1. photon
  2. Commit 7b0f3a58d8531d33d4dd1cde60684affb5ef555f
Browse code

Linux: Enable pci-hyperv support and apply relevant patches

Change-Id: I8a40ecdfd6b3d1d1d5243caf3e603e06c01d722f
Reviewed-on: http://photon-jenkins.eng.vmware.com:8082/6379
Reviewed-by: Srivatsa S. Bhat <srivatsab@vmware.com>
Tested-by: gerrit-photon <photon-checkins@vmware.com>

Ajay Kaher authored on 2018/12/18 03:27:59
Showing 5 changed files
SPECS/linux/0001_PCI_hv_Allocate_physically_contiguous_hypercall_params_buffer.patch
History View file @ 7b0f3a5
1 1 
new file mode 100644
... ... 
@@ -0,0 +1,97 @@
0 
+From 0de8ce3ee8e38cc66683438f715c79a2cc69539e Mon Sep 17 00:00:00 2001
1 
+From: Long Li <longli@microsoft.com>
2 
+Date: Tue, 8 Nov 2016 14:04:38 -0800
3 
+Subject: PCI: hv: Allocate physically contiguous hypercall params buffer
4 
+
5 
+hv_do_hypercall() assumes that we pass a segment from a physically
6 
+contiguous buffer.  A buffer allocated on the stack may not work if
7 
+CONFIG_VMAP_STACK=y is set.
8 
+
9 
+Use kmalloc() to allocate this buffer.
10 
+
11 
+Reported-by: Haiyang Zhang <haiyangz@microsoft.com>
12 
+Signed-off-by: Long Li <longli@microsoft.com>
13 
+Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
14 
+Acked-by: K. Y. Srinivasan <kys@microsoft.com>
15 
+Signed-off-by: Ajay Kaher <akaher@vmware.com>
16 
+---
17 
+ drivers/pci/host/pci-hyperv.c | 29 +++++++++++++++++++----------
18 
+ 1 file changed, 19 insertions(+), 10 deletions(-)
19 
+
20 
+(limited to 'drivers/pci/host/pci-hyperv.c')
21 
+
22 
+diff --git a/drivers/pci/host/pci-hyperv.c b/drivers/pci/host/pci-hyperv.c
23 
+index a63c3a4f6c7c..3efcc7bdc5fb 100644
24 
+--- a/drivers/pci/host/pci-hyperv.c
25 
+@@ -378,6 +378,8 @@ struct hv_pcibus_device {
26 
+ 	struct msi_domain_info msi_info;
27 
+ 	struct msi_controller msi_chip;
28 
+ 	struct irq_domain *irq_domain;
29 
++	struct retarget_msi_interrupt retarget_msi_interrupt_params;
30 
++	spinlock_t retarget_msi_interrupt_lock;
31 
+ };
32 
+
33 
+ /*
34 
+@@ -774,34 +776,40 @@ static void hv_irq_unmask(struct irq_data *data)
35 
+ {
36 
+ 	struct msi_desc *msi_desc = irq_data_get_msi_desc(data);
37 
+ 	struct irq_cfg *cfg = irqd_cfg(data);
38 
+-	struct retarget_msi_interrupt params;
39 
++	struct retarget_msi_interrupt *params;
40 
+ 	struct hv_pcibus_device *hbus;
41 
+ 	struct cpumask *dest;
42 
+ 	struct pci_bus *pbus;
43 
+ 	struct pci_dev *pdev;
44 
+ 	int cpu;
45 
++	unsigned long flags;
46 
+
47 
+ 	dest = irq_data_get_affinity_mask(data);
48 
+ 	pdev = msi_desc_to_pci_dev(msi_desc);
49 
+ 	pbus = pdev->bus;
50 
+ 	hbus = container_of(pbus->sysdata, struct hv_pcibus_device, sysdata);
51 
+
52 
+-	memset(&params, 0, sizeof(params));
53 
+-	params.partition_id = HV_PARTITION_ID_SELF;
54 
+-	params.source = 1; /* MSI(-X) */
55 
+-	params.address = msi_desc->msg.address_lo;
56 
+-	params.data = msi_desc->msg.data;
57 
+-	params.device_id = (hbus->hdev->dev_instance.b[5] << 24) |
58 
++	spin_lock_irqsave(&hbus->retarget_msi_interrupt_lock, flags);
59 
++
60 
++	params = &hbus->retarget_msi_interrupt_params;
61 
++	memset(params, 0, sizeof(*params));
62 
++	params->partition_id = HV_PARTITION_ID_SELF;
63 
++	params->source = 1; /* MSI(-X) */
64 
++	params->address = msi_desc->msg.address_lo;
65 
++	params->data = msi_desc->msg.data;
66 
++	params->device_id = (hbus->hdev->dev_instance.b[5] << 24) |
67 
+ 			   (hbus->hdev->dev_instance.b[4] << 16) |
68 
+ 			   (hbus->hdev->dev_instance.b[7] << 8) |
69 
+ 			   (hbus->hdev->dev_instance.b[6] & 0xf8) |
70 
+ 			   PCI_FUNC(pdev->devfn);
71 
+-	params.vector = cfg->vector;
72 
++	params->vector = cfg->vector;
73 
+
74 
+ 	for_each_cpu_and(cpu, dest, cpu_online_mask)
75 
+-		params.vp_mask |= (1ULL << vmbus_cpu_number_to_vp_number(cpu));
76 
++		params->vp_mask |= (1ULL << vmbus_cpu_number_to_vp_number(cpu));
77 
++
78 
++	hv_do_hypercall(HVCALL_RETARGET_INTERRUPT, params, NULL);
79 
+
80 
+-	hv_do_hypercall(HVCALL_RETARGET_INTERRUPT, &params, NULL);
81 
++	spin_unlock_irqrestore(&hbus->retarget_msi_interrupt_lock, flags);
82 
+
83 
+ 	pci_msi_unmask_irq(data);
84 
+ }
85 
+@@ -2186,6 +2194,7 @@ static int hv_pci_probe(struct hv_device *hdev,
86 
+ 	INIT_LIST_HEAD(&hbus->resources_for_children);
87 
+ 	spin_lock_init(&hbus->config_lock);
88 
+ 	spin_lock_init(&hbus->device_list_lock);
89 
++	spin_lock_init(&hbus->retarget_msi_interrupt_lock);
90 
+ 	sema_init(&hbus->enum_sem, 1);
91 
+ 	init_completion(&hbus->remove_event);
92 
+
93 
+--
94 
+cgit 1.2-0.3.lf.el7
95 
+
SPECS/linux/0002_PCI_hv_Add_vPCI_version_protocol_negotiation.patch
History View file @ 7b0f3a5
0 96 
new file mode 100644
... ... 
