xref: /freebsd/sys/dev/hyperv/pcib/vmbus_pcib.c (revision 2baed46e85d33b1f99e6f96033acc85a9a6fbba4)
18c582c7cSDexuan Cui /*-
293b4e111SSepherosa Ziehau  * Copyright (c) 2016-2017 Microsoft Corp.
38c582c7cSDexuan Cui  * All rights reserved.
48c582c7cSDexuan Cui  *
58c582c7cSDexuan Cui  * Redistribution and use in source and binary forms, with or without
68c582c7cSDexuan Cui  * modification, are permitted provided that the following conditions
78c582c7cSDexuan Cui  * are met:
88c582c7cSDexuan Cui  * 1. Redistributions of source code must retain the above copyright
98c582c7cSDexuan Cui  *    notice, this list of conditions and the following disclaimer.
108c582c7cSDexuan Cui  * 2. Redistributions in binary form must reproduce the above copyright
118c582c7cSDexuan Cui  *    notice, this list of conditions and the following disclaimer in the
128c582c7cSDexuan Cui  *    documentation and/or other materials provided with the distribution.
138c582c7cSDexuan Cui  *
148c582c7cSDexuan Cui  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
158c582c7cSDexuan Cui  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
168c582c7cSDexuan Cui  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
178c582c7cSDexuan Cui  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
188c582c7cSDexuan Cui  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
198c582c7cSDexuan Cui  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
208c582c7cSDexuan Cui  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
218c582c7cSDexuan Cui  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
228c582c7cSDexuan Cui  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
238c582c7cSDexuan Cui  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
248c582c7cSDexuan Cui  * SUCH DAMAGE.
258c582c7cSDexuan Cui  */
268c582c7cSDexuan Cui 
278c582c7cSDexuan Cui #include <sys/cdefs.h>
28cdb316eeSDexuan Cui #ifdef NEW_PCIB
296e5b082cSWei Hu #include "opt_acpi.h"
30cdb316eeSDexuan Cui 
318c582c7cSDexuan Cui #include <sys/param.h>
328c582c7cSDexuan Cui #include <sys/systm.h>
338c582c7cSDexuan Cui #include <sys/types.h>
348c582c7cSDexuan Cui #include <sys/malloc.h>
358c582c7cSDexuan Cui #include <sys/module.h>
368c582c7cSDexuan Cui #include <sys/kernel.h>
378c582c7cSDexuan Cui #include <sys/queue.h>
388c582c7cSDexuan Cui #include <sys/lock.h>
398c582c7cSDexuan Cui #include <sys/sx.h>
408c582c7cSDexuan Cui #include <sys/smp.h>
418c582c7cSDexuan Cui #include <sys/sysctl.h>
428c582c7cSDexuan Cui #include <sys/bus.h>
438c582c7cSDexuan Cui #include <sys/rman.h>
448c582c7cSDexuan Cui #include <sys/mutex.h>
458c582c7cSDexuan Cui #include <sys/errno.h>
468c582c7cSDexuan Cui 
478c582c7cSDexuan Cui #include <vm/vm.h>
488c582c7cSDexuan Cui #include <vm/vm_param.h>
498c582c7cSDexuan Cui #include <vm/vm_kern.h>
508c582c7cSDexuan Cui #include <vm/pmap.h>
518c582c7cSDexuan Cui 
526e5b082cSWei Hu #if defined(__aarch64__)
536e5b082cSWei Hu #include <arm64/include/intr.h>
546e5b082cSWei Hu #endif
558c582c7cSDexuan Cui #include <machine/atomic.h>
568c582c7cSDexuan Cui #include <machine/bus.h>
578c582c7cSDexuan Cui #include <machine/frame.h>
588c582c7cSDexuan Cui #include <machine/pci_cfgreg.h>
598c582c7cSDexuan Cui #include <machine/resource.h>
608c582c7cSDexuan Cui 
618c582c7cSDexuan Cui #include <sys/pciio.h>
628c582c7cSDexuan Cui #include <dev/pci/pcireg.h>
638c582c7cSDexuan Cui #include <dev/pci/pcivar.h>
648c582c7cSDexuan Cui #include <dev/pci/pci_private.h>
658c582c7cSDexuan Cui #include <dev/pci/pcib_private.h>
668c582c7cSDexuan Cui #include "pcib_if.h"
676e5b082cSWei Hu #if defined(__i386__) || defined(__amd64__)
688c582c7cSDexuan Cui #include <machine/intr_machdep.h>
698c582c7cSDexuan Cui #include <x86/apicreg.h>
70999174baSWei Hu #include <x86/apicvar.h>
716e5b082cSWei Hu #endif
726e5b082cSWei Hu #if defined(__aarch64__)
736e5b082cSWei Hu #include <contrib/dev/acpica/include/acpi.h>
746e5b082cSWei Hu #include <contrib/dev/acpica/include/accommon.h>
756e5b082cSWei Hu #include <dev/acpica/acpivar.h>
766e5b082cSWei Hu #include <dev/acpica/acpi_pcibvar.h>
776e5b082cSWei Hu #endif
788c582c7cSDexuan Cui #include <dev/hyperv/include/hyperv.h>
798c582c7cSDexuan Cui #include <dev/hyperv/include/vmbus_xact.h>
808c582c7cSDexuan Cui #include <dev/hyperv/vmbus/vmbus_reg.h>
818c582c7cSDexuan Cui #include <dev/hyperv/vmbus/vmbus_chanvar.h>
828c582c7cSDexuan Cui 
838c582c7cSDexuan Cui #include "vmbus_if.h"
848c582c7cSDexuan Cui 
858c582c7cSDexuan Cui struct completion {
868c582c7cSDexuan Cui 	unsigned int done;
878c582c7cSDexuan Cui 	struct mtx lock;
888c582c7cSDexuan Cui };
898c582c7cSDexuan Cui 
908c582c7cSDexuan Cui static void
918c582c7cSDexuan Cui init_completion(struct completion *c)
928c582c7cSDexuan Cui {
938c582c7cSDexuan Cui 	memset(c, 0, sizeof(*c));
948c582c7cSDexuan Cui 	mtx_init(&c->lock, "hvcmpl", NULL, MTX_DEF);
958c582c7cSDexuan Cui 	c->done = 0;
968c582c7cSDexuan Cui }
97ea11861eSWei Hu static void
98ea11861eSWei Hu reinit_completion(struct completion *c)
99ea11861eSWei Hu {
100ea11861eSWei Hu 	c->done = 0;
101ea11861eSWei Hu }
1028c582c7cSDexuan Cui static void
1038c582c7cSDexuan Cui free_completion(struct completion *c)
1048c582c7cSDexuan Cui {
1058c582c7cSDexuan Cui 	mtx_destroy(&c->lock);
1068c582c7cSDexuan Cui }
1078c582c7cSDexuan Cui 
1088c582c7cSDexuan Cui static void
1098c582c7cSDexuan Cui complete(struct completion *c)
1108c582c7cSDexuan Cui {
1118c582c7cSDexuan Cui 	mtx_lock(&c->lock);
1128c582c7cSDexuan Cui 	c->done++;
1138c582c7cSDexuan Cui 	mtx_unlock(&c->lock);
1148c582c7cSDexuan Cui 	wakeup(c);
1158c582c7cSDexuan Cui }
1168c582c7cSDexuan Cui 
1178c582c7cSDexuan Cui static void
1188c582c7cSDexuan Cui wait_for_completion(struct completion *c)
1198c582c7cSDexuan Cui {
1208c582c7cSDexuan Cui 	mtx_lock(&c->lock);
1218c582c7cSDexuan Cui 	while (c->done == 0)
1228c582c7cSDexuan Cui 		mtx_sleep(c, &c->lock, 0, "hvwfc", 0);
1238c582c7cSDexuan Cui 	c->done--;
1248c582c7cSDexuan Cui 	mtx_unlock(&c->lock);
1258c582c7cSDexuan Cui }
1268c582c7cSDexuan Cui 
12775c2786cSWei Hu /*
12875c2786cSWei Hu  * Return: 0 if completed, a non-zero value if timed out.
12975c2786cSWei Hu  */
13075c2786cSWei Hu static int
13175c2786cSWei Hu wait_for_completion_timeout(struct completion *c, int timeout)
13275c2786cSWei Hu {
13375c2786cSWei Hu 	int ret;
13475c2786cSWei Hu 
13575c2786cSWei Hu 	mtx_lock(&c->lock);
13675c2786cSWei Hu 
13775c2786cSWei Hu 	if (c->done == 0)
13875c2786cSWei Hu 		mtx_sleep(c, &c->lock, 0, "hvwfc", timeout);
13975c2786cSWei Hu 
14075c2786cSWei Hu 	if (c->done > 0) {
14175c2786cSWei Hu 		c->done--;
14275c2786cSWei Hu 		ret = 0;
14375c2786cSWei Hu 	} else {
14475c2786cSWei Hu 		ret = 1;
14575c2786cSWei Hu 	}
14675c2786cSWei Hu 
14775c2786cSWei Hu 	mtx_unlock(&c->lock);
14875c2786cSWei Hu 
14975c2786cSWei Hu 	return (ret);
15075c2786cSWei Hu }
15175c2786cSWei Hu 
152ea11861eSWei Hu #define ARRAY_SIZE(x) (sizeof(x) / sizeof((x)[0]))
1536e5b082cSWei Hu #define PCI_MAKE_VERSION(major, minor) ((uint32_t)(((major) << 16) | (minor)))
1548c582c7cSDexuan Cui 
155ea11861eSWei Hu enum pci_protocol_version_t {
1568c582c7cSDexuan Cui 	PCI_PROTOCOL_VERSION_1_1 = PCI_MAKE_VERSION(1, 1),
157ea11861eSWei Hu 	PCI_PROTOCOL_VERSION_1_4 = PCI_MAKE_VERSION(1, 4),
158ea11861eSWei Hu };
159ea11861eSWei Hu 
160ea11861eSWei Hu static enum pci_protocol_version_t pci_protocol_versions[] = {
161ea11861eSWei Hu 	PCI_PROTOCOL_VERSION_1_4,
162ea11861eSWei Hu 	PCI_PROTOCOL_VERSION_1_1,
1638c582c7cSDexuan Cui };
1648c582c7cSDexuan Cui 
1658c582c7cSDexuan Cui #define PCI_CONFIG_MMIO_LENGTH	0x2000
1668c582c7cSDexuan Cui #define CFG_PAGE_OFFSET 0x1000
1678c582c7cSDexuan Cui #define CFG_PAGE_SIZE (PCI_CONFIG_MMIO_LENGTH - CFG_PAGE_OFFSET)
1688c582c7cSDexuan Cui 
1698c582c7cSDexuan Cui /*
1708c582c7cSDexuan Cui  * Message Types
1718c582c7cSDexuan Cui  */
1728c582c7cSDexuan Cui 
1738c582c7cSDexuan Cui enum pci_message_type {
1748c582c7cSDexuan Cui 	/*
1758c582c7cSDexuan Cui 	 * Version 1.1
1768c582c7cSDexuan Cui 	 */
1778c582c7cSDexuan Cui 	PCI_MESSAGE_BASE                = 0x42490000,
1788c582c7cSDexuan Cui 	PCI_BUS_RELATIONS               = PCI_MESSAGE_BASE + 0,
1798c582c7cSDexuan Cui 	PCI_QUERY_BUS_RELATIONS         = PCI_MESSAGE_BASE + 1,
1808c582c7cSDexuan Cui 	PCI_POWER_STATE_CHANGE          = PCI_MESSAGE_BASE + 4,
1818c582c7cSDexuan Cui 	PCI_QUERY_RESOURCE_REQUIREMENTS = PCI_MESSAGE_BASE + 5,
1828c582c7cSDexuan Cui 	PCI_QUERY_RESOURCE_RESOURCES    = PCI_MESSAGE_BASE + 6,
1838c582c7cSDexuan Cui 	PCI_BUS_D0ENTRY                 = PCI_MESSAGE_BASE + 7,
1848c582c7cSDexuan Cui 	PCI_BUS_D0EXIT                  = PCI_MESSAGE_BASE + 8,
1858c582c7cSDexuan Cui 	PCI_READ_BLOCK                  = PCI_MESSAGE_BASE + 9,
1868c582c7cSDexuan Cui 	PCI_WRITE_BLOCK                 = PCI_MESSAGE_BASE + 0xA,
1878c582c7cSDexuan Cui 	PCI_EJECT                       = PCI_MESSAGE_BASE + 0xB,
1888c582c7cSDexuan Cui 	PCI_QUERY_STOP                  = PCI_MESSAGE_BASE + 0xC,
1898c582c7cSDexuan Cui 	PCI_REENABLE                    = PCI_MESSAGE_BASE + 0xD,
1908c582c7cSDexuan Cui 	PCI_QUERY_STOP_FAILED           = PCI_MESSAGE_BASE + 0xE,
1918c582c7cSDexuan Cui 	PCI_EJECTION_COMPLETE           = PCI_MESSAGE_BASE + 0xF,
1928c582c7cSDexuan Cui 	PCI_RESOURCES_ASSIGNED          = PCI_MESSAGE_BASE + 0x10,
1938c582c7cSDexuan Cui 	PCI_RESOURCES_RELEASED          = PCI_MESSAGE_BASE + 0x11,
1948c582c7cSDexuan Cui 	PCI_INVALIDATE_BLOCK            = PCI_MESSAGE_BASE + 0x12,
1958c582c7cSDexuan Cui 	PCI_QUERY_PROTOCOL_VERSION      = PCI_MESSAGE_BASE + 0x13,
1968c582c7cSDexuan Cui 	PCI_CREATE_INTERRUPT_MESSAGE    = PCI_MESSAGE_BASE + 0x14,
1978c582c7cSDexuan Cui 	PCI_DELETE_INTERRUPT_MESSAGE    = PCI_MESSAGE_BASE + 0x15,
1986e5b082cSWei Hu 	PCI_RESOURCES_ASSIGNED2         = PCI_MESSAGE_BASE + 0x16,
1996e5b082cSWei Hu 	PCI_CREATE_INTERRUPT_MESSAGE2   = PCI_MESSAGE_BASE + 0x17,
2006e5b082cSWei Hu 	PCI_DELETE_INTERRUPT_MESSAGE2   = PCI_MESSAGE_BASE + 0x18, /* unused */
2016e5b082cSWei Hu 	PCI_BUS_RELATIONS2              = PCI_MESSAGE_BASE + 0x19,
2026e5b082cSWei Hu 	PCI_RESOURCES_ASSIGNED3         = PCI_MESSAGE_BASE + 0x1A,
2036e5b082cSWei Hu 	PCI_CREATE_INTERRUPT_MESSAGE3   = PCI_MESSAGE_BASE + 0x1B,
2048c582c7cSDexuan Cui 	PCI_MESSAGE_MAXIMUM
2058c582c7cSDexuan Cui };
2068c582c7cSDexuan Cui 
207ea11861eSWei Hu #define STATUS_REVISION_MISMATCH 0xC0000059
208ea11861eSWei Hu 
2098c582c7cSDexuan Cui /*
2108c582c7cSDexuan Cui  * Structures defining the virtual PCI Express protocol.
2118c582c7cSDexuan Cui  */
2128c582c7cSDexuan Cui 
2138c582c7cSDexuan Cui union pci_version {
2148c582c7cSDexuan Cui 	struct {
2158c582c7cSDexuan Cui 		uint16_t minor_version;
2168c582c7cSDexuan Cui 		uint16_t major_version;
2178c582c7cSDexuan Cui 	} parts;
2188c582c7cSDexuan Cui 	uint32_t version;
2198c582c7cSDexuan Cui } __packed;
2208c582c7cSDexuan Cui 
2218c582c7cSDexuan Cui /*
2228c582c7cSDexuan Cui  * This representation is the one used in Windows, which is
2238c582c7cSDexuan Cui  * what is expected when sending this back and forth with
2248c582c7cSDexuan Cui  * the Hyper-V parent partition.
