1 // Initialize PCI devices (on emulators)
2 //
3 // Copyright (C) 2008  Kevin O'Connor <kevin@koconnor.net>
4 // Copyright (C) 2006 Fabrice Bellard
5 //
6 // This file may be distributed under the terms of the GNU LGPLv3 license.
7 
8 #include "byteorder.h" // le64_to_cpu
9 #include "config.h" // CONFIG_*
10 #include "dev-q35.h" // Q35_HOST_BRIDGE_PCIEXBAR_ADDR
11 #include "dev-piix.h" // PIIX_*
12 #include "e820map.h" // e820_add
13 #include "hw/ata.h" // PORT_ATA1_CMD_BASE
14 #include "hw/pci.h" // pci_config_readl
15 #include "hw/pcidevice.h" // pci_probe_devices
16 #include "hw/pci_ids.h" // PCI_VENDOR_ID_INTEL
17 #include "hw/pci_regs.h" // PCI_COMMAND
18 #include "fw/dev-pci.h" // REDHAT_CAP_RESOURCE_RESERVE
19 #include "list.h" // struct hlist_node
20 #include "malloc.h" // free
21 #include "output.h" // dprintf
22 #include "paravirt.h" // RamSize
23 #include "romfile.h" // romfile_loadint
24 #include "string.h" // memset
25 #include "util.h" // pci_setup
26 #include "x86.h" // outb
27 
28 #define PCI_DEVICE_MEM_MIN    (1<<12)  // 4k == page size
29 #define PCI_BRIDGE_MEM_MIN    (1<<21)  // 2M == hugepage size
30 #define PCI_BRIDGE_IO_MIN      0x1000  // mandated by pci bridge spec
31 
32 #define PCI_ROM_SLOT 6
33 #define PCI_NUM_REGIONS 7
34 #define PCI_BRIDGE_NUM_REGIONS 2
35 
36 enum pci_region_type {
37     PCI_REGION_TYPE_IO,
38     PCI_REGION_TYPE_MEM,
39     PCI_REGION_TYPE_PREFMEM,
40     PCI_REGION_TYPE_COUNT,
41 };
42 
43 static const char *region_type_name[] = {
44     [ PCI_REGION_TYPE_IO ]      = "io",
45     [ PCI_REGION_TYPE_MEM ]     = "mem",
46     [ PCI_REGION_TYPE_PREFMEM ] = "prefmem",
47 };
48 
49 u64 pcimem_start   = BUILD_PCIMEM_START;
50 u64 pcimem_end     = BUILD_PCIMEM_END;
51 u64 pcimem64_start = BUILD_PCIMEM64_START;
52 u64 pcimem64_end   = BUILD_PCIMEM64_END;
53 u64 pci_io_low_end = 0xa000;
54 
55 struct pci_region_entry {
56     struct pci_device *dev;
57     int bar;
58     u64 size;
59     u64 align;
60     int is64;
61     enum pci_region_type type;
62     struct hlist_node node;
63 };
64 
65 struct pci_region {
66     /* pci region assignments */
67     u64 base;
68     struct hlist_head list;
69 };
70 
71 struct pci_bus {
72     struct pci_region r[PCI_REGION_TYPE_COUNT];
73     struct pci_device *bus_dev;
74 };
75 
pci_bar(struct pci_device * pci,int region_num)76 static u32 pci_bar(struct pci_device *pci, int region_num)
77 {
78     if (region_num != PCI_ROM_SLOT) {
79         return PCI_BASE_ADDRESS_0 + region_num * 4;
80     }
81 
82 #define PCI_HEADER_TYPE_MULTI_FUNCTION 0x80
83     u8 type = pci->header_type & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
84     return type == PCI_HEADER_TYPE_BRIDGE ? PCI_ROM_ADDRESS1 : PCI_ROM_ADDRESS;
85 }
86 
87 static void
pci_set_io_region_addr(struct pci_device * pci,int bar,u64 addr,int is64)88 pci_set_io_region_addr(struct pci_device *pci, int bar, u64 addr, int is64)
89 {
90     u32 ofs = pci_bar(pci, bar);
91     pci_config_writel(pci->bdf, ofs, addr);
92     if (is64)
93         pci_config_writel(pci->bdf, ofs + 4, addr >> 32);
94 }
95 
96 
97 /****************************************************************
98  * Misc. device init
99  ****************************************************************/
100 
101 /* host irqs corresponding to PCI irqs A-D */
102 const u8 pci_irqs[4] = {
103     10, 10, 11, 11
104 };
105 
dummy_pci_slot_get_irq(struct pci_device * pci,int pin)106 static int dummy_pci_slot_get_irq(struct pci_device *pci, int pin)
107 {
108     dprintf(1, "pci_slot_get_irq called with unknown routing\n");
109 
110     return 0xff; /* PCI defined "unknown" or "no connection" for x86 */
111 }
112 
113 static int (*pci_slot_get_irq)(struct pci_device *pci, int pin) =
114     dummy_pci_slot_get_irq;
115 
116 // Return the global irq number corresponding to a host bus device irq pin.
