qom: Put name parameter before value / visitor parameter
[qemu/ar7.git] / hw / ppc / spapr_pci.c
blob223bb8f46493505d3817d0f2000c4f85b0ad9020
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.
26 #include "qemu/osdep.h"
27 #include "qapi/error.h"
28 #include "cpu.h"
29 #include "hw/irq.h"
30 #include "hw/sysbus.h"
31 #include "migration/vmstate.h"
32 #include "hw/pci/pci.h"
33 #include "hw/pci/msi.h"
34 #include "hw/pci/msix.h"
35 #include "hw/pci/pci_host.h"
36 #include "hw/ppc/spapr.h"
37 #include "hw/pci-host/spapr.h"
38 #include "exec/address-spaces.h"
39 #include "exec/ram_addr.h"
40 #include <libfdt.h>
41 #include "trace.h"
42 #include "qemu/error-report.h"
43 #include "qemu/module.h"
44 #include "qapi/qmp/qerror.h"
45 #include "hw/ppc/fdt.h"
46 #include "hw/pci/pci_bridge.h"
47 #include "hw/pci/pci_bus.h"
48 #include "hw/pci/pci_ids.h"
49 #include "hw/ppc/spapr_drc.h"
50 #include "hw/qdev-properties.h"
51 #include "sysemu/device_tree.h"
52 #include "sysemu/kvm.h"
53 #include "sysemu/hostmem.h"
54 #include "sysemu/numa.h"
56 /* Copied from the kernel arch/powerpc/platforms/pseries/msi.c */
57 #define RTAS_QUERY_FN 0
58 #define RTAS_CHANGE_FN 1
59 #define RTAS_RESET_FN 2
60 #define RTAS_CHANGE_MSI_FN 3
61 #define RTAS_CHANGE_MSIX_FN 4
63 /* Interrupt types to return on RTAS_CHANGE_* */
64 #define RTAS_TYPE_MSI 1
65 #define RTAS_TYPE_MSIX 2
67 SpaprPhbState *spapr_pci_find_phb(SpaprMachineState *spapr, uint64_t buid)
69 SpaprPhbState *sphb;
71 QLIST_FOREACH(sphb, &spapr->phbs, list) {
72 if (sphb->buid != buid) {
73 continue;
75 return sphb;
78 return NULL;
81 PCIDevice *spapr_pci_find_dev(SpaprMachineState *spapr, uint64_t buid,
82 uint32_t config_addr)
84 SpaprPhbState *sphb = spapr_pci_find_phb(spapr, buid);
85 PCIHostState *phb = PCI_HOST_BRIDGE(sphb);
86 int bus_num = (config_addr >> 16) & 0xFF;
87 int devfn = (config_addr >> 8) & 0xFF;
89 if (!phb) {
90 return NULL;
93 return pci_find_device(phb->bus, bus_num, devfn);
96 static uint32_t rtas_pci_cfgaddr(uint32_t arg)
98 /* This handles the encoding of extended config space addresses */
99 return ((arg >> 20) & 0xf00) | (arg & 0xff);
102 static void finish_read_pci_config(SpaprMachineState *spapr, uint64_t buid,
103 uint32_t addr, uint32_t size,
104 target_ulong rets)
106 PCIDevice *pci_dev;
107 uint32_t val;
109 if ((size != 1) && (size != 2) && (size != 4)) {
110 /* access must be 1, 2 or 4 bytes */
111 rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
112 return;
115 pci_dev = spapr_pci_find_dev(spapr, buid, addr);
116 addr = rtas_pci_cfgaddr(addr);
118 if (!pci_dev || (addr % size) || (addr >= pci_config_size(pci_dev))) {
119 /* Access must be to a valid device, within bounds and
120 * naturally aligned */
121 rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
122 return;
125 val = pci_host_config_read_common(pci_dev, addr,
126 pci_config_size(pci_dev), size);
128 rtas_st(rets, 0, RTAS_OUT_SUCCESS);
129 rtas_st(rets, 1, val);
132 static void rtas_ibm_read_pci_config(PowerPCCPU *cpu, SpaprMachineState *spapr,
133 uint32_t token, uint32_t nargs,
134 target_ulong args,
135 uint32_t nret, target_ulong rets)
137 uint64_t buid;
138 uint32_t size, addr;
140 if ((nargs != 4) || (nret != 2)) {
141 rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
142 return;
145 buid = rtas_ldq(args, 1);
146 size = rtas_ld(args, 3);
147 addr = rtas_ld(args, 0);
149 finish_read_pci_config(spapr, buid, addr, size, rets);
152 static void rtas_read_pci_config(PowerPCCPU *cpu, SpaprMachineState *spapr,
153 uint32_t token, uint32_t nargs,
154 target_ulong args,
155 uint32_t nret, target_ulong rets)
157 uint32_t size, addr;
159 if ((nargs != 2) || (nret != 2)) {
160 rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
161 return;
164 size = rtas_ld(args, 1);
165 addr = rtas_ld(args, 0);
167 finish_read_pci_config(spapr, 0, addr, size, rets);
170 static void finish_write_pci_config(SpaprMachineState *spapr, uint64_t buid,
171 uint32_t addr, uint32_t size,
172 uint32_t val, target_ulong rets)
174 PCIDevice *pci_dev;
176 if ((size != 1) && (size != 2) && (size != 4)) {
177 /* access must be 1, 2 or 4 bytes */
178 rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
179 return;
182 pci_dev = spapr_pci_find_dev(spapr, buid, addr);
183 addr = rtas_pci_cfgaddr(addr);
185 if (!pci_dev || (addr % size) || (addr >= pci_config_size(pci_dev))) {
186 /* Access must be to a valid device, within bounds and
187 * naturally aligned */
188 rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
189 return;
192 pci_host_config_write_common(pci_dev, addr, pci_config_size(pci_dev),
193 val, size);
195 rtas_st(rets, 0, RTAS_OUT_SUCCESS);
198 static void rtas_ibm_write_pci_config(PowerPCCPU *cpu, SpaprMachineState *spapr,
199 uint32_t token, uint32_t nargs,
200 target_ulong args,
201 uint32_t nret, target_ulong rets)
203 uint64_t buid;
204 uint32_t val, size, addr;
206 if ((nargs != 5) || (nret != 1)) {
207 rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
208 return;
211 buid = rtas_ldq(args, 1);
212 val = rtas_ld(args, 4);
213 size = rtas_ld(args, 3);
214 addr = rtas_ld(args, 0);
216 finish_write_pci_config(spapr, buid, addr, size, val, rets);
219 static void rtas_write_pci_config(PowerPCCPU *cpu, SpaprMachineState *spapr,
220 uint32_t token, uint32_t nargs,
221 target_ulong args,
222 uint32_t nret, target_ulong rets)
224 uint32_t val, size, addr;
226 if ((nargs != 3) || (nret != 1)) {
227 rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
228 return;
232 val = rtas_ld(args, 2);
233 size = rtas_ld(args, 1);
234 addr = rtas_ld(args, 0);
236 finish_write_pci_config(spapr, 0, addr, size, val, rets);
240 * Set MSI/MSIX message data.
241 * This is required for msi_notify()/msix_notify() which
242 * will write at the addresses via spapr_msi_write().
244 * If hwaddr == 0, all entries will have .data == first_irq i.e.
245 * table will be reset.
247 static void spapr_msi_setmsg(PCIDevice *pdev, hwaddr addr, bool msix,
248 unsigned first_irq, unsigned req_num)
250 unsigned i;
251 MSIMessage msg = { .address = addr, .data = first_irq };
253 if (!msix) {
254 msi_set_message(pdev, msg);
255 trace_spapr_pci_msi_setup(pdev->name, 0, msg.address);
256 return;
259 for (i = 0; i < req_num; ++i) {
260 msix_set_message(pdev, i, msg);
261 trace_spapr_pci_msi_setup(pdev->name, i, msg.address);
262 if (addr) {
263 ++msg.data;
268 static void rtas_ibm_change_msi(PowerPCCPU *cpu, SpaprMachineState *spapr,
269 uint32_t token, uint32_t nargs,
270 target_ulong args, uint32_t nret,
271 target_ulong rets)
273 SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
274 uint32_t config_addr = rtas_ld(args, 0);
275 uint64_t buid = rtas_ldq(args, 1);
276 unsigned int func = rtas_ld(args, 3);
277 unsigned int req_num = rtas_ld(args, 4); /* 0 == remove all */
278 unsigned int seq_num = rtas_ld(args, 5);
279 unsigned int ret_intr_type;
280 unsigned int irq, max_irqs = 0;
281 SpaprPhbState *phb = NULL;
282 PCIDevice *pdev = NULL;
283 SpaprPciMsi *msi;
284 int *config_addr_key;
285 Error *err = NULL;
286 int i;
288 /* Fins SpaprPhbState */
289 phb = spapr_pci_find_phb(spapr, buid);
290 if (phb) {
291 pdev = spapr_pci_find_dev(spapr, buid, config_addr);
293 if (!phb || !pdev) {
294 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
295 return;
298 switch (func) {
299 case RTAS_CHANGE_FN:
300 if (msi_present(pdev)) {
301 ret_intr_type = RTAS_TYPE_MSI;
302 } else if (msix_present(pdev)) {
303 ret_intr_type = RTAS_TYPE_MSIX;
304 } else {
305 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
306 return;
308 break;
309 case RTAS_CHANGE_MSI_FN:
310 if (msi_present(pdev)) {
311 ret_intr_type = RTAS_TYPE_MSI;
312 } else {
313 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
314 return;
316 break;
317 case RTAS_CHANGE_MSIX_FN:
318 if (msix_present(pdev)) {
319 ret_intr_type = RTAS_TYPE_MSIX;
320 } else {
321 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
322 return;
324 break;
325 default:
326 error_report("rtas_ibm_change_msi(%u) is not implemented", func);
327 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
328 return;
331 msi = (SpaprPciMsi *) g_hash_table_lookup(phb->msi, &config_addr);
333 /* Releasing MSIs */
334 if (!req_num) {
335 if (!msi) {
336 trace_spapr_pci_msi("Releasing wrong config", config_addr);
337 rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
338 return;
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 (smc->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 if (i) {
396 spapr_irq_free(spapr, irq, i);
398 if (!smc->legacy_irq_allocation) {
399 spapr_irq_msi_free(spapr, irq, req_num);
401 error_reportf_err(err, "Can't allocate MSIs for device %x: ",
402 config_addr);
403 rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
404 return;
408 /* Release previous MSIs */
409 if (msi) {
410 g_hash_table_remove(phb->msi, &config_addr);
413 /* Setup MSI/MSIX vectors in the device (via cfgspace or MSIX BAR) */
414 spapr_msi_setmsg(pdev, SPAPR_PCI_MSI_WINDOW, ret_intr_type == RTAS_TYPE_MSIX,
415 irq, req_num);
417 /* Add MSI device to cache */
418 msi = g_new(SpaprPciMsi, 1);
419 msi->first_irq = irq;
420 msi->num = req_num;
421 config_addr_key = g_new(int, 1);
422 *config_addr_key = config_addr;
423 g_hash_table_insert(phb->msi, config_addr_key, msi);
425 out:
426 rtas_st(rets, 0, RTAS_OUT_SUCCESS);
427 rtas_st(rets, 1, req_num);
428 rtas_st(rets, 2, ++seq_num);
429 if (nret > 3) {
430 rtas_st(rets, 3, ret_intr_type);
433 trace_spapr_pci_rtas_ibm_change_msi(config_addr, func, req_num, irq);
436 static void rtas_ibm_query_interrupt_source_number(PowerPCCPU *cpu,
437 SpaprMachineState *spapr,
438 uint32_t token,
439 uint32_t nargs,
440 target_ulong args,
441 uint32_t nret,
442 target_ulong rets)
444 uint32_t config_addr = rtas_ld(args, 0);
445 uint64_t buid = rtas_ldq(args, 1);
446 unsigned int intr_src_num = -1, ioa_intr_num = rtas_ld(args, 3);
447 SpaprPhbState *phb = NULL;
448 PCIDevice *pdev = NULL;
449 SpaprPciMsi *msi;
451 /* Find SpaprPhbState */
452 phb = spapr_pci_find_phb(spapr, buid);
453 if (phb) {
454 pdev = spapr_pci_find_dev(spapr, buid, config_addr);
456 if (!phb || !pdev) {
457 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
458 return;
461 /* Find device descriptor and start IRQ */
462 msi = (SpaprPciMsi *) g_hash_table_lookup(phb->msi, &config_addr);
463 if (!msi || !msi->first_irq || !msi->num || (ioa_intr_num >= msi->num)) {
464 trace_spapr_pci_msi("Failed to return vector", config_addr);
465 rtas_st(rets, 0, RTAS_OUT_HW_ERROR);
466 return;
468 intr_src_num = msi->first_irq + ioa_intr_num;
469 trace_spapr_pci_rtas_ibm_query_interrupt_source_number(ioa_intr_num,
470 intr_src_num);
472 rtas_st(rets, 0, RTAS_OUT_SUCCESS);
473 rtas_st(rets, 1, intr_src_num);
474 rtas_st(rets, 2, 1);/* 0 == level; 1 == edge */
477 static void rtas_ibm_set_eeh_option(PowerPCCPU *cpu,
478 SpaprMachineState *spapr,
479 uint32_t token, uint32_t nargs,
480 target_ulong args, uint32_t nret,
481 target_ulong rets)
483 SpaprPhbState *sphb;
484 uint32_t addr, option;
485 uint64_t buid;
486 int ret;
488 if ((nargs != 4) || (nret != 1)) {
489 goto param_error_exit;
492 buid = rtas_ldq(args, 1);
493 addr = rtas_ld(args, 0);
494 option = rtas_ld(args, 3);
496 sphb = spapr_pci_find_phb(spapr, buid);
497 if (!sphb) {
498 goto param_error_exit;
501 if (!spapr_phb_eeh_available(sphb)) {
502 goto param_error_exit;
505 ret = spapr_phb_vfio_eeh_set_option(sphb, addr, option);
506 rtas_st(rets, 0, ret);
507 return;
509 param_error_exit:
510 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
513 static void rtas_ibm_get_config_addr_info2(PowerPCCPU *cpu,
514 SpaprMachineState *spapr,
515 uint32_t token, uint32_t nargs,
516 target_ulong args, uint32_t nret,
517 target_ulong rets)
519 SpaprPhbState *sphb;
520 PCIDevice *pdev;
521 uint32_t addr, option;
522 uint64_t buid;
524 if ((nargs != 4) || (nret != 2)) {
525 goto param_error_exit;
528 buid = rtas_ldq(args, 1);
529 sphb = spapr_pci_find_phb(spapr, buid);
530 if (!sphb) {
531 goto param_error_exit;
534 if (!spapr_phb_eeh_available(sphb)) {
535 goto param_error_exit;
539 * We always have PE address of form "00BB0001". "BB"
540 * represents the bus number of PE's primary bus.