2258c582c7cSDexuan Cui  */
2268c582c7cSDexuan Cui union win_slot_encoding {
2278c582c7cSDexuan Cui 	struct {
2288c582c7cSDexuan Cui 		uint32_t	slot:5;
2298c582c7cSDexuan Cui 		uint32_t	func:3;
2308c582c7cSDexuan Cui 		uint32_t	reserved:24;
2318c582c7cSDexuan Cui 	} bits;
2328c582c7cSDexuan Cui 	uint32_t val;
2338c582c7cSDexuan Cui } __packed;
2348c582c7cSDexuan Cui 
2358c582c7cSDexuan Cui struct pci_func_desc {
2368c582c7cSDexuan Cui 	uint16_t	v_id;	/* vendor ID */
2378c582c7cSDexuan Cui 	uint16_t	d_id;	/* device ID */
2388c582c7cSDexuan Cui 	uint8_t		rev;
2398c582c7cSDexuan Cui 	uint8_t		prog_intf;
2408c582c7cSDexuan Cui 	uint8_t		subclass;
2418c582c7cSDexuan Cui 	uint8_t		base_class;
2428c582c7cSDexuan Cui 	uint32_t	subsystem_id;
2438c582c7cSDexuan Cui 	union win_slot_encoding wslot;
2448c582c7cSDexuan Cui 	uint32_t	ser;	/* serial number */
2458c582c7cSDexuan Cui } __packed;
2468c582c7cSDexuan Cui 
2476e5b082cSWei Hu struct pci_func_desc2 {
2486e5b082cSWei Hu 	uint16_t	v_id;	/* vendor ID */
2496e5b082cSWei Hu 	uint16_t	d_id;	/* device ID */
2506e5b082cSWei Hu 	uint8_t		rev;
2516e5b082cSWei Hu 	uint8_t		prog_intf;
2526e5b082cSWei Hu 	uint8_t		subclass;
2536e5b082cSWei Hu 	uint8_t		base_class;
2546e5b082cSWei Hu 	uint32_t	subsystem_id;
2556e5b082cSWei Hu 	union		win_slot_encoding wslot;
2566e5b082cSWei Hu 	uint32_t	ser;	/* serial number */
2576e5b082cSWei Hu 	uint32_t	flags;
2586e5b082cSWei Hu 	uint16_t	virtual_numa_node;
2596e5b082cSWei Hu 	uint16_t	reserved;
2606e5b082cSWei Hu } __packed;
2616e5b082cSWei Hu 
2626e5b082cSWei Hu 
2638c582c7cSDexuan Cui struct hv_msi_desc {
2648c582c7cSDexuan Cui 	uint8_t		vector;
2658c582c7cSDexuan Cui 	uint8_t		delivery_mode;
2668c582c7cSDexuan Cui 	uint16_t	vector_count;
2678c582c7cSDexuan Cui 	uint32_t	reserved;
2688c582c7cSDexuan Cui 	uint64_t	cpu_mask;
2698c582c7cSDexuan Cui } __packed;
2708c582c7cSDexuan Cui 
2716e5b082cSWei Hu struct hv_msi_desc3 {
2726e5b082cSWei Hu 	uint32_t	vector;
2736e5b082cSWei Hu 	uint8_t		delivery_mode;
2746e5b082cSWei Hu 	uint8_t		reserved;
2756e5b082cSWei Hu 	uint16_t	vector_count;
2766e5b082cSWei Hu 	uint16_t	processor_count;
2776e5b082cSWei Hu 	uint16_t	processor_array[32];
2786e5b082cSWei Hu } __packed;
2796e5b082cSWei Hu 
2808c582c7cSDexuan Cui struct tran_int_desc {
2818c582c7cSDexuan Cui 	uint16_t	reserved;
2828c582c7cSDexuan Cui 	uint16_t	vector_count;
2838c582c7cSDexuan Cui 	uint32_t	data;
2848c582c7cSDexuan Cui 	uint64_t	address;
2858c582c7cSDexuan Cui } __packed;
2868c582c7cSDexuan Cui 
2878c582c7cSDexuan Cui struct pci_message {
2888c582c7cSDexuan Cui 	uint32_t type;
2898c582c7cSDexuan Cui } __packed;
2908c582c7cSDexuan Cui 
2918c582c7cSDexuan Cui struct pci_child_message {
2928c582c7cSDexuan Cui 	struct pci_message message_type;
2938c582c7cSDexuan Cui 	union win_slot_encoding wslot;
2948c582c7cSDexuan Cui } __packed;
2958c582c7cSDexuan Cui 
2968c582c7cSDexuan Cui struct pci_incoming_message {
2978c582c7cSDexuan Cui 	struct vmbus_chanpkt_hdr hdr;
2988c582c7cSDexuan Cui 	struct pci_message message_type;
2998c582c7cSDexuan Cui } __packed;
3008c582c7cSDexuan Cui 
3018c582c7cSDexuan Cui struct pci_response {
3028c582c7cSDexuan Cui 	struct vmbus_chanpkt_hdr hdr;
3038c582c7cSDexuan Cui 	int32_t status;	/* negative values are failures */
3048c582c7cSDexuan Cui } __packed;
3058c582c7cSDexuan Cui 
3068c582c7cSDexuan Cui struct pci_packet {
3078c582c7cSDexuan Cui 	void (*completion_func)(void *context, struct pci_response *resp,
3088c582c7cSDexuan Cui 	    int resp_packet_size);
3098c582c7cSDexuan Cui 	void *compl_ctxt;
3108c582c7cSDexuan Cui 
3118c582c7cSDexuan Cui 	struct pci_message message[0];
3128c582c7cSDexuan Cui };
3138c582c7cSDexuan Cui 
3148c582c7cSDexuan Cui /*
3158c582c7cSDexuan Cui  * Specific message types supporting the PCI protocol.
3168c582c7cSDexuan Cui  */
3178c582c7cSDexuan Cui 
3188c582c7cSDexuan Cui struct pci_version_request {
3198c582c7cSDexuan Cui 	struct pci_message message_type;
3208c582c7cSDexuan Cui 	uint32_t protocol_version;
3218c582c7cSDexuan Cui 	uint32_t reservedz:31;
3228c582c7cSDexuan Cui } __packed;
3238c582c7cSDexuan Cui 
3248c582c7cSDexuan Cui struct pci_bus_d0_entry {
3258c582c7cSDexuan Cui 	struct pci_message message_type;
3268c582c7cSDexuan Cui 	uint32_t reserved;
3278c582c7cSDexuan Cui 	uint64_t mmio_base;
3288c582c7cSDexuan Cui } __packed;
3298c582c7cSDexuan Cui 
3308c582c7cSDexuan Cui struct pci_bus_relations {
3318c582c7cSDexuan Cui 	struct pci_incoming_message incoming;
3328c582c7cSDexuan Cui 	uint32_t device_count;
3338c582c7cSDexuan Cui 	struct pci_func_desc func[0];
3348c582c7cSDexuan Cui } __packed;
3358c582c7cSDexuan Cui 
3366e5b082cSWei Hu struct pci_bus_relations2 {
3376e5b082cSWei Hu 	struct pci_incoming_message incoming;
3386e5b082cSWei Hu 	uint32_t device_count;
3396e5b082cSWei Hu 	struct pci_func_desc2 func[0];
3406e5b082cSWei Hu } __packed;
3416e5b082cSWei Hu 
3428c582c7cSDexuan Cui #define MAX_NUM_BARS	(PCIR_MAX_BAR_0 + 1)
3438c582c7cSDexuan Cui struct pci_q_res_req_response {
3448c582c7cSDexuan Cui 	struct vmbus_chanpkt_hdr hdr;
3458c582c7cSDexuan Cui 	int32_t status; /* negative values are failures */
3468c582c7cSDexuan Cui 	uint32_t probed_bar[MAX_NUM_BARS];
3478c582c7cSDexuan Cui } __packed;
3488c582c7cSDexuan Cui 
3498c582c7cSDexuan Cui struct pci_resources_assigned {
3508c582c7cSDexuan Cui 	struct pci_message message_type;
3518c582c7cSDexuan Cui 	union win_slot_encoding wslot;
3528c582c7cSDexuan Cui 	uint8_t memory_range[0x14][MAX_NUM_BARS]; /* unused here */
3538c582c7cSDexuan Cui 	uint32_t msi_descriptors;
3548c582c7cSDexuan Cui 	uint32_t reserved[4];
3558c582c7cSDexuan Cui } __packed;
3568c582c7cSDexuan Cui 
3576e5b082cSWei Hu struct pci_resources_assigned2 {
3586e5b082cSWei Hu 	struct pci_message message_type;
3596e5b082cSWei Hu 	union win_slot_encoding wslot;
3606e5b082cSWei Hu 	uint8_t memory_range[0x14][6];   /* not used here */
3616e5b082cSWei Hu 	uint32_t msi_descriptor_count;
3626e5b082cSWei Hu 	uint8_t reserved[70];
3636e5b082cSWei Hu } __packed;
3646e5b082cSWei Hu 
3658c582c7cSDexuan Cui struct pci_create_interrupt {
3668c582c7cSDexuan Cui 	struct pci_message message_type;
3678c582c7cSDexuan Cui 	union win_slot_encoding wslot;
3688c582c7cSDexuan Cui 	struct hv_msi_desc int_desc;
3698c582c7cSDexuan Cui } __packed;
3708c582c7cSDexuan Cui 
3716e5b082cSWei Hu struct pci_create_interrupt3 {
3726e5b082cSWei Hu 	struct pci_message message_type;
3736e5b082cSWei Hu 	union win_slot_encoding wslot;
3746e5b082cSWei Hu 	struct hv_msi_desc3 int_desc;
3756e5b082cSWei Hu } __packed;
3766e5b082cSWei Hu 
3778c582c7cSDexuan Cui struct pci_create_int_response {
3788c582c7cSDexuan Cui 	struct pci_response response;
3798c582c7cSDexuan Cui 	uint32_t reserved;
3808c582c7cSDexuan Cui 	struct tran_int_desc int_desc;
3818c582c7cSDexuan Cui } __packed;
3828c582c7cSDexuan Cui 
3838c582c7cSDexuan Cui struct pci_delete_interrupt {
3848c582c7cSDexuan Cui 	struct pci_message message_type;
3858c582c7cSDexuan Cui 	union win_slot_encoding wslot;
3868c582c7cSDexuan Cui 	struct tran_int_desc int_desc;
3878c582c7cSDexuan Cui } __packed;
3888c582c7cSDexuan Cui 
3898c582c7cSDexuan Cui struct pci_dev_incoming {
3908c582c7cSDexuan Cui 	struct pci_incoming_message incoming;
3918c582c7cSDexuan Cui 	union win_slot_encoding wslot;
3928c582c7cSDexuan Cui } __packed;
3938c582c7cSDexuan Cui 
3948c582c7cSDexuan Cui struct pci_eject_response {
3958c582c7cSDexuan Cui 	struct pci_message message_type;
3968c582c7cSDexuan Cui 	union win_slot_encoding wslot;
3978c582c7cSDexuan Cui 	uint32_t status;
3988c582c7cSDexuan Cui } __packed;
3998c582c7cSDexuan Cui 
4008c582c7cSDexuan Cui /*
4018c582c7cSDexuan Cui  * Driver specific state.
4028c582c7cSDexuan Cui  */
4038c582c7cSDexuan Cui 
4048c582c7cSDexuan Cui enum hv_pcibus_state {
4058c582c7cSDexuan Cui 	hv_pcibus_init = 0,
4068c582c7cSDexuan Cui 	hv_pcibus_installed,
4078c582c7cSDexuan Cui };
4088c582c7cSDexuan Cui 
4098c582c7cSDexuan Cui struct hv_pcibus {
4108c582c7cSDexuan Cui 	device_t pcib;
4118c582c7cSDexuan Cui 	device_t pci_bus;
4128c582c7cSDexuan Cui 	struct vmbus_pcib_softc *sc;
4138c582c7cSDexuan Cui 
4148c582c7cSDexuan Cui 	uint16_t pci_domain;
4158c582c7cSDexuan Cui 
4168c582c7cSDexuan Cui 	enum hv_pcibus_state state;
4178c582c7cSDexuan Cui 
4188c582c7cSDexuan Cui 	struct resource *cfg_res;
4198c582c7cSDexuan Cui 
4208c582c7cSDexuan Cui 	struct completion query_completion, *query_comp;
4218c582c7cSDexuan Cui 
4228c582c7cSDexuan Cui 	struct mtx config_lock; /* Avoid two threads writing index page */
4238c582c7cSDexuan Cui 	struct mtx device_list_lock;    /* Protect lists below */
4246e5b082cSWei Hu 	uint32_t protocol_version;
4258c582c7cSDexuan Cui 	TAILQ_HEAD(, hv_pci_dev) children;
4268c582c7cSDexuan Cui 	TAILQ_HEAD(, hv_dr_state) dr_list;
4278c582c7cSDexuan Cui 
4288c582c7cSDexuan Cui 	volatile int detaching;
4298c582c7cSDexuan Cui };
4308c582c7cSDexuan Cui 
4316e5b082cSWei Hu struct hv_pcidev_desc {
4326e5b082cSWei Hu 	uint16_t v_id;	/* vendor ID */
4336e5b082cSWei Hu 	uint16_t d_id;	/* device ID */
4346e5b082cSWei Hu 	uint8_t rev;
4356e5b082cSWei Hu 	uint8_t prog_intf;
4366e5b082cSWei Hu 	uint8_t subclass;
4376e5b082cSWei Hu 	uint8_t base_class;
4386e5b082cSWei Hu 	uint32_t subsystem_id;
4396e5b082cSWei Hu 	union win_slot_encoding wslot;
4406e5b082cSWei Hu 	uint32_t ser;	/* serial number */
4416e5b082cSWei Hu 	uint32_t flags;
4426e5b082cSWei Hu 	uint16_t virtual_numa_node;
4436e5b082cSWei Hu } __packed;
4446e5b082cSWei Hu 
4458c582c7cSDexuan Cui struct hv_pci_dev {
4468c582c7cSDexuan Cui 	TAILQ_ENTRY(hv_pci_dev) link;
4478c582c7cSDexuan Cui 
448ea11861eSWei Hu 	struct hv_pcidev_desc desc;
4498c582c7cSDexuan Cui 
4508c582c7cSDexuan Cui 	bool reported_missing;
4518c582c7cSDexuan Cui 
4528c582c7cSDexuan Cui 	struct hv_pcibus *hbus;
4538c582c7cSDexuan Cui 	struct task eject_task;
4548c582c7cSDexuan Cui 
4558c582c7cSDexuan Cui 	TAILQ_HEAD(, hv_irq_desc) irq_desc_list;
4568c582c7cSDexuan Cui 
4578c582c7cSDexuan Cui 	/*
4588c582c7cSDexuan Cui 	 * What would be observed if one wrote 0xFFFFFFFF to a BAR and then
4598c582c7cSDexuan Cui 	 * read it back, for each of the BAR offsets within config space.
4608c582c7cSDexuan Cui 	 */
4618c582c7cSDexuan Cui 	uint32_t probed_bar[MAX_NUM_BARS];
4628c582c7cSDexuan Cui };
4638c582c7cSDexuan Cui 
4648c582c7cSDexuan Cui /*
4658c582c7cSDexuan Cui  * Tracks "Device Relations" messages from the host, which must be both
4668c582c7cSDexuan Cui  * processed in order.
4678c582c7cSDexuan Cui  */
4688c582c7cSDexuan Cui struct hv_dr_work {
4698c582c7cSDexuan Cui 	struct task task;
4708c582c7cSDexuan Cui 	struct hv_pcibus *bus;
4718c582c7cSDexuan Cui };
4728c582c7cSDexuan Cui 
4738c582c7cSDexuan Cui struct hv_dr_state {
4748c582c7cSDexuan Cui 	TAILQ_ENTRY(hv_dr_state) link;
4758c582c7cSDexuan Cui 	uint32_t device_count;
476ea11861eSWei Hu 	struct hv_pcidev_desc func[0];
4778c582c7cSDexuan Cui };
4788c582c7cSDexuan Cui 
4798c582c7cSDexuan Cui struct hv_irq_desc {
4808c582c7cSDexuan Cui 	TAILQ_ENTRY(hv_irq_desc) link;
4818c582c7cSDexuan Cui 	struct tran_int_desc desc;
4828c582c7cSDexuan Cui 	int irq;
4838c582c7cSDexuan Cui };
4848c582c7cSDexuan Cui 
4858c582c7cSDexuan Cui #define PCI_DEVFN(slot, func)   ((((slot) & 0x1f) << 3) | ((func) & 0x07))
4868c582c7cSDexuan Cui #define PCI_SLOT(devfn)         (((devfn) >> 3) & 0x1f)
4878c582c7cSDexuan Cui #define PCI_FUNC(devfn)         ((devfn) & 0x07)
4888c582c7cSDexuan Cui 
4898c582c7cSDexuan Cui static uint32_t
4908c582c7cSDexuan Cui devfn_to_wslot(unsigned int devfn)
4918c582c7cSDexuan Cui {
4928c582c7cSDexuan Cui 	union win_slot_encoding wslot;
4938c582c7cSDexuan Cui 
4948c582c7cSDexuan Cui 	wslot.val = 0;
4958c582c7cSDexuan Cui 	wslot.bits.slot = PCI_SLOT(devfn);
4968c582c7cSDexuan Cui 	wslot.bits.func = PCI_FUNC(devfn);
4978c582c7cSDexuan Cui 
4988c582c7cSDexuan Cui 	return (wslot.val);
4998c582c7cSDexuan Cui }
5008c582c7cSDexuan Cui 
5018c582c7cSDexuan Cui static unsigned int
5028c582c7cSDexuan Cui wslot_to_devfn(uint32_t wslot)
5038c582c7cSDexuan Cui {
5048c582c7cSDexuan Cui 	union win_slot_encoding encoding;
5058c582c7cSDexuan Cui 	unsigned int slot;
5068c582c7cSDexuan Cui 	unsigned int func;
5078c582c7cSDexuan Cui 
5088c582c7cSDexuan Cui 	encoding.val = wslot;
5098c582c7cSDexuan Cui 
5108c582c7cSDexuan Cui 	slot = encoding.bits.slot;
5118c582c7cSDexuan Cui 	func = encoding.bits.func;
5128c582c7cSDexuan Cui 
5138c582c7cSDexuan Cui 	return (PCI_DEVFN(slot, func));
5148c582c7cSDexuan Cui }
5158c582c7cSDexuan Cui 
5168c582c7cSDexuan Cui struct vmbus_pcib_softc {
5178c582c7cSDexuan Cui 	struct vmbus_channel	*chan;
5188c582c7cSDexuan Cui 	void *rx_buf;
5198c582c7cSDexuan Cui 
5208c582c7cSDexuan Cui 	struct taskqueue	*taskq;
5218c582c7cSDexuan Cui 
5228c582c7cSDexuan Cui 	struct hv_pcibus	*hbus;
5238c582c7cSDexuan Cui };
5248c582c7cSDexuan Cui 
5258c582c7cSDexuan Cui /* {44C4F61D-4444-4400-9D52-802E27EDE19F} */
5268c582c7cSDexuan Cui static const struct hyperv_guid g_pass_through_dev_type = {
5278c582c7cSDexuan Cui 	.hv_guid = {0x1D, 0xF6, 0xC4, 0x44, 0x44, 0x44, 0x00, 0x44,
5288c582c7cSDexuan Cui 	    0x9D, 0x52, 0x80, 0x2E, 0x27, 0xED, 0xE1, 0x9F}
5298c582c7cSDexuan Cui };
5308c582c7cSDexuan Cui 
5318c582c7cSDexuan Cui struct hv_pci_compl {
5328c582c7cSDexuan Cui 	struct completion host_event;
5338c582c7cSDexuan Cui 	int32_t completion_status;
5348c582c7cSDexuan Cui };
5358c582c7cSDexuan Cui 
5368c582c7cSDexuan Cui struct q_res_req_compl {
5378c582c7cSDexuan Cui 	struct completion host_event;
5388c582c7cSDexuan Cui 	struct hv_pci_dev *hpdev;
5398c582c7cSDexuan Cui };
5408c582c7cSDexuan Cui 
5418c582c7cSDexuan Cui struct compose_comp_ctxt {
5428c582c7cSDexuan Cui 	struct hv_pci_compl comp_pkt;
5438c582c7cSDexuan Cui 	struct tran_int_desc int_desc;
5448c582c7cSDexuan Cui };
5458c582c7cSDexuan Cui 
54675c2786cSWei Hu /*
54775c2786cSWei Hu  * It is possible the device is revoked during initialization.