piix_pci_slot_get_irq(struct pci_device * pci,int pin)117 static int piix_pci_slot_get_irq(struct pci_device *pci, int pin)
118 {
119     int slot_addend = 0;
120 
121     while (pci->parent != NULL) {
122         slot_addend += pci_bdf_to_dev(pci->bdf);
123         pci = pci->parent;
124     }
125     slot_addend += pci_bdf_to_dev(pci->bdf) - 1;
126     return pci_irqs[(pin - 1 + slot_addend) & 3];
127 }
128 
mch_pci_slot_get_irq(struct pci_device * pci,int pin)129 static int mch_pci_slot_get_irq(struct pci_device *pci, int pin)
130 {
131     int pin_addend = 0;
132     while (pci->parent != NULL) {
133         pin_addend += pci_bdf_to_dev(pci->bdf);
134         pci = pci->parent;
135     }
136     u8 slot = pci_bdf_to_dev(pci->bdf);
137     if (slot <= 24)
138         /* Slots 0-24 rotate slot:pin mapping similar to piix above, but
139            with a different starting index - see q35-acpi-dsdt.dsl */
140         return pci_irqs[(pin - 1 + pin_addend + slot) & 3];
141     /* Slots 25-31 all use LNKA mapping (or LNKE, but A:D = E:H) */
142     return pci_irqs[(pin - 1 + pin_addend) & 3];
143 }
144 
145 /* PIIX3/PIIX4 PCI to ISA bridge */
piix_isa_bridge_setup(struct pci_device * pci,void * arg)146 static void piix_isa_bridge_setup(struct pci_device *pci, void *arg)
147 {
148     int i, irq;
149     u8 elcr[2];
150 
151     elcr[0] = 0x00;
152     elcr[1] = 0x00;
153     for (i = 0; i < 4; i++) {
154         irq = pci_irqs[i];
155         /* set to trigger level */
156         elcr[irq >> 3] |= (1 << (irq & 7));
157         /* activate irq remapping in PIIX */
158         pci_config_writeb(pci->bdf, 0x60 + i, irq);
159     }
160     outb(elcr[0], PIIX_PORT_ELCR1);
161     outb(elcr[1], PIIX_PORT_ELCR2);
162     dprintf(1, "PIIX3/PIIX4 init: elcr=%02x %02x\n", elcr[0], elcr[1]);
163 }
164 
mch_isa_lpc_setup(u16 bdf)165 static void mch_isa_lpc_setup(u16 bdf)
166 {
167     /* pm io base */
168     pci_config_writel(bdf, ICH9_LPC_PMBASE,
169                       acpi_pm_base | ICH9_LPC_PMBASE_RTE);
170 
171     /* acpi enable, SCI: IRQ9 000b = irq9*/
172     pci_config_writeb(bdf, ICH9_LPC_ACPI_CTRL, ICH9_LPC_ACPI_CTRL_ACPI_EN);
173 
174     /* set root complex register block BAR */
175     pci_config_writel(bdf, ICH9_LPC_RCBA,
176                       ICH9_LPC_RCBA_ADDR | ICH9_LPC_RCBA_EN);
177 }
178 
179 static int ICH9LpcBDF = -1;
180 
181 /* ICH9 LPC PCI to ISA bridge */
182 /* PCI_VENDOR_ID_INTEL && PCI_DEVICE_ID_INTEL_ICH9_LPC */
mch_isa_bridge_setup(struct pci_device * dev,void * arg)183 static void mch_isa_bridge_setup(struct pci_device *dev, void *arg)
184 {
185     u16 bdf = dev->bdf;
186     int i, irq;
187     u8 elcr[2];
188 
189     elcr[0] = 0x00;
190     elcr[1] = 0x00;
191 
192     for (i = 0; i < 4; i++) {
193         irq = pci_irqs[i];
194         /* set to trigger level */
195         elcr[irq >> 3] |= (1 << (irq & 7));
196 
197         /* activate irq remapping in LPC */
198 
199         /* PIRQ[A-D] routing */
200         pci_config_writeb(bdf, ICH9_LPC_PIRQA_ROUT + i, irq);
201         /* PIRQ[E-H] routing */
202         pci_config_writeb(bdf, ICH9_LPC_PIRQE_ROUT + i, irq);
203     }
204     outb(elcr[0], ICH9_LPC_PORT_ELCR1);
205     outb(elcr[1], ICH9_LPC_PORT_ELCR2);
206     dprintf(1, "Q35 LPC init: elcr=%02x %02x\n", elcr[0], elcr[1]);
207 
208     ICH9LpcBDF = bdf;
209 
210     mch_isa_lpc_setup(bdf);
211 
212     e820_add(ICH9_LPC_RCBA_ADDR, 16*1024, E820_RESERVED);
213 
214     acpi_pm1a_cnt = acpi_pm_base + 0x04;
215     pmtimer_setup(acpi_pm_base + 0x08);
216 }
217 
storage_ide_setup(struct pci_device * pci,void * arg)218 static void storage_ide_setup(struct pci_device *pci, void *arg)
219 {
220     /* IDE: we map it as in ISA mode */
221     pci_set_io_region_addr(pci, 0, PORT_ATA1_CMD_BASE, 0);
222     pci_set_io_region_addr(pci, 1, PORT_ATA1_CTRL_BASE, 0);
223     pci_set_io_region_addr(pci, 2, PORT_ATA2_CMD_BASE, 0);
224     pci_set_io_region_addr(pci, 3, PORT_ATA2_CTRL_BASE, 0);
225 }
226 
227 /* PIIX3/PIIX4 IDE */
piix_ide_setup(struct pci_device * pci,void * arg)228 static void piix_ide_setup(struct pci_device *pci, void *arg)
229 {
230     u16 bdf = pci->bdf;
231     pci_config_writew(bdf, 0x40, 0x8000); // enable IDE0
232     pci_config_writew(bdf, 0x42, 0x8000); // enable IDE1
233 }
234 
pic_ibm_setup(struct pci_device * pci,void * arg)235 static void pic_ibm_setup(struct pci_device *pci, void *arg)
236 {
237     /* PIC, IBM, MPIC & MPIC2 */
238     pci_set_io_region_addr(pci, 0, 0x80800000 + 0x00040000, 0);
239 }
240 
apple_macio_setup(struct pci_device * pci,void * arg)241 static void apple_macio_setup(struct pci_device *pci, void *arg)
242 {
243     /* macio bridge */
244     pci_set_io_region_addr(pci, 0, 0x80800000, 0);
245 }
246 
piix4_pm_config_setup(u16 bdf)247 static void piix4_pm_config_setup(u16 bdf)
248 {
249     // acpi sci is hardwired to 9
250     pci_config_writeb(bdf, PCI_INTERRUPT_LINE, 9);
251 
252     pci_config_writel(bdf, PIIX_PMBASE, acpi_pm_base | 1);
253     pci_config_writeb(bdf, PIIX_PMREGMISC, 0x01); /* enable PM io space */
254     pci_config_writel(bdf, PIIX_SMBHSTBASE, (acpi_pm_base + 0x100) | 1);
255     pci_config_writeb(bdf, PIIX_SMBHSTCFG, 0x09); /* enable SMBus io space */
256 }
257 
258 static int PiixPmBDF = -1;
259 
260 /* PIIX4 Power Management device (for ACPI) */
piix4_pm_setup(struct pci_device * pci,void * arg)261 static void piix4_pm_setup(struct pci_device *pci, void *arg)
262 {
263     PiixPmBDF = pci->bdf;
264     piix4_pm_config_setup(pci->bdf);
265 
266     acpi_pm1a_cnt = acpi_pm_base + 0x04;
267     pmtimer_setup(acpi_pm_base + 0x08);
268 }
269 
ich9_smbus_enable(u16 bdf)270 static void ich9_smbus_enable(u16 bdf)
271 {
272     /* map smbus into io space */
273     pci_config_writel(bdf, ICH9_SMB_SMB_BASE,
274                       (acpi_pm_base + 0x100) | PCI_BASE_ADDRESS_SPACE_IO);
275 
276     /* enable SMBus */