@@ -0,0 +1,136 @@
0 
+From b1db7e7e1d70035cbd0a7be32006af7714737157 Mon Sep 17 00:00:00 2001
1 
+From: Jork Loeser <jloeser@microsoft.com>
2 
+Date: Wed, 24 May 2017 13:41:27 -0700
3 
+Subject: PCI: hv: Add vPCI version protocol negotiation
4 
+
5 
+Hyper-V vPCI offers different protocol versions.  Add the infra for
6 
+negotiating the one to use.
7 
+
8 
+Signed-off-by: Jork Loeser <jloeser@microsoft.com>
9 
+Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
10 
+Reviewed-by: K. Y. Srinivasan <kys@microsoft.com>
11 
+Acked-by: K. Y. Srinivasan <kys@microsoft.com>
12 
+Signed-off-by: Ajay Kaher <akaher@vmware.com>
13 
+---
14 
+ drivers/pci/host/pci-hyperv.c | 72 +++++++++++++++++++++++++++++++------------
15 
+ 1 file changed, 53 insertions(+), 19 deletions(-)
16 
+
17 
+(limited to 'drivers/pci/host/pci-hyperv.c')
18 
+
19 
+diff --git a/drivers/pci/host/pci-hyperv.c b/drivers/pci/host/pci-hyperv.c
20 
+index cf70fbc8e929..4a8a21e37b8e 100644
21 
+--- a/drivers/pci/host/pci-hyperv.c
22 
+@@ -64,22 +64,37 @@
23 
+  * major version.
24 
+  */
25 
+
26 
+-#define PCI_MAKE_VERSION(major, minor) ((u32)(((major) << 16) | (major)))
27 
++#define PCI_MAKE_VERSION(major, minor) ((u32)(((major) << 16) | (minor)))
28 
+ #define PCI_MAJOR_VERSION(version) ((u32)(version) >> 16)
29 
+ #define PCI_MINOR_VERSION(version) ((u32)(version) & 0xff)
30 
+
31 
+-enum {
32 
+-	PCI_PROTOCOL_VERSION_1_1 = PCI_MAKE_VERSION(1, 1),
33 
+-	PCI_PROTOCOL_VERSION_CURRENT = PCI_PROTOCOL_VERSION_1_1
34 
++enum pci_protocol_version_t {
35 
++	PCI_PROTOCOL_VERSION_1_1 = PCI_MAKE_VERSION(1, 1),	/* Win10 */
36 
+ };
37 
+
38 
+ #define CPU_AFFINITY_ALL	-1ULL
39 
++
40 
++/*
41 
++ * Supported protocol versions in the order of probing - highest go
42 
++ * first.
43 
++ */
44 
++static enum pci_protocol_version_t pci_protocol_versions[] = {
45 
++	PCI_PROTOCOL_VERSION_1_1,
46 
++};
47 
++
48 
++/*
49 
++ * Protocol version negotiated by hv_pci_protocol_negotiation().