54875c2786cSWei Hu  * Check if this happens during wait.
54975c2786cSWei Hu  * Return: 0 if response arrived, ENODEV if device revoked.
55075c2786cSWei Hu  */
55175c2786cSWei Hu static int
55275c2786cSWei Hu wait_for_response(struct hv_pcibus *hbus, struct completion *c)
55375c2786cSWei Hu {
55475c2786cSWei Hu 	do {
55575c2786cSWei Hu 		if (vmbus_chan_is_revoked(hbus->sc->chan)) {
55675c2786cSWei Hu 			device_printf(hbus->pcib,
55775c2786cSWei Hu 			    "The device is revoked.\n");
55875c2786cSWei Hu 			return (ENODEV);
55975c2786cSWei Hu 		}
56075c2786cSWei Hu 	} while (wait_for_completion_timeout(c, hz /10) != 0);
56175c2786cSWei Hu 
56275c2786cSWei Hu 	return 0;
56375c2786cSWei Hu }
56475c2786cSWei Hu 
5658c582c7cSDexuan Cui static void
5668c582c7cSDexuan Cui hv_pci_generic_compl(void *context, struct pci_response *resp,
5678c582c7cSDexuan Cui     int resp_packet_size)
5688c582c7cSDexuan Cui {
5698c582c7cSDexuan Cui 	struct hv_pci_compl *comp_pkt = context;
5708c582c7cSDexuan Cui 
5718c582c7cSDexuan Cui 	if (resp_packet_size >= sizeof(struct pci_response))
5728c582c7cSDexuan Cui 		comp_pkt->completion_status = resp->status;
5738c582c7cSDexuan Cui 	else
5748c582c7cSDexuan Cui 		comp_pkt->completion_status = -1;
5758c582c7cSDexuan Cui 
5768c582c7cSDexuan Cui 	complete(&comp_pkt->host_event);
5778c582c7cSDexuan Cui }
5788c582c7cSDexuan Cui 
5798c582c7cSDexuan Cui static void
5808c582c7cSDexuan Cui q_resource_requirements(void *context, struct pci_response *resp,
5818c582c7cSDexuan Cui     int resp_packet_size)
5828c582c7cSDexuan Cui {
5838c582c7cSDexuan Cui 	struct q_res_req_compl *completion = context;
5848c582c7cSDexuan Cui 	struct pci_q_res_req_response *q_res_req =
5858c582c7cSDexuan Cui 	    (struct pci_q_res_req_response *)resp;
5868c582c7cSDexuan Cui 	int i;
5878c582c7cSDexuan Cui 
5888c582c7cSDexuan Cui 	if (resp->status < 0) {
5898c582c7cSDexuan Cui 		printf("vmbus_pcib: failed to query resource requirements\n");
5908c582c7cSDexuan Cui 	} else {
5918c582c7cSDexuan Cui 		for (i = 0; i < MAX_NUM_BARS; i++)
5928c582c7cSDexuan Cui 			completion->hpdev->probed_bar[i] =
5938c582c7cSDexuan Cui 			    q_res_req->probed_bar[i];
5948c582c7cSDexuan Cui 	}
5958c582c7cSDexuan Cui 
5968c582c7cSDexuan Cui 	complete(&completion->host_event);
5978c582c7cSDexuan Cui }
5988c582c7cSDexuan Cui 
5998c582c7cSDexuan Cui static void
6008c582c7cSDexuan Cui hv_pci_compose_compl(void *context, struct pci_response *resp,
6018c582c7cSDexuan Cui     int resp_packet_size)
6028c582c7cSDexuan Cui {
6038c582c7cSDexuan Cui 	struct compose_comp_ctxt *comp_pkt = context;
6048c582c7cSDexuan Cui 	struct pci_create_int_response *int_resp =
6058c582c7cSDexuan Cui 	    (struct pci_create_int_response *)resp;
6068c582c7cSDexuan Cui 
6078c582c7cSDexuan Cui 	comp_pkt->comp_pkt.completion_status = resp->status;
6088c582c7cSDexuan Cui 	comp_pkt->int_desc = int_resp->int_desc;
6098c582c7cSDexuan Cui 	complete(&comp_pkt->comp_pkt.host_event);
6108c582c7cSDexuan Cui }
6118c582c7cSDexuan Cui 
6128c582c7cSDexuan Cui static void
6138c582c7cSDexuan Cui hv_int_desc_free(struct hv_pci_dev *hpdev, struct hv_irq_desc *hid)
6148c582c7cSDexuan Cui {
6158c582c7cSDexuan Cui 	struct pci_delete_interrupt *int_pkt;
6168c582c7cSDexuan Cui 	struct {
6178c582c7cSDexuan Cui 		struct pci_packet pkt;
6188c582c7cSDexuan Cui 		uint8_t buffer[sizeof(struct pci_delete_interrupt)];
6198c582c7cSDexuan Cui 	} ctxt;
6208c582c7cSDexuan Cui 
6218c582c7cSDexuan Cui 	memset(&ctxt, 0, sizeof(ctxt));
6228c582c7cSDexuan Cui 	int_pkt = (struct pci_delete_interrupt *)&ctxt.pkt.message;
6238c582c7cSDexuan Cui 	int_pkt->message_type.type = PCI_DELETE_INTERRUPT_MESSAGE;
6248c582c7cSDexuan Cui 	int_pkt->wslot.val = hpdev->desc.wslot.val;
6258c582c7cSDexuan Cui 	int_pkt->int_desc = hid->desc;
6268c582c7cSDexuan Cui 
6278c582c7cSDexuan Cui 	vmbus_chan_send(hpdev->hbus->sc->chan, VMBUS_CHANPKT_TYPE_INBAND, 0,
6288c582c7cSDexuan Cui 	    int_pkt, sizeof(*int_pkt), 0);
6298c582c7cSDexuan Cui 
6308c582c7cSDexuan Cui 	free(hid, M_DEVBUF);
6318c582c7cSDexuan Cui }
6328c582c7cSDexuan Cui 
6338c582c7cSDexuan Cui static void
6348c582c7cSDexuan Cui hv_pci_delete_device(struct hv_pci_dev *hpdev)
6358c582c7cSDexuan Cui {
6368c582c7cSDexuan Cui 	struct hv_pcibus *hbus = hpdev->hbus;
6378c582c7cSDexuan Cui 	struct hv_irq_desc *hid, *tmp_hid;
6388c582c7cSDexuan Cui 	device_t pci_dev;
6398c582c7cSDexuan Cui 	int devfn;
6408c582c7cSDexuan Cui 
6418c582c7cSDexuan Cui 	devfn = wslot_to_devfn(hpdev->desc.wslot.val);
6428c582c7cSDexuan Cui 
643c6df6f53SWarner Losh 	bus_topo_lock();
6448c582c7cSDexuan Cui 
6458c582c7cSDexuan Cui 	pci_dev = pci_find_dbsf(hbus->pci_domain,
6468c582c7cSDexuan Cui 	    0, PCI_SLOT(devfn), PCI_FUNC(devfn));
6478c582c7cSDexuan Cui 	if (pci_dev)
6488c582c7cSDexuan Cui 		device_delete_child(hbus->pci_bus, pci_dev);
6498c582c7cSDexuan Cui 
650c6df6f53SWarner Losh 	bus_topo_unlock();
6518c582c7cSDexuan Cui 
6528c582c7cSDexuan Cui 	mtx_lock(&hbus->device_list_lock);
6538c582c7cSDexuan Cui 	TAILQ_REMOVE(&hbus->children, hpdev, link);
6548c582c7cSDexuan Cui 	mtx_unlock(&hbus->device_list_lock);
6558c582c7cSDexuan Cui 
6568c582c7cSDexuan Cui 	TAILQ_FOREACH_SAFE(hid, &hpdev->irq_desc_list, link, tmp_hid)
6578c582c7cSDexuan Cui 		hv_int_desc_free(hpdev, hid);
6588c582c7cSDexuan Cui 
6598c582c7cSDexuan Cui 	free(hpdev, M_DEVBUF);
6608c582c7cSDexuan Cui }
6618c582c7cSDexuan Cui 
6628c582c7cSDexuan Cui static struct hv_pci_dev *
663ea11861eSWei Hu new_pcichild_device(struct hv_pcibus *hbus, struct hv_pcidev_desc *desc)
6648c582c7cSDexuan Cui {
6658c582c7cSDexuan Cui 	struct hv_pci_dev *hpdev;
6668c582c7cSDexuan Cui 	struct pci_child_message *res_req;
6678c582c7cSDexuan Cui 	struct q_res_req_compl comp_pkt;
6688c582c7cSDexuan Cui 	struct {
6698c582c7cSDexuan Cui 		struct pci_packet pkt;
6708c582c7cSDexuan Cui 		uint8_t buffer[sizeof(struct pci_child_message)];
6718c582c7cSDexuan Cui 	} ctxt;
6728c582c7cSDexuan Cui 	int ret;
6738c582c7cSDexuan Cui 
6748c582c7cSDexuan Cui 	hpdev = malloc(sizeof(*hpdev), M_DEVBUF, M_WAITOK | M_ZERO);
6758c582c7cSDexuan Cui 	hpdev->hbus = hbus;
6768c582c7cSDexuan Cui 
6778c582c7cSDexuan Cui 	TAILQ_INIT(&hpdev->irq_desc_list);
6788c582c7cSDexuan Cui 
6798c582c7cSDexuan Cui 	init_completion(&comp_pkt.host_event);
6808c582c7cSDexuan Cui 	comp_pkt.hpdev = hpdev;
6818c582c7cSDexuan Cui 
6828c582c7cSDexuan Cui 	ctxt.pkt.compl_ctxt = &comp_pkt;
6838c582c7cSDexuan Cui 	ctxt.pkt.completion_func = q_resource_requirements;
6848c582c7cSDexuan Cui 
6858c582c7cSDexuan Cui 	res_req = (struct pci_child_message *)&ctxt.pkt.message;
6868c582c7cSDexuan Cui 	res_req->message_type.type = PCI_QUERY_RESOURCE_REQUIREMENTS;
6878c582c7cSDexuan Cui 	res_req->wslot.val = desc->wslot.val;
6888c582c7cSDexuan Cui 
6898c582c7cSDexuan Cui 	ret = vmbus_chan_send(hbus->sc->chan,
6908c582c7cSDexuan Cui 	    VMBUS_CHANPKT_TYPE_INBAND, VMBUS_CHANPKT_FLAG_RC,
691aaf13123SDimitry Andric 	    res_req, sizeof(*res_req), (uint64_t)(uintptr_t)&ctxt.pkt);
6928c582c7cSDexuan Cui 	if (ret)
6938c582c7cSDexuan Cui 		goto err;
6948c582c7cSDexuan Cui 
69575c2786cSWei Hu 	if (wait_for_response(hbus, &comp_pkt.host_event))
69675c2786cSWei Hu 		goto err;
69775c2786cSWei Hu 
6988c582c7cSDexuan Cui 	free_completion(&comp_pkt.host_event);
6998c582c7cSDexuan Cui 
7008c582c7cSDexuan Cui 	hpdev->desc = *desc;
7018c582c7cSDexuan Cui 
7028c582c7cSDexuan Cui 	mtx_lock(&hbus->device_list_lock);
7036944b2e6SDexuan Cui 	if (TAILQ_EMPTY(&hbus->children))
7046944b2e6SDexuan Cui 		hbus->pci_domain = desc->ser & 0xFFFF;
7058c582c7cSDexuan Cui 	TAILQ_INSERT_TAIL(&hbus->children, hpdev, link);
7068c582c7cSDexuan Cui 	mtx_unlock(&hbus->device_list_lock);
7078c582c7cSDexuan Cui 	return (hpdev);
7088c582c7cSDexuan Cui err:
7098c582c7cSDexuan Cui 	free_completion(&comp_pkt.host_event);
7108c582c7cSDexuan Cui 	free(hpdev, M_DEVBUF);
7118c582c7cSDexuan Cui 	return (NULL);
7128c582c7cSDexuan Cui }
7138c582c7cSDexuan Cui 
7148c582c7cSDexuan Cui static int
7158c582c7cSDexuan Cui pci_rescan(device_t dev)
7168c582c7cSDexuan Cui {
7178c582c7cSDexuan Cui 	return (BUS_RESCAN(dev));
7188c582c7cSDexuan Cui }
7198c582c7cSDexuan Cui 
7208c582c7cSDexuan Cui static void
7218c582c7cSDexuan Cui pci_devices_present_work(void *arg, int pending __unused)
7228c582c7cSDexuan Cui {
7238c582c7cSDexuan Cui 	struct hv_dr_work *dr_wrk = arg;
7248c582c7cSDexuan Cui 	struct hv_dr_state *dr = NULL;
7258c582c7cSDexuan Cui 	struct hv_pcibus *hbus;
7268c582c7cSDexuan Cui 	uint32_t child_no;
7278c582c7cSDexuan Cui 	bool found;
728ea11861eSWei Hu 	struct hv_pcidev_desc *new_desc;
7298c582c7cSDexuan Cui 	struct hv_pci_dev *hpdev, *tmp_hpdev;
7308c582c7cSDexuan Cui 	struct completion *query_comp;
7318c582c7cSDexuan Cui 	bool need_rescan = false;
7328c582c7cSDexuan Cui 
7338c582c7cSDexuan Cui 	hbus = dr_wrk->bus;
7348c582c7cSDexuan Cui 	free(dr_wrk, M_DEVBUF);
7358c582c7cSDexuan Cui 
7368c582c7cSDexuan Cui 	/* Pull this off the queue and process it if it was the last one. */
7378c582c7cSDexuan Cui 	mtx_lock(&hbus->device_list_lock);
7388c582c7cSDexuan Cui 	while (!TAILQ_EMPTY(&hbus->dr_list)) {
7398c582c7cSDexuan Cui 		dr = TAILQ_FIRST(&hbus->dr_list);
7408c582c7cSDexuan Cui 		TAILQ_REMOVE(&hbus->dr_list, dr, link);
7418c582c7cSDexuan Cui 
7428c582c7cSDexuan Cui 		/* Throw this away if the list still has stuff in it. */
7438c582c7cSDexuan Cui 		if (!TAILQ_EMPTY(&hbus->dr_list)) {
7448c582c7cSDexuan Cui 			free(dr, M_DEVBUF);
7458c582c7cSDexuan Cui 			continue;
7468c582c7cSDexuan Cui 		}
7478c582c7cSDexuan Cui 	}
7488c582c7cSDexuan Cui 	mtx_unlock(&hbus->device_list_lock);
7498c582c7cSDexuan Cui 
7508c582c7cSDexuan Cui 	if (!dr)
7518c582c7cSDexuan Cui 		return;
7528c582c7cSDexuan Cui 
7538c582c7cSDexuan Cui 	/* First, mark all existing children as reported missing. */
7548c582c7cSDexuan Cui 	mtx_lock(&hbus->device_list_lock);
7558c582c7cSDexuan Cui 	TAILQ_FOREACH(hpdev, &hbus->children, link)
7568c582c7cSDexuan Cui 		hpdev->reported_missing = true;
7578c582c7cSDexuan Cui 	mtx_unlock(&hbus->device_list_lock);
7588c582c7cSDexuan Cui 
7598c582c7cSDexuan Cui 	/* Next, add back any reported devices. */
7608c582c7cSDexuan Cui 	for (child_no = 0; child_no < dr->device_count; child_no++) {
7618c582c7cSDexuan Cui 		found = false;
7628c582c7cSDexuan Cui 		new_desc = &dr->func[child_no];
7638c582c7cSDexuan Cui 
7648c582c7cSDexuan Cui 		mtx_lock(&hbus->device_list_lock);
7658c582c7cSDexuan Cui 		TAILQ_FOREACH(hpdev, &hbus->children, link) {
7668c582c7cSDexuan Cui 			if ((hpdev->desc.wslot.val ==
7678c582c7cSDexuan Cui 			    new_desc->wslot.val) &&
7688c582c7cSDexuan Cui 			    (hpdev->desc.v_id == new_desc->v_id) &&
7698c582c7cSDexuan Cui 			    (hpdev->desc.d_id == new_desc->d_id) &&
7708c582c7cSDexuan Cui 			    (hpdev->desc.ser == new_desc->ser)) {
7718c582c7cSDexuan Cui 				hpdev->reported_missing = false;
7728c582c7cSDexuan Cui 				found = true;
7738c582c7cSDexuan Cui 				break;
7748c582c7cSDexuan Cui 			}
7758c582c7cSDexuan Cui 		}
7768c582c7cSDexuan Cui 		mtx_unlock(&hbus->device_list_lock);
7778c582c7cSDexuan Cui 
7788c582c7cSDexuan Cui 		if (!found) {
7798c582c7cSDexuan Cui 			if (!need_rescan)
7808c582c7cSDexuan Cui 				need_rescan = true;
7818c582c7cSDexuan Cui 
7828c582c7cSDexuan Cui 			hpdev = new_pcichild_device(hbus, new_desc);
7838c582c7cSDexuan Cui 			if (!hpdev)
7848c582c7cSDexuan Cui 				printf("vmbus_pcib: failed to add a child\n");
7858c582c7cSDexuan Cui 		}
7868c582c7cSDexuan Cui 	}
7878c582c7cSDexuan Cui 
7888c582c7cSDexuan Cui 	/* Remove missing device(s), if any */
7898c582c7cSDexuan Cui 	TAILQ_FOREACH_SAFE(hpdev, &hbus->children, link, tmp_hpdev) {
7908c582c7cSDexuan Cui 		if (hpdev->reported_missing)
7918c582c7cSDexuan Cui 			hv_pci_delete_device(hpdev);
7928c582c7cSDexuan Cui 	}
7938c582c7cSDexuan Cui 
7948c582c7cSDexuan Cui 	/* Rescan the bus to find any new device, if necessary. */
7958c582c7cSDexuan Cui 	if (hbus->state == hv_pcibus_installed && need_rescan)
7968c582c7cSDexuan Cui 		pci_rescan(hbus->pci_bus);
7978c582c7cSDexuan Cui 
7988c582c7cSDexuan Cui 	/* Wake up hv_pci_query_relations(), if it's waiting. */
7998c582c7cSDexuan Cui 	query_comp = hbus->query_comp;
8008c582c7cSDexuan Cui 	if (query_comp) {
8018c582c7cSDexuan Cui 		hbus->query_comp = NULL;
8028c582c7cSDexuan Cui 		complete(query_comp);
8038c582c7cSDexuan Cui 	}
8048c582c7cSDexuan Cui 
