1 // SPDX-License-Identifier: GPL-2.0
2
3 /*
4 * Hyper-V stub IOMMU driver.
5 *
6 * Copyright (C) 2019, Microsoft, Inc.
7 *
8 * Author : Lan Tianyu <Tianyu.Lan@microsoft.com>
9 */
10
11 #include <linux/types.h>
12 #include <linux/interrupt.h>
13 #include <linux/irq.h>
14 #include <linux/iommu.h>
15 #include <linux/module.h>
16
17 #include <asm/apic.h>
18 #include <asm/cpu.h>
19 #include <asm/hw_irq.h>
20 #include <asm/io_apic.h>
21 #include <asm/irq_remapping.h>
22 #include <asm/hypervisor.h>
23 #include <asm/mshyperv.h>
24
25 #include "irq_remapping.h"
26
27 #ifdef CONFIG_IRQ_REMAP
28
29 /*
30 * According 82093AA IO-APIC spec , IO APIC has a 24-entry Interrupt
31 * Redirection Table. Hyper-V exposes one single IO-APIC and so define
32 * 24 IO APIC remmapping entries.
33 */
34 #define IOAPIC_REMAPPING_ENTRY 24
35
36 static cpumask_t ioapic_max_cpumask = { CPU_BITS_NONE };
37 static struct irq_domain *ioapic_ir_domain;
38
hyperv_ir_set_affinity(struct irq_data * data,const struct cpumask * mask,bool force)39 static int hyperv_ir_set_affinity(struct irq_data *data,
40 const struct cpumask *mask, bool force)
41 {
42 struct irq_data *parent = data->parent_data;
43 struct irq_cfg *cfg = irqd_cfg(data);
44 int ret;
45
46 /* Return error If new irq affinity is out of ioapic_max_cpumask. */
47 if (!cpumask_subset(mask, &ioapic_max_cpumask))
48 return -EINVAL;
49
50 ret = parent->chip->irq_set_affinity(parent, mask, force);
51 if (ret < 0 || ret == IRQ_SET_MASK_OK_DONE)
52 return ret;
53
54 send_cleanup_vector(cfg);
55
56 return 0;
57 }
58
59 static struct irq_chip hyperv_ir_chip = {
60 .name = "HYPERV-IR",
61 .irq_ack = apic_ack_irq,
62 .irq_set_affinity = hyperv_ir_set_affinity,
63 };
64
hyperv_irq_remapping_alloc(struct irq_domain * domain,unsigned int virq,unsigned int nr_irqs,void * arg)65 static int hyperv_irq_remapping_alloc(struct irq_domain *domain,
66 unsigned int virq, unsigned int nr_irqs,
67 void *arg)
68 {
69 struct irq_alloc_info *info = arg;
70 struct irq_data *irq_data;
71 int ret = 0;
72
73 if (!info || info->type != X86_IRQ_ALLOC_TYPE_IOAPIC || nr_irqs > 1)
74 return -EINVAL;
75
76 ret = irq_domain_alloc_irqs_parent(domain, virq, nr_irqs, arg);
77 if (ret < 0)
78 return ret;
79
80 irq_data = irq_domain_get_irq_data(domain, virq);
81 if (!irq_data) {
82 irq_domain_free_irqs_common(domain, virq, nr_irqs);
83 return -EINVAL;
84 }
85
86 irq_data->chip = &hyperv_ir_chip;
87
88 /*
89 * Hypver-V IO APIC irq affinity should be in the scope of
90 * ioapic_max_cpumask because no irq remapping support.