50 
++ */
51 
++static enum pci_protocol_version_t pci_protocol_version;
52 
++
53 
+ #define PCI_CONFIG_MMIO_LENGTH	0x2000
54 
+ #define CFG_PAGE_OFFSET 0x1000
55 
+ #define CFG_PAGE_SIZE (PCI_CONFIG_MMIO_LENGTH - CFG_PAGE_OFFSET)
56 
+
57 
+ #define MAX_SUPPORTED_MSI_MESSAGES 0x400
58 
+
59 
++#define STATUS_REVISION_MISMATCH 0xC0000059
60 
++
61 
+ /*
62 
+  * Message Types
63 
+  */
64 
+@@ -1849,6 +1864,7 @@ static int hv_pci_protocol_negotiation(struct hv_device *hdev)
65 
+ 	struct hv_pci_compl comp_pkt;
66 
+ 	struct pci_packet *pkt;
67 
+ 	int ret;
68 
++	int i;
69 
+
70 
+ 	/*
71 
+ 	 * Initiate the handshake with the host and negotiate
72 
+@@ -1865,26 +1881,44 @@ static int hv_pci_protocol_negotiation(struct hv_device *hdev)
73 
+ 	pkt->compl_ctxt = &comp_pkt;
74 
+ 	version_req = (struct pci_version_request *)&pkt->message;
75 
+ 	version_req->message_type.type = PCI_QUERY_PROTOCOL_VERSION;
76 
+-	version_req->protocol_version = PCI_PROTOCOL_VERSION_CURRENT;
77 
+
78 
+-	ret = vmbus_sendpacket(hdev->channel, version_req,
79 
+-			       sizeof(struct pci_version_request),
80 
+-			       (unsigned long)pkt, VM_PKT_DATA_INBAND,
81 
+-			       VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
82 
+-	if (ret)
83 
+-		goto exit;
84 
++	for (i = 0; i < ARRAY_SIZE(pci_protocol_versions); i++) {
85 
++		version_req->protocol_version = pci_protocol_versions[i];
86 
++		ret = vmbus_sendpacket(hdev->channel, version_req,
87 
++				sizeof(struct pci_version_request),
88 
++				(unsigned long)pkt, VM_PKT_DATA_INBAND,
89 
++				VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
90 
++		if (ret) {
91 
++			dev_err(&hdev->device,
92 
++				"PCI Pass-through VSP failed sending version reqquest: %#x",
93 
++				ret);
94 
++			goto exit;
95 
++		}
96 
+
97 
+-	wait_for_completion(&comp_pkt.host_event);
98 
++		wait_for_completion(&comp_pkt.host_event);
99 
+
100 
+-	if (comp_pkt.completion_status < 0) {
101 
+-		dev_err(&hdev->device,
102 
+-			"PCI Pass-through VSP failed version request %x\n",
103 
+-			comp_pkt.completion_status);
104 
+-		ret = -EPROTO;
105 
+-		goto exit;
106 
++		if (comp_pkt.completion_status >= 0) {
107 
++			pci_protocol_version = pci_protocol_versions[i];
108 
++			dev_info(&hdev->device,
109 
++				"PCI VMBus probing: Using version %#x\n",
110 
++				pci_protocol_version);
111 
++			goto exit;
112 
++		}
113 
++
114 
++		if (comp_pkt.completion_status != STATUS_REVISION_MISMATCH) {
115 
++			dev_err(&hdev->device,
116 
++				"PCI Pass-through VSP failed version request: %#x",
117 
++				comp_pkt.completion_status);
118 
++			ret = -EPROTO;
119 
++			goto exit;
120 
++		}
121 
++
122 
++		reinit_completion(&comp_pkt.host_event);
123 
+ 	}
124 
+
125 
+-	ret = 0;
126 
++	dev_err(&hdev->device,
127 
++		"PCI pass-through VSP failed to find supported version");
128 
++	ret = -EPROTO;
129 
+
130 
+ exit:
131 
+ 	kfree(pkt);
132 
+--
133 
+cgit 1.2-0.3.lf.el7
134 
+
SPECS/linux/0003_PCI_hv_Use_vPCI_protocol_version_1.2_v4.9.patch
History View file @ 7b0f3a5
0 135 
new file mode 100755
... ... 
@@ -0,0 +1,465 @@
0 
+From 7dcf90e9e032432e91ce77dd872d2227e9d5b741 Mon Sep 17 00:00:00 2001
1 
+From: Jork Loeser <jloeser@microsoft.com>
2 
+Date: Wed, 24 May 2017 13:41:28 -0700
3 
+Subject: PCI: hv: Use vPCI protocol version 1.2
4 
+
5 
+Update the Hyper-V vPCI driver to use the Server-2016 version of the vPCI
6 
+protocol, fixing MSI creation and retargeting issues.
7 
+
8 
+Replaced hv_tmp_cpu_nr_to_vp_nr() with vmbus_cpu_number_to_vp_number()
9 
+to make this patch compatibale for linux v4.9.
10 
+
11 
+Signed-off-by: Jork Loeser <jloeser@microsoft.com>
12 
+Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
13 
+Reviewed-by: K. Y. Srinivasan <kys@microsoft.com>
14 
+Acked-by: K. Y. Srinivasan <kys@microsoft.com>
15 
+Signed-off-by: Ajay Kaher <akaher@vmware.com>
16 
+---
17 
+ drivers/pci/host/pci-hyperv.c | 300 ++++++++++++++++++++++++++++++++++--------
18 
+ 1 file changed, 246 insertions(+), 54 deletions(-)
19 
+
20 
+(limited to 'drivers/pci/host/pci-hyperv.c')
21 
+
22 
+diff --git a/drivers/pci/host/pci-hyperv.c b/drivers/pci/host/pci-hyperv.c
23 
+index 4a8a21e37b8e..415dcc69a502 100644
24 
+--- a/drivers/pci/host/pci-hyperv.c
25 
+@@ -70,6 +70,7 @@
26 
+
27 
+ enum pci_protocol_version_t {
28 
+ 	PCI_PROTOCOL_VERSION_1_1 = PCI_MAKE_VERSION(1, 1),	/* Win10 */
29 
++	PCI_PROTOCOL_VERSION_1_2 = PCI_MAKE_VERSION(1, 2),	/* RS1 */
30 
+ };
31 
+
32 
+ #define CPU_AFFINITY_ALL	-1ULL
33 
+@@ -79,6 +80,7 @@ enum pci_protocol_version_t {
34 
+  * first.