277     pci_config_writeb(bdf, ICH9_SMB_HOSTC, ICH9_SMB_HOSTC_HST_EN);
278 }
279 
280 static int ICH9SmbusBDF = -1;
281 
282 /* ICH9 SMBUS */
283 /* PCI_VENDOR_ID_INTEL && PCI_DEVICE_ID_INTEL_ICH9_SMBUS */
ich9_smbus_setup(struct pci_device * dev,void * arg)284 static void ich9_smbus_setup(struct pci_device *dev, void *arg)
285 {
286     ICH9SmbusBDF = dev->bdf;
287 
288     ich9_smbus_enable(dev->bdf);
289 }
290 
intel_igd_setup(struct pci_device * dev,void * arg)291 static void intel_igd_setup(struct pci_device *dev, void *arg)
292 {
293     struct romfile_s *opregion = romfile_find("etc/igd-opregion");
294     u64 bdsm_size = le64_to_cpu(romfile_loadint("etc/igd-bdsm-size", 0));
295 
296     /* Apply OpRegion to any Intel VGA device, more than one is undefined */
297     if (opregion && opregion->size) {
298         void *addr = memalign_high(PAGE_SIZE, opregion->size);
299         if (!addr) {
300             warn_noalloc();
301             return;
302         }
303 
304         if (opregion->copy(opregion, addr, opregion->size) < 0) {
305             free(addr);
306             return;
307         }
308 
309         pci_config_writel(dev->bdf, 0xFC, cpu_to_le32((u32)addr));
310 
311         dprintf(1, "Intel IGD OpRegion enabled at 0x%08x, size %dKB, dev %pP\n"
312                 , (u32)addr, opregion->size >> 10, dev);
313     }
314 
315     /* Apply BDSM only to Intel VGA at 00:02.0 */
316     if (bdsm_size && (dev->bdf == pci_to_bdf(0, 2, 0))) {
317         void *addr = memalign_tmphigh(1024 * 1024, bdsm_size);
318         if (!addr) {
319             warn_noalloc();
320             return;
321         }
322 
323         e820_add((u32)addr, bdsm_size, E820_RESERVED);
324 
325         pci_config_writel(dev->bdf, 0x5C, cpu_to_le32((u32)addr));
326 
327         dprintf(1, "Intel IGD BDSM enabled at 0x%08x, size %lldMB, dev %pP\n"
328                 , (u32)addr, bdsm_size >> 20, dev);
329     }
330 }
331 
332 static const struct pci_device_id pci_device_tbl[] = {
333     /* PIIX3/PIIX4 PCI to ISA bridge */
334     PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_82371SB_0,
335                piix_isa_bridge_setup),
336     PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_82371AB_0,
337                piix_isa_bridge_setup),
338     PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_ICH9_LPC,
339                mch_isa_bridge_setup),
340 
341     /* STORAGE IDE */
342     PCI_DEVICE_CLASS(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_82371SB_1,
343                      PCI_CLASS_STORAGE_IDE, piix_ide_setup),
344     PCI_DEVICE_CLASS(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_82371AB,
345                      PCI_CLASS_STORAGE_IDE, piix_ide_setup),
346     PCI_DEVICE_CLASS(PCI_ANY_ID, PCI_ANY_ID, PCI_CLASS_STORAGE_IDE,
347                      storage_ide_setup),
348 
349     /* PIC, IBM, MPIC & MPIC2 */
350     PCI_DEVICE_CLASS(PCI_VENDOR_ID_IBM, 0x0046, PCI_CLASS_SYSTEM_PIC,
351                      pic_ibm_setup),
352     PCI_DEVICE_CLASS(PCI_VENDOR_ID_IBM, 0xFFFF, PCI_CLASS_SYSTEM_PIC,
353                      pic_ibm_setup),
354 
355     /* PIIX4 Power Management device (for ACPI) */
356     PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_82371AB_3,
357                piix4_pm_setup),
358     PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_ICH9_SMBUS,
359                ich9_smbus_setup),
360 
361     /* 0xff00 */
362     PCI_DEVICE_CLASS(PCI_VENDOR_ID_APPLE, 0x0017, 0xff00, apple_macio_setup),
363     PCI_DEVICE_CLASS(PCI_VENDOR_ID_APPLE, 0x0022, 0xff00, apple_macio_setup),
364 
365     /* Intel IGD OpRegion setup */
366     PCI_DEVICE_CLASS(PCI_VENDOR_ID_INTEL, PCI_ANY_ID, PCI_CLASS_DISPLAY_VGA,
367                      intel_igd_setup),
368 
369     PCI_DEVICE_END,
370 };
371 
372 static int MCHMmcfgBDF = -1;
373 static void mch_mmconfig_setup(u16 bdf);
374 
pci_resume(void)375 void pci_resume(void)
376 {
377     if (!CONFIG_QEMU) {
378         return;
379     }
380 
381     if (PiixPmBDF >= 0) {
382         piix4_pm_config_setup(PiixPmBDF);
383     }
384 
385     if (ICH9LpcBDF >= 0) {
386         mch_isa_lpc_setup(ICH9LpcBDF);
387     }
388 
389     if (ICH9SmbusBDF >= 0) {
390         ich9_smbus_enable(ICH9SmbusBDF);
391     }
392 
393     if(MCHMmcfgBDF >= 0) {
394         mch_mmconfig_setup(MCHMmcfgBDF);
395     }
396 }
397 
pci_bios_init_device(struct pci_device * pci)398 static void pci_bios_init_device(struct pci_device *pci)
399 {
400     dprintf(1, "PCI: init bdf=%pP id=%04x:%04x\n"
401             , pci, pci->vendor, pci->device);
402 
403     /* map the interrupt */
404     u16 bdf = pci->bdf;
405     int pin = pci_config_readb(bdf, PCI_INTERRUPT_PIN);
406     if (pin != 0)
407         pci_config_writeb(bdf, PCI_INTERRUPT_LINE, pci_slot_get_irq(pci, pin));
408 
409     pci_init_device(pci_device_tbl, pci, NULL);
410 
411     /* enable memory mappings */
412     pci_config_maskw(bdf, PCI_COMMAND, 0,
413                      PCI_COMMAND_IO | PCI_COMMAND_MEMORY | PCI_COMMAND_SERR);
414     /* enable SERR# for forwarding */
415     if (pci->header_type & PCI_HEADER_TYPE_BRIDGE)
416         pci_config_maskw(bdf, PCI_BRIDGE_CONTROL, 0,
417                          PCI_BRIDGE_CTL_SERR);
418 }
419 
pci_bios_init_devices(void)420 static void pci_bios_init_devices(void)
421 {
422     struct pci_device *pci;
423     foreachpci(pci) {
424         pci_bios_init_device(pci);
425     }
426 }
427 
pci_enable_default_vga(void)428 static void pci_enable_default_vga(void)
429 {
430     struct pci_device *pci;
431 
432     foreachpci(pci) {
433         if (is_pci_vga(pci)) {
434             dprintf(1, "PCI: Using %pP for primary VGA\n", pci);
435             return;
436         }
437     }
438 
439     pci = pci_find_class(PCI_CLASS_DISPLAY_VGA);
440     if (!pci) {
441         dprintf(1, "PCI: No VGA devices found\n");
442         return;
443     }
444 
