2071 int spapr_populate_pci_dt(sPAPRPHBState *phb,
2072 uint32_t xics_phandle,
2073 void *fdt)
2075 int bus_off, i, j, ret;
2076 gchar *nodename;
2077 uint32_t bus_range[] = { cpu_to_be32(0), cpu_to_be32(0xff) };
2078 struct {
2079 uint32_t hi;
2080 uint64_t child;
2081 uint64_t parent;
2082 uint64_t size;
2083 } QEMU_PACKED ranges[] = {
2085 cpu_to_be32(b_ss(1)), cpu_to_be64(0),
2086 cpu_to_be64(phb->io_win_addr),
2087 cpu_to_be64(memory_region_size(&phb->iospace)),
2090 cpu_to_be32(b_ss(2)), cpu_to_be64(SPAPR_PCI_MEM_WIN_BUS_OFFSET),
2091 cpu_to_be64(phb->mem_win_addr),
2092 cpu_to_be64(phb->mem_win_size),
2095 cpu_to_be32(b_ss(3)), cpu_to_be64(phb->mem64_win_pciaddr),
2096 cpu_to_be64(phb->mem64_win_addr),
2097 cpu_to_be64(phb->mem64_win_size),
2100 const unsigned sizeof_ranges =
2101 (phb->mem64_win_size ? 3 : 2) * sizeof(ranges[0]);
2102 uint64_t bus_reg[] = { cpu_to_be64(phb->buid), 0 };
2103 uint32_t interrupt_map_mask[] = {
2104 cpu_to_be32(b_ddddd(-1)|b_fff(0)), 0x0, 0x0, cpu_to_be32(-1)};
2105 uint32_t interrupt_map[PCI_SLOT_MAX * PCI_NUM_PINS][7];
2106 uint32_t ddw_applicable[] = {
2107 cpu_to_be32(RTAS_IBM_QUERY_PE_DMA_WINDOW),
2108 cpu_to_be32(RTAS_IBM_CREATE_PE_DMA_WINDOW),
2109 cpu_to_be32(RTAS_IBM_REMOVE_PE_DMA_WINDOW)
2111 uint32_t ddw_extensions[] = {
2112 cpu_to_be32(1),
2113 cpu_to_be32(RTAS_IBM_RESET_PE_DMA_WINDOW)
2115 uint32_t associativity[] = {cpu_to_be32(0x4),
2116 cpu_to_be32(0x0),
2117 cpu_to_be32(0x0),
2118 cpu_to_be32(0x0),
2119 cpu_to_be32(phb->numa_node)};
2120 sPAPRTCETable *tcet;
2121 PCIBus *bus = PCI_HOST_BRIDGE(phb)->bus;
2122 sPAPRFDT s_fdt;
2124 /* Start populating the FDT */
2125 nodename = g_strdup_printf("pci@%" PRIx64, phb->buid);
2126 _FDT(bus_off = fdt_add_subnode(fdt, 0, nodename));
2127 g_free(nodename);
2129 /* Write PHB properties */
2130 _FDT(fdt_setprop_string(fdt, bus_off, "device_type", "pci"));
2131 _FDT(fdt_setprop_string(fdt, bus_off, "compatible", "IBM,Logical_PHB"));
2132 _FDT(fdt_setprop_cell(fdt, bus_off, "#address-cells", 0x3));
2133 _FDT(fdt_setprop_cell(fdt, bus_off, "#size-cells", 0x2));
2134 _FDT(fdt_setprop_cell(fdt, bus_off, "#interrupt-cells", 0x1));
2135 _FDT(fdt_setprop(fdt, bus_off, "used-by-rtas", NULL, 0));
2136 _FDT(fdt_setprop(fdt, bus_off, "bus-range", &bus_range, sizeof(bus_range)));
2137 _FDT(fdt_setprop(fdt, bus_off, "ranges", &ranges, sizeof_ranges));
2138 _FDT(fdt_setprop(fdt, bus_off, "reg", &bus_reg, sizeof(bus_reg)));
2139 _FDT(fdt_setprop_cell(fdt, bus_off, "ibm,pci-config-space-type", 0x1));
2140 /* TODO: fine tune the total count of allocatable MSIs per PHB */
2141 _FDT(fdt_setprop_cell(fdt, bus_off, "ibm,pe-total-#msi", XICS_IRQS_SPAPR));
2143 /* Dynamic DMA window */
2144 if (phb->ddw_enabled) {
2145 _FDT(fdt_setprop(fdt, bus_off, "ibm,ddw-applicable", &ddw_applicable,
2146 sizeof(ddw_applicable)));
2147 _FDT(fdt_setprop(fdt, bus_off, "ibm,ddw-extensions",
2148 &ddw_extensions, sizeof(ddw_extensions)));
2151 /* Advertise NUMA via ibm,associativity */
2152 if (phb->numa_node != -1) {
2153 _FDT(fdt_setprop(fdt, bus_off, "ibm,associativity", associativity,
2154 sizeof(associativity)));
2157 /* Build the interrupt-map, this must matches what is done
2158 * in pci_spapr_map_irq
2160 _FDT(fdt_setprop(fdt, bus_off, "interrupt-map-mask",
2161 &interrupt_map_mask, sizeof(interrupt_map_mask)));
2162 for (i = 0; i < PCI_SLOT_MAX; i++) {
2163 for (j = 0; j < PCI_NUM_PINS; j++) {
2164 uint32_t *irqmap = interrupt_map[i*PCI_NUM_PINS + j];
2165 int lsi_num = pci_spapr_swizzle(i, j);