35 
+  */
36 
+ static enum pci_protocol_version_t pci_protocol_versions[] = {
37 
++	PCI_PROTOCOL_VERSION_1_2,
38 
+ 	PCI_PROTOCOL_VERSION_1_1,
39 
+ };
40 
+
41 
+@@ -124,6 +126,9 @@ enum pci_message_type {
42 
+ 	PCI_QUERY_PROTOCOL_VERSION      = PCI_MESSAGE_BASE + 0x13,
43 
+ 	PCI_CREATE_INTERRUPT_MESSAGE    = PCI_MESSAGE_BASE + 0x14,
44 
+ 	PCI_DELETE_INTERRUPT_MESSAGE    = PCI_MESSAGE_BASE + 0x15,
45 
++	PCI_RESOURCES_ASSIGNED2		= PCI_MESSAGE_BASE + 0x16,
46 
++	PCI_CREATE_INTERRUPT_MESSAGE2	= PCI_MESSAGE_BASE + 0x17,
47 
++	PCI_DELETE_INTERRUPT_MESSAGE2	= PCI_MESSAGE_BASE + 0x18, /* unused */
48 
+ 	PCI_MESSAGE_MAXIMUM
49 
+ };
50 
+
51 
+@@ -193,6 +198,30 @@ struct hv_msi_desc {
52 
+ 	u64	cpu_mask;
53 
+ } __packed;
54 
+
55 
++/**
56 
++ * struct hv_msi_desc2 - 1.2 version of hv_msi_desc
57 
++ * @vector:		IDT entry
58 
++ * @delivery_mode:	As defined in Intel's Programmer's
59 
++ *			Reference Manual, Volume 3, Chapter 8.
60 
++ * @vector_count:	Number of contiguous entries in the
61 
++ *			Interrupt Descriptor Table that are
62 
++ *			occupied by this Message-Signaled
63 
++ *			Interrupt. For "MSI", as first defined
64 
++ *			in PCI 2.2, this can be between 1 and
65 
++ *			32. For "MSI-X," as first defined in PCI
66 
++ *			3.0, this must be 1, as each MSI-X table
67 
++ *			entry would have its own descriptor.
68 
++ * @processor_count:	number of bits enabled in array.
69 
++ * @processor_array:	All the target virtual processors.
70 
++ */
71 
++struct hv_msi_desc2 {
72 
++	u8	vector;
73 
++	u8	delivery_mode;
74 
++	u16	vector_count;
75 
++	u16	processor_count;
76 
++	u16	processor_array[32];
77 
++} __packed;
78 
++
79 
+ /**
80 
+  * struct tran_int_desc
81 
+  * @reserved:		unused, padding
82 
+@@ -309,6 +338,14 @@ struct pci_resources_assigned {
83 
+ 	u32 reserved[4];
84 
+ } __packed;
85 
+
86 
++struct pci_resources_assigned2 {
87 
++	struct pci_message message_type;
88 
++	union win_slot_encoding wslot;
89 
++	u8 memory_range[0x14][6];	/* not used here */
90 
++	u32 msi_descriptor_count;
91 
++	u8 reserved[70];
92 
++} __packed;
93 
++
94 
+ struct pci_create_interrupt {
95 
+ 	struct pci_message message_type;
96 
+ 	union win_slot_encoding wslot;
97 
+@@ -321,6 +358,12 @@ struct pci_create_int_response {
98 
+ 	struct tran_int_desc int_desc;
99 
+ } __packed;
100 
+
101 
++struct pci_create_interrupt2 {
102 
++	struct pci_message message_type;
103 
++	union win_slot_encoding wslot;
104 
++	struct hv_msi_desc2 int_desc;
105 
++} __packed;
106 
++
107 
+ struct pci_delete_interrupt {
108 
+ 	struct pci_message message_type;
109 
+ 	union win_slot_encoding wslot;
110 
+@@ -346,17 +389,42 @@ static int pci_ring_size = (4 * PAGE_SIZE);
111 
+ #define HV_PARTITION_ID_SELF		((u64)-1)
112 
+ #define HVCALL_RETARGET_INTERRUPT	0x7e
113 
+
114 
+-struct retarget_msi_interrupt {
115 
+-	u64	partition_id;		/* use "self" */
116 
+-	u64	device_id;
117 
++struct hv_interrupt_entry {
118 
+ 	u32	source;			/* 1 for MSI(-X) */
119 
+ 	u32	reserved1;
120 
+ 	u32	address;
121 
+ 	u32	data;
122 
+-	u64	reserved2;
123 
++};
124 
++
125 
++#define HV_VP_SET_BANK_COUNT_MAX	5 /* current implementation limit */
126 
++
127 
++struct hv_vp_set {
128 
++	u64	format;			/* 0 (HvGenericSetSparse4k) */
129 
++	u64	valid_banks;
130 
++	u64	masks[HV_VP_SET_BANK_COUNT_MAX];
131 
++};
132 
++
133 
++/*
134 
++ * flags for hv_device_interrupt_target.flags
135 
++ */
136 
++#define HV_DEVICE_INTERRUPT_TARGET_MULTICAST		1
137 
++#define HV_DEVICE_INTERRUPT_TARGET_PROCESSOR_SET	2
138 
++
139 
++struct hv_device_interrupt_target {
140 
+ 	u32	vector;
141 
+ 	u32	flags;
142 
+-	u64	vp_mask;
143 
++	union {
144 
++		u64		 vp_mask;
145 
++		struct hv_vp_set vp_set;
146 
++	};
147 
++};
148 
++
149 
++struct retarget_msi_interrupt {
150 
++	u64	partition_id;		/* use "self" */
151 
++	u64	device_id;
152 
++	struct hv_interrupt_entry int_entry;
153 
++	u64	reserved2;
154 
++	struct hv_device_interrupt_target int_target;
155 
+ } __packed;
156 
+
157 
+ /*
158 
+@@ -850,8 +918,11 @@ static void hv_irq_unmask(struct irq_data *data)
159 
+ 	struct cpumask *dest;