542 option = rtas_ld(args, 3);
543 switch (option) {
544 case RTAS_GET_PE_ADDR:
545 addr = rtas_ld(args, 0);
546 pdev = spapr_pci_find_dev(spapr, buid, addr);
547 if (!pdev) {
548 goto param_error_exit;
551 rtas_st(rets, 1, (pci_bus_num(pci_get_bus(pdev)) << 16) + 1);
552 break;
553 case RTAS_GET_PE_MODE:
554 rtas_st(rets, 1, RTAS_PE_MODE_SHARED);
555 break;
556 default:
557 goto param_error_exit;
560 rtas_st(rets, 0, RTAS_OUT_SUCCESS);
561 return;
563 param_error_exit:
564 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
567 static void rtas_ibm_read_slot_reset_state2(PowerPCCPU *cpu,
568 SpaprMachineState *spapr,
569 uint32_t token, uint32_t nargs,
570 target_ulong args, uint32_t nret,
571 target_ulong rets)
573 SpaprPhbState *sphb;
574 uint64_t buid;
575 int state, ret;
577 if ((nargs != 3) || (nret != 4 && nret != 5)) {
578 goto param_error_exit;
581 buid = rtas_ldq(args, 1);
582 sphb = spapr_pci_find_phb(spapr, buid);
583 if (!sphb) {
584 goto param_error_exit;
587 if (!spapr_phb_eeh_available(sphb)) {
588 goto param_error_exit;
591 ret = spapr_phb_vfio_eeh_get_state(sphb, &state);
592 rtas_st(rets, 0, ret);
593 if (ret != RTAS_OUT_SUCCESS) {
594 return;
597 rtas_st(rets, 1, state);
598 rtas_st(rets, 2, RTAS_EEH_SUPPORT);
599 rtas_st(rets, 3, RTAS_EEH_PE_UNAVAIL_INFO);
600 if (nret >= 5) {
601 rtas_st(rets, 4, RTAS_EEH_PE_RECOVER_INFO);
603 return;
605 param_error_exit:
606 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
609 static void rtas_ibm_set_slot_reset(PowerPCCPU *cpu,
610 SpaprMachineState *spapr,
611 uint32_t token, uint32_t nargs,
612 target_ulong args, uint32_t nret,
613 target_ulong rets)
615 SpaprPhbState *sphb;
616 uint32_t option;
617 uint64_t buid;
618 int ret;
620 if ((nargs != 4) || (nret != 1)) {
621 goto param_error_exit;
624 buid = rtas_ldq(args, 1);
625 option = rtas_ld(args, 3);
626 sphb = spapr_pci_find_phb(spapr, buid);
627 if (!sphb) {
628 goto param_error_exit;
631 if (!spapr_phb_eeh_available(sphb)) {
632 goto param_error_exit;
635 ret = spapr_phb_vfio_eeh_reset(sphb, option);
636 rtas_st(rets, 0, ret);
637 return;
639 param_error_exit:
640 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
643 static void rtas_ibm_configure_pe(PowerPCCPU *cpu,
644 SpaprMachineState *spapr,
645 uint32_t token, uint32_t nargs,
646 target_ulong args, uint32_t nret,
647 target_ulong rets)
649 SpaprPhbState *sphb;
650 uint64_t buid;
651 int ret;
653 if ((nargs != 3) || (nret != 1)) {
654 goto param_error_exit;
657 buid = rtas_ldq(args, 1);
658 sphb = spapr_pci_find_phb(spapr, buid);
659 if (!sphb) {
660 goto param_error_exit;
663 if (!spapr_phb_eeh_available(sphb)) {
664 goto param_error_exit;
667 ret = spapr_phb_vfio_eeh_configure(sphb);
668 rtas_st(rets, 0, ret);
669 return;
671 param_error_exit:
672 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
675 /* To support it later */
676 static void rtas_ibm_slot_error_detail(PowerPCCPU *cpu,
677 SpaprMachineState *spapr,
678 uint32_t token, uint32_t nargs,
679 target_ulong args, uint32_t nret,
680 target_ulong rets)
682 SpaprPhbState *sphb;
683 int option;
684 uint64_t buid;
686 if ((nargs != 8) || (nret != 1)) {
687 goto param_error_exit;
690 buid = rtas_ldq(args, 1);
691 sphb = spapr_pci_find_phb(spapr, buid);
692 if (!sphb) {
693 goto param_error_exit;
696 if (!spapr_phb_eeh_available(sphb)) {
697 goto param_error_exit;
700 option = rtas_ld(args, 7);
701 switch (option) {
702 case RTAS_SLOT_TEMP_ERR_LOG:
703 case RTAS_SLOT_PERM_ERR_LOG:
704 break;
705 default:
706 goto param_error_exit;
709 /* We don't have error log yet */
710 rtas_st(rets, 0, RTAS_OUT_NO_ERRORS_FOUND);
711 return;
713 param_error_exit:
714 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
717 static void pci_spapr_set_irq(void *opaque, int irq_num, int level)
720 * Here we use the number returned by pci_swizzle_map_irq_fn to find a
721 * corresponding qemu_irq.
723 SpaprPhbState *phb = opaque;
724 SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
726 trace_spapr_pci_lsi_set(phb->dtbusname, irq_num, phb->lsi_table[irq_num].irq);
727 qemu_set_irq(spapr_qirq(spapr, phb->lsi_table[irq_num].irq), level);
730 static PCIINTxRoute spapr_route_intx_pin_to_irq(void *opaque, int pin)
732 SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(opaque);
733 PCIINTxRoute route;
735 route.mode = PCI_INTX_ENABLED;
736 route.irq = sphb->lsi_table[pin].irq;
738 return route;
742 * MSI/MSIX memory region implementation.
743 * The handler handles both MSI and MSIX.
744 * The vector number is encoded in least bits in data.
746 static void spapr_msi_write(void *opaque, hwaddr addr,
747 uint64_t data, unsigned size)
749 SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
750 uint32_t irq = data;
752 trace_spapr_pci_msi_write(addr, data, irq);
754 qemu_irq_pulse(spapr_qirq(spapr, irq));
757 static const MemoryRegionOps spapr_msi_ops = {
758 /* There is no .read as the read result is undefined by PCI spec */
759 .read = NULL,
760 .write = spapr_msi_write,
761 .endianness = DEVICE_LITTLE_ENDIAN
765 * PHB PCI device
767 static AddressSpace *spapr_pci_dma_iommu(PCIBus *bus, void *opaque, int devfn)
769 SpaprPhbState *phb = opaque;
771 return &phb->iommu_as;
774 static char *spapr_phb_vfio_get_loc_code(SpaprPhbState *sphb, PCIDevice *pdev)
776 char *path = NULL, *buf = NULL, *host = NULL;
778 /* Get the PCI VFIO host id */
779 host = object_property_get_str(OBJECT(pdev), "host", NULL);
780 if (!host) {
781 goto err_out;
784 /* Construct the path of the file that will give us the DT location */
785 path = g_strdup_printf("/sys/bus/pci/devices/%s/devspec", host);
786 g_free(host);
787 if (!g_file_get_contents(path, &buf, NULL, NULL)) {
788 goto err_out;
790 g_free(path);
792 /* Construct and read from host device tree the loc-code */
793 path = g_strdup_printf("/proc/device-tree%s/ibm,loc-code", buf);
794 g_free(buf);
795 if (!g_file_get_contents(path, &buf, NULL, NULL)) {
796 goto err_out;
798 return buf;
800 err_out:
801 g_free(path);
802 return NULL;
805 static char *spapr_phb_get_loc_code(SpaprPhbState *sphb, PCIDevice *pdev)
807 char *buf;
808 const char *devtype = "qemu";
809 uint32_t busnr = pci_bus_num(PCI_BUS(qdev_get_parent_bus(DEVICE(pdev))));
811 if (object_dynamic_cast(OBJECT(pdev), "vfio-pci")) {
812 buf = spapr_phb_vfio_get_loc_code(sphb, pdev);
813 if (buf) {
814 return buf;
816 devtype = "vfio";
819 * For emulated devices and VFIO-failure case, make up
820 * the loc-code.