8058c582c7cSDexuan Cui 	free(dr, M_DEVBUF);
8068c582c7cSDexuan Cui }
8078c582c7cSDexuan Cui 
8088c582c7cSDexuan Cui static struct hv_pci_dev *
8098c582c7cSDexuan Cui get_pcichild_wslot(struct hv_pcibus *hbus, uint32_t wslot)
8108c582c7cSDexuan Cui {
8118c582c7cSDexuan Cui 	struct hv_pci_dev *hpdev, *ret = NULL;
8128c582c7cSDexuan Cui 
8138c582c7cSDexuan Cui 	mtx_lock(&hbus->device_list_lock);
8148c582c7cSDexuan Cui 	TAILQ_FOREACH(hpdev, &hbus->children, link) {
8158c582c7cSDexuan Cui 		if (hpdev->desc.wslot.val == wslot) {
8168c582c7cSDexuan Cui 			ret = hpdev;
8178c582c7cSDexuan Cui 			break;
8188c582c7cSDexuan Cui 		}
8198c582c7cSDexuan Cui 	}
8208c582c7cSDexuan Cui 	mtx_unlock(&hbus->device_list_lock);
8218c582c7cSDexuan Cui 
8228c582c7cSDexuan Cui 	return (ret);
8238c582c7cSDexuan Cui }
8248c582c7cSDexuan Cui 
8258c582c7cSDexuan Cui static void
8268c582c7cSDexuan Cui hv_pci_devices_present(struct hv_pcibus *hbus,
8278c582c7cSDexuan Cui     struct pci_bus_relations *relations)
8288c582c7cSDexuan Cui {
8298c582c7cSDexuan Cui 	struct hv_dr_state *dr;
8308c582c7cSDexuan Cui 	struct hv_dr_work *dr_wrk;
8318c582c7cSDexuan Cui 	unsigned long dr_size;
8328c582c7cSDexuan Cui 
8338c582c7cSDexuan Cui 	if (hbus->detaching && relations->device_count > 0)
8348c582c7cSDexuan Cui 		return;
8358c582c7cSDexuan Cui 
8368c582c7cSDexuan Cui 	dr_size = offsetof(struct hv_dr_state, func) +
8378c582c7cSDexuan Cui 	    (sizeof(struct pci_func_desc) * relations->device_count);
8388c582c7cSDexuan Cui 	dr = malloc(dr_size, M_DEVBUF, M_WAITOK | M_ZERO);
8398c582c7cSDexuan Cui 
8408c582c7cSDexuan Cui 	dr->device_count = relations->device_count;
8418c582c7cSDexuan Cui 	if (dr->device_count != 0)
8428c582c7cSDexuan Cui 		memcpy(dr->func, relations->func,
843ea11861eSWei Hu 		    sizeof(struct hv_pcidev_desc) * dr->device_count);
844ea11861eSWei Hu 
845ea11861eSWei Hu 	mtx_lock(&hbus->device_list_lock);
846ea11861eSWei Hu 	TAILQ_INSERT_TAIL(&hbus->dr_list, dr, link);
847ea11861eSWei Hu 	mtx_unlock(&hbus->device_list_lock);
848ea11861eSWei Hu 
849ea11861eSWei Hu 	dr_wrk = malloc(sizeof(*dr_wrk), M_DEVBUF, M_WAITOK | M_ZERO);
850ea11861eSWei Hu 	dr_wrk->bus = hbus;
851ea11861eSWei Hu 	TASK_INIT(&dr_wrk->task, 0, pci_devices_present_work, dr_wrk);
852ea11861eSWei Hu 	taskqueue_enqueue(hbus->sc->taskq, &dr_wrk->task);
853ea11861eSWei Hu }
854ea11861eSWei Hu 
855ea11861eSWei Hu static void
856ea11861eSWei Hu hv_pci_devices_present2(struct hv_pcibus *hbus,
857ea11861eSWei Hu     struct pci_bus_relations2 *relations)
858ea11861eSWei Hu {
859ea11861eSWei Hu 	struct hv_dr_state *dr;
860ea11861eSWei Hu 	struct hv_dr_work *dr_wrk;
861ea11861eSWei Hu 	unsigned long dr_size;
862ea11861eSWei Hu 
863ea11861eSWei Hu 	if (hbus->detaching && relations->device_count > 0)
864ea11861eSWei Hu 		return;
865ea11861eSWei Hu 
866ea11861eSWei Hu 	dr_size = offsetof(struct hv_dr_state, func) +
867ea11861eSWei Hu 	    (sizeof(struct pci_func_desc2) * relations->device_count);
868ea11861eSWei Hu 	dr = malloc(dr_size, M_DEVBUF, M_WAITOK | M_ZERO);
869ea11861eSWei Hu 
870ea11861eSWei Hu 	dr->device_count = relations->device_count;
871ea11861eSWei Hu 	if (dr->device_count != 0)
872ea11861eSWei Hu 		memcpy(dr->func, relations->func,
873ea11861eSWei Hu 		    sizeof(struct pci_func_desc2) * dr->device_count);
8748c582c7cSDexuan Cui 
8758c582c7cSDexuan Cui 	mtx_lock(&hbus->device_list_lock);
8768c582c7cSDexuan Cui 	TAILQ_INSERT_TAIL(&hbus->dr_list, dr, link);
8778c582c7cSDexuan Cui 	mtx_unlock(&hbus->device_list_lock);
8788c582c7cSDexuan Cui 
8798c582c7cSDexuan Cui 	dr_wrk = malloc(sizeof(*dr_wrk), M_DEVBUF, M_WAITOK | M_ZERO);
8808c582c7cSDexuan Cui 	dr_wrk->bus = hbus;
8818c582c7cSDexuan Cui 	TASK_INIT(&dr_wrk->task, 0, pci_devices_present_work, dr_wrk);
8828c582c7cSDexuan Cui 	taskqueue_enqueue(hbus->sc->taskq, &dr_wrk->task);
8838c582c7cSDexuan Cui }
8848c582c7cSDexuan Cui 
8858c582c7cSDexuan Cui static void
8868c582c7cSDexuan Cui hv_eject_device_work(void *arg, int pending __unused)
8878c582c7cSDexuan Cui {
8888c582c7cSDexuan Cui 	struct hv_pci_dev *hpdev = arg;
8898c582c7cSDexuan Cui 	union win_slot_encoding wslot = hpdev->desc.wslot;
8908c582c7cSDexuan Cui 	struct hv_pcibus *hbus = hpdev->hbus;
8918c582c7cSDexuan Cui 	struct pci_eject_response *eject_pkt;
8928c582c7cSDexuan Cui 	struct {
8938c582c7cSDexuan Cui 		struct pci_packet pkt;
8948c582c7cSDexuan Cui 		uint8_t buffer[sizeof(struct pci_eject_response)];
8958c582c7cSDexuan Cui 	} ctxt;
8968c582c7cSDexuan Cui 
8978c582c7cSDexuan Cui 	hv_pci_delete_device(hpdev);
8988c582c7cSDexuan Cui 
8998c582c7cSDexuan Cui 	memset(&ctxt, 0, sizeof(ctxt));
9008c582c7cSDexuan Cui 	eject_pkt = (struct pci_eject_response *)&ctxt.pkt.message;
9018c582c7cSDexuan Cui 	eject_pkt->message_type.type = PCI_EJECTION_COMPLETE;
9028c582c7cSDexuan Cui 	eject_pkt->wslot.val = wslot.val;
9038c582c7cSDexuan Cui 	vmbus_chan_send(hbus->sc->chan, VMBUS_CHANPKT_TYPE_INBAND, 0,
9048c582c7cSDexuan Cui 	    eject_pkt, sizeof(*eject_pkt), 0);
9058c582c7cSDexuan Cui }
9068c582c7cSDexuan Cui 
9078c582c7cSDexuan Cui static void
9088c582c7cSDexuan Cui hv_pci_eject_device(struct hv_pci_dev *hpdev)
9098c582c7cSDexuan Cui {
9108c582c7cSDexuan Cui 	struct hv_pcibus *hbus = hpdev->hbus;
9118c582c7cSDexuan Cui 	struct taskqueue *taskq;
9128c582c7cSDexuan Cui 
9138c582c7cSDexuan Cui 	if (hbus->detaching)
9148c582c7cSDexuan Cui 		return;
9158c582c7cSDexuan Cui 
9168c582c7cSDexuan Cui 	/*
9178c582c7cSDexuan Cui 	 * Push this task into the same taskqueue on which
9188c582c7cSDexuan Cui 	 * vmbus_pcib_attach() runs, so we're sure this task can't run
9198c582c7cSDexuan Cui 	 * concurrently with vmbus_pcib_attach().
9208c582c7cSDexuan Cui 	 */
9218c582c7cSDexuan Cui 	TASK_INIT(&hpdev->eject_task, 0, hv_eject_device_work, hpdev);
9228c582c7cSDexuan Cui 	taskq = vmbus_chan_mgmt_tq(hbus->sc->chan);
9238c582c7cSDexuan Cui 	taskqueue_enqueue(taskq, &hpdev->eject_task);
9248c582c7cSDexuan Cui }
9258c582c7cSDexuan Cui 
9268c582c7cSDexuan Cui #define PCIB_PACKET_SIZE	0x100
9278c582c7cSDexuan Cui 
9288c582c7cSDexuan Cui static void
9298c582c7cSDexuan Cui vmbus_pcib_on_channel_callback(struct vmbus_channel *chan, void *arg)
9308c582c7cSDexuan Cui {
9318c582c7cSDexuan Cui 	struct vmbus_pcib_softc *sc = arg;
9328c582c7cSDexuan Cui 	struct hv_pcibus *hbus = sc->hbus;
9338c582c7cSDexuan Cui 
9348c582c7cSDexuan Cui 	void *buffer;
9358c582c7cSDexuan Cui 	int bufferlen = PCIB_PACKET_SIZE;
9368c582c7cSDexuan Cui 
9378c582c7cSDexuan Cui 	struct pci_packet *comp_packet;
9388c582c7cSDexuan Cui 	struct pci_response *response;
9398c582c7cSDexuan Cui 	struct pci_incoming_message *new_msg;
9408c582c7cSDexuan Cui 	struct pci_bus_relations *bus_rel;
941ea11861eSWei Hu 	struct pci_bus_relations2 *bus_rel2;
9428c582c7cSDexuan Cui 	struct pci_dev_incoming *dev_msg;
9438c582c7cSDexuan Cui 	struct hv_pci_dev *hpdev;
9448c582c7cSDexuan Cui 
9458c582c7cSDexuan Cui 	buffer = sc->rx_buf;
9468c582c7cSDexuan Cui 	do {
9478c582c7cSDexuan Cui 		struct vmbus_chanpkt_hdr *pkt = buffer;
9488c582c7cSDexuan Cui 		uint32_t bytes_rxed;
9498c582c7cSDexuan Cui 		int ret;
9508c582c7cSDexuan Cui 
9518c582c7cSDexuan Cui 		bytes_rxed = bufferlen;
9528c582c7cSDexuan Cui 		ret = vmbus_chan_recv_pkt(chan, pkt, &bytes_rxed);
9538c582c7cSDexuan Cui 
9548c582c7cSDexuan Cui 		if (ret == ENOBUFS) {
9558c582c7cSDexuan Cui 			/* Handle large packet */
9568c582c7cSDexuan Cui 			if (bufferlen > PCIB_PACKET_SIZE) {
9578c582c7cSDexuan Cui 				free(buffer, M_DEVBUF);
9588c582c7cSDexuan Cui 				buffer = NULL;
9598c582c7cSDexuan Cui 			}
9608c582c7cSDexuan Cui 
9618c582c7cSDexuan Cui 			/* alloc new buffer */
962ea11861eSWei Hu 			buffer =
963ea11861eSWei Hu 			    malloc(bytes_rxed, M_DEVBUF, M_WAITOK | M_ZERO);
9648c582c7cSDexuan Cui 			bufferlen = bytes_rxed;
9658c582c7cSDexuan Cui 
9668c582c7cSDexuan Cui 			continue;
9678c582c7cSDexuan Cui 		}
9688c582c7cSDexuan Cui 
9698c582c7cSDexuan Cui 		if (ret != 0) {
9708c582c7cSDexuan Cui 			/* ignore EIO or EAGAIN */
9718c582c7cSDexuan Cui 			break;
9728c582c7cSDexuan Cui 		}
9738c582c7cSDexuan Cui 
9748c582c7cSDexuan Cui 		if (bytes_rxed <= sizeof(struct pci_response))
9758c582c7cSDexuan Cui 			continue;
9768c582c7cSDexuan Cui 
9778c582c7cSDexuan Cui 		switch (pkt->cph_type) {
9788c582c7cSDexuan Cui 		case VMBUS_CHANPKT_TYPE_COMP:
979aaf13123SDimitry Andric 			comp_packet =
980aaf13123SDimitry Andric 			    (struct pci_packet *)(uintptr_t)pkt->cph_xactid;
9818c582c7cSDexuan Cui 			response = (struct pci_response *)pkt;
9828c582c7cSDexuan Cui 			comp_packet->completion_func(comp_packet->compl_ctxt,
9838c582c7cSDexuan Cui 			    response, bytes_rxed);
9848c582c7cSDexuan Cui 			break;
9858c582c7cSDexuan Cui 		case VMBUS_CHANPKT_TYPE_INBAND:
9868c582c7cSDexuan Cui 			new_msg = (struct pci_incoming_message *)buffer;
9878c582c7cSDexuan Cui 
9888c582c7cSDexuan Cui 			switch (new_msg->message_type.type) {
9898c582c7cSDexuan Cui 			case PCI_BUS_RELATIONS:
9908c582c7cSDexuan Cui 				bus_rel = (struct pci_bus_relations *)buffer;
9918c582c7cSDexuan Cui 
9928c582c7cSDexuan Cui 				if (bus_rel->device_count == 0)
9938c582c7cSDexuan Cui 					break;
9948c582c7cSDexuan Cui 
9958c582c7cSDexuan Cui 				if (bytes_rxed <
9968c582c7cSDexuan Cui 				    offsetof(struct pci_bus_relations, func) +
9978c582c7cSDexuan Cui 				        (sizeof(struct pci_func_desc) *
9988c582c7cSDexuan Cui 				            (bus_rel->device_count)))
9998c582c7cSDexuan Cui 					break;
10008c582c7cSDexuan Cui 
10018c582c7cSDexuan Cui 				hv_pci_devices_present(hbus, bus_rel);
10028c582c7cSDexuan Cui 				break;
10038c582c7cSDexuan Cui 
1004ea11861eSWei Hu 			case PCI_BUS_RELATIONS2:
1005ea11861eSWei Hu 				bus_rel2 = (struct pci_bus_relations2 *)buffer;
1006ea11861eSWei Hu 
1007ea11861eSWei Hu 				if (bus_rel2->device_count == 0)
1008ea11861eSWei Hu 					break;
1009ea11861eSWei Hu 
1010ea11861eSWei Hu 				if (bytes_rxed <
1011ea11861eSWei Hu 				    offsetof(struct pci_bus_relations2, func) +
1012ea11861eSWei Hu 				    (sizeof(struct pci_func_desc2) *
1013ea11861eSWei Hu 				    (bus_rel2->device_count)))
1014ea11861eSWei Hu 					break;
1015ea11861eSWei Hu 
1016ea11861eSWei Hu 				hv_pci_devices_present2(hbus, bus_rel2);
1017ea11861eSWei Hu 
10188c582c7cSDexuan Cui 			case PCI_EJECT:
10198c582c7cSDexuan Cui 				dev_msg = (struct pci_dev_incoming *)buffer;
10208c582c7cSDexuan Cui 				hpdev = get_pcichild_wslot(hbus,
10218c582c7cSDexuan Cui 				    dev_msg->wslot.val);
10228c582c7cSDexuan Cui 
10238c582c7cSDexuan Cui 				if (hpdev)
10248c582c7cSDexuan Cui 					hv_pci_eject_device(hpdev);
10258c582c7cSDexuan Cui 
10268c582c7cSDexuan Cui 				break;
10278c582c7cSDexuan Cui 			default:
10288c582c7cSDexuan Cui 				printf("vmbus_pcib: Unknown msg type 0x%x\n",
10298c582c7cSDexuan Cui 				    new_msg->message_type.type);
10308c582c7cSDexuan Cui 				break;
10318c582c7cSDexuan Cui 			}
10328c582c7cSDexuan Cui 			break;
10338c582c7cSDexuan Cui 		default:
10348c582c7cSDexuan Cui 			printf("vmbus_pcib: Unknown VMBus msg type %hd\n",
10358c582c7cSDexuan Cui 			    pkt->cph_type);
10368c582c7cSDexuan Cui 			break;
10378c582c7cSDexuan Cui 		}
10388c582c7cSDexuan Cui 	} while (1);
10398c582c7cSDexuan Cui 
10408c582c7cSDexuan Cui 	if (bufferlen > PCIB_PACKET_SIZE)
10418c582c7cSDexuan Cui 		free(buffer, M_DEVBUF);
10428c582c7cSDexuan Cui }
10438c582c7cSDexuan Cui 
10448c582c7cSDexuan Cui static int
1045ea11861eSWei Hu hv_pci_protocol_negotiation(struct hv_pcibus *hbus,
1046ea11861eSWei Hu     enum pci_protocol_version_t version[],
1047ea11861eSWei Hu     int num_version)
10488c582c7cSDexuan Cui {
10498c582c7cSDexuan Cui 	struct pci_version_request *version_req;
10508c582c7cSDexuan Cui 	struct hv_pci_compl comp_pkt;
10518c582c7cSDexuan Cui 	struct {
10528c582c7cSDexuan Cui 		struct pci_packet pkt;
10538c582c7cSDexuan Cui 		uint8_t buffer[sizeof(struct pci_version_request)];
10548c582c7cSDexuan Cui 	} ctxt;
10558c582c7cSDexuan Cui 	int ret;
1056ea11861eSWei Hu 	int i;
10578c582c7cSDexuan Cui 
10588c582c7cSDexuan Cui 	init_completion(&comp_pkt.host_event);
10598c582c7cSDexuan Cui 
10608c582c7cSDexuan Cui 	ctxt.pkt.completion_func = hv_pci_generic_compl;
10618c582c7cSDexuan Cui 	ctxt.pkt.compl_ctxt = &comp_pkt;
10628c582c7cSDexuan Cui 	version_req = (struct pci_version_request *)&ctxt.pkt.message;
10638c582c7cSDexuan Cui 	version_req->message_type.type = PCI_QUERY_PROTOCOL_VERSION;
