1 | // SPDX-License-Identifier: GPL-2.0 |
2 | /* |
3 | * Copyright (C) 2021 Western Digital Corporation or its affiliates. |
4 | * Copyright (C) 2022 Ventana Micro Systems Inc. |
5 | */ |
6 | |
7 | #define pr_fmt(fmt) "riscv-imsic: " fmt |
8 | #include <linux/acpi.h> |
9 | #include <linux/bitmap.h> |
10 | #include <linux/cpu.h> |
11 | #include <linux/interrupt.h> |
12 | #include <linux/io.h> |
13 | #include <linux/irq.h> |
14 | #include <linux/irqchip.h> |
15 | #include <linux/irqdomain.h> |
16 | #include <linux/module.h> |
17 | #include <linux/msi.h> |
18 | #include <linux/pci.h> |
19 | #include <linux/platform_device.h> |
20 | #include <linux/spinlock.h> |
21 | #include <linux/smp.h> |
22 | |
23 | #include <linux/irqchip/irq-msi-lib.h> |
24 | #include "irq-riscv-imsic-state.h" |
25 | |
26 | static bool imsic_cpu_page_phys(unsigned int cpu, unsigned int guest_index, |
27 | phys_addr_t *out_msi_pa) |
28 | { |
29 | struct imsic_global_config *global; |
30 | struct imsic_local_config *local; |
31 | |
32 | global = &imsic->global; |
33 | local = per_cpu_ptr(global->local, cpu); |
34 | |
35 | if (BIT(global->guest_index_bits) <= guest_index) |
36 | return false; |
37 | |
38 | if (out_msi_pa) |
39 | *out_msi_pa = local->msi_pa + (guest_index * IMSIC_MMIO_PAGE_SZ); |
40 | |
41 | return true; |
42 | } |
43 | |
44 | static void imsic_irq_mask(struct irq_data *d) |
45 | { |
46 | imsic_vector_mask(vec: irq_data_get_irq_chip_data(d)); |
47 | } |
48 | |
49 | static void imsic_irq_unmask(struct irq_data *d) |
50 | { |
51 | imsic_vector_unmask(vec: irq_data_get_irq_chip_data(d)); |
52 | } |
53 | |
54 | static int imsic_irq_retrigger(struct irq_data *d) |
55 | { |
56 | struct imsic_vector *vec = irq_data_get_irq_chip_data(d); |
57 | struct imsic_local_config *local; |
58 | |
59 | if (WARN_ON(!vec)) |
60 | return -ENOENT; |
61 | |
62 | local = per_cpu_ptr(imsic->global.local, vec->cpu); |
63 | writel_relaxed(vec->local_id, local->msi_va); |
64 | return 0; |
65 | } |
66 | |
67 | static void imsic_irq_ack(struct irq_data *d) |
68 | { |
69 | irq_move_irq(data: d); |
70 | } |
71 | |
72 | static void imsic_irq_compose_vector_msg(struct imsic_vector *vec, struct msi_msg *msg) |
73 | { |
74 | phys_addr_t msi_addr; |
75 | |
76 | if (WARN_ON(!vec)) |
77 | return; |
78 | |
79 | if (WARN_ON(!imsic_cpu_page_phys(vec->cpu, 0, &msi_addr))) |
80 | return; |
81 | |
82 | msg->address_hi = upper_32_bits(msi_addr); |
83 | msg->address_lo = lower_32_bits(msi_addr); |
84 | msg->data = vec->local_id; |
85 | } |
86 | |
87 | static void imsic_irq_compose_msg(struct irq_data *d, struct msi_msg *msg) |
88 | { |
89 | imsic_irq_compose_vector_msg(vec: irq_data_get_irq_chip_data(d), msg); |
90 | } |
91 | |
92 | #ifdef CONFIG_SMP |
93 | static void imsic_msi_update_msg(struct irq_data *d, struct imsic_vector *vec) |
94 | { |
95 | struct msi_msg msg = { }; |
96 | |
97 | imsic_irq_compose_vector_msg(vec, msg: &msg); |
98 | irq_data_get_irq_chip(d)->irq_write_msi_msg(d, &msg); |
99 | } |
100 | |
101 | static int imsic_irq_set_affinity(struct irq_data *d, const struct cpumask *mask_val, |
102 | bool force) |
103 | { |
104 | struct imsic_vector *old_vec, *new_vec; |
105 | struct imsic_vector tmp_vec; |
106 | |
107 | /* |
108 | * Requirements for the downstream irqdomains (or devices): |
109 | * |
110 | * 1) Downstream irqdomains (or devices) with atomic MSI update can |
111 | * happily do imsic_irq_set_affinity() in the process-context on |
112 | * any CPU so the irqchip of such irqdomains must not set the |
113 | * IRQCHIP_MOVE_DEFERRED flag. |
114 | * |
115 | * 2) Downstream irqdomains (or devices) with non-atomic MSI update |
116 | * must use imsic_irq_set_affinity() in nterrupt-context upon |
117 | * the next device interrupt so the irqchip of such irqdomains |
118 | * must set the IRQCHIP_MOVE_DEFERRED flag. |
119 | */ |
120 | |