160 
+ 	struct pci_bus *pbus;
161 
+ 	struct pci_dev *pdev;
162 
+-	int cpu;
163 
+ 	unsigned long flags;
164 
++	u32 var_size = 0;
165 
++	int cpu_vmbus;
166 
++	int cpu;
167 
++	u64 res;
168 
+
169 
+ 	dest = irq_data_get_affinity_mask(data);
170 
+ 	pdev = msi_desc_to_pci_dev(msi_desc);
171 
+@@ -863,23 +934,74 @@ static void hv_irq_unmask(struct irq_data *data)
172 
+ 	params = &hbus->retarget_msi_interrupt_params;
173 
+ 	memset(params, 0, sizeof(*params));
174 
+ 	params->partition_id = HV_PARTITION_ID_SELF;
175 
+-	params->source = 1; /* MSI(-X) */
176 
+-	params->address = msi_desc->msg.address_lo;
177 
+-	params->data = msi_desc->msg.data;
178 
++	params->int_entry.source = 1; /* MSI(-X) */
179 
++	params->int_entry.address = msi_desc->msg.address_lo;
180 
++	params->int_entry.data = msi_desc->msg.data;
181 
+ 	params->device_id = (hbus->hdev->dev_instance.b[5] << 24) |
182 
+ 			   (hbus->hdev->dev_instance.b[4] << 16) |
183 
+ 			   (hbus->hdev->dev_instance.b[7] << 8) |
184 
+ 			   (hbus->hdev->dev_instance.b[6] & 0xf8) |
185 
+ 			   PCI_FUNC(pdev->devfn);
186 
+-	params->vector = cfg->vector;
187 
++	params->int_target.vector = cfg->vector;
188 
++
189 
++	/*
190 
++	 * Honoring apic->irq_delivery_mode set to dest_Fixed by
191 
++	 * setting the HV_DEVICE_INTERRUPT_TARGET_MULTICAST flag results in a
192 
++	 * spurious interrupt storm. Not doing so does not seem to have a
193 
++	 * negative effect (yet?).
194 
++	 */
195 
++
196 
++	if (pci_protocol_version >= PCI_PROTOCOL_VERSION_1_2) {
197 
++		/*
198 
++		 * PCI_PROTOCOL_VERSION_1_2 supports the VP_SET version of the
199 
++		 * HVCALL_RETARGET_INTERRUPT hypercall, which also coincides
200 
++		 * with >64 VP support.
201 
++		 * ms_hyperv.hints & HV_X64_EX_PROCESSOR_MASKS_RECOMMENDED
202 
++		 * is not sufficient for this hypercall.
203 
++		 */
204 
++		params->int_target.flags |=
205 
++			HV_DEVICE_INTERRUPT_TARGET_PROCESSOR_SET;
206 
++		params->int_target.vp_set.valid_banks =
207 
++			(1ull << HV_VP_SET_BANK_COUNT_MAX) - 1;
208 
++
209 
++		/*
210 
++		 * var-sized hypercall, var-size starts after vp_mask (thus
211 
++		 * vp_set.format does not count, but vp_set.valid_banks does).
212 
++		 */
213 
++		var_size = 1 + HV_VP_SET_BANK_COUNT_MAX;
214 
++
215 
++		for_each_cpu_and(cpu, dest, cpu_online_mask) {
216 
++			cpu_vmbus = vmbus_cpu_number_to_vp_number(cpu);
217 
++
218 
++			if (cpu_vmbus >= HV_VP_SET_BANK_COUNT_MAX * 64) {
219 
++				dev_err(&hbus->hdev->device,
220 
++					"too high CPU %d", cpu_vmbus);
221 
++				res = 1;
222 
++				goto exit_unlock;
223 
++			}
224 
+
225 
+-	for_each_cpu_and(cpu, dest, cpu_online_mask)
226 
+-		params->vp_mask |= (1ULL << vmbus_cpu_number_to_vp_number(cpu));
227 
++			params->int_target.vp_set.masks[cpu_vmbus / 64] |=
228 
++				(1ULL << (cpu_vmbus & 63));
229 
++		}
230 
++	} else {
231 
++		for_each_cpu_and(cpu, dest, cpu_online_mask) {
232 
++			params->int_target.vp_mask |=
233 
++				(1ULL << vmbus_cpu_number_to_vp_number(cpu));
234 
++		}
235 
++	}
236 
+
237 
+-	hv_do_hypercall(HVCALL_RETARGET_INTERRUPT, params, NULL);
238 
++	res = hv_do_hypercall(HVCALL_RETARGET_INTERRUPT | (var_size << 17),
239 
++			      params, NULL);
240 
+
241 
++exit_unlock:
242 
+ 	spin_unlock_irqrestore(&hbus->retarget_msi_interrupt_lock, flags);
243 
+
244 
++	if (res) {
245 
++		dev_err(&hbus->hdev->device,
246 
++			"%s() failed: %#llx", __func__, res);
247 
++		return;
248 
++	}
249 
++
250 
+ 	pci_msi_unmask_irq(data);
251 
+ }
252 
+
253 
+@@ -900,6 +1022,53 @@ static void hv_pci_compose_compl(void *context, struct pci_response *resp,
254 
+ 	complete(&comp_pkt->comp_pkt.host_event);
255 
+ }
256 
+
257 
++static u32 hv_compose_msi_req_v1(
258 
++	struct pci_create_interrupt *int_pkt, struct cpumask *affinity,
259 
++	u32 slot, u8 vector)
260 
++{
261 
++	int_pkt->message_type.type = PCI_CREATE_INTERRUPT_MESSAGE;
262 
++	int_pkt->wslot.slot = slot;
263 
++	int_pkt->int_desc.vector = vector;
264 
++	int_pkt->int_desc.vector_count = 1;
265 
++	int_pkt->int_desc.delivery_mode =
266 
++		(apic->irq_delivery_mode == dest_LowestPrio) ?