822 buf = g_strdup_printf("%s_%s:%04x:%02x:%02x.%x",
823 devtype, pdev->name, sphb->index, busnr,
824 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn));
825 return buf;
828 /* Macros to operate with address in OF binding to PCI */
829 #define b_x(x, p, l) (((x) & ((1<<(l))-1)) << (p))
830 #define b_n(x) b_x((x), 31, 1) /* 0 if relocatable */
831 #define b_p(x) b_x((x), 30, 1) /* 1 if prefetchable */
832 #define b_t(x) b_x((x), 29, 1) /* 1 if the address is aliased */
833 #define b_ss(x) b_x((x), 24, 2) /* the space code */
834 #define b_bbbbbbbb(x) b_x((x), 16, 8) /* bus number */
835 #define b_ddddd(x) b_x((x), 11, 5) /* device number */
836 #define b_fff(x) b_x((x), 8, 3) /* function number */
837 #define b_rrrrrrrr(x) b_x((x), 0, 8) /* register number */
839 /* for 'reg' OF properties */
840 #define RESOURCE_CELLS_SIZE 2
841 #define RESOURCE_CELLS_ADDRESS 3
843 typedef struct ResourceFields {
844 uint32_t phys_hi;
845 uint32_t phys_mid;
846 uint32_t phys_lo;
847 uint32_t size_hi;
848 uint32_t size_lo;
849 } QEMU_PACKED ResourceFields;
851 typedef struct ResourceProps {
852 ResourceFields reg[8];
853 uint32_t reg_len;
854 } ResourceProps;
856 /* fill in the 'reg' OF properties for
857 * a PCI device. 'reg' describes resource requirements for a
858 * device's IO/MEM regions.
860 * the property is an array of ('phys-addr', 'size') pairs describing
861 * the addressable regions of the PCI device, where 'phys-addr' is a
862 * RESOURCE_CELLS_ADDRESS-tuple of 32-bit integers corresponding to
863 * (phys.hi, phys.mid, phys.lo), and 'size' is a
864 * RESOURCE_CELLS_SIZE-tuple corresponding to (size.hi, size.lo).
866 * phys.hi = 0xYYXXXXZZ, where:
867 * 0xYY = npt000ss
868 * ||| |
869 * ||| +-- space code
870 * ||| |
871 * ||| + 00 if configuration space
872 * ||| + 01 if IO region,
873 * ||| + 10 if 32-bit MEM region
874 * ||| + 11 if 64-bit MEM region
875 * |||
876 * ||+------ for non-relocatable IO: 1 if aliased
877 * || for relocatable IO: 1 if below 64KB
878 * || for MEM: 1 if below 1MB
879 * |+------- 1 if region is prefetchable
880 * +-------- 1 if region is non-relocatable
881 * 0xXXXX = bbbbbbbb dddddfff, encoding bus, slot, and function
882 * bits respectively
883 * 0xZZ = rrrrrrrr, the register number of the BAR corresponding
884 * to the region
886 * phys.mid and phys.lo correspond respectively to the hi/lo portions
887 * of the actual address of the region.
889 * note also that addresses defined in this property are, at least
890 * for PAPR guests, relative to the PHBs IO/MEM windows, and
891 * correspond directly to the addresses in the BARs.
893 * in accordance with PCI Bus Binding to Open Firmware,
894 * IEEE Std 1275-1994, section 4.1.1, as implemented by PAPR+ v2.7,
895 * Appendix C.
897 static void populate_resource_props(PCIDevice *d, ResourceProps *rp)
899 int bus_num = pci_bus_num(PCI_BUS(qdev_get_parent_bus(DEVICE(d))));
900 uint32_t dev_id = (b_bbbbbbbb(bus_num) |
901 b_ddddd(PCI_SLOT(d->devfn)) |
902 b_fff(PCI_FUNC(d->devfn)));
903 ResourceFields *reg;
904 int i, reg_idx = 0;
906 /* config space region */
907 reg = &rp->reg[reg_idx++];
908 reg->phys_hi = cpu_to_be32(dev_id);
909 reg->phys_mid = 0;
910 reg->phys_lo = 0;
911 reg->size_hi = 0;
912 reg->size_lo = 0;
914 for (i = 0; i < PCI_NUM_REGIONS; i++) {
915 if (!d->io_regions[i].size) {
916 continue;
919 reg = &rp->reg[reg_idx++];
921 reg->phys_hi = cpu_to_be32(dev_id | b_rrrrrrrr(pci_bar(d, i)));
922 if (d->io_regions[i].type & PCI_BASE_ADDRESS_SPACE_IO) {
923 reg->phys_hi |= cpu_to_be32(b_ss(1));
924 } else if (d->io_regions[i].type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
925 reg->phys_hi |= cpu_to_be32(b_ss(3));
926 } else {
927 reg->phys_hi |= cpu_to_be32(b_ss(2));
929 reg->phys_mid = 0;
930 reg->phys_lo = 0;
931 reg->size_hi = cpu_to_be32(d->io_regions[i].size >> 32);
932 reg->size_lo = cpu_to_be32(d->io_regions[i].size);
935 rp->reg_len = reg_idx * sizeof(ResourceFields);
938 typedef struct PCIClass PCIClass;
939 typedef struct PCISubClass PCISubClass;
940 typedef struct PCIIFace PCIIFace;
942 struct PCIIFace {
943 int iface;
944 const char *name;
947 struct PCISubClass {
948 int subclass;
949 const char *name;
950 const PCIIFace *iface;
953 struct PCIClass {
954 const char *name;
955 const PCISubClass *subc;
958 static const PCISubClass undef_subclass[] = {
959 { PCI_CLASS_NOT_DEFINED_VGA, "display", NULL },
960 { 0xFF, NULL, NULL },
963 static const PCISubClass mass_subclass[] = {
964 { PCI_CLASS_STORAGE_SCSI, "scsi", NULL },
965 { PCI_CLASS_STORAGE_IDE, "ide", NULL },
966 { PCI_CLASS_STORAGE_FLOPPY, "fdc", NULL },
967 { PCI_CLASS_STORAGE_IPI, "ipi", NULL },
968 { PCI_CLASS_STORAGE_RAID, "raid", NULL },
969 { PCI_CLASS_STORAGE_ATA, "ata", NULL },
970 { PCI_CLASS_STORAGE_SATA, "sata", NULL },
971 { PCI_CLASS_STORAGE_SAS, "sas", NULL },
972 { 0xFF, NULL, NULL },
975 static const PCISubClass net_subclass[] = {
976 { PCI_CLASS_NETWORK_ETHERNET, "ethernet", NULL },
977 { PCI_CLASS_NETWORK_TOKEN_RING, "token-ring", NULL },
978 { PCI_CLASS_NETWORK_FDDI, "fddi", NULL },
979 { PCI_CLASS_NETWORK_ATM, "atm", NULL },
980 { PCI_CLASS_NETWORK_ISDN, "isdn", NULL },
981 { PCI_CLASS_NETWORK_WORLDFIP, "worldfip", NULL },
982 { PCI_CLASS_NETWORK_PICMG214, "picmg", NULL },
983 { 0xFF, NULL, NULL },
986 static const PCISubClass displ_subclass[] = {
987 { PCI_CLASS_DISPLAY_VGA, "vga", NULL },
988 { PCI_CLASS_DISPLAY_XGA, "xga", NULL },
989 { PCI_CLASS_DISPLAY_3D, "3d-controller", NULL },
990 { 0xFF, NULL, NULL },
993 static const PCISubClass media_subclass[] = {
994 { PCI_CLASS_MULTIMEDIA_VIDEO, "video", NULL },
995 { PCI_CLASS_MULTIMEDIA_AUDIO, "sound", NULL },
996 { PCI_CLASS_MULTIMEDIA_PHONE, "telephony", NULL },
997 { 0xFF, NULL, NULL },
1000 static const PCISubClass mem_subclass[] = {
1001 { PCI_CLASS_MEMORY_RAM, "memory", NULL },
1002 { PCI_CLASS_MEMORY_FLASH, "flash", NULL },
1003 { 0xFF, NULL, NULL },
1006 static const PCISubClass bridg_subclass[] = {
1007 { PCI_CLASS_BRIDGE_HOST, "host", NULL },
1008 { PCI_CLASS_BRIDGE_ISA, "isa", NULL },
1009 { PCI_CLASS_BRIDGE_EISA, "eisa", NULL },
1010 { PCI_CLASS_BRIDGE_MC, "mca", NULL },
1011 { PCI_CLASS_BRIDGE_PCI, "pci", NULL },
1012 { PCI_CLASS_BRIDGE_PCMCIA, "pcmcia", NULL },
1013 { PCI_CLASS_BRIDGE_NUBUS, "nubus", NULL },
1014 { PCI_CLASS_BRIDGE_CARDBUS, "cardbus", NULL },
1015 { PCI_CLASS_BRIDGE_RACEWAY, "raceway", NULL },
1016 { PCI_CLASS_BRIDGE_PCI_SEMITP, "semi-transparent-pci", NULL },
1017 { PCI_CLASS_BRIDGE_IB_PCI, "infiniband", NULL },
1018 { 0xFF, NULL, NULL },
1021 static const PCISubClass comm_subclass[] = {
1022 { PCI_CLASS_COMMUNICATION_SERIAL, "serial", NULL },
1023 { PCI_CLASS_COMMUNICATION_PARALLEL, "parallel", NULL },
1024 { PCI_CLASS_COMMUNICATION_MULTISERIAL, "multiport-serial", NULL },
1025 { PCI_CLASS_COMMUNICATION_MODEM, "modem", NULL },
1026 { PCI_CLASS_COMMUNICATION_GPIB, "gpib", NULL },
1027 { PCI_CLASS_COMMUNICATION_SC, "smart-card", NULL },
1028 { 0xFF, NULL, NULL, },
1031 static const PCIIFace pic_iface[] = {
1032 { PCI_CLASS_SYSTEM_PIC_IOAPIC, "io-apic" },
1033 { PCI_CLASS_SYSTEM_PIC_IOXAPIC, "io-xapic" },
1034 { 0xFF, NULL },
1037 static const PCISubClass sys_subclass[] = {
1038 { PCI_CLASS_SYSTEM_PIC, "interrupt-controller", pic_iface },
1039 { PCI_CLASS_SYSTEM_DMA, "dma-controller", NULL },
1040 { PCI_CLASS_SYSTEM_TIMER, "timer", NULL },
1041 { PCI_CLASS_SYSTEM_RTC, "rtc", NULL },
1042 { PCI_CLASS_SYSTEM_PCI_HOTPLUG, "hot-plug-controller", NULL },
1043 { PCI_CLASS_SYSTEM_SDHCI, "sd-host-controller", NULL },
1044 { 0xFF, NULL, NULL },
1047 static const PCISubClass inp_subclass[] = {
1048 { PCI_CLASS_INPUT_KEYBOARD, "keyboard", NULL },
1049 { PCI_CLASS_INPUT_PEN, "pen", NULL },
1050 { PCI_CLASS_INPUT_MOUSE, "mouse", NULL },
1051 { PCI_CLASS_INPUT_SCANNER, "scanner", NULL },
1052 { PCI_CLASS_INPUT_GAMEPORT, "gameport", NULL },
1053 { 0xFF, NULL, NULL },
1056 static const PCISubClass dock_subclass[] = {
1057 { PCI_CLASS_DOCKING_GENERIC, "dock", NULL },
1058 { 0xFF, NULL, NULL },
1061 static const PCISubClass cpu_subclass[] = {
1062 { PCI_CLASS_PROCESSOR_PENTIUM, "pentium", NULL },
1063 { PCI_CLASS_PROCESSOR_POWERPC, "powerpc", NULL },
1064 { PCI_CLASS_PROCESSOR_MIPS, "mips", NULL },
1065 { PCI_CLASS_PROCESSOR_CO, "co-processor", NULL },
1066 { 0xFF, NULL, NULL },
1069 static const PCIIFace usb_iface[] = {
1070 { PCI_CLASS_SERIAL_USB_UHCI, "usb-uhci" },
1071 { PCI_CLASS_SERIAL_USB_OHCI, "usb-ohci", },