445     dprintf(1, "PCI: Enabling %pP for primary VGA\n", pci);
446 
447     pci_config_maskw(pci->bdf, PCI_COMMAND, 0,
448                      PCI_COMMAND_IO | PCI_COMMAND_MEMORY);
449 
450     while (pci->parent) {
451         pci = pci->parent;
452 
453         dprintf(1, "PCI: Setting VGA enable on bridge %pP\n", pci);
454 
455         pci_config_maskw(pci->bdf, PCI_BRIDGE_CONTROL, 0, PCI_BRIDGE_CTL_VGA);
456         pci_config_maskw(pci->bdf, PCI_COMMAND, 0,
457                          PCI_COMMAND_IO | PCI_COMMAND_MEMORY);
458     }
459 }
460 
461 /****************************************************************
462  * Platform device initialization
463  ****************************************************************/
464 
i440fx_mem_addr_setup(struct pci_device * dev,void * arg)465 static void i440fx_mem_addr_setup(struct pci_device *dev, void *arg)
466 {
467     if (RamSize <= 0x80000000)
468         pcimem_start = 0x80000000;
469     else if (RamSize <= 0xc0000000)
470         pcimem_start = 0xc0000000;
471 
472     pci_slot_get_irq = piix_pci_slot_get_irq;
473 }
474 
mch_mmconfig_setup(u16 bdf)475 static void mch_mmconfig_setup(u16 bdf)
476 {
477     u64 addr = Q35_HOST_BRIDGE_PCIEXBAR_ADDR;
478     u32 upper = addr >> 32;
479     u32 lower = (addr & 0xffffffff) | Q35_HOST_BRIDGE_PCIEXBAREN;
480     pci_config_writel(bdf, Q35_HOST_BRIDGE_PCIEXBAR, 0);
481     pci_config_writel(bdf, Q35_HOST_BRIDGE_PCIEXBAR + 4, upper);
482     pci_config_writel(bdf, Q35_HOST_BRIDGE_PCIEXBAR, lower);
483 }
484 
mch_mem_addr_setup(struct pci_device * dev,void * arg)485 static void mch_mem_addr_setup(struct pci_device *dev, void *arg)
486 {
487     u64 addr = Q35_HOST_BRIDGE_PCIEXBAR_ADDR;
488     u32 size = Q35_HOST_BRIDGE_PCIEXBAR_SIZE;
489 
490     /* setup mmconfig */
491     MCHMmcfgBDF = dev->bdf;
492     mch_mmconfig_setup(dev->bdf);
493     e820_add(addr, size, E820_RESERVED);
494 
495     /* setup pci i/o window (above mmconfig) */
496     pcimem_start = addr + size;
497 
498     pci_slot_get_irq = mch_pci_slot_get_irq;
499 
500     /* setup io address space */
501     if (acpi_pm_base < 0x1000)
502         pci_io_low_end = 0x10000;
503     else
504         pci_io_low_end = acpi_pm_base;
505 }
506 
507 static const struct pci_device_id pci_platform_tbl[] = {
508     PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_82441,
509                i440fx_mem_addr_setup),
510     PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_Q35_MCH,
511                mch_mem_addr_setup),
512     PCI_DEVICE_END
513 };
514 
pci_bios_init_platform(void)515 static void pci_bios_init_platform(void)
516 {
517     struct pci_device *pci;
518     foreachpci(pci) {
519         pci_init_device(pci_platform_tbl, pci, NULL);
520     }
521 }
522 
pci_find_resource_reserve_capability(u16 bdf)523 static u8 pci_find_resource_reserve_capability(u16 bdf)
524 {
525     u16 device_id;
526 
527     if (pci_config_readw(bdf, PCI_VENDOR_ID) != PCI_VENDOR_ID_REDHAT) {
528         dprintf(3, "PCI: This is non-QEMU bridge.\n");
529         return 0;
530     }
531 
532     device_id = pci_config_readw(bdf, PCI_DEVICE_ID);
533 
534     if (device_id != PCI_DEVICE_ID_REDHAT_ROOT_PORT &&
535         device_id != PCI_DEVICE_ID_REDHAT_BRIDGE) {
536         dprintf(1, "PCI: QEMU resource reserve cap device ID doesn't match.\n");
537         return 0;
538     }
539     u8 cap = 0;
540 
541     do {
542         cap = pci_find_capability(bdf, PCI_CAP_ID_VNDR, cap);
543     } while (cap &&
544              pci_config_readb(bdf, cap + PCI_CAP_REDHAT_TYPE_OFFSET) !=
545                               REDHAT_CAP_RESOURCE_RESERVE);
546     if (cap) {
547         u8 cap_len = pci_config_readb(bdf, cap + PCI_CAP_FLAGS);
548         if (cap_len < RES_RESERVE_CAP_SIZE) {
549             dprintf(1, "PCI: QEMU resource reserve cap length %d is invalid\n",
550                     cap_len);
551             return 0;
552         }
553     } else {
554         dprintf(1, "PCI: QEMU resource reserve cap not found\n");
555     }
556     return cap;
557 }
558 
559 /****************************************************************
560  * Bus initialization
561  ****************************************************************/
562 
563 static void
pci_bios_init_bus_rec(int bus,u8 * pci_bus)564 pci_bios_init_bus_rec(int bus, u8 *pci_bus)
565 {
566     int bdf;
567     u16 class;
568 
569     dprintf(1, "PCI: %s bus = 0x%x\n", __func__, bus);
570 
571     /* prevent accidental access to unintended devices */
572     foreachbdf(bdf, bus) {
573         class = pci_config_readw(bdf, PCI_CLASS_DEVICE);
574         if (class == PCI_CLASS_BRIDGE_PCI) {
575             pci_config_writeb(bdf, PCI_SECONDARY_BUS, 255);
576             pci_config_writeb(bdf, PCI_SUBORDINATE_BUS, 0);
577         }
578     }
579 
580     foreachbdf(bdf, bus) {
581         class = pci_config_readw(bdf, PCI_CLASS_DEVICE);
582         if (class != PCI_CLASS_BRIDGE_PCI) {
583             continue;
584         }
585         dprintf(1, "PCI: %s bdf = 0x%x\n", __func__, bdf);
586 
587         u8 pribus = pci_config_readb(bdf, PCI_PRIMARY_BUS);
588         if (pribus != bus) {
589             dprintf(1, "PCI: primary bus = 0x%x -> 0x%x\n", pribus, bus);
590             pci_config_writeb(bdf, PCI_PRIMARY_BUS, bus);
591         } else {
592             dprintf(1, "PCI: primary bus = 0x%x\n", pribus);
593         }
594 
595         u8 secbus = pci_config_readb(bdf, PCI_SECONDARY_BUS);
596         (*pci_bus)++;
597         if (*pci_bus != secbus) {
598             dprintf(1, "PCI: secondary bus = 0x%x -> 0x%x\n",
599                     secbus, *pci_bus);
600             secbus = *pci_bus;
601             pci_config_writeb(bdf, PCI_SECONDARY_BUS, secbus);
602         } else {
603             dprintf(1, "PCI: secondary bus = 0x%x\n", secbus);
604         }
605 
606         /* set to max for access to all subordinate buses.