267 
++			dest_LowestPrio : dest_Fixed;
268 
++
269 
++	/*
270 
++	 * Create MSI w/ dummy vCPU set, overwritten by subsequent retarget in
271 
++	 * hv_irq_unmask().
272 
++	 */
273 
++	int_pkt->int_desc.cpu_mask = CPU_AFFINITY_ALL;
274 
++
275 
++	return sizeof(*int_pkt);
276 
++}
277 
++
278 
++static u32 hv_compose_msi_req_v2(
279 
++	struct pci_create_interrupt2 *int_pkt, struct cpumask *affinity,
280 
++	u32 slot, u8 vector)
281 
++{
282 
++	int cpu;
283 
++
284 
++	int_pkt->message_type.type = PCI_CREATE_INTERRUPT_MESSAGE2;
285 
++	int_pkt->wslot.slot = slot;
286 
++	int_pkt->int_desc.vector = vector;
287 
++	int_pkt->int_desc.vector_count = 1;
288 
++	int_pkt->int_desc.delivery_mode =
289 
++		(apic->irq_delivery_mode == dest_LowestPrio) ?
290 
++			dest_LowestPrio : dest_Fixed;
291 
++
292 
++	/*
293 
++	 * Create MSI w/ dummy vCPU set targeting just one vCPU, overwritten
294 
++	 * by subsequent retarget in hv_irq_unmask().
295 
++	 */
296 
++	cpu = cpumask_first_and(affinity, cpu_online_mask);
297 
++	int_pkt->int_desc.processor_array[0] =
298 
++		vmbus_cpu_number_to_vp_number(cpu);
299 
++	int_pkt->int_desc.processor_count = 1;
300 
++
301 
++	return sizeof(*int_pkt);
302 
++}
303 
++
304 
+ /**
305 
+  * hv_compose_msi_msg() - Supplies a valid MSI address/data
306 
+  * @data:	Everything about this MSI
307 
+@@ -918,15 +1087,17 @@ static void hv_compose_msi_msg(struct irq_data *data, struct msi_msg *msg)
308 
+ 	struct hv_pci_dev *hpdev;
309 
+ 	struct pci_bus *pbus;
310 
+ 	struct pci_dev *pdev;
311 
+-	struct pci_create_interrupt *int_pkt;
312 
+ 	struct compose_comp_ctxt comp;
313 
+ 	struct tran_int_desc *int_desc;
314 
+-	struct cpumask *affinity;
315 
+ 	struct {
316 
+-		struct pci_packet pkt;
317 
+-		u8 buffer[sizeof(struct pci_create_interrupt)];
318 
+-	} ctxt;
319 
+-	int cpu;
320 
++		struct pci_packet pci_pkt;
321 
++		union {
322 
++			struct pci_create_interrupt v1;
323 
++			struct pci_create_interrupt2 v2;
324 
++		} int_pkts;
325 
++	} __packed ctxt;
326 
++
327 
++	u32 size;
328 
+ 	int ret;
329 
+
330 
+ 	pdev = msi_desc_to_pci_dev(irq_data_get_msi_desc(data));
331 
+@@ -949,36 +1120,44 @@ static void hv_compose_msi_msg(struct irq_data *data, struct msi_msg *msg)
332 
+
333 
+ 	memset(&ctxt, 0, sizeof(ctxt));
334 
+ 	init_completion(&comp.comp_pkt.host_event);
335 
+-	ctxt.pkt.completion_func = hv_pci_compose_compl;
336 
+-	ctxt.pkt.compl_ctxt = &comp;
337 
+-	int_pkt = (struct pci_create_interrupt *)&ctxt.pkt.message;
338 
+-	int_pkt->message_type.type = PCI_CREATE_INTERRUPT_MESSAGE;
339 
+-	int_pkt->wslot.slot = hpdev->desc.win_slot.slot;
340 
+-	int_pkt->int_desc.vector = cfg->vector;
341 
+-	int_pkt->int_desc.vector_count = 1;
342 
+-	int_pkt->int_desc.delivery_mode =
343 
+-		(apic->irq_delivery_mode == dest_LowestPrio) ? 1 : 0;
344 
++	ctxt.pci_pkt.completion_func = hv_pci_compose_compl;
345 
++	ctxt.pci_pkt.compl_ctxt = &comp;
346 
++
347 
++	switch (pci_protocol_version) {
348 
++	case PCI_PROTOCOL_VERSION_1_1:
349 
++		size = hv_compose_msi_req_v1(&ctxt.int_pkts.v1,
350 
++					irq_data_get_affinity_mask(data),
351 
++					hpdev->desc.win_slot.slot,
352 
++					cfg->vector);
353 
++		break;
354 
+
355 
+-	/*
356 
+-	 * This bit doesn't have to work on machines with more than 64
357 
+-	 * processors because Hyper-V only supports 64 in a guest.