1072 { PCI_CLASS_SERIAL_USB_EHCI, "usb-ehci" },
1073 { PCI_CLASS_SERIAL_USB_XHCI, "usb-xhci" },
1074 { PCI_CLASS_SERIAL_USB_UNKNOWN, "usb-unknown" },
1075 { PCI_CLASS_SERIAL_USB_DEVICE, "usb-device" },
1076 { 0xFF, NULL },
1079 static const PCISubClass ser_subclass[] = {
1080 { PCI_CLASS_SERIAL_FIREWIRE, "firewire", NULL },
1081 { PCI_CLASS_SERIAL_ACCESS, "access-bus", NULL },
1082 { PCI_CLASS_SERIAL_SSA, "ssa", NULL },
1083 { PCI_CLASS_SERIAL_USB, "usb", usb_iface },
1084 { PCI_CLASS_SERIAL_FIBER, "fibre-channel", NULL },
1085 { PCI_CLASS_SERIAL_SMBUS, "smb", NULL },
1086 { PCI_CLASS_SERIAL_IB, "infiniband", NULL },
1087 { PCI_CLASS_SERIAL_IPMI, "ipmi", NULL },
1088 { PCI_CLASS_SERIAL_SERCOS, "sercos", NULL },
1089 { PCI_CLASS_SERIAL_CANBUS, "canbus", NULL },
1090 { 0xFF, NULL, NULL },
1093 static const PCISubClass wrl_subclass[] = {
1094 { PCI_CLASS_WIRELESS_IRDA, "irda", NULL },
1095 { PCI_CLASS_WIRELESS_CIR, "consumer-ir", NULL },
1096 { PCI_CLASS_WIRELESS_RF_CONTROLLER, "rf-controller", NULL },
1097 { PCI_CLASS_WIRELESS_BLUETOOTH, "bluetooth", NULL },
1098 { PCI_CLASS_WIRELESS_BROADBAND, "broadband", NULL },
1099 { 0xFF, NULL, NULL },
1102 static const PCISubClass sat_subclass[] = {
1103 { PCI_CLASS_SATELLITE_TV, "satellite-tv", NULL },
1104 { PCI_CLASS_SATELLITE_AUDIO, "satellite-audio", NULL },
1105 { PCI_CLASS_SATELLITE_VOICE, "satellite-voice", NULL },
1106 { PCI_CLASS_SATELLITE_DATA, "satellite-data", NULL },
1107 { 0xFF, NULL, NULL },
1110 static const PCISubClass crypt_subclass[] = {
1111 { PCI_CLASS_CRYPT_NETWORK, "network-encryption", NULL },
1112 { PCI_CLASS_CRYPT_ENTERTAINMENT,
1113 "entertainment-encryption", NULL },
1114 { 0xFF, NULL, NULL },
1117 static const PCISubClass spc_subclass[] = {
1118 { PCI_CLASS_SP_DPIO, "dpio", NULL },
1119 { PCI_CLASS_SP_PERF, "counter", NULL },
1120 { PCI_CLASS_SP_SYNCH, "measurement", NULL },
1121 { PCI_CLASS_SP_MANAGEMENT, "management-card", NULL },
1122 { 0xFF, NULL, NULL },
1125 static const PCIClass pci_classes[] = {
1126 { "legacy-device", undef_subclass },
1127 { "mass-storage", mass_subclass },
1128 { "network", net_subclass },
1129 { "display", displ_subclass, },
1130 { "multimedia-device", media_subclass },
1131 { "memory-controller", mem_subclass },
1132 { "unknown-bridge", bridg_subclass },
1133 { "communication-controller", comm_subclass},
1134 { "system-peripheral", sys_subclass },
1135 { "input-controller", inp_subclass },
1136 { "docking-station", dock_subclass },
1137 { "cpu", cpu_subclass },
1138 { "serial-bus", ser_subclass },
1139 { "wireless-controller", wrl_subclass },
1140 { "intelligent-io", NULL },
1141 { "satellite-device", sat_subclass },
1142 { "encryption", crypt_subclass },
1143 { "data-processing-controller", spc_subclass },
1146 static const char *dt_name_from_class(uint8_t class, uint8_t subclass,
1147 uint8_t iface)
1149 const PCIClass *pclass;
1150 const PCISubClass *psubclass;
1151 const PCIIFace *piface;
1152 const char *name;
1154 if (class >= ARRAY_SIZE(pci_classes)) {
1155 return "pci";
1158 pclass = pci_classes + class;
1159 name = pclass->name;
1161 if (pclass->subc == NULL) {
1162 return name;
1165 psubclass = pclass->subc;
1166 while ((psubclass->subclass & 0xff) != 0xff) {
1167 if ((psubclass->subclass & 0xff) == subclass) {
1168 name = psubclass->name;
1169 break;
1171 psubclass++;
1174 piface = psubclass->iface;
1175 if (piface == NULL) {
1176 return name;
1178 while ((piface->iface & 0xff) != 0xff) {
1179 if ((piface->iface & 0xff) == iface) {
1180 name = piface->name;
1181 break;
1183 piface++;
1186 return name;
1190 * DRC helper functions
1193 static uint32_t drc_id_from_devfn(SpaprPhbState *phb,
1194 uint8_t chassis, int32_t devfn)
1196 return (phb->index << 16) | (chassis << 8) | devfn;
1199 static SpaprDrc *drc_from_devfn(SpaprPhbState *phb,
1200 uint8_t chassis, int32_t devfn)
1202 return spapr_drc_by_id(TYPE_SPAPR_DRC_PCI,
1203 drc_id_from_devfn(phb, chassis, devfn));
1206 static uint8_t chassis_from_bus(PCIBus *bus)
1208 if (pci_bus_is_root(bus)) {
1209 return 0;
1210 } else {
1211 PCIDevice *bridge = pci_bridge_get_device(bus);
1213 return object_property_get_uint(OBJECT(bridge), "chassis_nr",
1214 &error_abort);
1218 static SpaprDrc *drc_from_dev(SpaprPhbState *phb, PCIDevice *dev)
1220 uint8_t chassis = chassis_from_bus(pci_get_bus(dev));
1222 return drc_from_devfn(phb, chassis, dev->devfn);
1225 static void add_drcs(SpaprPhbState *phb, PCIBus *bus)
1227 Object *owner;
1228 int i;
1229 uint8_t chassis;
1231 if (!phb->dr_enabled) {
1232 return;
1235 chassis = chassis_from_bus(bus);
1237 if (pci_bus_is_root(bus)) {
1238 owner = OBJECT(phb);
1239 } else {
1240 owner = OBJECT(pci_bridge_get_device(bus));
1243 for (i = 0; i < PCI_SLOT_MAX * PCI_FUNC_MAX; i++) {
1244 spapr_dr_connector_new(owner, TYPE_SPAPR_DRC_PCI,
1245 drc_id_from_devfn(phb, chassis, i));
1249 static void remove_drcs(SpaprPhbState *phb, PCIBus *bus)
1251 int i;
1252 uint8_t chassis;
1254 if (!phb->dr_enabled) {
1255 return;
1258 chassis = chassis_from_bus(bus);
1260 for (i = PCI_SLOT_MAX * PCI_FUNC_MAX - 1; i >= 0; i--) {
1261 SpaprDrc *drc = drc_from_devfn(phb, chassis, i);
1263 if (drc) {
1264 object_unparent(OBJECT(drc));
1269 typedef struct PciWalkFdt {
1270 void *fdt;
1271 int offset;
1272 SpaprPhbState *sphb;
1273 int err;
1274 } PciWalkFdt;
1276 static int spapr_dt_pci_device(SpaprPhbState *sphb, PCIDevice *dev,
1277 void *fdt, int parent_offset);
1279 static void spapr_dt_pci_device_cb(PCIBus *bus, PCIDevice *pdev,
1280 void *opaque)
1282 PciWalkFdt *p = opaque;
1283 int err;
1285 if (p->err) {
1286 /* Something's already broken, don't keep going */
1287 return;
1290 err = spapr_dt_pci_device(p->sphb, pdev, p->fdt, p->offset);
1291 if (err < 0) {
1292 p->err = err;
1296 /* Augment PCI device node with bridge specific information */
1297 static int spapr_dt_pci_bus(SpaprPhbState *sphb, PCIBus *bus,
1298 void *fdt, int offset)
1300 Object *owner;
1301 PciWalkFdt cbinfo = {
1302 .fdt = fdt,
1303 .offset = offset,
1304 .sphb = sphb,
1305 .err = 0,
1307 int ret;
1309 _FDT(fdt_setprop_cell(fdt, offset, "#address-cells",
1310 RESOURCE_CELLS_ADDRESS));
1311 _FDT(fdt_setprop_cell(fdt, offset, "#size-cells",
1312 RESOURCE_CELLS_SIZE));
1314 assert(bus);
1315 pci_for_each_device_reverse(bus, pci_bus_num(bus),
1316 spapr_dt_pci_device_cb, &cbinfo);
1317 if (cbinfo.err) {
1318 return cbinfo.err;
1321 if (pci_bus_is_root(bus)) {
1322 owner = OBJECT(sphb);
1323 } else {
1324 owner = OBJECT(pci_bridge_get_device(bus));
1327 ret = spapr_dt_drc(fdt, offset, owner,
1328 SPAPR_DR_CONNECTOR_TYPE_PCI);
1329 if (ret) {
1330 return ret;
1333 return offset;
1336 /* create OF node for pci device and required OF DT properties */
1337 static int spapr_dt_pci_device(SpaprPhbState *sphb, PCIDevice *dev,
1338 void *fdt, int parent_offset)
1340 int offset;
1341 const gchar *basename;
1342 gchar *nodename;
1343 int slot = PCI_SLOT(dev->devfn);
1344 int func = PCI_FUNC(dev->devfn);
1345 PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(dev);
1346 ResourceProps rp;
1347 SpaprDrc *drc = drc_from_dev(sphb, dev);
1348 uint32_t vendor_id = pci_default_read_config(dev, PCI_VENDOR_ID, 2);
1349 uint32_t device_id = pci_default_read_config(dev, PCI_DEVICE_ID, 2);
1350 uint32_t revision_id = pci_default_read_config(dev, PCI_REVISION_ID, 1);
1351 uint32_t ccode = pci_default_read_config(dev, PCI_CLASS_PROG, 3);
1352 uint32_t irq_pin = pci_default_read_config(dev, PCI_INTERRUPT_PIN, 1);
1353 uint32_t subsystem_id = pci_default_read_config(dev, PCI_SUBSYSTEM_ID, 2);
1354 uint32_t subsystem_vendor_id =
1355 pci_default_read_config(dev, PCI_SUBSYSTEM_VENDOR_ID, 2);
1356 uint32_t cache_line_size =
1357 pci_default_read_config(dev, PCI_CACHE_LINE_SIZE, 1);
1358 uint32_t pci_status = pci_default_read_config(dev, PCI_STATUS, 2);
1359 gchar *loc_code;
1361 basename = dt_name_from_class((ccode >> 16) & 0xff, (ccode >> 8) & 0xff,
1362 ccode & 0xff);
1364 if (func != 0) {
1365 nodename = g_strdup_printf("%s@%x,%x", basename, slot, func);
1366 } else {
1367 nodename = g_strdup_printf("%s@%x", basename, slot);
1370 _FDT(offset = fdt_add_subnode(fdt, parent_offset, nodename));
1372 g_free(nodename);
1374 /* in accordance with PAPR+ v2.7 13.6.3, Table 181 */
1375 _FDT(fdt_setprop_cell(fdt, offset, "vendor-id", vendor_id));
1376 _FDT(fdt_setprop_cell(fdt, offset, "device-id", device_id));