10648c582c7cSDexuan Cui 
1065ea11861eSWei Hu 	for(i=0; i< num_version; i++) {
1066ea11861eSWei Hu 		version_req->protocol_version = version[i];
1067ea11861eSWei Hu 		ret = vmbus_chan_send(hbus->sc->chan,
1068ea11861eSWei Hu 		    VMBUS_CHANPKT_TYPE_INBAND, VMBUS_CHANPKT_FLAG_RC,
1069ea11861eSWei Hu 		    version_req, sizeof(*version_req),
1070aaf13123SDimitry Andric 		    (uint64_t)(uintptr_t)&ctxt.pkt);
107175c2786cSWei Hu 		if (!ret)
107275c2786cSWei Hu 			ret = wait_for_response(hbus, &comp_pkt.host_event);
10738c582c7cSDexuan Cui 
107475c2786cSWei Hu 		if (ret) {
107575c2786cSWei Hu 			device_printf(hbus->pcib,
107675c2786cSWei Hu 				"vmbus_pcib failed to request version: %d\n",
107775c2786cSWei Hu 				ret);
107875c2786cSWei Hu 			goto out;
107975c2786cSWei Hu 		}
10808c582c7cSDexuan Cui 
1081ea11861eSWei Hu 		if (comp_pkt.completion_status >= 0) {
1082ea11861eSWei Hu 			hbus->protocol_version = version[i];
1083ea11861eSWei Hu 			device_printf(hbus->pcib,
1084ea11861eSWei Hu 				"PCI VMBus using version 0x%x\n",
1085ea11861eSWei Hu 				hbus->protocol_version);
1086ea11861eSWei Hu 			ret = 0;
1087ea11861eSWei Hu 			goto out;
1088ea11861eSWei Hu 		}
1089ea11861eSWei Hu 
1090ea11861eSWei Hu 		if (comp_pkt.completion_status != STATUS_REVISION_MISMATCH) {
10918c582c7cSDexuan Cui 			device_printf(hbus->pcib,
10928c582c7cSDexuan Cui 				"vmbus_pcib version negotiation failed: %x\n",
10938c582c7cSDexuan Cui 				comp_pkt.completion_status);
10948c582c7cSDexuan Cui 			ret = EPROTO;
1095ea11861eSWei Hu 			goto out;
10968c582c7cSDexuan Cui 		}
1097ea11861eSWei Hu 		reinit_completion(&comp_pkt.host_event);
1098ea11861eSWei Hu 	}
1099ea11861eSWei Hu 
1100ea11861eSWei Hu 	device_printf(hbus->pcib,
1101ea11861eSWei Hu 		"PCI pass-trhpugh VSP failed to find supported version\n");
11028c582c7cSDexuan Cui out:
11038c582c7cSDexuan Cui 	free_completion(&comp_pkt.host_event);
11048c582c7cSDexuan Cui 	return (ret);
11058c582c7cSDexuan Cui }
11068c582c7cSDexuan Cui 
11078c582c7cSDexuan Cui /* Ask the host to send along the list of child devices */
11088c582c7cSDexuan Cui static int
11098c582c7cSDexuan Cui hv_pci_query_relations(struct hv_pcibus *hbus)
11108c582c7cSDexuan Cui {
11118c582c7cSDexuan Cui 	struct pci_message message;
11128c582c7cSDexuan Cui 	int ret;
11138c582c7cSDexuan Cui 
11148c582c7cSDexuan Cui 	message.type = PCI_QUERY_BUS_RELATIONS;
11158c582c7cSDexuan Cui 	ret = vmbus_chan_send(hbus->sc->chan, VMBUS_CHANPKT_TYPE_INBAND, 0,
11168c582c7cSDexuan Cui 	    &message, sizeof(message), 0);
11178c582c7cSDexuan Cui 	return (ret);
11188c582c7cSDexuan Cui }
11198c582c7cSDexuan Cui 
11208c582c7cSDexuan Cui static int
11218c582c7cSDexuan Cui hv_pci_enter_d0(struct hv_pcibus *hbus)
11228c582c7cSDexuan Cui {
11238c582c7cSDexuan Cui 	struct pci_bus_d0_entry *d0_entry;
11248c582c7cSDexuan Cui 	struct hv_pci_compl comp_pkt;
11258c582c7cSDexuan Cui 	struct {
11268c582c7cSDexuan Cui 		struct pci_packet pkt;
11278c582c7cSDexuan Cui 		uint8_t buffer[sizeof(struct pci_bus_d0_entry)];
11288c582c7cSDexuan Cui 	} ctxt;
11298c582c7cSDexuan Cui 	int ret;
11308c582c7cSDexuan Cui 
11318c582c7cSDexuan Cui 	/*
11328c582c7cSDexuan Cui 	 * Tell the host that the bus is ready to use, and moved into the
11338c582c7cSDexuan Cui 	 * powered-on state.  This includes telling the host which region
11348c582c7cSDexuan Cui 	 * of memory-mapped I/O space has been chosen for configuration space
11358c582c7cSDexuan Cui 	 * access.
11368c582c7cSDexuan Cui 	 */
11378c582c7cSDexuan Cui 	init_completion(&comp_pkt.host_event);
11388c582c7cSDexuan Cui 
11398c582c7cSDexuan Cui 	ctxt.pkt.completion_func = hv_pci_generic_compl;
11408c582c7cSDexuan Cui 	ctxt.pkt.compl_ctxt = &comp_pkt;
11418c582c7cSDexuan Cui 
11428c582c7cSDexuan Cui 	d0_entry = (struct pci_bus_d0_entry *)&ctxt.pkt.message;
11438c582c7cSDexuan Cui 	memset(d0_entry, 0, sizeof(*d0_entry));
11448c582c7cSDexuan Cui 	d0_entry->message_type.type = PCI_BUS_D0ENTRY;
11458c582c7cSDexuan Cui 	d0_entry->mmio_base = rman_get_start(hbus->cfg_res);
11468c582c7cSDexuan Cui 
11478c582c7cSDexuan Cui 	ret = vmbus_chan_send(hbus->sc->chan, VMBUS_CHANPKT_TYPE_INBAND,
11488c582c7cSDexuan Cui 	    VMBUS_CHANPKT_FLAG_RC, d0_entry, sizeof(*d0_entry),
1149aaf13123SDimitry Andric 	    (uint64_t)(uintptr_t)&ctxt.pkt);
115075c2786cSWei Hu 	if (!ret)
115175c2786cSWei Hu 		ret = wait_for_response(hbus, &comp_pkt.host_event);
115275c2786cSWei Hu 
11538c582c7cSDexuan Cui 	if (ret)
11548c582c7cSDexuan Cui 		goto out;
11558c582c7cSDexuan Cui 
11568c582c7cSDexuan Cui 	if (comp_pkt.completion_status < 0) {
11578c582c7cSDexuan Cui 		device_printf(hbus->pcib, "vmbus_pcib failed to enable D0\n");
11588c582c7cSDexuan Cui 		ret = EPROTO;
11598c582c7cSDexuan Cui 	} else {
11608c582c7cSDexuan Cui 		ret = 0;
11618c582c7cSDexuan Cui 	}
11628c582c7cSDexuan Cui 
11638c582c7cSDexuan Cui out:
11648c582c7cSDexuan Cui 	free_completion(&comp_pkt.host_event);
11658c582c7cSDexuan Cui 	return (ret);
11668c582c7cSDexuan Cui }
11678c582c7cSDexuan Cui 
11688c582c7cSDexuan Cui /*
11698c582c7cSDexuan Cui  * It looks this is only needed by Windows VM, but let's send the message too
11708c582c7cSDexuan Cui  * just to make the host happy.
11718c582c7cSDexuan Cui  */
11728c582c7cSDexuan Cui static int
11738c582c7cSDexuan Cui hv_send_resources_allocated(struct hv_pcibus *hbus)
11748c582c7cSDexuan Cui {
11758c582c7cSDexuan Cui 	struct pci_resources_assigned *res_assigned;
1176ea11861eSWei Hu 	struct pci_resources_assigned2 *res_assigned2;
11778c582c7cSDexuan Cui 	struct hv_pci_compl comp_pkt;
11788c582c7cSDexuan Cui 	struct hv_pci_dev *hpdev;
11798c582c7cSDexuan Cui 	struct pci_packet *pkt;
11808c582c7cSDexuan Cui 	uint32_t wslot;
11818c582c7cSDexuan Cui 	int ret = 0;
1182ea11861eSWei Hu 	size_t size_res;
11838c582c7cSDexuan Cui 
1184ea11861eSWei Hu 	size_res = (hbus->protocol_version < PCI_PROTOCOL_VERSION_1_4)
1185ea11861eSWei Hu 			? sizeof(*res_assigned) : sizeof(*res_assigned2);
1186ea11861eSWei Hu 	pkt = malloc(sizeof(*pkt) + size_res,
11878c582c7cSDexuan Cui 	    M_DEVBUF, M_WAITOK | M_ZERO);
11888c582c7cSDexuan Cui 
11898c582c7cSDexuan Cui 	for (wslot = 0; wslot < 256; wslot++) {
11908c582c7cSDexuan Cui 		hpdev = get_pcichild_wslot(hbus, wslot);
11918c582c7cSDexuan Cui 		if (!hpdev)
11928c582c7cSDexuan Cui 			continue;
11938c582c7cSDexuan Cui 
11948c582c7cSDexuan Cui 		init_completion(&comp_pkt.host_event);
11958c582c7cSDexuan Cui 
1196ea11861eSWei Hu 		memset(pkt, 0, sizeof(*pkt) + size_res);
11978c582c7cSDexuan Cui 		pkt->completion_func = hv_pci_generic_compl;
11988c582c7cSDexuan Cui 		pkt->compl_ctxt = &comp_pkt;
11998c582c7cSDexuan Cui 
1200ea11861eSWei Hu 		if (hbus->protocol_version < PCI_PROTOCOL_VERSION_1_4) {
1201ea11861eSWei Hu 			res_assigned =
1202ea11861eSWei Hu 			    (struct pci_resources_assigned *)&pkt->message;
1203ea11861eSWei Hu 			res_assigned->message_type.type =
1204ea11861eSWei Hu 			    PCI_RESOURCES_ASSIGNED;
12058c582c7cSDexuan Cui 			res_assigned->wslot.val = hpdev->desc.wslot.val;
1206ea11861eSWei Hu 		} else {
1207ea11861eSWei Hu 			res_assigned2 =
1208ea11861eSWei Hu 			    (struct pci_resources_assigned2 *)&pkt->message;
1209ea11861eSWei Hu 			res_assigned2->message_type.type =
1210ea11861eSWei Hu 			    PCI_RESOURCES_ASSIGNED2;
1211ea11861eSWei Hu 			res_assigned2->wslot.val = hpdev->desc.wslot.val;
1212ea11861eSWei Hu 		}
12138c582c7cSDexuan Cui 
12148c582c7cSDexuan Cui 		ret = vmbus_chan_send(hbus->sc->chan,
12158c582c7cSDexuan Cui 		    VMBUS_CHANPKT_TYPE_INBAND, VMBUS_CHANPKT_FLAG_RC,
1216ea11861eSWei Hu 		    &pkt->message, size_res,
1217aaf13123SDimitry Andric 		    (uint64_t)(uintptr_t)pkt);
121875c2786cSWei Hu 		if (!ret)
121975c2786cSWei Hu 			ret = wait_for_response(hbus, &comp_pkt.host_event);
12208c582c7cSDexuan Cui 
12218c582c7cSDexuan Cui 		free_completion(&comp_pkt.host_event);
12228c582c7cSDexuan Cui 
122375c2786cSWei Hu 		if (ret)
122475c2786cSWei Hu 			break;
122575c2786cSWei Hu 
12268c582c7cSDexuan Cui 		if (comp_pkt.completion_status < 0) {
12278c582c7cSDexuan Cui 			ret = EPROTO;
12288c582c7cSDexuan Cui 			device_printf(hbus->pcib,
12298c582c7cSDexuan Cui 			    "failed to send PCI_RESOURCES_ASSIGNED\n");
12308c582c7cSDexuan Cui 			break;
12318c582c7cSDexuan Cui 		}
12328c582c7cSDexuan Cui 	}
12338c582c7cSDexuan Cui 
12348c582c7cSDexuan Cui 	free(pkt, M_DEVBUF);
12358c582c7cSDexuan Cui 	return (ret);
12368c582c7cSDexuan Cui }
12378c582c7cSDexuan Cui 
12388c582c7cSDexuan Cui static int
12398c582c7cSDexuan Cui hv_send_resources_released(struct hv_pcibus *hbus)
12408c582c7cSDexuan Cui {
12418c582c7cSDexuan Cui 	struct pci_child_message pkt;
12428c582c7cSDexuan Cui 	struct hv_pci_dev *hpdev;
12438c582c7cSDexuan Cui 	uint32_t wslot;
12448c582c7cSDexuan Cui 	int ret;
12458c582c7cSDexuan Cui 
12468c582c7cSDexuan Cui 	for (wslot = 0; wslot < 256; wslot++) {
12478c582c7cSDexuan Cui 		hpdev = get_pcichild_wslot(hbus, wslot);
12488c582c7cSDexuan Cui 		if (!hpdev)
12498c582c7cSDexuan Cui 			continue;
12508c582c7cSDexuan Cui 
12518c582c7cSDexuan Cui 		pkt.message_type.type = PCI_RESOURCES_RELEASED;
12528c582c7cSDexuan Cui 		pkt.wslot.val = hpdev->desc.wslot.val;
12538c582c7cSDexuan Cui 
12548c582c7cSDexuan Cui 		ret = vmbus_chan_send(hbus->sc->chan,
12558c582c7cSDexuan Cui 		    VMBUS_CHANPKT_TYPE_INBAND, 0, &pkt, sizeof(pkt), 0);
12568c582c7cSDexuan Cui 		if (ret)
12578c582c7cSDexuan Cui 			return (ret);
12588c582c7cSDexuan Cui 	}
12598c582c7cSDexuan Cui 
12608c582c7cSDexuan Cui 	return (0);
12618c582c7cSDexuan Cui }
12628c582c7cSDexuan Cui 
12638c582c7cSDexuan Cui #define hv_cfg_read(x, s)						\
12648c582c7cSDexuan Cui static inline uint##x##_t hv_cfg_read_##s(struct hv_pcibus *bus,	\
12658c582c7cSDexuan Cui     bus_size_t offset)							\
12668c582c7cSDexuan Cui {									\
12678c582c7cSDexuan Cui 	return (bus_read_##s(bus->cfg_res, offset));			\
12688c582c7cSDexuan Cui }
12698c582c7cSDexuan Cui 
12708c582c7cSDexuan Cui #define hv_cfg_write(x, s)						\
12718c582c7cSDexuan Cui static inline void hv_cfg_write_##s(struct hv_pcibus *bus,		\
12728c582c7cSDexuan Cui     bus_size_t offset, uint##x##_t val)					\
12738c582c7cSDexuan Cui {									\
12748c582c7cSDexuan Cui 	return (bus_write_##s(bus->cfg_res, offset, val));		\
12758c582c7cSDexuan Cui }
12768c582c7cSDexuan Cui 
12778c582c7cSDexuan Cui hv_cfg_read(8, 1)
12788c582c7cSDexuan Cui hv_cfg_read(16, 2)
12798c582c7cSDexuan Cui hv_cfg_read(32, 4)
12808c582c7cSDexuan Cui 
12818c582c7cSDexuan Cui hv_cfg_write(8, 1)
12828c582c7cSDexuan Cui hv_cfg_write(16, 2)
12838c582c7cSDexuan Cui hv_cfg_write(32, 4)
12848c582c7cSDexuan Cui 
12858c582c7cSDexuan Cui static void
12868c582c7cSDexuan Cui _hv_pcifront_read_config(struct hv_pci_dev *hpdev, int where, int size,
12878c582c7cSDexuan Cui     uint32_t *val)
12888c582c7cSDexuan Cui {
12898c582c7cSDexuan Cui 	struct hv_pcibus *hbus = hpdev->hbus;
12908c582c7cSDexuan Cui 	bus_size_t addr = CFG_PAGE_OFFSET + where;
12918c582c7cSDexuan Cui 
12928c582c7cSDexuan Cui 	/*
12938c582c7cSDexuan Cui 	 * If the attempt is to read the IDs or the ROM BAR, simulate that.
12948c582c7cSDexuan Cui 	 */
12958c582c7cSDexuan Cui 	if (where + size <= PCIR_COMMAND) {
12968c582c7cSDexuan Cui 		memcpy(val, ((uint8_t *)&hpdev->desc.v_id) + where, size);
12978c582c7cSDexuan Cui 	} else if (where >= PCIR_REVID && where + size <=
12988c582c7cSDexuan Cui 		   PCIR_CACHELNSZ) {
12998c582c7cSDexuan Cui 		memcpy(val, ((uint8_t *)&hpdev->desc.rev) + where -
13008c582c7cSDexuan Cui 		       PCIR_REVID, size);
13018c582c7cSDexuan Cui 	} else if (where >= PCIR_SUBVEND_0 && where + size <=
13028c582c7cSDexuan Cui 		   PCIR_BIOS) {
13038c582c7cSDexuan Cui 		memcpy(val, (uint8_t *)&hpdev->desc.subsystem_id + where -
13048c582c7cSDexuan Cui 		       PCIR_SUBVEND_0, size);
13058c582c7cSDexuan Cui 	} else if (where >= PCIR_BIOS && where + size <=
13068c582c7cSDexuan Cui 		   PCIR_CAP_PTR) {
13078c582c7cSDexuan Cui 		/* ROM BARs are unimplemented */
13088c582c7cSDexuan Cui 		*val = 0;
13098c582c7cSDexuan Cui 	} else if ((where >= PCIR_INTLINE && where + size <=
13108c582c7cSDexuan Cui 		   PCIR_INTPIN) ||(where == PCIR_INTPIN && size == 1)) {
13118c582c7cSDexuan Cui 		/*
13128c582c7cSDexuan Cui 		 * Interrupt Line and Interrupt PIN are hard-wired to zero
13138c582c7cSDexuan Cui 		 * because this front-end only supports message-signaled
13148c582c7cSDexuan Cui 		 * interrupts.
