1// SPDX-License-Identifier: GPL-2.0
2
3/*
4 * Irqdomain for Linux to run as the root partition on Microsoft Hypervisor.
5 *
6 * Authors:
7 * Sunil Muthuswamy <sunilmut@microsoft.com>
8 * Wei Liu <wei.liu@kernel.org>
9 */
10
11#include <linux/pci.h>
12#include <linux/irq.h>
13#include <asm/mshyperv.h>
14
15static int hv_map_interrupt(union hv_device_id device_id, bool level,
16 int cpu, int vector, struct hv_interrupt_entry *entry)
17{
18 struct hv_input_map_device_interrupt *input;
19 struct hv_output_map_device_interrupt *output;
20 struct hv_device_interrupt_descriptor *intr_desc;
21 unsigned long flags;
22 u64 status;
23 int nr_bank, var_size;
24
25 local_irq_save(flags);
26
27 input = *this_cpu_ptr(hyperv_pcpu_input_arg);
28 output = *this_cpu_ptr(hyperv_pcpu_output_arg);
29
30 intr_desc = &input->interrupt_descriptor;
31 memset(input, 0, sizeof(*input));
32 input->partition_id = hv_current_partition_id;
33 input->device_id = device_id.as_uint64;
34 intr_desc->interrupt_type = HV_X64_INTERRUPT_TYPE_FIXED;
35 intr_desc->vector_count = 1;
36 intr_desc->target.vector = vector;
37
38 if (level)
39 intr_desc->trigger_mode = HV_INTERRUPT_TRIGGER_MODE_LEVEL;
40 else
41 intr_desc->trigger_mode = HV_INTERRUPT_TRIGGER_MODE_EDGE;
42
43 intr_desc->target.vp_set.valid_bank_mask = 0;
44 intr_desc->target.vp_set.format = HV_GENERIC_SET_SPARSE_4K;
45 nr_bank = cpumask_to_vpset(vpset: &(intr_desc->target.vp_set), cpumask_of(cpu));
46 if (nr_bank < 0) {
47 local_irq_restore(flags);
48 pr_err("%s: unable to generate VP set\n", __func__);
49 return EINVAL;
50 }
51 intr_desc->target.flags = HV_DEVICE_INTERRUPT_TARGET_PROCESSOR_SET;
52
53 /*
54 * var-sized hypercall, var-size starts after vp_mask (thus
55 * vp_set.format does not count, but vp_set.valid_bank_mask
56 * does).
57 */
58 var_size = nr_bank + 1;
59
60 status = hv_do_rep_hypercall(HVCALL_MAP_DEVICE_INTERRUPT, rep_count: 0, varhead_size: var_size,
61 input, output);
62 *entry = output->interrupt_entry;
63
64 local_irq_restore(flags);
65
66 if (!hv_result_success(status))
67 pr_err("%s: hypercall failed, status %lld\n", __func__, status);
68
69 return hv_result(status);
70}
71
72static int hv_unmap_interrupt(u64 id, struct hv_interrupt_entry *old_entry)
73{
74 unsigned long flags;
75 struct hv_input_unmap_device_interrupt *input;
76 struct hv_interrupt_entry *intr_entry;
77 u64 status;
78
79 local_irq_save(flags);
80 input = *this_cpu_ptr(hyperv_pcpu_input_arg);
81
82 memset(input, 0, sizeof(*input));
83 intr_entry = &input->interrupt_entry;
84 input->partition_id = hv_current_partition_id;
85 input->device_id = id;
86 *intr_entry = *old_entry;
87
88 status = hv_do_hypercall(HVCALL_UNMAP_DEVICE_INTERRUPT, inputaddr: input, NULL);
89 local_irq_restore(flags);
90
91 return hv_result(status);
92}
93
94#ifdef CONFIG_PCI_MSI
95struct rid_data {
96 struct pci_dev *bridge;
97 u32 rid;
98};
99
100static int get_rid_cb(struct pci_dev *pdev, u16 alias, void *data)
101{
102 struct rid_data *rd = data;
103 u8 bus = PCI_BUS_NUM(rd->rid);
104
105 if (pdev->bus->number != bus || PCI_BUS_NUM(alias) != bus) {
106 rd->bridge = pdev;
107 rd->rid = alias;
108 }
109
110 return 0;
111}
112
113static union hv_device_id hv_build_pci_dev_id(struct pci_dev *dev)
114{
115 union hv_device_id dev_id;
116 struct rid_data data = {
117 .bridge = NULL,
118 .rid = PCI_DEVID(dev->bus->number, dev->devfn)
119 };
120
121 pci_for_each_dma_alias(pdev: dev, fn: get_rid_cb, data: &data);
122
123 dev_id.as_uint64 = 0;
124 dev_id.device_type = HV_DEVICE_TYPE_PCI;
125 dev_id.pci.segment = pci_domain_nr(bus: dev->bus);
126
127 dev_id.pci.bdf.bus = PCI_BUS_NUM(data.rid);
128 dev_id.pci.bdf.device = PCI_SLOT(data.rid);
129 dev_id.pci.bdf.function = PCI_FUNC(data.rid);
130 dev_id.pci.source_shadow = HV_SOURCE_SHADOW_NONE;
131
132 if (data.bridge) {
133 int pos;
134
135 /*
136 * Microsoft Hypervisor requires a bus range when the bridge is
137 * running in PCI-X mode.