91 */
92 irq_data_update_affinity(irq_data, &ioapic_max_cpumask);
93
94 return 0;
95 }
96
hyperv_irq_remapping_free(struct irq_domain * domain,unsigned int virq,unsigned int nr_irqs)97 static void hyperv_irq_remapping_free(struct irq_domain *domain,
98 unsigned int virq, unsigned int nr_irqs)
99 {
100 irq_domain_free_irqs_common(domain, virq, nr_irqs);
101 }
102
hyperv_irq_remapping_select(struct irq_domain * d,struct irq_fwspec * fwspec,enum irq_domain_bus_token bus_token)103 static int hyperv_irq_remapping_select(struct irq_domain *d,
104 struct irq_fwspec *fwspec,
105 enum irq_domain_bus_token bus_token)
106 {
107 /* Claim the only I/O APIC emulated by Hyper-V */
108 return x86_fwspec_is_ioapic(fwspec);
109 }
110
111 static const struct irq_domain_ops hyperv_ir_domain_ops = {
112 .select = hyperv_irq_remapping_select,
113 .alloc = hyperv_irq_remapping_alloc,
114 .free = hyperv_irq_remapping_free,
115 };
116
117 static const struct irq_domain_ops hyperv_root_ir_domain_ops;
hyperv_prepare_irq_remapping(void)118 static int __init hyperv_prepare_irq_remapping(void)
119 {
120 struct fwnode_handle *fn;
121 int i;
122 const char *name;
123 const struct irq_domain_ops *ops;
124
125 if (!hypervisor_is_type(X86_HYPER_MS_HYPERV) ||
126 x86_init.hyper.msi_ext_dest_id() ||
127 !x2apic_supported())
128 return -ENODEV;
129
130 if (hv_root_partition) {
131 name = "HYPERV-ROOT-IR";
132 ops = &hyperv_root_ir_domain_ops;
133 } else {
134 name = "HYPERV-IR";
135 ops = &hyperv_ir_domain_ops;
136 }
137
138 fn = irq_domain_alloc_named_id_fwnode(name, 0);
139 if (!fn)
140 return -ENOMEM;
141
142 ioapic_ir_domain =
143 irq_domain_create_hierarchy(arch_get_ir_parent_domain(),
144 0, IOAPIC_REMAPPING_ENTRY, fn, ops, NULL);
145
146 if (!ioapic_ir_domain) {
147 irq_domain_free_fwnode(fn);
148 return -ENOMEM;
149 }
150
151 if (hv_root_partition)
152 return 0; /* The rest is only relevant to guests */
153
154 /*
155 * Hyper-V doesn't provide irq remapping function for
156 * IO-APIC and so IO-APIC only accepts 8-bit APIC ID.
157 * Cpu's APIC ID is read from ACPI MADT table and APIC IDs
158 * in the MADT table on Hyper-v are sorted monotonic increasingly.
159 * APIC ID reflects cpu topology. There maybe some APIC ID
160 * gaps when cpu number in a socket is not power of two. Prepare
161 * max cpu affinity for IOAPIC irqs. Scan cpu 0-255 and set cpu
162 * into ioapic_max_cpumask if its APIC ID is less than 256.
163 */
164 for (i = min_t(unsigned int, num_possible_cpus() - 1, 255); i >= 0; i--)
165 if (cpu_physical_id(i) < 256)
166 cpumask_set_cpu(i, &ioapic_max_cpumask);
167
168 return 0;
169 }
170
hyperv_enable_irq_remapping(void)171 static int __init hyperv_enable_irq_remapping(void)
172 {
173 return IRQ_REMAP_X2APIC_MODE;
174 }
175
176 struct irq_remap_ops hyperv_irq_remap_ops = {
177 .prepare = hyperv_prepare_irq_remapping,
178 .enable = hyperv_enable_irq_remapping,
179 };
180
181 /* IRQ remapping domain when Linux runs as the root partition */
182 struct hyperv_root_ir_data {
183 u8 ioapic_id;
184 bool is_level;
185 struct hv_interrupt_entry entry;
186 };
187
188 static void
hyperv_root_ir_compose_msi_msg(struct irq_data * irq_data,struct msi_msg * msg)189 hyperv_root_ir_compose_msi_msg(struct irq_data *irq_data, struct msi_msg *msg)
190 {
191 u64 status;
192 u32 vector;
193 struct irq_cfg *cfg;
194 int ioapic_id;
195 const struct cpumask *affinity;
196 int cpu;
197 struct hv_interrupt_entry entry;
198 struct hyperv_root_ir_data *data = irq_data->chip_data;
199 struct IO_APIC_route_entry e;
200
201 cfg = irqd_cfg(irq_data);
202 affinity = irq_data_get_effective_affinity_mask(irq_data);
203 cpu = cpumask_first_and(affinity, cpu_online_mask);
204
205 vector = cfg->vector;
206 ioapic_id = data->ioapic_id;
207
208 if (data->entry.source == HV_DEVICE_TYPE_IOAPIC
209 && data->entry.ioapic_rte.as_uint64) {