13158c582c7cSDexuan Cui 		 */
13168c582c7cSDexuan Cui 		*val = 0;
13178c582c7cSDexuan Cui 	} else if (where + size <= CFG_PAGE_SIZE) {
13188c582c7cSDexuan Cui 		mtx_lock(&hbus->config_lock);
13198c582c7cSDexuan Cui 
13208c582c7cSDexuan Cui 		/* Choose the function to be read. */
13218c582c7cSDexuan Cui 		hv_cfg_write_4(hbus, 0, hpdev->desc.wslot.val);
13228c582c7cSDexuan Cui 
13238c582c7cSDexuan Cui 		/* Make sure the function was chosen before we start reading.*/
13248c582c7cSDexuan Cui 		mb();
13258c582c7cSDexuan Cui 
13268c582c7cSDexuan Cui 		/* Read from that function's config space. */
13278c582c7cSDexuan Cui 		switch (size) {
13288c582c7cSDexuan Cui 		case 1:
13298c582c7cSDexuan Cui 			*((uint8_t *)val) = hv_cfg_read_1(hbus, addr);
13308c582c7cSDexuan Cui 			break;
13318c582c7cSDexuan Cui 		case 2:
13328c582c7cSDexuan Cui 			*((uint16_t *)val) = hv_cfg_read_2(hbus, addr);
13338c582c7cSDexuan Cui 			break;
13348c582c7cSDexuan Cui 		default:
13358c582c7cSDexuan Cui 			*((uint32_t *)val) = hv_cfg_read_4(hbus, addr);
13368c582c7cSDexuan Cui 			break;
13378c582c7cSDexuan Cui 		}
13388c582c7cSDexuan Cui 		/*
13398c582c7cSDexuan Cui 		 * Make sure the write was done before we release the lock,
13408c582c7cSDexuan Cui 		 * allowing consecutive reads/writes.
13418c582c7cSDexuan Cui 		 */
13428c582c7cSDexuan Cui 		mb();
13438c582c7cSDexuan Cui 
13448c582c7cSDexuan Cui 		mtx_unlock(&hbus->config_lock);
13458c582c7cSDexuan Cui 	} else {
13468c582c7cSDexuan Cui 		/* Invalid config read: it's unlikely to reach here. */
13478c582c7cSDexuan Cui 		memset(val, 0, size);
13488c582c7cSDexuan Cui 	}
13498c582c7cSDexuan Cui }
13508c582c7cSDexuan Cui 
13518c582c7cSDexuan Cui static void
13528c582c7cSDexuan Cui _hv_pcifront_write_config(struct hv_pci_dev *hpdev, int where, int size,
13538c582c7cSDexuan Cui     uint32_t val)
13548c582c7cSDexuan Cui {
13558c582c7cSDexuan Cui 	struct hv_pcibus *hbus = hpdev->hbus;
13568c582c7cSDexuan Cui 	bus_size_t addr = CFG_PAGE_OFFSET + where;
13578c582c7cSDexuan Cui 
13588c582c7cSDexuan Cui 	/* SSIDs and ROM BARs are read-only */
13598c582c7cSDexuan Cui 	if (where >= PCIR_SUBVEND_0 && where + size <= PCIR_CAP_PTR)
13608c582c7cSDexuan Cui 		return;
13618c582c7cSDexuan Cui 
13628c582c7cSDexuan Cui 	if (where >= PCIR_COMMAND && where + size <= CFG_PAGE_SIZE) {
13638c582c7cSDexuan Cui 		mtx_lock(&hbus->config_lock);
13648c582c7cSDexuan Cui 
13658c582c7cSDexuan Cui 		/* Choose the function to be written. */
13668c582c7cSDexuan Cui 		hv_cfg_write_4(hbus, 0, hpdev->desc.wslot.val);
13678c582c7cSDexuan Cui 
13688c582c7cSDexuan Cui 		/* Make sure the function was chosen before we start writing.*/
13698c582c7cSDexuan Cui 		wmb();
13708c582c7cSDexuan Cui 
13718c582c7cSDexuan Cui 		/* Write to that function's config space. */
13728c582c7cSDexuan Cui 		switch (size) {
13738c582c7cSDexuan Cui 		case 1:
13748c582c7cSDexuan Cui 			hv_cfg_write_1(hbus, addr, (uint8_t)val);
13758c582c7cSDexuan Cui 			break;
13768c582c7cSDexuan Cui 		case 2:
13778c582c7cSDexuan Cui 			hv_cfg_write_2(hbus, addr, (uint16_t)val);
13788c582c7cSDexuan Cui 			break;
13798c582c7cSDexuan Cui 		default:
13808c582c7cSDexuan Cui 			hv_cfg_write_4(hbus, addr, (uint32_t)val);
13818c582c7cSDexuan Cui 			break;
13828c582c7cSDexuan Cui 		}
13838c582c7cSDexuan Cui 
13848c582c7cSDexuan Cui 		/*
13858c582c7cSDexuan Cui 		 * Make sure the write was done before we release the lock,
13868c582c7cSDexuan Cui 		 * allowing consecutive reads/writes.
13878c582c7cSDexuan Cui 		 */
13888c582c7cSDexuan Cui 		mb();
13898c582c7cSDexuan Cui 
13908c582c7cSDexuan Cui 		mtx_unlock(&hbus->config_lock);
13918c582c7cSDexuan Cui 	} else {
13928c582c7cSDexuan Cui 		/* Invalid config write: it's unlikely to reach here. */
13938c582c7cSDexuan Cui 		return;
13948c582c7cSDexuan Cui 	}
13958c582c7cSDexuan Cui }
13968c582c7cSDexuan Cui 
139775412a52SWei Hu /*
139875412a52SWei Hu  * The vPCI in some Hyper-V releases do not initialize the last 4
139975412a52SWei Hu  * bit of BAR registers. This could result weird problems causing PCI
140075412a52SWei Hu  * code fail to configure BAR correctly.
140175412a52SWei Hu  *
140275412a52SWei Hu  * Just write all 1's to those BARs whose probed values are not zero.
140375412a52SWei Hu  * This seems to make the Hyper-V vPCI and pci_write_bar() to cooperate
140475412a52SWei Hu  * correctly.
140575412a52SWei Hu  */
140675412a52SWei Hu 
140775412a52SWei Hu static void
140875412a52SWei Hu vmbus_pcib_prepopulate_bars(struct hv_pcibus *hbus)
140975412a52SWei Hu {
141075412a52SWei Hu 	struct hv_pci_dev *hpdev;
141175412a52SWei Hu 	int i;
141275412a52SWei Hu 
141375412a52SWei Hu 	mtx_lock(&hbus->device_list_lock);
141475412a52SWei Hu 	TAILQ_FOREACH(hpdev, &hbus->children, link) {
141575412a52SWei Hu 		for (i = 0; i < 6; i++) {
141675412a52SWei Hu 			/* Ignore empty bar */
141775412a52SWei Hu 			if (hpdev->probed_bar[i] == 0)
141875412a52SWei Hu 				continue;
141975412a52SWei Hu 
142075412a52SWei Hu 			uint32_t bar_val = 0;
142175412a52SWei Hu 
142275412a52SWei Hu 			_hv_pcifront_read_config(hpdev, PCIR_BAR(i),
142375412a52SWei Hu 			    4, &bar_val);
142475412a52SWei Hu 
142575412a52SWei Hu 			if (hpdev->probed_bar[i] != bar_val) {
142675412a52SWei Hu 				if (bootverbose)
142775412a52SWei Hu 					printf("vmbus_pcib: initialize bar %d "
142875412a52SWei Hu 					    "by writing all 1s\n", i);
142975412a52SWei Hu 
143075412a52SWei Hu 				_hv_pcifront_write_config(hpdev, PCIR_BAR(i),
143175412a52SWei Hu 				    4, 0xffffffff);
14325473dee7SWei Hu 
14335473dee7SWei Hu 				/* Now write the original value back */
14345473dee7SWei Hu 				_hv_pcifront_write_config(hpdev, PCIR_BAR(i),
14355473dee7SWei Hu 				    4, bar_val);
143675412a52SWei Hu 			}
143775412a52SWei Hu 		}
143875412a52SWei Hu 	}
143975412a52SWei Hu 	mtx_unlock(&hbus->device_list_lock);
144075412a52SWei Hu }
144175412a52SWei Hu 
14428c582c7cSDexuan Cui static void
14438c582c7cSDexuan Cui vmbus_pcib_set_detaching(void *arg, int pending __unused)
14448c582c7cSDexuan Cui {
14458c582c7cSDexuan Cui 	struct hv_pcibus *hbus = arg;
14468c582c7cSDexuan Cui 
14478c582c7cSDexuan Cui 	atomic_set_int(&hbus->detaching, 1);
14488c582c7cSDexuan Cui }
14498c582c7cSDexuan Cui 
14508c582c7cSDexuan Cui static void
14518c582c7cSDexuan Cui vmbus_pcib_pre_detach(struct hv_pcibus *hbus)
14528c582c7cSDexuan Cui {
14538c582c7cSDexuan Cui 	struct task task;
14548c582c7cSDexuan Cui 
14558c582c7cSDexuan Cui 	TASK_INIT(&task, 0, vmbus_pcib_set_detaching, hbus);
14568c582c7cSDexuan Cui 
14578c582c7cSDexuan Cui 	/*
14588c582c7cSDexuan Cui 	 * Make sure the channel callback won't push any possible new
14598c582c7cSDexuan Cui 	 * PCI_BUS_RELATIONS and PCI_EJECT tasks to sc->taskq.
14608c582c7cSDexuan Cui 	 */
14618c582c7cSDexuan Cui 	vmbus_chan_run_task(hbus->sc->chan, &task);
14628c582c7cSDexuan Cui 
14638c582c7cSDexuan Cui 	taskqueue_drain_all(hbus->sc->taskq);
14648c582c7cSDexuan Cui }
14658c582c7cSDexuan Cui 
14668c582c7cSDexuan Cui 
14678c582c7cSDexuan Cui /*
14688c582c7cSDexuan Cui  * Standard probe entry point.
14698c582c7cSDexuan Cui  *
14708c582c7cSDexuan Cui  */
14718c582c7cSDexuan Cui static int
14728c582c7cSDexuan Cui vmbus_pcib_probe(device_t dev)
14738c582c7cSDexuan Cui {
14748c582c7cSDexuan Cui 	if (VMBUS_PROBE_GUID(device_get_parent(dev), dev,
14758c582c7cSDexuan Cui 	    &g_pass_through_dev_type) == 0) {
14768c582c7cSDexuan Cui 		device_set_desc(dev, "Hyper-V PCI Express Pass Through");
14778c582c7cSDexuan Cui 		return (BUS_PROBE_DEFAULT);
14788c582c7cSDexuan Cui 	}
14798c582c7cSDexuan Cui 	return (ENXIO);
14808c582c7cSDexuan Cui }
14818c582c7cSDexuan Cui 
14828c582c7cSDexuan Cui /*
14838c582c7cSDexuan Cui  * Standard attach entry point.
14848c582c7cSDexuan Cui  *
14858c582c7cSDexuan Cui  */
14868c582c7cSDexuan Cui static int
14878c582c7cSDexuan Cui vmbus_pcib_attach(device_t dev)
14888c582c7cSDexuan Cui {
14898c582c7cSDexuan Cui 	const int pci_ring_size = (4 * PAGE_SIZE);
14908c582c7cSDexuan Cui 	const struct hyperv_guid *inst_guid;
14918c582c7cSDexuan Cui 	struct vmbus_channel *channel;
14928c582c7cSDexuan Cui 	struct vmbus_pcib_softc *sc;
14938c582c7cSDexuan Cui 	struct hv_pcibus *hbus;
14948c582c7cSDexuan Cui 	int rid = 0;
14958c582c7cSDexuan Cui 	int ret;
14968c582c7cSDexuan Cui 
14978c582c7cSDexuan Cui 	hbus = malloc(sizeof(*hbus), M_DEVBUF, M_WAITOK | M_ZERO);
14988c582c7cSDexuan Cui 	hbus->pcib = dev;
14998c582c7cSDexuan Cui 
15008c582c7cSDexuan Cui 	channel = vmbus_get_channel(dev);
15018c582c7cSDexuan Cui 	inst_guid = vmbus_chan_guid_inst(channel);
15028c582c7cSDexuan Cui 	hbus->pci_domain = inst_guid->hv_guid[9] |
15038c582c7cSDexuan Cui 			  (inst_guid->hv_guid[8] << 8);
15048c582c7cSDexuan Cui 
15058c582c7cSDexuan Cui 	mtx_init(&hbus->config_lock, "hbcfg", NULL, MTX_DEF);
15068c582c7cSDexuan Cui 	mtx_init(&hbus->device_list_lock, "hbdl", NULL, MTX_DEF);
15078c582c7cSDexuan Cui 	TAILQ_INIT(&hbus->children);
15088c582c7cSDexuan Cui 	TAILQ_INIT(&hbus->dr_list);
15098c582c7cSDexuan Cui 
15108c582c7cSDexuan Cui 	hbus->cfg_res = bus_alloc_resource(dev, SYS_RES_MEMORY, &rid,
15118c582c7cSDexuan Cui 	    0, RM_MAX_END, PCI_CONFIG_MMIO_LENGTH,
15128c582c7cSDexuan Cui 	    RF_ACTIVE | rman_make_alignment_flags(PAGE_SIZE));
15138c582c7cSDexuan Cui 
15148c582c7cSDexuan Cui 	if (!hbus->cfg_res) {
15158c582c7cSDexuan Cui 		device_printf(dev, "failed to get resource for cfg window\n");
15168c582c7cSDexuan Cui 		ret = ENXIO;
15178c582c7cSDexuan Cui 		goto free_bus;
15188c582c7cSDexuan Cui 	}
15198c582c7cSDexuan Cui 
15208c582c7cSDexuan Cui 	sc = device_get_softc(dev);
15218c582c7cSDexuan Cui 	sc->chan = channel;
15228c582c7cSDexuan Cui 	sc->rx_buf = malloc(PCIB_PACKET_SIZE, M_DEVBUF, M_WAITOK | M_ZERO);
15238c582c7cSDexuan Cui 	sc->hbus = hbus;
15248c582c7cSDexuan Cui 
15258c582c7cSDexuan Cui 	/*
15268c582c7cSDexuan Cui 	 * The taskq is used to handle PCI_BUS_RELATIONS and PCI_EJECT
15278c582c7cSDexuan Cui 	 * messages. NB: we can't handle the messages in the channel callback
15288c582c7cSDexuan Cui 	 * directly, because the message handlers need to send new messages
15298c582c7cSDexuan Cui 	 * to the host and waits for the host's completion messages, which
15308c582c7cSDexuan Cui 	 * must also be handled by the channel callback.