121 | old_vec = irq_data_get_irq_chip_data(d); |
122 | if (WARN_ON(!old_vec)) |
123 | return -ENOENT; |
124 | |
125 | /* If old vector cpu belongs to the target cpumask then do nothing */ |
126 | if (cpumask_test_cpu(cpu: old_vec->cpu, cpumask: mask_val)) |
127 | return IRQ_SET_MASK_OK_DONE; |
128 | |
129 | /* If move is already in-flight then return failure */ |
130 | if (imsic_vector_get_move(vec: old_vec)) |
131 | return -EBUSY; |
132 | |
133 | /* Get a new vector on the desired set of CPUs */ |
134 | new_vec = imsic_vector_alloc(irq: old_vec->irq, mask: mask_val); |
135 | if (!new_vec) |
136 | return -ENOSPC; |
137 | |
138 | /* |
139 | * Device having non-atomic MSI update might see an intermediate |
140 | * state when changing target IMSIC vector from one CPU to another. |
141 | * |
142 | * To avoid losing interrupt to such intermediate state, do the |
143 | * following (just like x86 APIC): |
144 | * |
145 | * 1) First write a temporary IMSIC vector to the device which |
146 | * has MSI address same as the old IMSIC vector but MSI data |
147 | * matches the new IMSIC vector. |
148 | * |
149 | * 2) Next write the new IMSIC vector to the device. |
150 | * |
151 | * Based on the above, __imsic_local_sync() must check pending |
152 | * status of both old MSI data and new MSI data on the old CPU. |
153 | */ |
154 | if (!irq_can_move_in_process_context(data: d) && |
155 | new_vec->local_id != old_vec->local_id) { |
156 | /* Setup temporary vector */ |
157 | tmp_vec.cpu = old_vec->cpu; |
158 | tmp_vec.local_id = new_vec->local_id; |
159 | |
160 | /* Point device to the temporary vector */ |
161 | imsic_msi_update_msg(d: irq_get_irq_data(irq: d->irq), vec: &tmp_vec); |
162 | } |
163 | |
164 | /* Point device to the new vector */ |
165 | imsic_msi_update_msg(d: irq_get_irq_data(irq: d->irq), vec: new_vec); |
166 | |
167 | /* Update irq descriptors with the new vector */ |
168 | d->chip_data = new_vec; |
169 | |
170 | /* Update effective affinity */ |
171 | irq_data_update_effective_affinity(d, cpumask_of(new_vec->cpu)); |
172 | |
173 | /* Move state of the old vector to the new vector */ |
174 | imsic_vector_move(old_vec, new_vec); |
175 | |
176 | return IRQ_SET_MASK_OK_DONE; |
177 | } |
178 | |
179 | static void imsic_irq_force_complete_move(struct irq_data *d) |
180 | { |
181 | struct imsic_vector *mvec, *vec = irq_data_get_irq_chip_data(d); |
182 | unsigned int cpu = smp_processor_id(); |
183 | |
184 | if (WARN_ON(!vec)) |
185 | return; |
186 | |
187 | /* Do nothing if there is no in-flight move */ |
188 | mvec = imsic_vector_get_move(vec); |
189 | if (!mvec) |
190 | return; |
191 | |
192 | /* Do nothing if the old IMSIC vector does not belong to current CPU */ |
193 | if (mvec->cpu != cpu) |
194 | return; |
195 | |
196 | /* |
197 | * The best we can do is force cleanup the old IMSIC vector. |
198 | * |
199 | * The challenges over here are same as x86 vector domain so |
200 | * refer to the comments in irq_force_complete_move() function |
201 | * implemented at arch/x86/kernel/apic/vector.c. |
202 | */ |
203 | |
204 | /* Force cleanup in-flight move */ |
205 | pr_info("IRQ fixup: irq %d move in progress, old vector cpu %d local_id %d\n" , |
206 | d->irq, mvec->cpu, mvec->local_id); |
207 | imsic_vector_force_move_cleanup(vec); |
208 | } |
209 | #endif |
210 | |
211 | static struct irq_chip imsic_irq_base_chip = { |
212 | .name = "IMSIC" , |
213 | .irq_mask = imsic_irq_mask, |
214 | .irq_unmask = imsic_irq_unmask, |
215 | #ifdef CONFIG_SMP |
216 | .irq_set_affinity = imsic_irq_set_affinity, |
217 | .irq_force_complete_move = imsic_irq_force_complete_move, |
218 | #endif |
219 | .irq_retrigger = imsic_irq_retrigger, |
220 | .irq_ack = imsic_irq_ack, |
221 | .irq_compose_msi_msg = imsic_irq_compose_msg, |
222 | .flags = IRQCHIP_SKIP_SET_WAKE | IRQCHIP_MASK_ON_SUSPEND, |
223 | }; |
224 | |