1377 _FDT(fdt_setprop_cell(fdt, offset, "revision-id", revision_id));
1379 _FDT(fdt_setprop_cell(fdt, offset, "class-code", ccode));
1380 if (irq_pin) {
1381 _FDT(fdt_setprop_cell(fdt, offset, "interrupts", irq_pin));
1384 if (subsystem_id) {
1385 _FDT(fdt_setprop_cell(fdt, offset, "subsystem-id", subsystem_id));
1388 if (subsystem_vendor_id) {
1389 _FDT(fdt_setprop_cell(fdt, offset, "subsystem-vendor-id",
1390 subsystem_vendor_id));
1393 _FDT(fdt_setprop_cell(fdt, offset, "cache-line-size", cache_line_size));
1396 /* the following fdt cells are masked off the pci status register */
1397 _FDT(fdt_setprop_cell(fdt, offset, "devsel-speed",
1398 PCI_STATUS_DEVSEL_MASK & pci_status));
1400 if (pci_status & PCI_STATUS_FAST_BACK) {
1401 _FDT(fdt_setprop(fdt, offset, "fast-back-to-back", NULL, 0));
1403 if (pci_status & PCI_STATUS_66MHZ) {
1404 _FDT(fdt_setprop(fdt, offset, "66mhz-capable", NULL, 0));
1406 if (pci_status & PCI_STATUS_UDF) {
1407 _FDT(fdt_setprop(fdt, offset, "udf-supported", NULL, 0));
1410 loc_code = spapr_phb_get_loc_code(sphb, dev);
1411 _FDT(fdt_setprop_string(fdt, offset, "ibm,loc-code", loc_code));
1412 g_free(loc_code);
1414 if (drc) {
1415 _FDT(fdt_setprop_cell(fdt, offset, "ibm,my-drc-index",
1416 spapr_drc_index(drc)));
1419 if (msi_present(dev)) {
1420 uint32_t max_msi = msi_nr_vectors_allocated(dev);
1421 if (max_msi) {
1422 _FDT(fdt_setprop_cell(fdt, offset, "ibm,req#msi", max_msi));
1425 if (msix_present(dev)) {
1426 uint32_t max_msix = dev->msix_entries_nr;
1427 if (max_msix) {
1428 _FDT(fdt_setprop_cell(fdt, offset, "ibm,req#msi-x", max_msix));
1432 populate_resource_props(dev, &rp);
1433 _FDT(fdt_setprop(fdt, offset, "reg", (uint8_t *)rp.reg, rp.reg_len));
1435 if (sphb->pcie_ecs && pci_is_express(dev)) {
1436 _FDT(fdt_setprop_cell(fdt, offset, "ibm,pci-config-space-type", 0x1));
1439 spapr_phb_nvgpu_populate_pcidev_dt(dev, fdt, offset, sphb);
1441 if (!pc->is_bridge) {
1442 /* Properties only for non-bridges */
1443 uint32_t min_grant = pci_default_read_config(dev, PCI_MIN_GNT, 1);
1444 uint32_t max_latency = pci_default_read_config(dev, PCI_MAX_LAT, 1);
1445 _FDT(fdt_setprop_cell(fdt, offset, "min-grant", min_grant));
1446 _FDT(fdt_setprop_cell(fdt, offset, "max-latency", max_latency));
1447 return offset;
1448 } else {
1449 PCIBus *sec_bus = pci_bridge_get_sec_bus(PCI_BRIDGE(dev));
1451 return spapr_dt_pci_bus(sphb, sec_bus, fdt, offset);
1455 /* Callback to be called during DRC release. */
1456 void spapr_phb_remove_pci_device_cb(DeviceState *dev)
1458 HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
1460 hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
1461 object_unparent(OBJECT(dev));
1464 int spapr_pci_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
1465 void *fdt, int *fdt_start_offset, Error **errp)
1467 HotplugHandler *plug_handler = qdev_get_hotplug_handler(drc->dev);
1468 SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(plug_handler);
1469 PCIDevice *pdev = PCI_DEVICE(drc->dev);
1471 *fdt_start_offset = spapr_dt_pci_device(sphb, pdev, fdt, 0);
1472 return 0;
1475 static void spapr_pci_bridge_plug(SpaprPhbState *phb,
1476 PCIBridge *bridge)
1478 PCIBus *bus = pci_bridge_get_sec_bus(bridge);
1480 add_drcs(phb, bus);
1483 static void spapr_pci_plug(HotplugHandler *plug_handler,
1484 DeviceState *plugged_dev, Error **errp)
1486 SpaprPhbState *phb = SPAPR_PCI_HOST_BRIDGE(DEVICE(plug_handler));
1487 PCIDevice *pdev = PCI_DEVICE(plugged_dev);
1488 PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(plugged_dev);
1489 SpaprDrc *drc = drc_from_dev(phb, pdev);
1490 Error *local_err = NULL;
1491 PCIBus *bus = PCI_BUS(qdev_get_parent_bus(DEVICE(pdev)));
1492 uint32_t slotnr = PCI_SLOT(pdev->devfn);
1494 /* if DR is disabled we don't need to do anything in the case of
1495 * hotplug or coldplug callbacks
1497 if (!phb->dr_enabled) {
1498 /* if this is a hotplug operation initiated by the user
1499 * we need to let them know it's not enabled
1501 if (plugged_dev->hotplugged) {
1502 error_setg(&local_err, QERR_BUS_NO_HOTPLUG,
1503 object_get_typename(OBJECT(phb)));
1505 goto out;
1508 g_assert(drc);
1510 if (pc->is_bridge) {
1511 spapr_pci_bridge_plug(phb, PCI_BRIDGE(plugged_dev));
1514 /* Following the QEMU convention used for PCIe multifunction
1515 * hotplug, we do not allow functions to be hotplugged to a
1516 * slot that already has function 0 present
1518 if (plugged_dev->hotplugged && bus->devices[PCI_DEVFN(slotnr, 0)] &&
1519 PCI_FUNC(pdev->devfn) != 0) {
1520 error_setg(&local_err, "PCI: slot %d function 0 already ocuppied by %s,"
1521 " additional functions can no longer be exposed to guest.",
1522 slotnr, bus->devices[PCI_DEVFN(slotnr, 0)]->name);
1523 goto out;
1526 spapr_drc_attach(drc, DEVICE(pdev), &local_err);
1527 if (local_err) {
1528 goto out;
1531 /* If this is function 0, signal hotplug for all the device functions.
1532 * Otherwise defer sending the hotplug event.
1534 if (!spapr_drc_hotplugged(plugged_dev)) {
1535 spapr_drc_reset(drc);
1536 } else if (PCI_FUNC(pdev->devfn) == 0) {
1537 int i;
1538 uint8_t chassis = chassis_from_bus(pci_get_bus(pdev));
1540 for (i = 0; i < 8; i++) {
1541 SpaprDrc *func_drc;
1542 SpaprDrcClass *func_drck;
1543 SpaprDREntitySense state;
1545 func_drc = drc_from_devfn(phb, chassis, PCI_DEVFN(slotnr, i));
1546 func_drck = SPAPR_DR_CONNECTOR_GET_CLASS(func_drc);
1547 state = func_drck->dr_entity_sense(func_drc);
1549 if (state == SPAPR_DR_ENTITY_SENSE_PRESENT) {
1550 spapr_hotplug_req_add_by_index(func_drc);
1555 out:
1556 error_propagate(errp, local_err);
1559 static void spapr_pci_bridge_unplug(SpaprPhbState *phb,
1560 PCIBridge *bridge)
1562 PCIBus *bus = pci_bridge_get_sec_bus(bridge);
1564 remove_drcs(phb, bus);
1567 static void spapr_pci_unplug(HotplugHandler *plug_handler,
1568 DeviceState *plugged_dev, Error **errp)
1570 PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(plugged_dev);
1571 SpaprPhbState *phb = SPAPR_PCI_HOST_BRIDGE(DEVICE(plug_handler));
1573 /* some version guests do not wait for completion of a device
1574 * cleanup (generally done asynchronously by the kernel) before
1575 * signaling to QEMU that the device is safe, but instead sleep
1576 * for some 'safe' period of time. unfortunately on a busy host
1577 * this sleep isn't guaranteed to be long enough, resulting in
1578 * bad things like IRQ lines being left asserted during final
1579 * device removal. to deal with this we call reset just prior
1580 * to finalizing the device, which will put the device back into
1581 * an 'idle' state, as the device cleanup code expects.
1583 pci_device_reset(PCI_DEVICE(plugged_dev));
1585 if (pc->is_bridge) {
1586 spapr_pci_bridge_unplug(phb, PCI_BRIDGE(plugged_dev));
1587 return;
1590 qdev_unrealize(plugged_dev);
1593 static void spapr_pci_unplug_request(HotplugHandler *plug_handler,
1594 DeviceState *plugged_dev, Error **errp)
1596 SpaprPhbState *phb = SPAPR_PCI_HOST_BRIDGE(DEVICE(plug_handler));
1597 PCIDevice *pdev = PCI_DEVICE(plugged_dev);
1598 SpaprDrc *drc = drc_from_dev(phb, pdev);
1600 if (!phb->dr_enabled) {
1601 error_setg(errp, QERR_BUS_NO_HOTPLUG,
1602 object_get_typename(OBJECT(phb)));
1603 return;
1606 g_assert(drc);
1607 g_assert(drc->dev == plugged_dev);
1609 if (!spapr_drc_unplug_requested(drc)) {
1610 PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(plugged_dev);
1611 uint32_t slotnr = PCI_SLOT(pdev->devfn);
1612 SpaprDrc *func_drc;
1613 SpaprDrcClass *func_drck;
1614 SpaprDREntitySense state;
1615 int i;
1616 uint8_t chassis = chassis_from_bus(pci_get_bus(pdev));
1618 if (pc->is_bridge) {
1619 error_setg(errp, "PCI: Hot unplug of PCI bridges not supported");
1620 return;
1622 if (object_property_get_uint(OBJECT(pdev), "nvlink2-tgt", NULL)) {
1623 error_setg(errp, "PCI: Cannot unplug NVLink2 devices");
1624 return;
1627 /* ensure any other present functions are pending unplug */
1628 if (PCI_FUNC(pdev->devfn) == 0) {
1629 for (i = 1; i < 8; i++) {
1630 func_drc = drc_from_devfn(phb, chassis, PCI_DEVFN(slotnr, i));
1631 func_drck = SPAPR_DR_CONNECTOR_GET_CLASS(func_drc);
1632 state = func_drck->dr_entity_sense(func_drc);
1633 if (state == SPAPR_DR_ENTITY_SENSE_PRESENT
1634 && !spapr_drc_unplug_requested(func_drc)) {
1636 * Attempting to remove function 0 of a multifunction
1637 * device will will cascade into removing all child
1638 * functions, even if their unplug weren't requested
1639 * beforehand.