15318c582c7cSDexuan Cui 	 */
15328c582c7cSDexuan Cui 	sc->taskq = taskqueue_create("vmbus_pcib_tq", M_WAITOK,
15338c582c7cSDexuan Cui 	    taskqueue_thread_enqueue, &sc->taskq);
15348c582c7cSDexuan Cui 	taskqueue_start_threads(&sc->taskq, 1, PI_NET, "vmbus_pcib_tq");
15358c582c7cSDexuan Cui 
15368c582c7cSDexuan Cui 	hbus->sc = sc;
15378c582c7cSDexuan Cui 
15388c582c7cSDexuan Cui 	init_completion(&hbus->query_completion);
15398c582c7cSDexuan Cui 	hbus->query_comp = &hbus->query_completion;
15408c582c7cSDexuan Cui 
15418c582c7cSDexuan Cui 	ret = vmbus_chan_open(sc->chan, pci_ring_size, pci_ring_size,
15428c582c7cSDexuan Cui 		NULL, 0, vmbus_pcib_on_channel_callback, sc);
15438c582c7cSDexuan Cui 	if (ret)
15448c582c7cSDexuan Cui 		goto free_res;
15458c582c7cSDexuan Cui 
1546ea11861eSWei Hu 	ret = hv_pci_protocol_negotiation(hbus, pci_protocol_versions,
1547ea11861eSWei Hu 	    ARRAY_SIZE(pci_protocol_versions));
15488c582c7cSDexuan Cui 	if (ret)
15498c582c7cSDexuan Cui 		goto vmbus_close;
15508c582c7cSDexuan Cui 
15518c582c7cSDexuan Cui 	ret = hv_pci_query_relations(hbus);
155275c2786cSWei Hu 	if (!ret)
155375c2786cSWei Hu 		ret = wait_for_response(hbus, hbus->query_comp);
155475c2786cSWei Hu 
15558c582c7cSDexuan Cui 	if (ret)
15568c582c7cSDexuan Cui 		goto vmbus_close;
15578c582c7cSDexuan Cui 
15588c582c7cSDexuan Cui 	ret = hv_pci_enter_d0(hbus);
15598c582c7cSDexuan Cui 	if (ret)
15608c582c7cSDexuan Cui 		goto vmbus_close;
15618c582c7cSDexuan Cui 
15628c582c7cSDexuan Cui 	ret = hv_send_resources_allocated(hbus);
15638c582c7cSDexuan Cui 	if (ret)
15648c582c7cSDexuan Cui 		goto vmbus_close;
15658c582c7cSDexuan Cui 
156675412a52SWei Hu 	vmbus_pcib_prepopulate_bars(hbus);
156775412a52SWei Hu 
15688c582c7cSDexuan Cui 	hbus->pci_bus = device_add_child(dev, "pci", -1);
15698c582c7cSDexuan Cui 	if (!hbus->pci_bus) {
15708c582c7cSDexuan Cui 		device_printf(dev, "failed to create pci bus\n");
15718c582c7cSDexuan Cui 		ret = ENXIO;
15728c582c7cSDexuan Cui 		goto vmbus_close;
15738c582c7cSDexuan Cui 	}
15748c582c7cSDexuan Cui 
15758c582c7cSDexuan Cui 	bus_generic_attach(dev);
15768c582c7cSDexuan Cui 
15778c582c7cSDexuan Cui 	hbus->state = hv_pcibus_installed;
15788c582c7cSDexuan Cui 
15798c582c7cSDexuan Cui 	return (0);
15808c582c7cSDexuan Cui 
15818c582c7cSDexuan Cui vmbus_close:
15828c582c7cSDexuan Cui 	vmbus_pcib_pre_detach(hbus);
15838c582c7cSDexuan Cui 	vmbus_chan_close(sc->chan);
15848c582c7cSDexuan Cui free_res:
15858c582c7cSDexuan Cui 	taskqueue_free(sc->taskq);
15868c582c7cSDexuan Cui 	free_completion(&hbus->query_completion);
15878c582c7cSDexuan Cui 	free(sc->rx_buf, M_DEVBUF);
15888c582c7cSDexuan Cui 	bus_release_resource(dev, SYS_RES_MEMORY, 0, hbus->cfg_res);
15898c582c7cSDexuan Cui free_bus:
15908c582c7cSDexuan Cui 	mtx_destroy(&hbus->device_list_lock);
15918c582c7cSDexuan Cui 	mtx_destroy(&hbus->config_lock);
15928c582c7cSDexuan Cui 	free(hbus, M_DEVBUF);
15938c582c7cSDexuan Cui 	return (ret);
15948c582c7cSDexuan Cui }
15958c582c7cSDexuan Cui 
15968c582c7cSDexuan Cui /*
15978c582c7cSDexuan Cui  * Standard detach entry point
15988c582c7cSDexuan Cui  */
15998c582c7cSDexuan Cui static int
16008c582c7cSDexuan Cui vmbus_pcib_detach(device_t dev)
16018c582c7cSDexuan Cui {
16028c582c7cSDexuan Cui 	struct vmbus_pcib_softc *sc = device_get_softc(dev);
16038c582c7cSDexuan Cui 	struct hv_pcibus *hbus = sc->hbus;
16048c582c7cSDexuan Cui 	struct pci_message teardown_packet;
16058c582c7cSDexuan Cui 	struct pci_bus_relations relations;
16068c582c7cSDexuan Cui 	int ret;
16078c582c7cSDexuan Cui 
16088c582c7cSDexuan Cui 	vmbus_pcib_pre_detach(hbus);
16098c582c7cSDexuan Cui 
16108c582c7cSDexuan Cui 	if (hbus->state == hv_pcibus_installed)
16118c582c7cSDexuan Cui 		bus_generic_detach(dev);
16128c582c7cSDexuan Cui 
16138c582c7cSDexuan Cui 	/* Delete any children which might still exist. */
16148c582c7cSDexuan Cui 	memset(&relations, 0, sizeof(relations));
16158c582c7cSDexuan Cui 	hv_pci_devices_present(hbus, &relations);
16168c582c7cSDexuan Cui 
16178c582c7cSDexuan Cui 	ret = hv_send_resources_released(hbus);
16188c582c7cSDexuan Cui 	if (ret)
16198c582c7cSDexuan Cui 		device_printf(dev, "failed to send PCI_RESOURCES_RELEASED\n");
16208c582c7cSDexuan Cui 
16218c582c7cSDexuan Cui 	teardown_packet.type = PCI_BUS_D0EXIT;
16228c582c7cSDexuan Cui 	ret = vmbus_chan_send(sc->chan, VMBUS_CHANPKT_TYPE_INBAND, 0,
16238c582c7cSDexuan Cui 	    &teardown_packet, sizeof(struct pci_message), 0);
16248c582c7cSDexuan Cui 	if (ret)
16258c582c7cSDexuan Cui 		device_printf(dev, "failed to send PCI_BUS_D0EXIT\n");
16268c582c7cSDexuan Cui 
16278c582c7cSDexuan Cui 	taskqueue_drain_all(hbus->sc->taskq);
16288c582c7cSDexuan Cui 	vmbus_chan_close(sc->chan);
16298c582c7cSDexuan Cui 	taskqueue_free(sc->taskq);
16308c582c7cSDexuan Cui 
16318c582c7cSDexuan Cui 	free_completion(&hbus->query_completion);
16328c582c7cSDexuan Cui 	free(sc->rx_buf, M_DEVBUF);
16338c582c7cSDexuan Cui 	bus_release_resource(dev, SYS_RES_MEMORY, 0, hbus->cfg_res);
16348c582c7cSDexuan Cui 
16358c582c7cSDexuan Cui 	mtx_destroy(&hbus->device_list_lock);
16368c582c7cSDexuan Cui 	mtx_destroy(&hbus->config_lock);
16378c582c7cSDexuan Cui 	free(hbus, M_DEVBUF);
16388c582c7cSDexuan Cui 
16398c582c7cSDexuan Cui 	return (0);
16408c582c7cSDexuan Cui }
16418c582c7cSDexuan Cui 
16428c582c7cSDexuan Cui static int
16438c582c7cSDexuan Cui vmbus_pcib_read_ivar(device_t dev, device_t child, int which, uintptr_t *val)
16448c582c7cSDexuan Cui {
16458c582c7cSDexuan Cui 	struct vmbus_pcib_softc *sc = device_get_softc(dev);
16468c582c7cSDexuan Cui 
16478c582c7cSDexuan Cui 	switch (which) {
16488c582c7cSDexuan Cui 	case PCIB_IVAR_DOMAIN:
16498c582c7cSDexuan Cui 		*val = sc->hbus->pci_domain;
16508c582c7cSDexuan Cui 		return (0);
16518c582c7cSDexuan Cui 
16528c582c7cSDexuan Cui 	case PCIB_IVAR_BUS:
16538c582c7cSDexuan Cui 		/* There is only bus 0. */
16548c582c7cSDexuan Cui 		*val = 0;
16558c582c7cSDexuan Cui 		return (0);
16568c582c7cSDexuan Cui 	}
16578c582c7cSDexuan Cui 	return (ENOENT);
16588c582c7cSDexuan Cui }
16598c582c7cSDexuan Cui 
16608c582c7cSDexuan Cui static int
16618c582c7cSDexuan Cui vmbus_pcib_write_ivar(device_t dev, device_t child, int which, uintptr_t val)
16628c582c7cSDexuan Cui {
16638c582c7cSDexuan Cui 	return (ENOENT);
16648c582c7cSDexuan Cui }
16658c582c7cSDexuan Cui 
16668c582c7cSDexuan Cui static struct resource *
16678c582c7cSDexuan Cui vmbus_pcib_alloc_resource(device_t dev, device_t child, int type, int *rid,
16688c582c7cSDexuan Cui 	rman_res_t start, rman_res_t end, rman_res_t count, u_int flags)
16698c582c7cSDexuan Cui {
16708c582c7cSDexuan Cui 	unsigned int bar_no;
16718c582c7cSDexuan Cui 	struct hv_pci_dev *hpdev;
16728c582c7cSDexuan Cui 	struct vmbus_pcib_softc *sc = device_get_softc(dev);
16738c582c7cSDexuan Cui 	struct resource *res;
16748c582c7cSDexuan Cui 	unsigned int devfn;
16758c582c7cSDexuan Cui 
16768c582c7cSDexuan Cui 	if (type == PCI_RES_BUS)
16778c582c7cSDexuan Cui 		return (pci_domain_alloc_bus(sc->hbus->pci_domain, child, rid,
16788c582c7cSDexuan Cui 		    start, end, count, flags));
16798c582c7cSDexuan Cui 
16808c582c7cSDexuan Cui 	/* Devices with port I/O BAR are not supported. */
16818c582c7cSDexuan Cui 	if (type == SYS_RES_IOPORT)
16828c582c7cSDexuan Cui 		return (NULL);
16838c582c7cSDexuan Cui 
16848c582c7cSDexuan Cui 	if (type == SYS_RES_MEMORY) {
16858c582c7cSDexuan Cui 		devfn = PCI_DEVFN(pci_get_slot(child),
16868c582c7cSDexuan Cui 		    pci_get_function(child));
16878c582c7cSDexuan Cui 		hpdev = get_pcichild_wslot(sc->hbus, devfn_to_wslot(devfn));
16888c582c7cSDexuan Cui 		if (!hpdev)
16898c582c7cSDexuan Cui 			return (NULL);
16908c582c7cSDexuan Cui 
16918c582c7cSDexuan Cui 		bar_no = PCI_RID2BAR(*rid);
16928c582c7cSDexuan Cui 		if (bar_no >= MAX_NUM_BARS)
16938c582c7cSDexuan Cui 			return (NULL);
16948c582c7cSDexuan Cui 
16958c582c7cSDexuan Cui 		/* Make sure a 32-bit BAR gets a 32-bit address */
16968c582c7cSDexuan Cui 		if (!(hpdev->probed_bar[bar_no] & PCIM_BAR_MEM_64))
16978c582c7cSDexuan Cui 			end = ulmin(end, 0xFFFFFFFF);
16988c582c7cSDexuan Cui 	}
16998c582c7cSDexuan Cui 
17008c582c7cSDexuan Cui 	res = bus_generic_alloc_resource(dev, child, type, rid,
17018c582c7cSDexuan Cui 		start, end, count, flags);
17028c582c7cSDexuan Cui 	/*
17038c582c7cSDexuan Cui 	 * If this is a request for a specific range, assume it is
17048c582c7cSDexuan Cui 	 * correct and pass it up to the parent.
17058c582c7cSDexuan Cui 	 */
17068c582c7cSDexuan Cui 	if (res == NULL && start + count - 1 == end)
17078c582c7cSDexuan Cui 		res = bus_generic_alloc_resource(dev, child, type, rid,
17088c582c7cSDexuan Cui 		    start, end, count, flags);
17094196186dSLi-Wen Hsu 	if (res == NULL)
17104196186dSLi-Wen Hsu 		device_printf(dev, "vmbus_pcib_alloc_resource failed\n");
17114196186dSLi-Wen Hsu 
17128c582c7cSDexuan Cui 	return (res);
17138c582c7cSDexuan Cui }
17148c582c7cSDexuan Cui 
17158c582c7cSDexuan Cui static int
1716fef01f04SJohn Baldwin vmbus_pcib_adjust_resource(device_t dev, device_t child,
17179c4effb1SJohn Baldwin     struct resource *r, rman_res_t start, rman_res_t end)
17189c4effb1SJohn Baldwin {
17199c4effb1SJohn Baldwin 	struct vmbus_pcib_softc *sc = device_get_softc(dev);
17209c4effb1SJohn Baldwin 
1721fef01f04SJohn Baldwin 	if (rman_get_type(r) == PCI_RES_BUS)
17229c4effb1SJohn Baldwin 		return (pci_domain_adjust_bus(sc->hbus->pci_domain, child, r,
17239c4effb1SJohn Baldwin 		    start, end));
1724fef01f04SJohn Baldwin 	return (bus_generic_adjust_resource(dev, child, r, start, end));
17259c4effb1SJohn Baldwin }
17269c4effb1SJohn Baldwin 
17279c4effb1SJohn Baldwin static int
17288c582c7cSDexuan Cui vmbus_pcib_release_resource(device_t dev, device_t child, int type, int rid,
17298c582c7cSDexuan Cui     struct resource *r)
17308c582c7cSDexuan Cui {
17318c582c7cSDexuan Cui 	struct vmbus_pcib_softc *sc = device_get_softc(dev);
17328c582c7cSDexuan Cui 
17338c582c7cSDexuan Cui 	if (type == PCI_RES_BUS)
17348c582c7cSDexuan Cui 		return (pci_domain_release_bus(sc->hbus->pci_domain, child,
17358c582c7cSDexuan Cui 		    rid, r));
17368c582c7cSDexuan Cui 
17378c582c7cSDexuan Cui 	if (type == SYS_RES_IOPORT)
17388c582c7cSDexuan Cui 		return (EINVAL);
17398c582c7cSDexuan Cui 
17408c582c7cSDexuan Cui 	return (bus_generic_release_resource(dev, child, type, rid, r));
17418c582c7cSDexuan Cui }
17428c582c7cSDexuan Cui 
17438c582c7cSDexuan Cui static int
1744*2baed46eSJohn Baldwin vmbus_pcib_activate_resource(device_t dev, device_t child, struct resource *r)
17459c4effb1SJohn Baldwin {
17469c4effb1SJohn Baldwin 	struct vmbus_pcib_softc *sc = device_get_softc(dev);
17479c4effb1SJohn Baldwin 
1748*2baed46eSJohn Baldwin 	if (rman_get_type(r) == PCI_RES_BUS)
17499c4effb1SJohn Baldwin 		return (pci_domain_activate_bus(sc->hbus->pci_domain, child,
1750*2baed46eSJohn Baldwin 		    r));
1751*2baed46eSJohn Baldwin 	return (bus_generic_activate_resource(dev, child, r));
17529c4effb1SJohn Baldwin }
17539c4effb1SJohn Baldwin 
17549c4effb1SJohn Baldwin static int
1755*2baed46eSJohn Baldwin vmbus_pcib_deactivate_resource(device_t dev, device_t child, struct resource *r)
17569c4effb1SJohn Baldwin {
17579c4effb1SJohn Baldwin 	struct vmbus_pcib_softc *sc = device_get_softc(dev);
17589c4effb1SJohn Baldwin 
1759*2baed46eSJohn Baldwin 	if (rman_get_type(r) == PCI_RES_BUS)
17609c4effb1SJohn Baldwin 		return (pci_domain_deactivate_bus(sc->hbus->pci_domain, child,
1761*2baed46eSJohn Baldwin 		    r));
1762*2baed46eSJohn Baldwin 	return (bus_generic_deactivate_resource(dev, child, r));
17639c4effb1SJohn Baldwin }
17649c4effb1SJohn Baldwin 
17659c4effb1SJohn Baldwin static int
17668c582c7cSDexuan Cui vmbus_pcib_get_cpus(device_t pcib, device_t dev, enum cpu_sets op,
17678c582c7cSDexuan Cui     size_t setsize, cpuset_t *cpuset)
17688c582c7cSDexuan Cui {
17698c582c7cSDexuan Cui 	return (bus_get_cpus(pcib, op, setsize, cpuset));
17708c582c7cSDexuan Cui }
17718c582c7cSDexuan Cui 
17728c582c7cSDexuan Cui static uint32_t
17738c582c7cSDexuan Cui vmbus_pcib_read_config(device_t dev, u_int bus, u_int slot, u_int func,
17748c582c7cSDexuan Cui     u_int reg, int bytes)
17758c582c7cSDexuan Cui {
17768c582c7cSDexuan Cui 	struct vmbus_pcib_softc *sc = device_get_softc(dev);
17778c582c7cSDexuan Cui 	struct hv_pci_dev *hpdev;
17788c582c7cSDexuan Cui 	unsigned int devfn = PCI_DEVFN(slot, func);
17798c582c7cSDexuan Cui 	uint32_t data = 0;
17808c582c7cSDexuan Cui 
17818c582c7cSDexuan Cui 	KASSERT(bus == 0, ("bus should be 0, but is %u", bus));
17828c582c7cSDexuan Cui 
17838c582c7cSDexuan Cui 	hpdev = get_pcichild_wslot(sc->hbus, devfn_to_wslot(devfn));
17848c582c7cSDexuan Cui 	if (!hpdev)
17858c582c7cSDexuan Cui 		return (~0);
17868c582c7cSDexuan Cui 
17878c582c7cSDexuan Cui 	_hv_pcifront_read_config(hpdev, reg, bytes, &data);
17888c582c7cSDexuan Cui 
17898c582c7cSDexuan Cui 	return (data);
17908c582c7cSDexuan Cui }
17918c582c7cSDexuan Cui 
17928c582c7cSDexuan Cui static void
17938c582c7cSDexuan Cui vmbus_pcib_write_config(device_t dev, u_int bus, u_int slot, u_int func,
17948c582c7cSDexuan Cui     u_int reg, uint32_t data, int bytes)
17958c582c7cSDexuan Cui {
17968c582c7cSDexuan Cui 	struct vmbus_pcib_softc *sc = device_get_softc(dev);