358 
+-	 */
359 
+-	affinity = irq_data_get_affinity_mask(data);
360 
+-	if (cpumask_weight(affinity) >= 32) {
361 
+-		int_pkt->int_desc.cpu_mask = CPU_AFFINITY_ALL;
362 
+-	} else {
363 
+-		for_each_cpu_and(cpu, affinity, cpu_online_mask) {
364 
+-			int_pkt->int_desc.cpu_mask |=
365 
+-				(1ULL << vmbus_cpu_number_to_vp_number(cpu));
366 
+-		}
367 
++	case PCI_PROTOCOL_VERSION_1_2:
368 
++		size = hv_compose_msi_req_v2(&ctxt.int_pkts.v2,
369 
++					irq_data_get_affinity_mask(data),
370 
++					hpdev->desc.win_slot.slot,
371 
++					cfg->vector);
372 
++		break;
373 
++
374 
++	default:
375 
++		/* As we only negotiate protocol versions known to this driver,
376 
++		 * this path should never hit. However, this is it not a hot
377 
++		 * path so we print a message to aid future updates.
378 
++		 */
379 
++		dev_err(&hbus->hdev->device,
380 
++			"Unexpected vPCI protocol, update driver.");
381 
++		goto free_int_desc;
382 
+ 	}
383 
+
384 
+-	ret = vmbus_sendpacket(hpdev->hbus->hdev->channel, int_pkt,
385 
+-			       sizeof(*int_pkt), (unsigned long)&ctxt.pkt,
386 
++	ret = vmbus_sendpacket(hpdev->hbus->hdev->channel, &ctxt.int_pkts,
387 
++			       size, (unsigned long)&ctxt.pci_pkt,
388 
+ 			       VM_PKT_DATA_INBAND,
389 
+ 			       VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
390 
+-	if (ret)
391 
++	if (ret) {
392 
++		dev_err(&hbus->hdev->device,
393 
++			"Sending request for interrupt failed: 0x%x",
394 
++			comp.comp_pkt.completion_status);
395 
+ 		goto free_int_desc;
396 
++	}
397 
+
398 
+ 	wait_for_completion(&comp.comp_pkt.host_event);
399 
+
400 
+@@ -2177,13 +2356,18 @@ static int hv_send_resources_allocated(struct hv_device *hdev)
401 
+ {
402 
+ 	struct hv_pcibus_device *hbus = hv_get_drvdata(hdev);
403 
+ 	struct pci_resources_assigned *res_assigned;
404 
++	struct pci_resources_assigned2 *res_assigned2;
405 
+ 	struct hv_pci_compl comp_pkt;
406 
+ 	struct hv_pci_dev *hpdev;
407 
+ 	struct pci_packet *pkt;
408 
++	size_t size_res;
409 
+ 	u32 wslot;
410 
+ 	int ret;
411 
+
412 
+-	pkt = kmalloc(sizeof(*pkt) + sizeof(*res_assigned), GFP_KERNEL);
413 
++	size_res = (pci_protocol_version < PCI_PROTOCOL_VERSION_1_2)
414 
++			? sizeof(*res_assigned) : sizeof(*res_assigned2);
415 
++
416 
++	pkt = kmalloc(sizeof(*pkt) + size_res, GFP_KERNEL);
417 
+ 	if (!pkt)
418 
+ 		return -ENOMEM;
419 
+
420 
+@@ -2194,22 +2378,30 @@ static int hv_send_resources_allocated(struct hv_device *hdev)
421 
+ 		if (!hpdev)
422 
+ 			continue;
423 
+
424 
+-		memset(pkt, 0, sizeof(*pkt) + sizeof(*res_assigned));
425 
++		memset(pkt, 0, sizeof(*pkt) + size_res);
426 
+ 		init_completion(&comp_pkt.host_event);
427 
+ 		pkt->completion_func = hv_pci_generic_compl;
428 
+ 		pkt->compl_ctxt = &comp_pkt;
429 
+-		res_assigned = (struct pci_resources_assigned *)&pkt->message;
430 
+-		res_assigned->message_type.type = PCI_RESOURCES_ASSIGNED;
431 
+-		res_assigned->wslot.slot = hpdev->desc.win_slot.slot;
432 
+
433 
++		if (pci_protocol_version < PCI_PROTOCOL_VERSION_1_2) {
434 
++			res_assigned =
435 
++				(struct pci_resources_assigned *)&pkt->message;
436 
++			res_assigned->message_type.type =
437 
++				PCI_RESOURCES_ASSIGNED;
438 
++			res_assigned->wslot.slot = hpdev->desc.win_slot.slot;
439 
++		} else {
440 
++			res_assigned2 =
441 
++				(struct pci_resources_assigned2 *)&pkt->message;
442 
++			res_assigned2->message_type.type =
443 
++				PCI_RESOURCES_ASSIGNED2;
444 
++			res_assigned2->wslot.slot = hpdev->desc.win_slot.slot;
445 
++		}
446 
+ 		put_pcichild(hpdev, hv_pcidev_ref_by_slot);
447 
+
448 
+-		ret = vmbus_sendpacket(
449 
+-			hdev->channel, &pkt->message,
450 
+-			sizeof(*res_assigned),
451 
+-			(unsigned long)pkt,
452 
+-			VM_PKT_DATA_INBAND,
453 
+-			VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
454 
++		ret = vmbus_sendpacket(hdev->channel, &pkt->message,
455 
++				size_res, (unsigned long)pkt,
456 
++				VM_PKT_DATA_INBAND,
457 
++				VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
458 
+ 		if (ret)
459 
+ 			break;
460 
+
461 
+--
462 
+cgit 1.2-0.3.lf.el7
463 
+
SPECS/linux/config
History View file @ 7b0f3a5
0 464 
old mode 100644
1 465 
new mode 100755
... ... 