1641 spapr_drc_detach(func_drc);
1646 spapr_drc_detach(drc);
1648 /* if this isn't func 0, defer unplug event. otherwise signal removal
1649 * for all present functions
1651 if (PCI_FUNC(pdev->devfn) == 0) {
1652 for (i = 7; i >= 0; i--) {
1653 func_drc = drc_from_devfn(phb, chassis, PCI_DEVFN(slotnr, i));
1654 func_drck = SPAPR_DR_CONNECTOR_GET_CLASS(func_drc);
1655 state = func_drck->dr_entity_sense(func_drc);
1656 if (state == SPAPR_DR_ENTITY_SENSE_PRESENT) {
1657 spapr_hotplug_req_remove_by_index(func_drc);
1664 static void spapr_phb_finalizefn(Object *obj)
1666 SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(obj);
1668 g_free(sphb->dtbusname);
1669 sphb->dtbusname = NULL;
1672 static void spapr_phb_unrealize(DeviceState *dev)
1674 SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
1675 SysBusDevice *s = SYS_BUS_DEVICE(dev);
1676 PCIHostState *phb = PCI_HOST_BRIDGE(s);
1677 SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(phb);
1678 SpaprTceTable *tcet;
1679 int i;
1680 const unsigned windows_supported = spapr_phb_windows_supported(sphb);
1682 spapr_phb_nvgpu_free(sphb);
1684 if (sphb->msi) {
1685 g_hash_table_unref(sphb->msi);
1686 sphb->msi = NULL;
1690 * Remove IO/MMIO subregions and aliases, rest should get cleaned
1691 * via PHB's unrealize->object_finalize
1693 for (i = windows_supported - 1; i >= 0; i--) {
1694 tcet = spapr_tce_find_by_liobn(sphb->dma_liobn[i]);
1695 if (tcet) {
1696 memory_region_del_subregion(&sphb->iommu_root,
1697 spapr_tce_get_iommu(tcet));
1701 remove_drcs(sphb, phb->bus);
1703 for (i = PCI_NUM_PINS - 1; i >= 0; i--) {
1704 if (sphb->lsi_table[i].irq) {
1705 spapr_irq_free(spapr, sphb->lsi_table[i].irq, 1);
1706 sphb->lsi_table[i].irq = 0;
1710 QLIST_REMOVE(sphb, list);
1712 memory_region_del_subregion(&sphb->iommu_root, &sphb->msiwindow);
1715 * An attached PCI device may have memory listeners, eg. VFIO PCI. We have
1716 * unmapped all sections. Remove the listeners now, before destroying the
1717 * address space.
1719 address_space_remove_listeners(&sphb->iommu_as);
1720 address_space_destroy(&sphb->iommu_as);
1722 qbus_set_hotplug_handler(BUS(phb->bus), NULL);
1723 pci_unregister_root_bus(phb->bus);
1725 memory_region_del_subregion(get_system_memory(), &sphb->iowindow);
1726 if (sphb->mem64_win_pciaddr != (hwaddr)-1) {
1727 memory_region_del_subregion(get_system_memory(), &sphb->mem64window);
1729 memory_region_del_subregion(get_system_memory(), &sphb->mem32window);
1732 static void spapr_phb_destroy_msi(gpointer opaque)
1734 SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
1735 SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
1736 SpaprPciMsi *msi = opaque;
1738 if (!smc->legacy_irq_allocation) {
1739 spapr_irq_msi_free(spapr, msi->first_irq, msi->num);
1741 spapr_irq_free(spapr, msi->first_irq, msi->num);
1742 g_free(msi);
1745 static void spapr_phb_realize(DeviceState *dev, Error **errp)
1747 /* We don't use SPAPR_MACHINE() in order to exit gracefully if the user
1748 * tries to add a sPAPR PHB to a non-pseries machine.
1750 SpaprMachineState *spapr =
1751 (SpaprMachineState *) object_dynamic_cast(qdev_get_machine(),
1752 TYPE_SPAPR_MACHINE);
1753 SpaprMachineClass *smc = spapr ? SPAPR_MACHINE_GET_CLASS(spapr) : NULL;
1754 SysBusDevice *s = SYS_BUS_DEVICE(dev);
1755 SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(s);
1756 PCIHostState *phb = PCI_HOST_BRIDGE(s);
1757 MachineState *ms = MACHINE(spapr);
1758 char *namebuf;
1759 int i;
1760 PCIBus *bus;
1761 uint64_t msi_window_size = 4096;
1762 SpaprTceTable *tcet;
1763 const unsigned windows_supported = spapr_phb_windows_supported(sphb);
1764 Error *local_err = NULL;
1766 if (!spapr) {
1767 error_setg(errp, TYPE_SPAPR_PCI_HOST_BRIDGE " needs a pseries machine");
1768 return;
1771 assert(sphb->index != (uint32_t)-1); /* checked in spapr_phb_pre_plug() */
1773 if (sphb->mem64_win_size != 0) {
1774 if (sphb->mem_win_size > SPAPR_PCI_MEM32_WIN_SIZE) {
1775 error_setg(errp, "32-bit memory window of size 0x%"HWADDR_PRIx
1776 " (max 2 GiB)", sphb->mem_win_size);
1777 return;
1780 /* 64-bit window defaults to identity mapping */
1781 sphb->mem64_win_pciaddr = sphb->mem64_win_addr;
1782 } else if (sphb->mem_win_size > SPAPR_PCI_MEM32_WIN_SIZE) {
1784 * For compatibility with old configuration, if no 64-bit MMIO
1785 * window is specified, but the ordinary (32-bit) memory
1786 * window is specified as > 2GiB, we treat it as a 2GiB 32-bit
1787 * window, with a 64-bit MMIO window following on immediately
1788 * afterwards
1790 sphb->mem64_win_size = sphb->mem_win_size - SPAPR_PCI_MEM32_WIN_SIZE;
1791 sphb->mem64_win_addr = sphb->mem_win_addr + SPAPR_PCI_MEM32_WIN_SIZE;
1792 sphb->mem64_win_pciaddr =
1793 SPAPR_PCI_MEM_WIN_BUS_OFFSET + SPAPR_PCI_MEM32_WIN_SIZE;
1794 sphb->mem_win_size = SPAPR_PCI_MEM32_WIN_SIZE;
1797 if (spapr_pci_find_phb(spapr, sphb->buid)) {
1798 SpaprPhbState *s;
1800 error_setg(errp, "PCI host bridges must have unique indexes");
1801 error_append_hint(errp, "The following indexes are already in use:");
1802 QLIST_FOREACH(s, &spapr->phbs, list) {
1803 error_append_hint(errp, " %d", s->index);
1805 error_append_hint(errp, "\nTry another value for the index property\n");
1806 return;
1809 if (sphb->numa_node != -1 &&
1810 (sphb->numa_node >= MAX_NODES ||
1811 !ms->numa_state->nodes[sphb->numa_node].present)) {
1812 error_setg(errp, "Invalid NUMA node ID for PCI host bridge");
1813 return;
1816 sphb->dtbusname = g_strdup_printf("pci@%" PRIx64, sphb->buid);
1818 /* Initialize memory regions */
1819 namebuf = g_strdup_printf("%s.mmio", sphb->dtbusname);
1820 memory_region_init(&sphb->memspace, OBJECT(sphb), namebuf, UINT64_MAX);
1821 g_free(namebuf);
1823 namebuf = g_strdup_printf("%s.mmio32-alias", sphb->dtbusname);
1824 memory_region_init_alias(&sphb->mem32window, OBJECT(sphb),
1825 namebuf, &sphb->memspace,
1826 SPAPR_PCI_MEM_WIN_BUS_OFFSET, sphb->mem_win_size);
1827 g_free(namebuf);
1828 memory_region_add_subregion(get_system_memory(), sphb->mem_win_addr,
1829 &sphb->mem32window);
1831 if (sphb->mem64_win_size != 0) {
1832 namebuf = g_strdup_printf("%s.mmio64-alias", sphb->dtbusname);
1833 memory_region_init_alias(&sphb->mem64window, OBJECT(sphb),
1834 namebuf, &sphb->memspace,
1835 sphb->mem64_win_pciaddr, sphb->mem64_win_size);
1836 g_free(namebuf);
1838 memory_region_add_subregion(get_system_memory(),
1839 sphb->mem64_win_addr,
1840 &sphb->mem64window);
1843 /* Initialize IO regions */
1844 namebuf = g_strdup_printf("%s.io", sphb->dtbusname);
1845 memory_region_init(&sphb->iospace, OBJECT(sphb),
1846 namebuf, SPAPR_PCI_IO_WIN_SIZE);
1847 g_free(namebuf);
1849 namebuf = g_strdup_printf("%s.io-alias", sphb->dtbusname);
1850 memory_region_init_alias(&sphb->iowindow, OBJECT(sphb), namebuf,
1851 &sphb->iospace, 0, SPAPR_PCI_IO_WIN_SIZE);
1852 g_free(namebuf);
1853 memory_region_add_subregion(get_system_memory(), sphb->io_win_addr,
1854 &sphb->iowindow);
1856 bus = pci_register_root_bus(dev, NULL,
1857 pci_spapr_set_irq, pci_swizzle_map_irq_fn, sphb,
1858 &sphb->memspace, &sphb->iospace,
1859 PCI_DEVFN(0, 0), PCI_NUM_PINS,
1860 TYPE_PCI_BUS);
1863 * Despite resembling a vanilla PCI bus in most ways, the PAPR
1864 * para-virtualized PCI bus *does* permit PCI-E extended config
1865 * space access
1867 if (sphb->pcie_ecs) {
1868 bus->flags |= PCI_BUS_EXTENDED_CONFIG_SPACE;
1870 phb->bus = bus;
1871 qbus_set_hotplug_handler(BUS(phb->bus), OBJECT(sphb));
1874 * Initialize PHB address space.
1875 * By default there will be at least one subregion for default
1876 * 32bit DMA window.
1877 * Later the guest might want to create another DMA window
1878 * which will become another memory subregion.
1880 namebuf = g_strdup_printf("%s.iommu-root", sphb->dtbusname);
1881 memory_region_init(&sphb->iommu_root, OBJECT(sphb),
1882 namebuf, UINT64_MAX);
1883 g_free(namebuf);
1884 address_space_init(&sphb->iommu_as, &sphb->iommu_root,
1885 sphb->dtbusname);
1888 * As MSI/MSIX interrupts trigger by writing at MSI/MSIX vectors,
1889 * we need to allocate some memory to catch those writes coming
1890 * from msi_notify()/msix_notify().
1891 * As MSIMessage:addr is going to be the same and MSIMessage:data
1892 * is going to be a VIRQ number, 4 bytes of the MSI MR will only
1893 * be used.
1895 * For KVM we want to ensure that this memory is a full page so that
1896 * our memory slot is of page size granularity.
1898 if (kvm_enabled()) {
1899 msi_window_size = qemu_real_host_page_size;
1902 memory_region_init_io(&sphb->msiwindow, OBJECT(sphb), &spapr_msi_ops, spapr,
1903 "msi", msi_window_size);
1904 memory_region_add_subregion(&sphb->iommu_root, SPAPR_PCI_MSI_WINDOW,
1905 &sphb->msiwindow);
1907 pci_setup_iommu(bus, spapr_pci_dma_iommu, sphb);
1909 pci_bus_set_route_irq_fn(bus, spapr_route_intx_pin_to_irq);
1911 QLIST_INSERT_HEAD(&spapr->phbs, sphb, list);
1913 /* Initialize the LSI table */
1914 for (i = 0; i < PCI_NUM_PINS; i++) {
1915 uint32_t irq = SPAPR_IRQ_PCI_LSI + sphb->index * PCI_NUM_PINS + i;
1917 if (smc->legacy_irq_allocation) {
1918 irq = spapr_irq_findone(spapr, &local_err);
1919 if (local_err) {
1920 error_propagate_prepend(errp, local_err,
1921 "can't allocate LSIs: ");
1923 * Older machines will never support PHB hotplug, ie, this is an
1924 * init only path and QEMU will terminate. No need to rollback.