210 entry = data->entry;
211
212 status = hv_unmap_ioapic_interrupt(ioapic_id, &entry);
213
214 if (status != HV_STATUS_SUCCESS)
215 pr_debug("%s: unexpected unmap status %lld\n", __func__, status);
216
217 data->entry.ioapic_rte.as_uint64 = 0;
218 data->entry.source = 0; /* Invalid source */
219 }
220
221
222 status = hv_map_ioapic_interrupt(ioapic_id, data->is_level, cpu,
223 vector, &entry);
224
225 if (status != HV_STATUS_SUCCESS) {
226 pr_err("%s: map hypercall failed, status %lld\n", __func__, status);
227 return;
228 }
229
230 data->entry = entry;
231
232 /* Turn it into an IO_APIC_route_entry, and generate MSI MSG. */
233 e.w1 = entry.ioapic_rte.low_uint32;
234 e.w2 = entry.ioapic_rte.high_uint32;
235
236 memset(msg, 0, sizeof(*msg));
237 msg->arch_data.vector = e.vector;
238 msg->arch_data.delivery_mode = e.delivery_mode;
239 msg->arch_addr_lo.dest_mode_logical = e.dest_mode_logical;
240 msg->arch_addr_lo.dmar_format = e.ir_format;
241 msg->arch_addr_lo.dmar_index_0_14 = e.ir_index_0_14;
242 }
243
hyperv_root_ir_set_affinity(struct irq_data * data,const struct cpumask * mask,bool force)244 static int hyperv_root_ir_set_affinity(struct irq_data *data,
245 const struct cpumask *mask, bool force)
246 {
247 struct irq_data *parent = data->parent_data;
248 struct irq_cfg *cfg = irqd_cfg(data);
249 int ret;
250
251 ret = parent->chip->irq_set_affinity(parent, mask, force);
252 if (ret < 0 || ret == IRQ_SET_MASK_OK_DONE)
253 return ret;
254
255 send_cleanup_vector(cfg);
256
257 return 0;
258 }
259
260 static struct irq_chip hyperv_root_ir_chip = {
261 .name = "HYPERV-ROOT-IR",
262 .irq_ack = apic_ack_irq,
263 .irq_set_affinity = hyperv_root_ir_set_affinity,
264 .irq_compose_msi_msg = hyperv_root_ir_compose_msi_msg,
265 };
266
hyperv_root_irq_remapping_alloc(struct irq_domain * domain,unsigned int virq,unsigned int nr_irqs,void * arg)267 static int hyperv_root_irq_remapping_alloc(struct irq_domain *domain,
268 unsigned int virq, unsigned int nr_irqs,
269 void *arg)
270 {
271 struct irq_alloc_info *info = arg;
272 struct irq_data *irq_data;
273 struct hyperv_root_ir_data *data;
274 int ret = 0;
275
276 if (!info || info->type != X86_IRQ_ALLOC_TYPE_IOAPIC || nr_irqs > 1)
277 return -EINVAL;
278
279 ret = irq_domain_alloc_irqs_parent(domain, virq, nr_irqs, arg);
280 if (ret < 0)
281 return ret;
282
283 data = kzalloc(sizeof(*data), GFP_KERNEL);
284 if (!data) {
285 irq_domain_free_irqs_common(domain, virq, nr_irqs);
286 return -ENOMEM;
287 }
288
289 irq_data = irq_domain_get_irq_data(domain, virq);
290 if (!irq_data) {
291 kfree(data);
292 irq_domain_free_irqs_common(domain, virq, nr_irqs);
293 return -EINVAL;
294 }
295
296 data->ioapic_id = info->devid;
297 data->is_level = info->ioapic.is_level;
298
299 irq_data->chip = &hyperv_root_ir_chip;
300 irq_data->chip_data = data;
301
302 return 0;
303 }
304
hyperv_root_irq_remapping_free(struct irq_domain * domain,unsigned int virq,unsigned int nr_irqs)305 static void hyperv_root_irq_remapping_free(struct irq_domain *domain,
306 unsigned int virq, unsigned int nr_irqs)
307 {
308 struct irq_data *irq_data;
309 struct hyperv_root_ir_data *data;
310 struct hv_interrupt_entry *e;
311 int i;
312
313 for (i = 0; i < nr_irqs; i++) {
314 irq_data = irq_domain_get_irq_data(domain, virq + i);
315
316 if (irq_data && irq_data->chip_data) {
317 data = irq_data->chip_data;
318 e = &data->entry;
319
320 if (e->source == HV_DEVICE_TYPE_IOAPIC
321 && e->ioapic_rte.as_uint64)
322 hv_unmap_ioapic_interrupt(data->ioapic_id,
323 &data->entry);
324
325 kfree(data);
326 }
327 }
328
329 irq_domain_free_irqs_common(domain, virq, nr_irqs);
330 }
331
332 static const struct irq_domain_ops hyperv_root_ir_domain_ops = {
333 .select = hyperv_irq_remapping_select,
334 .alloc = hyperv_root_irq_remapping_alloc,
335 .free = hyperv_root_irq_remapping_free,
336 };
337
338 #endif
339