138 *
139 * To distinguish conventional vs PCI-X bridge, we can check
140 * the bridge's PCI-X Secondary Status Register, Secondary Bus
141 * Mode and Frequency bits. See PCI Express to PCI/PCI-X Bridge
142 * Specification Revision 1.0 5.2.2.1.3.
143 *
144 * Value zero means it is in conventional mode, otherwise it is
145 * in PCI-X mode.
146 */
147
148 pos = pci_find_capability(dev: data.bridge, PCI_CAP_ID_PCIX);
149 if (pos) {
150 u16 status;
151
152 pci_read_config_word(dev: data.bridge, where: pos +
153 PCI_X_BRIDGE_SSTATUS, val: &status);
154
155 if (status & PCI_X_SSTATUS_FREQ) {
156 /* Non-zero, PCI-X mode */
157 u8 sec_bus, sub_bus;
158
159 dev_id.pci.source_shadow = HV_SOURCE_SHADOW_BRIDGE_BUS_RANGE;
160
161 pci_read_config_byte(dev: data.bridge, PCI_SECONDARY_BUS, val: &sec_bus);
162 dev_id.pci.shadow_bus_range.secondary_bus = sec_bus;
163 pci_read_config_byte(dev: data.bridge, PCI_SUBORDINATE_BUS, val: &sub_bus);
164 dev_id.pci.shadow_bus_range.subordinate_bus = sub_bus;
165 }
166 }
167 }
168
169 return dev_id;
170}
171
172static int hv_map_msi_interrupt(struct pci_dev *dev, int cpu, int vector,
173 struct hv_interrupt_entry *entry)
174{
175 union hv_device_id device_id = hv_build_pci_dev_id(dev);
176
177 return hv_map_interrupt(device_id, level: false, cpu, vector, entry);
178}
179
180static inline void entry_to_msi_msg(struct hv_interrupt_entry *entry, struct msi_msg *msg)
181{
182 /* High address is always 0 */
183 msg->address_hi = 0;
184 msg->address_lo = entry->msi_entry.address.as_uint32;
185 msg->data = entry->msi_entry.data.as_uint32;
186}
187
188static int hv_unmap_msi_interrupt(struct pci_dev *dev, struct hv_interrupt_entry *old_entry);
189static void hv_irq_compose_msi_msg(struct irq_data *data, struct msi_msg *msg)
190{
191 struct msi_desc *msidesc;
192 struct pci_dev *dev;
193 struct hv_interrupt_entry out_entry, *stored_entry;
194 struct irq_cfg *cfg = irqd_cfg(irq_data: data);
195 const cpumask_t *affinity;
196 int cpu;
197 u64 status;
198
199 msidesc = irq_data_get_msi_desc(d: data);
200 dev = msi_desc_to_pci_dev(desc: msidesc);
201
202 if (!cfg) {
203 pr_debug("%s: cfg is NULL", __func__);
204 return;
205 }
206
207 affinity = irq_data_get_effective_affinity_mask(d: data);
208 cpu = cpumask_first_and(srcp1: affinity, cpu_online_mask);
209
210 if (data->chip_data) {
211 /*
212 * This interrupt is already mapped. Let's unmap first.
213 *
214 * We don't use retarget interrupt hypercalls here because
215 * Microsoft Hypervisor doens't allow root to change the vector
216 * or specify VPs outside of the set that is initially used
217 * during mapping.