607            later set it to accurate value */
608         u8 subbus = pci_config_readb(bdf, PCI_SUBORDINATE_BUS);
609         pci_config_writeb(bdf, PCI_SUBORDINATE_BUS, 255);
610 
611         pci_bios_init_bus_rec(secbus, pci_bus);
612 
613         if (subbus != *pci_bus) {
614             u8 res_bus = *pci_bus;
615             u8 cap = pci_find_resource_reserve_capability(bdf);
616 
617             if (cap) {
618                 u32 tmp_res_bus = pci_config_readl(bdf,
619                         cap + RES_RESERVE_BUS_RES);
620                 if (tmp_res_bus != (u32)-1) {
621                     res_bus = tmp_res_bus & 0xFF;
622                     if ((u8)(res_bus + secbus) < secbus ||
623                             (u8)(res_bus + secbus) < res_bus) {
624                         dprintf(1, "PCI: bus_reserve value %d is invalid\n",
625                                 res_bus);
626                         res_bus = 0;
627                     }
628                     if (secbus + res_bus > *pci_bus) {
629                         dprintf(1, "PCI: QEMU resource reserve cap: bus = %u\n",
630                                 res_bus);
631                         res_bus = secbus + res_bus;
632                     }
633                 }
634             }
635             dprintf(1, "PCI: subordinate bus = 0x%x -> 0x%x\n",
636                     subbus, res_bus);
637             subbus = res_bus;
638             *pci_bus = res_bus;
639         } else {
640             dprintf(1, "PCI: subordinate bus = 0x%x\n", subbus);
641         }
642         pci_config_writeb(bdf, PCI_SUBORDINATE_BUS, subbus);
643     }
644 }
645 
646 static void
pci_bios_init_bus(void)647 pci_bios_init_bus(void)
648 {
649     u8 extraroots = romfile_loadint("etc/extra-pci-roots", 0);
650     u8 pci_bus = 0;
651 
652     pci_bios_init_bus_rec(0 /* host bus */, &pci_bus);
653 
654     if (extraroots) {
655         while (pci_bus < 0xff) {
656             pci_bus++;
657             pci_bios_init_bus_rec(pci_bus, &pci_bus);
658         }
659     }
660 }
661 
662 
663 /****************************************************************
664  * Bus sizing
665  ****************************************************************/
666 
667 static void
pci_bios_get_bar(struct pci_device * pci,int bar,int * ptype,u64 * psize,int * pis64)668 pci_bios_get_bar(struct pci_device *pci, int bar,
669                  int *ptype, u64 *psize, int *pis64)
670 {
671     u32 ofs = pci_bar(pci, bar);
672     u16 bdf = pci->bdf;
673     u32 old = pci_config_readl(bdf, ofs);
674     int is64 = 0, type = PCI_REGION_TYPE_MEM;
675     u64 mask;
676 
677     if (bar == PCI_ROM_SLOT) {
678         mask = PCI_ROM_ADDRESS_MASK;
679         pci_config_writel(bdf, ofs, mask);
680     } else {
681         if (old & PCI_BASE_ADDRESS_SPACE_IO) {
682             mask = PCI_BASE_ADDRESS_IO_MASK;
683             type = PCI_REGION_TYPE_IO;
684         } else {
685             mask = PCI_BASE_ADDRESS_MEM_MASK;
686             if (old & PCI_BASE_ADDRESS_MEM_PREFETCH)
687                 type = PCI_REGION_TYPE_PREFMEM;
688             is64 = ((old & PCI_BASE_ADDRESS_MEM_TYPE_MASK)
689                     == PCI_BASE_ADDRESS_MEM_TYPE_64);
690         }
691         pci_config_writel(bdf, ofs, ~0);
692     }
693     u64 val = pci_config_readl(bdf, ofs);
694     pci_config_writel(bdf, ofs, old);
695     if (is64) {
696         u32 hold = pci_config_readl(bdf, ofs + 4);
697         pci_config_writel(bdf, ofs + 4, ~0);
698         u32 high = pci_config_readl(bdf, ofs + 4);
699         pci_config_writel(bdf, ofs + 4, hold);
700         val |= ((u64)high << 32);
701         mask |= ((u64)0xffffffff << 32);
702         *psize = (~(val & mask)) + 1;
703     } else {
704         *psize = ((~(val & mask)) + 1) & 0xffffffff;
705     }
706     *ptype = type;
707     *pis64 = is64;
708 }
709 
pci_bios_bridge_region_is64(struct pci_region * r,struct pci_device * pci,int type)710 static int pci_bios_bridge_region_is64(struct pci_region *r,
711                                  struct pci_device *pci, int type)
712 {
713     if (type != PCI_REGION_TYPE_PREFMEM)
714         return 0;
715     u32 pmem = pci_config_readl(pci->bdf, PCI_PREF_MEMORY_BASE);
716     if (!pmem) {
717         pci_config_writel(pci->bdf, PCI_PREF_MEMORY_BASE, 0xfff0fff0);
718         pmem = pci_config_readl(pci->bdf, PCI_PREF_MEMORY_BASE);
719         pci_config_writel(pci->bdf, PCI_PREF_MEMORY_BASE, 0x0);
720     }
721     if ((pmem & PCI_PREF_RANGE_TYPE_MASK) != PCI_PREF_RANGE_TYPE_64)
722        return 0;
723     struct pci_region_entry *entry;
724     hlist_for_each_entry(entry, &r->list, node) {
725         if (!entry->is64)
726             return 0;
727     }
728     return 1;
729 }
730 
pci_region_align(struct pci_region * r)731 static u64 pci_region_align(struct pci_region *r)
732 {
733     struct pci_region_entry *entry;
734     hlist_for_each_entry(entry, &r->list, node) {
735         // The first entry in the sorted list has the largest alignment
736         return entry->align;
737     }
738     return 1;
739 }
740 
pci_region_sum(struct pci_region * r)741 static u64 pci_region_sum(struct pci_region *r)
742 {
743     u64 sum = 0;
744     struct pci_region_entry *entry;
745     hlist_for_each_entry(entry, &r->list, node) {
746         sum += entry->size;
747     }
748     return sum;
749 }
750 
pci_region_migrate_64bit_entries(struct pci_region * from,struct pci_region * to)751 static void pci_region_migrate_64bit_entries(struct pci_region *from,
752                                              struct pci_region *to)
753 {
754     struct hlist_node *n, **last = &to->list.first;
755     struct pci_region_entry *entry;
756     hlist_for_each_entry_safe(entry, n, &from->list, node) {
757         if (!entry->is64)
758             continue;
759         if (entry->dev->class == PCI_CLASS_SERIAL_USB)
760             continue;
761         // Move from source list to destination list.