1926 return;
1930 spapr_irq_claim(spapr, irq, true, &local_err);
1931 if (local_err) {
1932 error_propagate_prepend(errp, local_err, "can't allocate LSIs: ");
1933 goto unrealize;
1936 sphb->lsi_table[i].irq = irq;
1939 /* allocate connectors for child PCI devices */
1940 add_drcs(sphb, phb->bus);
1942 /* DMA setup */
1943 for (i = 0; i < windows_supported; ++i) {
1944 tcet = spapr_tce_new_table(DEVICE(sphb), sphb->dma_liobn[i]);
1945 if (!tcet) {
1946 error_setg(errp, "Creating window#%d failed for %s",
1947 i, sphb->dtbusname);
1948 goto unrealize;
1950 memory_region_add_subregion(&sphb->iommu_root, 0,
1951 spapr_tce_get_iommu(tcet));
1954 sphb->msi = g_hash_table_new_full(g_int_hash, g_int_equal, g_free,
1955 spapr_phb_destroy_msi);
1956 return;
1958 unrealize:
1959 spapr_phb_unrealize(dev);
1962 static int spapr_phb_children_reset(Object *child, void *opaque)
1964 DeviceState *dev = (DeviceState *) object_dynamic_cast(child, TYPE_DEVICE);
1966 if (dev) {
1967 device_legacy_reset(dev);
1970 return 0;
1973 void spapr_phb_dma_reset(SpaprPhbState *sphb)
1975 int i;
1976 SpaprTceTable *tcet;
1978 for (i = 0; i < SPAPR_PCI_DMA_MAX_WINDOWS; ++i) {
1979 tcet = spapr_tce_find_by_liobn(sphb->dma_liobn[i]);
1981 if (tcet && tcet->nb_table) {
1982 spapr_tce_table_disable(tcet);
1986 /* Register default 32bit DMA window */
1987 tcet = spapr_tce_find_by_liobn(sphb->dma_liobn[0]);
1988 spapr_tce_table_enable(tcet, SPAPR_TCE_PAGE_SHIFT, sphb->dma_win_addr,
1989 sphb->dma_win_size >> SPAPR_TCE_PAGE_SHIFT);
1992 static void spapr_phb_reset(DeviceState *qdev)
1994 SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(qdev);
1995 Error *err = NULL;
1997 spapr_phb_dma_reset(sphb);
1998 spapr_phb_nvgpu_free(sphb);
1999 spapr_phb_nvgpu_setup(sphb, &err);
2000 if (err) {
2001 error_report_err(err);
2004 /* Reset the IOMMU state */
2005 object_child_foreach(OBJECT(qdev), spapr_phb_children_reset, NULL);
2007 if (spapr_phb_eeh_available(SPAPR_PCI_HOST_BRIDGE(qdev))) {
2008 spapr_phb_vfio_reset(qdev);
2011 g_hash_table_remove_all(sphb->msi);
2014 static Property spapr_phb_properties[] = {
2015 DEFINE_PROP_UINT32("index", SpaprPhbState, index, -1),
2016 DEFINE_PROP_UINT64("mem_win_size", SpaprPhbState, mem_win_size,
2017 SPAPR_PCI_MEM32_WIN_SIZE),
2018 DEFINE_PROP_UINT64("mem64_win_size", SpaprPhbState, mem64_win_size,
2019 SPAPR_PCI_MEM64_WIN_SIZE),
2020 DEFINE_PROP_UINT64("io_win_size", SpaprPhbState, io_win_size,
2021 SPAPR_PCI_IO_WIN_SIZE),
2022 DEFINE_PROP_BOOL("dynamic-reconfiguration", SpaprPhbState, dr_enabled,
2023 true),
2024 /* Default DMA window is 0..1GB */
2025 DEFINE_PROP_UINT64("dma_win_addr", SpaprPhbState, dma_win_addr, 0),
2026 DEFINE_PROP_UINT64("dma_win_size", SpaprPhbState, dma_win_size, 0x40000000),
2027 DEFINE_PROP_UINT64("dma64_win_addr", SpaprPhbState, dma64_win_addr,
2028 0x800000000000000ULL),
2029 DEFINE_PROP_BOOL("ddw", SpaprPhbState, ddw_enabled, true),
2030 DEFINE_PROP_UINT64("pgsz", SpaprPhbState, page_size_mask,
2031 (1ULL << 12) | (1ULL << 16)
2032 | (1ULL << 21) | (1ULL << 24)),
2033 DEFINE_PROP_UINT32("numa_node", SpaprPhbState, numa_node, -1),
2034 DEFINE_PROP_BOOL("pre-2.8-migration", SpaprPhbState,
2035 pre_2_8_migration, false),
2036 DEFINE_PROP_BOOL("pcie-extended-configuration-space", SpaprPhbState,
2037 pcie_ecs, true),
2038 DEFINE_PROP_UINT64("gpa", SpaprPhbState, nv2_gpa_win_addr, 0),
2039 DEFINE_PROP_UINT64("atsd", SpaprPhbState, nv2_atsd_win_addr, 0),
2040 DEFINE_PROP_END_OF_LIST(),
2043 static const VMStateDescription vmstate_spapr_pci_lsi = {
2044 .name = "spapr_pci/lsi",
2045 .version_id = 1,
2046 .minimum_version_id = 1,
2047 .fields = (VMStateField[]) {
2048 VMSTATE_UINT32_EQUAL(irq, SpaprPciLsi, NULL),
2050 VMSTATE_END_OF_LIST()
2054 static const VMStateDescription vmstate_spapr_pci_msi = {
2055 .name = "spapr_pci/msi",
2056 .version_id = 1,
2057 .minimum_version_id = 1,
2058 .fields = (VMStateField []) {
2059 VMSTATE_UINT32(key, SpaprPciMsiMig),
2060 VMSTATE_UINT32(value.first_irq, SpaprPciMsiMig),
2061 VMSTATE_UINT32(value.num, SpaprPciMsiMig),
2062 VMSTATE_END_OF_LIST()
2066 static int spapr_pci_pre_save(void *opaque)
2068 SpaprPhbState *sphb = opaque;
2069 GHashTableIter iter;
2070 gpointer key, value;
2071 int i;
2073 if (sphb->pre_2_8_migration) {
2074 sphb->mig_liobn = sphb->dma_liobn[0];
2075 sphb->mig_mem_win_addr = sphb->mem_win_addr;
2076 sphb->mig_mem_win_size = sphb->mem_win_size;
2077 sphb->mig_io_win_addr = sphb->io_win_addr;
2078 sphb->mig_io_win_size = sphb->io_win_size;
2080 if ((sphb->mem64_win_size != 0)
2081 && (sphb->mem64_win_addr
2082 == (sphb->mem_win_addr + sphb->mem_win_size))) {
2083 sphb->mig_mem_win_size += sphb->mem64_win_size;
2087 g_free(sphb->msi_devs);
2088 sphb->msi_devs = NULL;
2089 sphb->msi_devs_num = g_hash_table_size(sphb->msi);
2090 if (!sphb->msi_devs_num) {
2091 return 0;
2093 sphb->msi_devs = g_new(SpaprPciMsiMig, sphb->msi_devs_num);
2095 g_hash_table_iter_init(&iter, sphb->msi);
2096 for (i = 0; g_hash_table_iter_next(&iter, &key, &value); ++i) {
2097 sphb->msi_devs[i].key = *(uint32_t *) key;
2098 sphb->msi_devs[i].value = *(SpaprPciMsi *) value;
2101 return 0;
2104 static int spapr_pci_post_load(void *opaque, int version_id)
2106 SpaprPhbState *sphb = opaque;
2107 gpointer key, value;
2108 int i;
2110 for (i = 0; i < sphb->msi_devs_num; ++i) {
2111 key = g_memdup(&sphb->msi_devs[i].key,
2112 sizeof(sphb->msi_devs[i].key));
2113 value = g_memdup(&sphb->msi_devs[i].value,
2114 sizeof(sphb->msi_devs[i].value));
2115 g_hash_table_insert(sphb->msi, key, value);
2117 g_free(sphb->msi_devs);
2118 sphb->msi_devs = NULL;
2119 sphb->msi_devs_num = 0;
2121 return 0;
2124 static bool pre_2_8_migration(void *opaque, int version_id)
2126 SpaprPhbState *sphb = opaque;
2128 return sphb->pre_2_8_migration;
2131 static const VMStateDescription vmstate_spapr_pci = {
2132 .name = "spapr_pci",
2133 .version_id = 2,
2134 .minimum_version_id = 2,
2135 .pre_save = spapr_pci_pre_save,
2136 .post_load = spapr_pci_post_load,
2137 .fields = (VMStateField[]) {
2138 VMSTATE_UINT64_EQUAL(buid, SpaprPhbState, NULL),
2139 VMSTATE_UINT32_TEST(mig_liobn, SpaprPhbState, pre_2_8_migration),
2140 VMSTATE_UINT64_TEST(mig_mem_win_addr, SpaprPhbState, pre_2_8_migration),
2141 VMSTATE_UINT64_TEST(mig_mem_win_size, SpaprPhbState, pre_2_8_migration),
2142 VMSTATE_UINT64_TEST(mig_io_win_addr, SpaprPhbState, pre_2_8_migration),
2143 VMSTATE_UINT64_TEST(mig_io_win_size, SpaprPhbState, pre_2_8_migration),
2144 VMSTATE_STRUCT_ARRAY(lsi_table, SpaprPhbState, PCI_NUM_PINS, 0,
2145 vmstate_spapr_pci_lsi, SpaprPciLsi),
2146 VMSTATE_INT32(msi_devs_num, SpaprPhbState),
2147 VMSTATE_STRUCT_VARRAY_ALLOC(msi_devs, SpaprPhbState, msi_devs_num, 0,
2148 vmstate_spapr_pci_msi, SpaprPciMsiMig),
2149 VMSTATE_END_OF_LIST()
2153 static const char *spapr_phb_root_bus_path(PCIHostState *host_bridge,
2154 PCIBus *rootbus)
2156 SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(host_bridge);
2158 return sphb->dtbusname;
2161 static void spapr_phb_class_init(ObjectClass *klass, void *data)
2163 PCIHostBridgeClass *hc = PCI_HOST_BRIDGE_CLASS(klass);
2164 DeviceClass *dc = DEVICE_CLASS(klass);
2165 HotplugHandlerClass *hp = HOTPLUG_HANDLER_CLASS(klass);
2167 hc->root_bus_path = spapr_phb_root_bus_path;
2168 dc->realize = spapr_phb_realize;
2169 dc->unrealize = spapr_phb_unrealize;
2170 device_class_set_props(dc, spapr_phb_properties);
2171 dc->reset = spapr_phb_reset;
2172 dc->vmsd = &vmstate_spapr_pci;
2173 /* Supported by TYPE_SPAPR_MACHINE */
2174 dc->user_creatable = true;
2175 set_bit(DEVICE_CATEGORY_BRIDGE, dc->categories);
2176 hp->plug = spapr_pci_plug;
2177 hp->unplug = spapr_pci_unplug;
2178 hp->unplug_request = spapr_pci_unplug_request;
2181 static const TypeInfo spapr_phb_info = {
2182 .name = TYPE_SPAPR_PCI_HOST_BRIDGE,
2183 .parent = TYPE_PCI_HOST_BRIDGE,
2184 .instance_size = sizeof(SpaprPhbState),
2185 .instance_finalize = spapr_phb_finalizefn,
2186 .class_init = spapr_phb_class_init,
2187 .interfaces = (InterfaceInfo[]) {