218 */
219 stored_entry = data->chip_data;
220 data->chip_data = NULL;
221
222 status = hv_unmap_msi_interrupt(dev, old_entry: stored_entry);
223
224 kfree(objp: stored_entry);
225
226 if (status != HV_STATUS_SUCCESS) {
227 pr_debug("%s: failed to unmap, status %lld", __func__, status);
228 return;
229 }
230 }
231
232 stored_entry = kzalloc(size: sizeof(*stored_entry), GFP_ATOMIC);
233 if (!stored_entry) {
234 pr_debug("%s: failed to allocate chip data\n", __func__);
235 return;
236 }
237
238 status = hv_map_msi_interrupt(dev, cpu, vector: cfg->vector, entry: &out_entry);
239 if (status != HV_STATUS_SUCCESS) {
240 kfree(objp: stored_entry);
241 return;
242 }
243
244 *stored_entry = out_entry;
245 data->chip_data = stored_entry;
246 entry_to_msi_msg(entry: &out_entry, msg);
247
248 return;
249}
250
251static int hv_unmap_msi_interrupt(struct pci_dev *dev, struct hv_interrupt_entry *old_entry)
252{
253 return hv_unmap_interrupt(id: hv_build_pci_dev_id(dev).as_uint64, old_entry);
254}
255
256static void hv_teardown_msi_irq(struct pci_dev *dev, struct irq_data *irqd)
257{
258 struct hv_interrupt_entry old_entry;
259 struct msi_msg msg;
260 u64 status;
261
262 if (!irqd->chip_data) {
263 pr_debug("%s: no chip data\n!", __func__);
264 return;
265 }
266
267 old_entry = *(struct hv_interrupt_entry *)irqd->chip_data;
268 entry_to_msi_msg(entry: &old_entry, msg: &msg);
269
270 kfree(objp: irqd->chip_data);
271 irqd->chip_data = NULL;
272
273 status = hv_unmap_msi_interrupt(dev, old_entry: &old_entry);
274
275 if (status != HV_STATUS_SUCCESS)
276 pr_err("%s: hypercall failed, status %lld\n", __func__, status);
277}
278
279static void hv_msi_free_irq(struct irq_domain *domain,
280 struct msi_domain_info *info, unsigned int virq)
281{
282 struct irq_data *irqd = irq_get_irq_data(irq: virq);
283 struct msi_desc *desc;
284
285 if (!irqd)
286 return;
287
288 desc = irq_data_get_msi_desc(d: irqd);
289 if (!desc || !desc->irq || WARN_ON_ONCE(!dev_is_pci(desc->dev)))
290 return;
291
292 hv_teardown_msi_irq(to_pci_dev(desc->dev), irqd);
293}
294
295/*
296 * IRQ Chip for MSI PCI/PCI-X/PCI-Express Devices,
297 * which implement the MSI or MSI-X Capability Structure.
298 */
299static struct irq_chip hv_pci_msi_controller = {
300 .name = "HV-PCI-MSI",
301 .irq_unmask = pci_msi_unmask_irq,
302 .irq_mask = pci_msi_mask_irq,
303 .irq_ack = irq_chip_ack_parent,
304 .irq_retrigger = irq_chip_retrigger_hierarchy,
305 .irq_compose_msi_msg = hv_irq_compose_msi_msg,
306 .irq_set_affinity = msi_domain_set_affinity,
307 .flags = IRQCHIP_SKIP_SET_WAKE,
308};
309
310static struct msi_domain_ops pci_msi_domain_ops = {
311 .msi_free = hv_msi_free_irq,
312 .msi_prepare = pci_msi_prepare,
313};
314
315static struct msi_domain_info hv_pci_msi_domain_info = {
316 .flags = MSI_FLAG_USE_DEF_DOM_OPS | MSI_FLAG_USE_DEF_CHIP_OPS |
317 MSI_FLAG_PCI_MSIX,
318 .ops = &pci_msi_domain_ops,
319 .chip = &hv_pci_msi_controller,
320 .handler = handle_edge_irq,
321 .handler_name = "edge",
322};
323
324struct irq_domain * __init hv_create_pci_msi_domain(void)
325{
326 struct irq_domain *d = NULL;
327 struct fwnode_handle *fn;
328
329 fn = irq_domain_alloc_named_fwnode(name: "HV-PCI-MSI");
330 if (fn)
331 d = pci_msi_create_irq_domain(fwnode: fn, info: &hv_pci_msi_domain_info, parent: x86_vector_domain);
332
333 /* No point in going further if we can't get an irq domain */
334 BUG_ON(!d);
335
336 return d;
337}
338
339#endif /* CONFIG_PCI_MSI */
340
341int hv_unmap_ioapic_interrupt(int ioapic_id, struct hv_interrupt_entry *entry)
342{
343 union hv_device_id device_id;
344
345 device_id.as_uint64 = 0;
346 device_id.device_type = HV_DEVICE_TYPE_IOAPIC;
347 device_id.ioapic.ioapic_id = (u8)ioapic_id;
348
349 return hv_unmap_interrupt(id: device_id.as_uint64, old_entry: entry);
350}
351EXPORT_SYMBOL_GPL(hv_unmap_ioapic_interrupt);
352
353int hv_map_ioapic_interrupt(int ioapic_id, bool level, int cpu, int vector,
354 struct hv_interrupt_entry *entry)
355{
356 union hv_device_id device_id;
357
358 device_id.as_uint64 = 0;
359 device_id.device_type = HV_DEVICE_TYPE_IOAPIC;
360 device_id.ioapic.ioapic_id = (u8)ioapic_id;
361
362 return hv_map_interrupt(device_id, level, cpu, vector, entry);
363}
364EXPORT_SYMBOL_GPL(hv_map_ioapic_interrupt);
365

source code of linux/arch/x86/hyperv/irqdomain.c