@@ -1,6 +1,6 @@
1 1 
 #
2 2 
 # Automatically generated file; DO NOT EDIT.
3 
-# Linux/x86 4.9.120 Kernel Configuration
3 
+# Linux/x86 4.9.140 Kernel Configuration
4 4 
 #
5 5 
 CONFIG_64BIT=y
6 6 
 CONFIG_X86_64=y
... ... 
@@ -770,7 +770,7 @@ CONFIG_PCI_IOV=y
770 770 
 CONFIG_PCI_PRI=y
771 771 
 CONFIG_PCI_PASID=y
772 772 
 CONFIG_PCI_LABEL=y
773 
-# CONFIG_PCI_HYPERV is not set
773 
+CONFIG_PCI_HYPERV=m
774 774 
 CONFIG_HOTPLUG_PCI=y
775 775 
 CONFIG_HOTPLUG_PCI_ACPI=y
776 776 
 CONFIG_HOTPLUG_PCI_ACPI_IBM=m
... ... 
@@ -1384,6 +1384,7 @@ CONFIG_VSOCKETS=m
1384 1384 
 CONFIG_VMWARE_VMCI_VSOCKETS=m
1385 1385 
 CONFIG_VIRTIO_VSOCKETS=m
1386 1386 
 CONFIG_VIRTIO_VSOCKETS_COMMON=m
1387 
+CONFIG_HYPERV_SOCK=m
1387 1388 
 CONFIG_NETLINK_DIAG=m
1388 1389 
 CONFIG_MPLS=y
1389 1390 
 CONFIG_NET_MPLS_GSO=m
SPECS/linux/linux.spec
History View file @ 7b0f3a5
... ... 
@@ -2,7 +2,7 @@
2 2 
 Summary:        Kernel
3 3 
 Name:           linux
4 4 
 Version:        4.9.140
5 
-Release:        1%{?kat_build:.%kat_build}%{?dist}
5 
+Release:        2%{?kat_build:.%kat_build}%{?dist}
6 6 
 License:    	GPLv2
7 7 
 URL:        	http://www.kernel.org/
8 8 
 Group:        	System Environment/Kernel
... ... 
@@ -71,6 +71,11 @@ Patch46:        0001-xfs-move-inode-fork-verifiers-to-xfs-dinode-verify.patch
71 71 
 Patch47:        0002-xfs-verify-dinode-header-first.patch
72 72 
 Patch48:        0003-xfs-enhance-dinode-verifier.patch
73 73
74 
+#HyperV PCI patches
75 
+Patch51:        0001_PCI_hv_Allocate_physically_contiguous_hypercall_params_buffer.patch
76 
+Patch52:        0002_PCI_hv_Add_vPCI_version_protocol_negotiation.patch
77 
+Patch53:        0003_PCI_hv_Use_vPCI_protocol_version_1.2_v4.9.patch
78 
+
74 79 
 # Out-of-tree patches from AppArmor:
75 80 
 Patch71: 0001-UBUNTU-SAUCE-AppArmor-basic-networking-rules.patch
76 81 
 Patch72: 0002-apparmor-Fix-quieting-of-audit-messages-for-network-.patch
... ... 
@@ -190,6 +195,10 @@ This package contains the 'perf' performance analysis tools for Linux kernel.
190 190 
 %patch47 -p1
191 191 
 %patch48 -p1
192 192
193 
+%patch51 -p1
194 
+%patch52 -p1
195 
+%patch53 -p1
196 
+
193 197 
 %patch71 -p1
194 198 
 %patch72 -p1
195 199 
 %patch73 -p1
... ... 
@@ -363,6 +372,8 @@ ln -sf %{name}-%{uname_r}.cfg /boot/photon.cfg
363 363 
 /usr/share/doc/*
364 364
365 365 
 %changelog
366 
+*   Mon Dec 17 2018 Ajay Kaher <akaher@vmware.com> 4.9.140-2
367 
+-   Enable pci-hyperv support and apply relevant patches
366 368 
 *   Mon Nov 26 2018 Srivatsa S. Bhat (VMware) <srivatsa@csail.mit.edu> 4.9.140-1
367 369 
 -   Update to version 4.9.140
368 370 
 *   Fri Nov 16 2018 Srivatsa S. Bhat (VMware) <srivatsa@csail.mit.edu> 4.9.137-1