2188 { TYPE_HOTPLUG_HANDLER },
2193 static void spapr_phb_pci_enumerate_bridge(PCIBus *bus, PCIDevice *pdev,
2194 void *opaque)
2196 unsigned int *bus_no = opaque;
2197 PCIBus *sec_bus = NULL;
2199 if ((pci_default_read_config(pdev, PCI_HEADER_TYPE, 1) !=
2200 PCI_HEADER_TYPE_BRIDGE)) {
2201 return;
2204 (*bus_no)++;
2205 pci_default_write_config(pdev, PCI_PRIMARY_BUS, pci_dev_bus_num(pdev), 1);
2206 pci_default_write_config(pdev, PCI_SECONDARY_BUS, *bus_no, 1);
2207 pci_default_write_config(pdev, PCI_SUBORDINATE_BUS, *bus_no, 1);
2209 sec_bus = pci_bridge_get_sec_bus(PCI_BRIDGE(pdev));
2210 if (!sec_bus) {
2211 return;
2214 pci_for_each_device(sec_bus, pci_bus_num(sec_bus),
2215 spapr_phb_pci_enumerate_bridge, bus_no);
2216 pci_default_write_config(pdev, PCI_SUBORDINATE_BUS, *bus_no, 1);
2219 static void spapr_phb_pci_enumerate(SpaprPhbState *phb)
2221 PCIBus *bus = PCI_HOST_BRIDGE(phb)->bus;
2222 unsigned int bus_no = 0;
2224 pci_for_each_device(bus, pci_bus_num(bus),
2225 spapr_phb_pci_enumerate_bridge,
2226 &bus_no);
2230 int spapr_dt_phb(SpaprMachineState *spapr, SpaprPhbState *phb,
2231 uint32_t intc_phandle, void *fdt, int *node_offset)
2233 int bus_off, i, j, ret;
2234 uint32_t bus_range[] = { cpu_to_be32(0), cpu_to_be32(0xff) };
2235 struct {
2236 uint32_t hi;
2237 uint64_t child;
2238 uint64_t parent;
2239 uint64_t size;
2240 } QEMU_PACKED ranges[] = {
2242 cpu_to_be32(b_ss(1)), cpu_to_be64(0),
2243 cpu_to_be64(phb->io_win_addr),
2244 cpu_to_be64(memory_region_size(&phb->iospace)),
2247 cpu_to_be32(b_ss(2)), cpu_to_be64(SPAPR_PCI_MEM_WIN_BUS_OFFSET),
2248 cpu_to_be64(phb->mem_win_addr),
2249 cpu_to_be64(phb->mem_win_size),
2252 cpu_to_be32(b_ss(3)), cpu_to_be64(phb->mem64_win_pciaddr),
2253 cpu_to_be64(phb->mem64_win_addr),
2254 cpu_to_be64(phb->mem64_win_size),
2257 const unsigned sizeof_ranges =
2258 (phb->mem64_win_size ? 3 : 2) * sizeof(ranges[0]);
2259 uint64_t bus_reg[] = { cpu_to_be64(phb->buid), 0 };
2260 uint32_t interrupt_map_mask[] = {
2261 cpu_to_be32(b_ddddd(-1)|b_fff(0)), 0x0, 0x0, cpu_to_be32(-1)};
2262 uint32_t interrupt_map[PCI_SLOT_MAX * PCI_NUM_PINS][7];
2263 uint32_t ddw_applicable[] = {
2264 cpu_to_be32(RTAS_IBM_QUERY_PE_DMA_WINDOW),
2265 cpu_to_be32(RTAS_IBM_CREATE_PE_DMA_WINDOW),
2266 cpu_to_be32(RTAS_IBM_REMOVE_PE_DMA_WINDOW)
2268 uint32_t ddw_extensions[] = {
2269 cpu_to_be32(1),
2270 cpu_to_be32(RTAS_IBM_RESET_PE_DMA_WINDOW)
2272 uint32_t associativity[] = {cpu_to_be32(0x4),
2273 cpu_to_be32(0x0),
2274 cpu_to_be32(0x0),
2275 cpu_to_be32(0x0),
2276 cpu_to_be32(phb->numa_node)};
2277 SpaprTceTable *tcet;
2278 SpaprDrc *drc;
2279 Error *err = NULL;
2281 /* Start populating the FDT */
2282 _FDT(bus_off = fdt_add_subnode(fdt, 0, phb->dtbusname));
2283 if (node_offset) {
2284 *node_offset = bus_off;
2287 /* Write PHB properties */
2288 _FDT(fdt_setprop_string(fdt, bus_off, "device_type", "pci"));
2289 _FDT(fdt_setprop_string(fdt, bus_off, "compatible", "IBM,Logical_PHB"));
2290 _FDT(fdt_setprop_cell(fdt, bus_off, "#interrupt-cells", 0x1));
2291 _FDT(fdt_setprop(fdt, bus_off, "used-by-rtas", NULL, 0));
2292 _FDT(fdt_setprop(fdt, bus_off, "bus-range", &bus_range, sizeof(bus_range)));
2293 _FDT(fdt_setprop(fdt, bus_off, "ranges", &ranges, sizeof_ranges));
2294 _FDT(fdt_setprop(fdt, bus_off, "reg", &bus_reg, sizeof(bus_reg)));
2295 _FDT(fdt_setprop_cell(fdt, bus_off, "ibm,pci-config-space-type", 0x1));
2296 _FDT(fdt_setprop_cell(fdt, bus_off, "ibm,pe-total-#msi",
2297 spapr_irq_nr_msis(spapr)));
2299 /* Dynamic DMA window */
2300 if (phb->ddw_enabled) {
2301 _FDT(fdt_setprop(fdt, bus_off, "ibm,ddw-applicable", &ddw_applicable,
2302 sizeof(ddw_applicable)));
2303 _FDT(fdt_setprop(fdt, bus_off, "ibm,ddw-extensions",
2304 &ddw_extensions, sizeof(ddw_extensions)));
2307 /* Advertise NUMA via ibm,associativity */
2308 if (phb->numa_node != -1) {
2309 _FDT(fdt_setprop(fdt, bus_off, "ibm,associativity", associativity,
2310 sizeof(associativity)));
2313 /* Build the interrupt-map, this must matches what is done
2314 * in pci_swizzle_map_irq_fn
2316 _FDT(fdt_setprop(fdt, bus_off, "interrupt-map-mask",
2317 &interrupt_map_mask, sizeof(interrupt_map_mask)));
2318 for (i = 0; i < PCI_SLOT_MAX; i++) {
2319 for (j = 0; j < PCI_NUM_PINS; j++) {
2320 uint32_t *irqmap = interrupt_map[i*PCI_NUM_PINS + j];
2321 int lsi_num = pci_swizzle(i, j);
2323 irqmap[0] = cpu_to_be32(b_ddddd(i)|b_fff(0));
2324 irqmap[1] = 0;
2325 irqmap[2] = 0;
2326 irqmap[3] = cpu_to_be32(j+1);
2327 irqmap[4] = cpu_to_be32(intc_phandle);
2328 spapr_dt_irq(&irqmap[5], phb->lsi_table[lsi_num].irq, true);
2331 /* Write interrupt map */
2332 _FDT(fdt_setprop(fdt, bus_off, "interrupt-map", &interrupt_map,
2333 sizeof(interrupt_map)));
2335 tcet = spapr_tce_find_by_liobn(phb->dma_liobn[0]);
2336 if (!tcet) {
2337 return -1;
2339 spapr_dma_dt(fdt, bus_off, "ibm,dma-window",
2340 tcet->liobn, tcet->bus_offset,
2341 tcet->nb_table << tcet->page_shift);
2343 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, phb->index);
2344 if (drc) {
2345 uint32_t drc_index = cpu_to_be32(spapr_drc_index(drc));
2347 _FDT(fdt_setprop(fdt, bus_off, "ibm,my-drc-index", &drc_index,
2348 sizeof(drc_index)));
2351 /* Walk the bridges and program the bus numbers*/
2352 spapr_phb_pci_enumerate(phb);
2353 _FDT(fdt_setprop_cell(fdt, bus_off, "qemu,phb-enumerated", 0x1));
2355 /* Walk the bridge and subordinate buses */
2356 ret = spapr_dt_pci_bus(phb, PCI_HOST_BRIDGE(phb)->bus, fdt, bus_off);
2357 if (ret < 0) {
2358 return ret;
2361 spapr_phb_nvgpu_populate_dt(phb, fdt, bus_off, &err);
2362 if (err) {
2363 error_report_err(err);
2365 spapr_phb_nvgpu_ram_populate_dt(phb, fdt);
2367 return 0;
2370 void spapr_pci_rtas_init(void)
2372 spapr_rtas_register(RTAS_READ_PCI_CONFIG, "read-pci-config",
2373 rtas_read_pci_config);
2374 spapr_rtas_register(RTAS_WRITE_PCI_CONFIG, "write-pci-config",
2375 rtas_write_pci_config);
2376 spapr_rtas_register(RTAS_IBM_READ_PCI_CONFIG, "ibm,read-pci-config",
2377 rtas_ibm_read_pci_config);
2378 spapr_rtas_register(RTAS_IBM_WRITE_PCI_CONFIG, "ibm,write-pci-config",
2379 rtas_ibm_write_pci_config);
2380 if (msi_nonbroken) {
2381 spapr_rtas_register(RTAS_IBM_QUERY_INTERRUPT_SOURCE_NUMBER,
2382 "ibm,query-interrupt-source-number",
2383 rtas_ibm_query_interrupt_source_number);
2384 spapr_rtas_register(RTAS_IBM_CHANGE_MSI, "ibm,change-msi",
2385 rtas_ibm_change_msi);
2388 spapr_rtas_register(RTAS_IBM_SET_EEH_OPTION,
2389 "ibm,set-eeh-option",
2390 rtas_ibm_set_eeh_option);
2391 spapr_rtas_register(RTAS_IBM_GET_CONFIG_ADDR_INFO2,
2392 "ibm,get-config-addr-info2",
2393 rtas_ibm_get_config_addr_info2);
2394 spapr_rtas_register(RTAS_IBM_READ_SLOT_RESET_STATE2,
2395 "ibm,read-slot-reset-state2",
2396 rtas_ibm_read_slot_reset_state2);
2397 spapr_rtas_register(RTAS_IBM_SET_SLOT_RESET,
2398 "ibm,set-slot-reset",
2399 rtas_ibm_set_slot_reset);
2400 spapr_rtas_register(RTAS_IBM_CONFIGURE_PE,
2401 "ibm,configure-pe",
2402 rtas_ibm_configure_pe);
2403 spapr_rtas_register(RTAS_IBM_SLOT_ERROR_DETAIL,
2404 "ibm,slot-error-detail",
2405 rtas_ibm_slot_error_detail);
2408 static void spapr_pci_register_types(void)
2410 type_register_static(&spapr_phb_info);
2413 type_init(spapr_pci_register_types)
2415 static int spapr_switch_one_vga(DeviceState *dev, void *opaque)
2417 bool be = *(bool *)opaque;
2419 if (object_dynamic_cast(OBJECT(dev), "VGA")
2420 || object_dynamic_cast(OBJECT(dev), "secondary-vga")) {
2421 object_property_set_bool(OBJECT(dev), "big-endian-framebuffer", be,
2422 &error_abort);
2424 return 0;
2427 void spapr_pci_switch_vga(bool big_endian)
2429 SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
2430 SpaprPhbState *sphb;
2433 * For backward compatibility with existing guests, we switch
2434 * the endianness of the VGA controller when changing the guest
2435 * interrupt mode
2437 QLIST_FOREACH(sphb, &spapr->phbs, list) {
2438 BusState *bus = &PCI_HOST_BRIDGE(sphb)->bus->qbus;
2439 qbus_walk_children(bus, spapr_switch_one_vga, NULL, NULL, NULL,
2440 &big_endian);