762         hlist_del(&entry->node);
763         hlist_add(&entry->node, last);
764         last = &entry->node.next;
765     }
766 }
767 
768 static struct pci_region_entry *
pci_region_create_entry(struct pci_bus * bus,struct pci_device * dev,int bar,u64 size,u64 align,int type,int is64)769 pci_region_create_entry(struct pci_bus *bus, struct pci_device *dev,
770                         int bar, u64 size, u64 align, int type, int is64)
771 {
772     struct pci_region_entry *entry = malloc_tmp(sizeof(*entry));
773     if (!entry) {
774         warn_noalloc();
775         return NULL;
776     }
777     memset(entry, 0, sizeof(*entry));
778     entry->dev = dev;
779     entry->bar = bar;
780     entry->size = size;
781     entry->align = align;
782     entry->is64 = is64;
783     entry->type = type;
784     // Insert into list in sorted order.
785     struct hlist_node **pprev;
786     struct pci_region_entry *pos;
787     hlist_for_each_entry_pprev(pos, pprev, &bus->r[type].list, node) {
788         if (pos->align < align || (pos->align == align && pos->size < size))
789             break;
790     }
791     hlist_add(&entry->node, pprev);
792     return entry;
793 }
794 
pci_bus_hotplug_support(struct pci_bus * bus,u8 pcie_cap)795 static int pci_bus_hotplug_support(struct pci_bus *bus, u8 pcie_cap)
796 {
797     u8 shpc_cap;
798 
799     if (pcie_cap) {
800         u16 pcie_flags = pci_config_readw(bus->bus_dev->bdf,
801                                           pcie_cap + PCI_EXP_FLAGS);
802         u8 port_type = ((pcie_flags & PCI_EXP_FLAGS_TYPE) >>
803                        (__builtin_ffs(PCI_EXP_FLAGS_TYPE) - 1));
804         u8 downstream_port = (port_type == PCI_EXP_TYPE_DOWNSTREAM) ||
805                              (port_type == PCI_EXP_TYPE_ROOT_PORT);
806         /*
807          * PCI Express SPEC, 7.8.2:
808          *   Slot Implemented – When Set, this bit indicates that the Link
809          *   HwInit associated with this Port is connected to a slot (as
810          *   compared to being connected to a system-integrated device or
811          *   being disabled).
812          *   This bit is valid for Downstream Ports. This bit is undefined
813          *   for Upstream Ports.
814          */
815         u16 slot_implemented = pcie_flags & PCI_EXP_FLAGS_SLOT;
816 
817         return downstream_port && slot_implemented;
818     }
819 
820     shpc_cap = pci_find_capability(bus->bus_dev->bdf, PCI_CAP_ID_SHPC, 0);
821     return !!shpc_cap;
822 }
823 
824 /* Test whether bridge support forwarding of transactions
825  * of a specific type.
826  * Note: disables bridge's window registers as a side effect.
827  */
pci_bridge_has_region(struct pci_device * pci,enum pci_region_type region_type)828 static int pci_bridge_has_region(struct pci_device *pci,
829                                  enum pci_region_type region_type)
830 {
831     u8 base;
832 
833     switch (region_type) {
834         case PCI_REGION_TYPE_IO:
835             base = PCI_IO_BASE;
836             break;
837         case PCI_REGION_TYPE_PREFMEM:
838             base = PCI_PREF_MEMORY_BASE;
839             break;
840         default:
841             /* Regular memory support is mandatory */
842             return 1;
843     }
844 
845     pci_config_writeb(pci->bdf, base, 0xFF);
846 
847     return pci_config_readb(pci->bdf, base) != 0;
848 }
849 
pci_bios_check_devices(struct pci_bus * busses)850 static int pci_bios_check_devices(struct pci_bus *busses)
851 {
852     dprintf(1, "PCI: check devices\n");
853 
854     // Calculate resources needed for regular (non-bus) devices.
855     struct pci_device *pci;
856     foreachpci(pci) {
857         if (pci->class == PCI_CLASS_BRIDGE_PCI)
858             busses[pci->secondary_bus].bus_dev = pci;
859 
860         struct pci_bus *bus = &busses[pci_bdf_to_bus(pci->bdf)];
861         if (!bus->bus_dev)
862             /*
863              * Resources for all root busses go in busses[0]
864              */
865             bus = &busses[0];
866         int i;
867         for (i = 0; i < PCI_NUM_REGIONS; i++) {
868             if ((pci->class == PCI_CLASS_BRIDGE_PCI) &&
869                 (i >= PCI_BRIDGE_NUM_REGIONS && i < PCI_ROM_SLOT))
870                 continue;
871             int type, is64;
872             u64 size;
873             pci_bios_get_bar(pci, i, &type, &size, &is64);
874             if (size == 0)
875                 continue;
876 
877             if (type != PCI_REGION_TYPE_IO && size < PCI_DEVICE_MEM_MIN)
878                 size = PCI_DEVICE_MEM_MIN;
879             struct pci_region_entry *entry = pci_region_create_entry(
880                 bus, pci, i, size, size, type, is64);
881             if (!entry)
882                 return -1;
883 
884             if (is64)
885                 i++;
886         }
887     }
888 
889     // Propagate required bus resources to parent busses.
890     int secondary_bus;
891     for (secondary_bus=MaxPCIBus; secondary_bus>0; secondary_bus--) {
892         struct pci_bus *s = &busses[secondary_bus];
893         if (!s->bus_dev)
894             continue;
895         struct pci_bus *parent = &busses[pci_bdf_to_bus(s->bus_dev->bdf)];
896         if (!parent->bus_dev)
897             /*
898              * Resources for all root busses go in busses[0]
899              */
900             parent = &busses[0];
901         int type;
902         u16 bdf = s->bus_dev->bdf;
903         u8 pcie_cap = pci_find_capability(bdf, PCI_CAP_ID_EXP, 0);
904         u8 qemu_cap = pci_find_resource_reserve_capability(bdf);
905 
906         int hotplug_support = pci_bus_hotplug_support(s, pcie_cap);
907         for (type = 0; type < PCI_REGION_TYPE_COUNT; type++) {
908             u64 align = (type == PCI_REGION_TYPE_IO) ?