2167 irqmap[0] = cpu_to_be32(b_ddddd(i)|b_fff(0));
2168 irqmap[1] = 0;
2169 irqmap[2] = 0;
2170 irqmap[3] = cpu_to_be32(j+1);
2171 irqmap[4] = cpu_to_be32(xics_phandle);
2172 spapr_dt_xics_irq(&irqmap[5], phb->lsi_table[lsi_num].irq, true);
2175 /* Write interrupt map */
2176 _FDT(fdt_setprop(fdt, bus_off, "interrupt-map", &interrupt_map,
2177 sizeof(interrupt_map)));
2179 tcet = spapr_tce_find_by_liobn(phb->dma_liobn[0]);
2180 if (!tcet) {
2181 return -1;
2183 spapr_dma_dt(fdt, bus_off, "ibm,dma-window",
2184 tcet->liobn, tcet->bus_offset,
2185 tcet->nb_table << tcet->page_shift);
2187 /* Walk the bridges and program the bus numbers*/
2188 spapr_phb_pci_enumerate(phb);
2189 _FDT(fdt_setprop_cell(fdt, bus_off, "qemu,phb-enumerated", 0x1));
2191 /* Populate tree nodes with PCI devices attached */
2192 s_fdt.fdt = fdt;
2193 s_fdt.node_off = bus_off;
2194 s_fdt.sphb = phb;
2195 pci_for_each_device_reverse(bus, pci_bus_num(bus),
2196 spapr_populate_pci_devices_dt,
2197 &s_fdt);
2199 ret = spapr_drc_populate_dt(fdt, bus_off, OBJECT(phb),
2200 SPAPR_DR_CONNECTOR_TYPE_PCI);
2201 if (ret) {
2202 return ret;
2205 return 0;
2208 void spapr_pci_rtas_init(void)
2210 spapr_rtas_register(RTAS_READ_PCI_CONFIG, "read-pci-config",
2211 rtas_read_pci_config);
2212 spapr_rtas_register(RTAS_WRITE_PCI_CONFIG, "write-pci-config",
2213 rtas_write_pci_config);
2214 spapr_rtas_register(RTAS_IBM_READ_PCI_CONFIG, "ibm,read-pci-config",
2215 rtas_ibm_read_pci_config);
2216 spapr_rtas_register(RTAS_IBM_WRITE_PCI_CONFIG, "ibm,write-pci-config",
2217 rtas_ibm_write_pci_config);
2218 if (msi_nonbroken) {
2219 spapr_rtas_register(RTAS_IBM_QUERY_INTERRUPT_SOURCE_NUMBER,
2220 "ibm,query-interrupt-source-number",
2221 rtas_ibm_query_interrupt_source_number);
2222 spapr_rtas_register(RTAS_IBM_CHANGE_MSI, "ibm,change-msi",
2223 rtas_ibm_change_msi);
2226 spapr_rtas_register(RTAS_IBM_SET_EEH_OPTION,
2227 "ibm,set-eeh-option",
2228 rtas_ibm_set_eeh_option);
2229 spapr_rtas_register(RTAS_IBM_GET_CONFIG_ADDR_INFO2,
2230 "ibm,get-config-addr-info2",
2231 rtas_ibm_get_config_addr_info2);
2232 spapr_rtas_register(RTAS_IBM_READ_SLOT_RESET_STATE2,
2233 "ibm,read-slot-reset-state2",
2234 rtas_ibm_read_slot_reset_state2);
2235 spapr_rtas_register(RTAS_IBM_SET_SLOT_RESET,
2236 "ibm,set-slot-reset",
2237 rtas_ibm_set_slot_reset);
2238 spapr_rtas_register(RTAS_IBM_CONFIGURE_PE,
2239 "ibm,configure-pe",
2240 rtas_ibm_configure_pe);
2241 spapr_rtas_register(RTAS_IBM_SLOT_ERROR_DETAIL,
2242 "ibm,slot-error-detail",
2243 rtas_ibm_slot_error_detail);
2246 static void spapr_pci_register_types(void)
2248 type_register_static(&spapr_phb_info);
2251 type_init(spapr_pci_register_types)
2253 static int spapr_switch_one_vga(DeviceState *dev, void *opaque)
2255 bool be = *(bool *)opaque;
2257 if (object_dynamic_cast(OBJECT(dev), "VGA")
2258 || object_dynamic_cast(OBJECT(dev), "secondary-vga")) {
2259 object_property_set_bool(OBJECT(dev), be, "big-endian-framebuffer",
2260 &error_abort);
2262 return 0;
2265 void spapr_pci_switch_vga(bool big_endian)
2267 sPAPRMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
2268 sPAPRPHBState *sphb;
2271 * For backward compatibility with existing guests, we switch
2272 * the endianness of the VGA controller when changing the guest
2273 * interrupt mode
2275 QLIST_FOREACH(sphb, &spapr->phbs, list) {
2276 BusState *bus = &PCI_HOST_BRIDGE(sphb)->bus->qbus;
2277 qbus_walk_children(bus, spapr_switch_one_vga, NULL, NULL, NULL,
2278 &big_endian);