17978c582c7cSDexuan Cui 	struct hv_pci_dev *hpdev;
17988c582c7cSDexuan Cui 	unsigned int devfn = PCI_DEVFN(slot, func);
17998c582c7cSDexuan Cui 
18008c582c7cSDexuan Cui 	KASSERT(bus == 0, ("bus should be 0, but is %u", bus));
18018c582c7cSDexuan Cui 
18028c582c7cSDexuan Cui 	hpdev = get_pcichild_wslot(sc->hbus, devfn_to_wslot(devfn));
18038c582c7cSDexuan Cui 	if (!hpdev)
18048c582c7cSDexuan Cui 		return;
18058c582c7cSDexuan Cui 
18068c582c7cSDexuan Cui 	_hv_pcifront_write_config(hpdev, reg, bytes, data);
18078c582c7cSDexuan Cui }
18088c582c7cSDexuan Cui 
18098c582c7cSDexuan Cui static int
18108c582c7cSDexuan Cui vmbus_pcib_route_intr(device_t pcib, device_t dev, int pin)
18118c582c7cSDexuan Cui {
18128c582c7cSDexuan Cui 	/* We only support MSI/MSI-X and don't support INTx interrupt. */
18138c582c7cSDexuan Cui 	return (PCI_INVALID_IRQ);
18148c582c7cSDexuan Cui }
18158c582c7cSDexuan Cui 
18168c582c7cSDexuan Cui static int
18178c582c7cSDexuan Cui vmbus_pcib_alloc_msi(device_t pcib, device_t dev, int count,
18188c582c7cSDexuan Cui     int maxcount, int *irqs)
18198c582c7cSDexuan Cui {
1820ea11861eSWei Hu #if defined(__amd64__) || defined(__i386__)
18218c582c7cSDexuan Cui 	return (PCIB_ALLOC_MSI(device_get_parent(pcib), dev, count, maxcount,
18228c582c7cSDexuan Cui 	    irqs));
1823ea11861eSWei Hu #endif
1824ea11861eSWei Hu #if defined(__aarch64__)
1825ea11861eSWei Hu 	return (intr_alloc_msi(pcib, dev, ACPI_MSI_XREF, count, maxcount,
1826ea11861eSWei Hu 	    irqs));
1827ea11861eSWei Hu #endif
18288c582c7cSDexuan Cui }
18298c582c7cSDexuan Cui 
18308c582c7cSDexuan Cui static int
18318c582c7cSDexuan Cui vmbus_pcib_release_msi(device_t pcib, device_t dev, int count, int *irqs)
18328c582c7cSDexuan Cui {
1833ea11861eSWei Hu #if defined(__amd64__) || defined(__i386__)
18348c582c7cSDexuan Cui 	return (PCIB_RELEASE_MSI(device_get_parent(pcib), dev, count, irqs));
1835ea11861eSWei Hu #endif
1836ea11861eSWei Hu #if defined(__aarch64__)
1837ea11861eSWei Hu 	return(intr_release_msi(pcib, dev, ACPI_MSI_XREF, count, irqs));
1838ea11861eSWei Hu #endif
18398c582c7cSDexuan Cui }
18408c582c7cSDexuan Cui 
18418c582c7cSDexuan Cui static int
18428c582c7cSDexuan Cui vmbus_pcib_alloc_msix(device_t pcib, device_t dev, int *irq)
18438c582c7cSDexuan Cui {
1844ea11861eSWei Hu #if defined(__aarch64__)
1845ea11861eSWei Hu 	int ret;
1846ea11861eSWei Hu #if defined(INTRNG)
1847ea11861eSWei Hu 	ret = intr_alloc_msix(pcib, dev, ACPI_MSI_XREF, irq);
1848ea11861eSWei Hu 	return ret;
1849ea11861eSWei Hu #else
1850ea11861eSWei Hu     return (ENXIO);
1851ea11861eSWei Hu #endif
1852ea11861eSWei Hu #else
18538c582c7cSDexuan Cui 	return (PCIB_ALLOC_MSIX(device_get_parent(pcib), dev, irq));
1854ea11861eSWei Hu #endif /* __aarch64__ */
18558c582c7cSDexuan Cui }
18568c582c7cSDexuan Cui 
18578c582c7cSDexuan Cui static int
18588c582c7cSDexuan Cui vmbus_pcib_release_msix(device_t pcib, device_t dev, int irq)
18598c582c7cSDexuan Cui {
1860ea11861eSWei Hu #if defined(__aarch64__)
1861ea11861eSWei Hu 	return (intr_release_msix(pcib, dev, ACPI_MSI_XREF, irq));
1862ea11861eSWei Hu #else
18638c582c7cSDexuan Cui 	return (PCIB_RELEASE_MSIX(device_get_parent(pcib), dev, irq));
1864ea11861eSWei Hu #endif /* __aarch64__ */
18658c582c7cSDexuan Cui }
18668c582c7cSDexuan Cui 
1867ea11861eSWei Hu #if defined(__aarch64__)
1868ea11861eSWei Hu #define	MSI_INTEL_ADDR_DEST	0x00000000
1869ea11861eSWei Hu #define	MSI_INTEL_DATA_DELFIXED 0x0
1870ea11861eSWei Hu #endif
1871ea11861eSWei Hu #if defined(__amd64__) || defined(__i386__)
18728c582c7cSDexuan Cui #define MSI_INTEL_ADDR_DEST 0x000ff000
18738c582c7cSDexuan Cui #define MSI_INTEL_DATA_INTVEC   IOART_INTVEC    /* Interrupt vector. */
18748c582c7cSDexuan Cui #define MSI_INTEL_DATA_DELFIXED IOART_DELFIXED
1875ea11861eSWei Hu #endif
18768c582c7cSDexuan Cui 
18778c582c7cSDexuan Cui static int
18788c582c7cSDexuan Cui vmbus_pcib_map_msi(device_t pcib, device_t child, int irq,
18798c582c7cSDexuan Cui     uint64_t *addr, uint32_t *data)
18808c582c7cSDexuan Cui {
18818c582c7cSDexuan Cui 	unsigned int devfn;
18828c582c7cSDexuan Cui 	struct hv_pci_dev *hpdev;
18838c582c7cSDexuan Cui 
18848c582c7cSDexuan Cui 	uint64_t v_addr;
18858c582c7cSDexuan Cui 	uint32_t v_data;
18868c582c7cSDexuan Cui 	struct hv_irq_desc *hid, *tmp_hid;
18878c582c7cSDexuan Cui 	unsigned int cpu, vcpu_id;
18888c582c7cSDexuan Cui 	unsigned int vector;
18898c582c7cSDexuan Cui 
18908c582c7cSDexuan Cui 	struct vmbus_pcib_softc *sc = device_get_softc(pcib);
18918c582c7cSDexuan Cui 	struct compose_comp_ctxt comp;
18928c582c7cSDexuan Cui 	struct {
18938c582c7cSDexuan Cui 		struct pci_packet pkt;
1894ea11861eSWei Hu 		union {
1895ea11861eSWei Hu 			struct pci_create_interrupt v1;
1896ea11861eSWei Hu 			struct pci_create_interrupt3 v3;
1897ea11861eSWei Hu 		}int_pkts;
18988c582c7cSDexuan Cui 	} ctxt;
18998c582c7cSDexuan Cui 	int ret;
1900ea11861eSWei Hu 	uint32_t size;
19018c582c7cSDexuan Cui 
19028c582c7cSDexuan Cui 	devfn = PCI_DEVFN(pci_get_slot(child), pci_get_function(child));
19038c582c7cSDexuan Cui 	hpdev = get_pcichild_wslot(sc->hbus, devfn_to_wslot(devfn));
19048c582c7cSDexuan Cui 	if (!hpdev)
19058c582c7cSDexuan Cui 		return (ENOENT);
1906ea11861eSWei Hu #if defined(__aarch64__)
1907ea11861eSWei Hu 	ret = intr_map_msi(pcib, child, ACPI_MSI_XREF, irq,
1908ea11861eSWei Hu 	    &v_addr, &v_data);
1909ea11861eSWei Hu #else
19108c582c7cSDexuan Cui 	ret = PCIB_MAP_MSI(device_get_parent(pcib), child, irq,
19118c582c7cSDexuan Cui             &v_addr, &v_data);
1912ea11861eSWei Hu #endif
19138c582c7cSDexuan Cui 	if (ret)
19148c582c7cSDexuan Cui 		return (ret);
19158c582c7cSDexuan Cui 
19168c582c7cSDexuan Cui 	TAILQ_FOREACH_SAFE(hid, &hpdev->irq_desc_list, link, tmp_hid) {
19178c582c7cSDexuan Cui 		if (hid->irq == irq) {
19188c582c7cSDexuan Cui 			TAILQ_REMOVE(&hpdev->irq_desc_list, hid, link);
19198c582c7cSDexuan Cui 			hv_int_desc_free(hpdev, hid);
19208c582c7cSDexuan Cui 			break;
19218c582c7cSDexuan Cui 		}
19228c582c7cSDexuan Cui 	}
19238c582c7cSDexuan Cui 
1924ea11861eSWei Hu #if defined(__aarch64__)
1925ea11861eSWei Hu 	cpu = 0;
1926ea11861eSWei Hu 	vcpu_id = VMBUS_GET_VCPU_ID(device_get_parent(pcib), pcib, cpu);
1927ea11861eSWei Hu 	vector = v_data;
1928ea11861eSWei Hu #else
1929999174baSWei Hu 	cpu = apic_cpuid((v_addr & MSI_INTEL_ADDR_DEST) >> 12);
19308c582c7cSDexuan Cui 	vcpu_id = VMBUS_GET_VCPU_ID(device_get_parent(pcib), pcib, cpu);
19318c582c7cSDexuan Cui 	vector = v_data & MSI_INTEL_DATA_INTVEC;
1932ea11861eSWei Hu #endif
19338c582c7cSDexuan Cui 
1934999174baSWei Hu 	if (hpdev->hbus->protocol_version < PCI_PROTOCOL_VERSION_1_4 &&
1935999174baSWei Hu 	    vcpu_id > 63) {
1936999174baSWei Hu 		/* We only support vcpu_id < 64 before vPCI version 1.4 */
1937999174baSWei Hu 		device_printf(pcib,
1938999174baSWei Hu 		    "Error: "
1939999174baSWei Hu 		    "vcpu_id %u overflowed on PCI VMBus version 0x%x\n",
1940999174baSWei Hu 		    vcpu_id, hpdev->hbus->protocol_version);
1941999174baSWei Hu 		return (ENODEV);
1942999174baSWei Hu 	}
1943999174baSWei Hu 
19448c582c7cSDexuan Cui 	init_completion(&comp.comp_pkt.host_event);
19458c582c7cSDexuan Cui 
19468c582c7cSDexuan Cui 	memset(&ctxt, 0, sizeof(ctxt));
19478c582c7cSDexuan Cui 	ctxt.pkt.completion_func = hv_pci_compose_compl;
19488c582c7cSDexuan Cui 	ctxt.pkt.compl_ctxt = &comp;
1949ea11861eSWei Hu 	switch (hpdev->hbus->protocol_version) {
1950ea11861eSWei Hu 	case PCI_PROTOCOL_VERSION_1_1:
1951ea11861eSWei Hu 		ctxt.int_pkts.v1.message_type.type =
1952ea11861eSWei Hu 		    PCI_CREATE_INTERRUPT_MESSAGE;
1953ea11861eSWei Hu 		ctxt.int_pkts.v1.wslot.val = hpdev->desc.wslot.val;
1954ea11861eSWei Hu 		ctxt.int_pkts.v1.int_desc.vector = vector;
1955ea11861eSWei Hu 		ctxt.int_pkts.v1.int_desc.vector_count = 1;
1956ea11861eSWei Hu 		ctxt.int_pkts.v1.int_desc.delivery_mode =
1957ea11861eSWei Hu 		    MSI_INTEL_DATA_DELFIXED;
1958ea11861eSWei Hu 		ctxt.int_pkts.v1.int_desc.cpu_mask = 1ULL << vcpu_id;
1959ea11861eSWei Hu 		size = sizeof(ctxt.int_pkts.v1);
1960ea11861eSWei Hu 		break;
19618c582c7cSDexuan Cui 
1962ea11861eSWei Hu 	case PCI_PROTOCOL_VERSION_1_4:
1963ea11861eSWei Hu 		ctxt.int_pkts.v3.message_type.type =
1964ea11861eSWei Hu 		    PCI_CREATE_INTERRUPT_MESSAGE3;
1965ea11861eSWei Hu 		ctxt.int_pkts.v3.wslot.val = hpdev->desc.wslot.val;
1966ea11861eSWei Hu 		ctxt.int_pkts.v3.int_desc.vector = vector;
1967ea11861eSWei Hu 		ctxt.int_pkts.v3.int_desc.vector_count = 1;
1968ea11861eSWei Hu 		ctxt.int_pkts.v3.int_desc.reserved = 0;
1969ea11861eSWei Hu 		ctxt.int_pkts.v3.int_desc.delivery_mode =
1970ea11861eSWei Hu 		    MSI_INTEL_DATA_DELFIXED;
1971ea11861eSWei Hu 		ctxt.int_pkts.v3.int_desc.processor_count = 1;
1972ea11861eSWei Hu 		ctxt.int_pkts.v3.int_desc.processor_array[0] = vcpu_id;
1973ea11861eSWei Hu 		size = sizeof(ctxt.int_pkts.v3);
1974ea11861eSWei Hu 		break;
1975ea11861eSWei Hu 	}
19768c582c7cSDexuan Cui 	ret = vmbus_chan_send(sc->chan,	VMBUS_CHANPKT_TYPE_INBAND,
1977ea11861eSWei Hu 	    VMBUS_CHANPKT_FLAG_RC, &ctxt.int_pkts, size,
1978aaf13123SDimitry Andric 	    (uint64_t)(uintptr_t)&ctxt.pkt);
19798c582c7cSDexuan Cui 	if (ret) {
19808c582c7cSDexuan Cui 		free_completion(&comp.comp_pkt.host_event);
19818c582c7cSDexuan Cui 		return (ret);
19828c582c7cSDexuan Cui 	}
19838c582c7cSDexuan Cui 
19848c582c7cSDexuan Cui 	wait_for_completion(&comp.comp_pkt.host_event);
19858c582c7cSDexuan Cui 	free_completion(&comp.comp_pkt.host_event);
19868c582c7cSDexuan Cui 
1987ea11861eSWei Hu 	if (comp.comp_pkt.completion_status < 0) {
1988ea11861eSWei Hu 		device_printf(pcib,
1989ea11861eSWei Hu 		    "vmbus_pcib_map_msi completion_status %d\n",
1990ea11861eSWei Hu 		    comp.comp_pkt.completion_status);
19918c582c7cSDexuan Cui 		return (EPROTO);
1992ea11861eSWei Hu 	}
19938c582c7cSDexuan Cui 
19948c582c7cSDexuan Cui 	*addr = comp.int_desc.address;
19958c582c7cSDexuan Cui 	*data = comp.int_desc.data;
19968c582c7cSDexuan Cui 
19978c582c7cSDexuan Cui 	hid = malloc(sizeof(struct hv_irq_desc), M_DEVBUF, M_WAITOK | M_ZERO);
19988c582c7cSDexuan Cui 	hid->irq = irq;
19998c582c7cSDexuan Cui 	hid->desc = comp.int_desc;
20008c582c7cSDexuan Cui 	TAILQ_INSERT_TAIL(&hpdev->irq_desc_list, hid, link);
20018c582c7cSDexuan Cui 
20028c582c7cSDexuan Cui 	return (0);
20038c582c7cSDexuan Cui }
20048c582c7cSDexuan Cui 
20058c582c7cSDexuan Cui static device_method_t vmbus_pcib_methods[] = {
20068c582c7cSDexuan Cui 	/* Device interface */
20078c582c7cSDexuan Cui 	DEVMETHOD(device_probe,         vmbus_pcib_probe),
20088c582c7cSDexuan Cui 	DEVMETHOD(device_attach,        vmbus_pcib_attach),
20098c582c7cSDexuan Cui 	DEVMETHOD(device_detach,        vmbus_pcib_detach),
20108c582c7cSDexuan Cui 	DEVMETHOD(device_shutdown,	bus_generic_shutdown),
20118c582c7cSDexuan Cui 	DEVMETHOD(device_suspend,	bus_generic_suspend),
20128c582c7cSDexuan Cui 	DEVMETHOD(device_resume,	bus_generic_resume),
20138c582c7cSDexuan Cui 
20148c582c7cSDexuan Cui 	/* Bus interface */
20158c582c7cSDexuan Cui 	DEVMETHOD(bus_read_ivar,		vmbus_pcib_read_ivar),
20168c582c7cSDexuan Cui 	DEVMETHOD(bus_write_ivar,		vmbus_pcib_write_ivar),
20178c582c7cSDexuan Cui 	DEVMETHOD(bus_alloc_resource,		vmbus_pcib_alloc_resource),
20189c4effb1SJohn Baldwin 	DEVMETHOD(bus_adjust_resource,		vmbus_pcib_adjust_resource),
20198c582c7cSDexuan Cui 	DEVMETHOD(bus_release_resource,		vmbus_pcib_release_resource),
20209c4effb1SJohn Baldwin 	DEVMETHOD(bus_activate_resource,   	vmbus_pcib_activate_resource),
20219c4effb1SJohn Baldwin 	DEVMETHOD(bus_deactivate_resource, 	vmbus_pcib_deactivate_resource),
20228c582c7cSDexuan Cui 	DEVMETHOD(bus_setup_intr,	   bus_generic_setup_intr),
20238c582c7cSDexuan Cui 	DEVMETHOD(bus_teardown_intr,	   bus_generic_teardown_intr),
20248c582c7cSDexuan Cui 	DEVMETHOD(bus_get_cpus,			vmbus_pcib_get_cpus),
20258c582c7cSDexuan Cui 
20268c582c7cSDexuan Cui 	/* pcib interface */
20278c582c7cSDexuan Cui 	DEVMETHOD(pcib_maxslots,		pcib_maxslots),
20288c582c7cSDexuan Cui 	DEVMETHOD(pcib_read_config,		vmbus_pcib_read_config),
20298c582c7cSDexuan Cui 	DEVMETHOD(pcib_write_config,		vmbus_pcib_write_config),
20308c582c7cSDexuan Cui 	DEVMETHOD(pcib_route_interrupt,		vmbus_pcib_route_intr),
20318c582c7cSDexuan Cui 	DEVMETHOD(pcib_alloc_msi,		vmbus_pcib_alloc_msi),
20328c582c7cSDexuan Cui 	DEVMETHOD(pcib_release_msi,		vmbus_pcib_release_msi),
20338c582c7cSDexuan Cui 	DEVMETHOD(pcib_alloc_msix,		vmbus_pcib_alloc_msix),
20348c582c7cSDexuan Cui 	DEVMETHOD(pcib_release_msix,		vmbus_pcib_release_msix),
20358c582c7cSDexuan Cui 	DEVMETHOD(pcib_map_msi,			vmbus_pcib_map_msi),
203628586889SWarner Losh 	DEVMETHOD(pcib_request_feature,		pcib_request_feature_allow),
20378c582c7cSDexuan Cui 
20388c582c7cSDexuan Cui 	DEVMETHOD_END
20398c582c7cSDexuan Cui };
20408c582c7cSDexuan Cui 
20418c582c7cSDexuan Cui DEFINE_CLASS_0(pcib, vmbus_pcib_driver, vmbus_pcib_methods,
20428c582c7cSDexuan Cui 		sizeof(struct vmbus_pcib_softc));
2043c1cef544SJohn Baldwin DRIVER_MODULE(vmbus_pcib, vmbus, vmbus_pcib_driver, 0, 0);
20448c582c7cSDexuan Cui MODULE_DEPEND(vmbus_pcib, vmbus, 1, 1, 1);
20458c582c7cSDexuan Cui MODULE_DEPEND(vmbus_pcib, pci, 1, 1, 1);
2046cdb316eeSDexuan Cui 
2047cdb316eeSDexuan Cui #endif /* NEW_PCIB */
2048