909                 PCI_BRIDGE_IO_MIN : PCI_BRIDGE_MEM_MIN;
910             if (!pci_bridge_has_region(s->bus_dev, type))
911                 continue;
912             u64 size = 0;
913             if (qemu_cap) {
914                 u32 tmp_size;
915                 u64 tmp_size_64;
916                 switch(type) {
917                 case PCI_REGION_TYPE_IO:
918                     tmp_size_64 = (pci_config_readl(bdf, qemu_cap + RES_RESERVE_IO) |
919                             (u64)pci_config_readl(bdf, qemu_cap + RES_RESERVE_IO + 4) << 32);
920                     if (tmp_size_64 != (u64)-1) {
921                         size = tmp_size_64;
922                     }
923                     break;
924                 case PCI_REGION_TYPE_MEM:
925                     tmp_size = pci_config_readl(bdf, qemu_cap + RES_RESERVE_MEM);
926                     if (tmp_size != (u32)-1) {
927                         size = tmp_size;
928                     }
929                     break;
930                 case PCI_REGION_TYPE_PREFMEM:
931                     tmp_size = pci_config_readl(bdf, qemu_cap + RES_RESERVE_PREF_MEM_32);
932                     tmp_size_64 = (pci_config_readl(bdf, qemu_cap + RES_RESERVE_PREF_MEM_64) |
933                             (u64)pci_config_readl(bdf, qemu_cap + RES_RESERVE_PREF_MEM_64 + 4) << 32);
934                     if (tmp_size != (u32)-1 && tmp_size_64 == (u64)-1) {
935                         size = tmp_size;
936                     } else if (tmp_size == (u32)-1 && tmp_size_64 != (u64)-1) {
937                         size = tmp_size_64;
938                     } else if (tmp_size != (u32)-1 && tmp_size_64 != (u64)-1) {
939                         dprintf(1, "PCI: resource reserve cap PREF32 and PREF64"
940                                 " conflict\n");
941                     }
942                     break;
943                 default:
944                     break;
945                 }
946             }
947             if (pci_region_align(&s->r[type]) > align)
948                  align = pci_region_align(&s->r[type]);
949             u64 sum = pci_region_sum(&s->r[type]);
950             int resource_optional = pcie_cap && (type == PCI_REGION_TYPE_IO);
951             if (!sum && hotplug_support && !resource_optional)
952                 sum = align; /* reserve min size for hot-plug */
953             if (size > sum) {
954                 dprintf(1, "PCI: QEMU resource reserve cap: "
955                         "size %08llx type %s\n",
956                         size, region_type_name[type]);
957                 if (type != PCI_REGION_TYPE_IO) {
958                     size = ALIGN(size, align);
959                 }
960             } else {
961                 size = ALIGN(sum, align);
962             }
963             int is64 = pci_bios_bridge_region_is64(&s->r[type],
964                                             s->bus_dev, type);
965             // entry->bar is -1 if the entry represents a bridge region
966             struct pci_region_entry *entry = pci_region_create_entry(
967                 parent, s->bus_dev, -1, size, align, type, is64);
968             if (!entry)
969                 return -1;
970             dprintf(1, "PCI: secondary bus %d size %08llx type %s\n",
971                       entry->dev->secondary_bus, size,
972                       region_type_name[entry->type]);
973         }
974     }
975     return 0;
976 }
977 
978 
979 /****************************************************************
980  * BAR assignment
981  ****************************************************************/
982 
983 // Setup region bases (given the regions' size and alignment)
pci_bios_init_root_regions_io(struct pci_bus * bus)984 static int pci_bios_init_root_regions_io(struct pci_bus *bus)
985 {
986     /*
987      * QEMU I/O address space usage:
988      *   0000 - 0fff    legacy isa, pci config, pci root bus, ...
989      *   1000 - 9fff    free
990      *   a000 - afff    hotplug (cpu, pci via acpi, i440fx/piix only)
991      *   b000 - bfff    power management (PORT_ACPI_PM_BASE)
992      *                  [ qemu 1.4+ implements pci config registers
993      *                    properly so guests can place the registers
994      *                    where they want, on older versions its fixed ]
995      *   c000 - ffff    free, traditionally used for pci io
996      */
997     struct pci_region *r_io = &bus->r[PCI_REGION_TYPE_IO];
998     u64 sum = pci_region_sum(r_io);
999     if (sum < 0x4000) {
1000         /* traditional region is big enougth, use it */
1001         r_io->base = 0xc000;
1002     } else if (sum < pci_io_low_end - 0x1000) {
1003         /* use the larger region at 0x1000 */
1004         r_io->base = 0x1000;
1005     } else {
1006         /* not enouth io address space -> error out */
1007         return -1;
1008     }
1009     dprintf(1, "PCI: IO: %4llx - %4llx\n", r_io->base, r_io->base + sum - 1);
1010     return 0;
1011 }
1012 
pci_bios_init_root_regions_mem(struct pci_bus * bus)1013 static int pci_bios_init_root_regions_mem(struct pci_bus *bus)
1014 {
1015     struct pci_region *r_end = &bus->r[PCI_REGION_TYPE_PREFMEM];
1016     struct pci_region *r_start = &bus->r[PCI_REGION_TYPE_MEM];
1017 
1018     if (pci_region_align(r_start) < pci_region_align(r_end)) {
1019         // Swap regions to improve alignment.
1020         r_end = r_start;
1021         r_start = &bus->r[PCI_REGION_TYPE_PREFMEM];
1022     }
1023     u64 sum = pci_region_sum(r_end);
1024     u64 align = pci_region_align(r_end);
1025     r_end->base = ALIGN_DOWN((pcimem_end - sum), align);
1026     sum = pci_region_sum(r_start);
1027     align = pci_region_align(r_start);
1028     r_start->base = ALIGN_DOWN((r_end->base - sum), align);
1029 
1030     if ((r_start->base < pcimem_start) ||
1031          (r_start->base > pcimem_end))
1032         // Memory range requested is larger than available.
1033         return -1;
1034     return 0;
1035 }
1036 
1037 #define PCI_IO_SHIFT            8
1038 #define PCI_MEMORY_SHIFT        16
1039 #define PCI_PREF_MEMORY_SHIFT   16
1040 
1041 static void
pci_region_map_one_entry(struct pci_region_entry * entry,u64 addr)1042 pci_region_map_one_entry(struct pci_region_entry *entry, u64 addr)
1043 {
1044     if (entry->bar >= 0) {
1045         dprintf(1, "PCI: map device bdf=%pP"
1046                 "  bar %d, addr %08llx, size %08llx [%s]\n",
1047                 entry->dev,
1048                 entry->bar, addr, entry->size, region_type_name[entry->type]);
1049 
1050         pci_set_io_region_addr(entry->dev, entry->bar, addr, entry->is64);
1051         return;
1052     }
1053 
1054     u16 bdf = entry->dev->bdf;
1055     u64 limit = addr + entry->size - 1;
1056     if (entry->type == PCI_REGION_TYPE_IO) {
1057         pci_config_writeb(bdf, PCI_IO_BASE, addr >> PCI_IO_SHIFT);
1058         pci_config_writew(bdf, PCI_IO_BASE_UPPER16, 0);
1059         pci_config_writeb(bdf, PCI_IO_LIMIT, limit >> PCI_IO_SHIFT);
1060         pci_config_writew(bdf, PCI_IO_LIMIT_UPPER16, 0);
1061     }
1062     if (entry->type == PCI_REGION_TYPE_MEM) {
1063         pci_config_writew(bdf, PCI_MEMORY_BASE, addr >> PCI_MEMORY_SHIFT);
1064         pci_config_writew(bdf, PCI_MEMORY_LIMIT, limit >> PCI_MEMORY_SHIFT);
1065     }
1066     if (entry->type == PCI_REGION_TYPE_PREFMEM) {
1067         pci_config_writew(bdf, PCI_PREF_MEMORY_BASE, addr >> PCI_PREF_MEMORY_SHIFT);
1068         pci_config_writew(bdf, PCI_PREF_MEMORY_LIMIT, limit >> PCI_PREF_MEMORY_SHIFT);
1069         pci_config_writel(bdf, PCI_PREF_BASE_UPPER32, addr >> 32);
1070         pci_config_writel(bdf, PCI_PREF_LIMIT_UPPER32, limit >> 32);
1071     }
1072 }
1073 
pci_region_map_entries(struct pci_bus * busses,struct pci_region * r)1074 static void pci_region_map_entries(struct pci_bus *busses, struct pci_region *r)
1075 {
1076     struct hlist_node *n;
1077     struct pci_region_entry *entry;
1078     hlist_for_each_entry_safe(entry, n, &r->list, node) {
1079         u64 addr = r->base;
1080         r->base += entry->size;
1081         if (entry->bar == -1)
1082             // Update bus base address if entry is a bridge region
1083             busses[entry->dev->secondary_bus].r[entry->type].base = addr;
1084         pci_region_map_one_entry(entry, addr);
1085         hlist_del(&entry->node);
1086         free(entry);
1087     }
1088 }
1089 
pci_bios_map_devices(struct pci_bus * busses)1090 static void pci_bios_map_devices(struct pci_bus *busses)
1091 {
1092     if (pci_bios_init_root_regions_io(busses))
1093         panic("PCI: out of I/O address space\n");
1094 
1095     dprintf(1, "PCI: 32: %016llx - %016llx\n", pcimem_start, pcimem_end);
1096     if (pci_bios_init_root_regions_mem(busses)) {
1097         struct pci_region r64_mem, r64_pref;
1098         r64_mem.list.first = NULL;
1099         r64_pref.list.first = NULL;
1100         pci_region_migrate_64bit_entries(&busses[0].r[PCI_REGION_TYPE_MEM],
1101                                          &r64_mem);
1102         pci_region_migrate_64bit_entries(&busses[0].r[PCI_REGION_TYPE_PREFMEM],
1103                                          &r64_pref);
1104 
1105         if (pci_bios_init_root_regions_mem(busses))
1106             panic("PCI: out of 32bit address space\n");
1107 
1108         u64 sum_mem = pci_region_sum(&r64_mem);
1109         u64 sum_pref = pci_region_sum(&r64_pref);
1110         u64 align_mem = pci_region_align(&r64_mem);
1111         u64 align_pref = pci_region_align(&r64_pref);
1112 
1113         r64_mem.base = le64_to_cpu(romfile_loadint("etc/reserved-memory-end", 0));
1114         if (r64_mem.base < 0x100000000LL + RamSizeOver4G)
1115             r64_mem.base = 0x100000000LL + RamSizeOver4G;
1116         r64_mem.base = ALIGN(r64_mem.base, align_mem);
1117         r64_mem.base = ALIGN(r64_mem.base, (1LL<<30));    // 1G hugepage
1118         r64_pref.base = r64_mem.base + sum_mem;
1119         r64_pref.base = ALIGN(r64_pref.base, align_pref);
1120         r64_pref.base = ALIGN(r64_pref.base, (1LL<<30));  // 1G hugepage
1121         pcimem64_start = r64_mem.base;
1122         pcimem64_end = r64_pref.base + sum_pref;
1123         pcimem64_end = ALIGN(pcimem64_end, (1LL<<30));    // 1G hugepage
1124         dprintf(1, "PCI: 64: %016llx - %016llx\n", pcimem64_start, pcimem64_end);
1125 
1126         pci_region_map_entries(busses, &r64_mem);
1127         pci_region_map_entries(busses, &r64_pref);
1128     } else {
1129         // no bars mapped high -> drop 64bit window (see dsdt)
1130         pcimem64_start = 0;
1131     }
1132     // Map regions on each device.
1133     int bus;
1134     for (bus = 0; bus<=MaxPCIBus; bus++) {
1135         int type;
1136         for (type = 0; type < PCI_REGION_TYPE_COUNT; type++)
1137             pci_region_map_entries(busses, &busses[bus].r[type]);
1138     }
1139 }
1140 
1141 
1142 /****************************************************************
1143  * Main setup code
1144  ****************************************************************/
1145 
1146 void
pci_setup(void)1147 pci_setup(void)
1148 {
1149     if (!CONFIG_QEMU)
1150         return;
1151 
1152     dprintf(3, "pci setup\n");
1153 
1154     dprintf(1, "=== PCI bus & bridge init ===\n");
1155     if (pci_probe_host() != 0) {
1156         return;
1157     }
1158     pci_bios_init_bus();
1159 
1160     dprintf(1, "=== PCI device probing ===\n");
1161     pci_probe_devices();
1162 
1163     pcimem_start = RamSize;
1164     pci_bios_init_platform();
1165 
1166     dprintf(1, "=== PCI new allocation pass #1 ===\n");
1167     struct pci_bus *busses = malloc_tmp(sizeof(*busses) * (MaxPCIBus + 1));
1168     if (!busses) {
1169         warn_noalloc();
1170         return;
1171     }
1172     memset(busses, 0, sizeof(*busses) * (MaxPCIBus + 1));
1173     if (pci_bios_check_devices(busses))
1174         return;
1175 
1176     dprintf(1, "=== PCI new allocation pass #2 ===\n");
1177     pci_bios_map_devices(busses);
1178 
1179     pci_bios_init_devices();
1180 
1181     free(busses);
1182 
1183     pci_enable_default_vga();
1184 }
1185