225 | static int imsic_irq_domain_alloc(struct irq_domain *domain, unsigned int virq, |
226 | unsigned int nr_irqs, void *args) |
227 | { |
228 | struct imsic_vector *vec; |
229 | |
230 | /* Multi-MSI is not supported yet. */ |
231 | if (nr_irqs > 1) |
232 | return -EOPNOTSUPP; |
233 | |
234 | vec = imsic_vector_alloc(irq: virq, cpu_online_mask); |
235 | if (!vec) |
236 | return -ENOSPC; |
237 | |
238 | irq_domain_set_info(domain, virq, hwirq: virq, chip: &imsic_irq_base_chip, chip_data: vec, |
239 | handler: handle_edge_irq, NULL, NULL); |
240 | irq_set_noprobe(irq: virq); |
241 | irq_set_affinity(irq: virq, cpu_online_mask); |
242 | irq_data_update_effective_affinity(d: irq_get_irq_data(irq: virq), cpumask_of(vec->cpu)); |
243 | |
244 | return 0; |
245 | } |
246 | |
247 | static void imsic_irq_domain_free(struct irq_domain *domain, unsigned int virq, |
248 | unsigned int nr_irqs) |
249 | { |
250 | struct irq_data *d = irq_domain_get_irq_data(domain, virq); |
251 | |
252 | imsic_vector_free(vector: irq_data_get_irq_chip_data(d)); |
253 | irq_domain_free_irqs_parent(domain, irq_base: virq, nr_irqs); |
254 | } |
255 | |
256 | #ifdef CONFIG_GENERIC_IRQ_DEBUGFS |
257 | static void imsic_irq_debug_show(struct seq_file *m, struct irq_domain *d, |
258 | struct irq_data *irqd, int ind) |
259 | { |
260 | if (!irqd) { |
261 | imsic_vector_debug_show_summary(m, ind); |
262 | return; |
263 | } |
264 | |
265 | imsic_vector_debug_show(m, vec: irq_data_get_irq_chip_data(d: irqd), ind); |
266 | } |
267 | #endif |
268 | |
269 | static const struct irq_domain_ops imsic_base_domain_ops = { |
270 | .alloc = imsic_irq_domain_alloc, |
271 | .free = imsic_irq_domain_free, |
272 | .select = msi_lib_irq_domain_select, |
273 | #ifdef CONFIG_GENERIC_IRQ_DEBUGFS |
274 | .debug_show = imsic_irq_debug_show, |
275 | #endif |
276 | }; |
277 | |
278 | static bool imsic_init_dev_msi_info(struct device *dev, struct irq_domain *domain, |
279 | struct irq_domain *real_parent, struct msi_domain_info *info) |
280 | { |
281 | if (!msi_lib_init_dev_msi_info(dev, domain, real_parent, info)) |
282 | return false; |
283 | |
284 | switch (info->bus_token) { |
285 | case DOMAIN_BUS_PCI_DEVICE_MSI: |
286 | case DOMAIN_BUS_PCI_DEVICE_MSIX: |
287 | info->chip->flags |= IRQCHIP_MOVE_DEFERRED; |
288 | break; |
289 | default: |
290 | break; |
291 | } |
292 | |
293 | return true; |
294 | } |
295 | |
296 | static const struct msi_parent_ops imsic_msi_parent_ops = { |
297 | .supported_flags = MSI_GENERIC_FLAGS_MASK | |
298 | MSI_FLAG_PCI_MSIX, |
299 | .required_flags = MSI_FLAG_USE_DEF_DOM_OPS | |
300 | MSI_FLAG_USE_DEF_CHIP_OPS | |
301 | MSI_FLAG_PCI_MSI_MASK_PARENT, |
302 | .chip_flags = MSI_CHIP_FLAG_SET_ACK, |
303 | .bus_select_token = DOMAIN_BUS_NEXUS, |
304 | .bus_select_mask = MATCH_PCI_MSI | MATCH_PLATFORM_MSI, |
305 | .init_dev_msi_info = imsic_init_dev_msi_info, |
306 | }; |
307 | |
308 | int imsic_irqdomain_init(void) |
309 | { |
310 | struct imsic_global_config *global; |
311 | |
312 | if (!imsic || !imsic->fwnode) { |
313 | pr_err("early driver not probed\n" ); |
314 | return -ENODEV; |
315 | } |
316 | |
317 | if (imsic->base_domain) { |
318 | pr_err("%pfwP: irq domain already created\n" , imsic->fwnode); |
319 | return -ENODEV; |
320 | } |
321 | |
322 | /* Create Base IRQ domain */ |
323 | imsic->base_domain = irq_domain_create_tree(fwnode: imsic->fwnode, |
324 | ops: &imsic_base_domain_ops, host_data: imsic); |
325 | if (!imsic->base_domain) { |
326 | pr_err("%pfwP: failed to create IMSIC base domain\n" , imsic->fwnode); |
327 | return -ENOMEM; |
328 | } |
329 | imsic->base_domain->flags |= IRQ_DOMAIN_FLAG_MSI_PARENT; |
330 | imsic->base_domain->msi_parent_ops = &imsic_msi_parent_ops; |
331 | |
332 | irq_domain_update_bus_token(domain: imsic->base_domain, bus_token: DOMAIN_BUS_NEXUS); |
333 | |
334 | global = &imsic->global; |
335 | pr_info("%pfwP: hart-index-bits: %d, guest-index-bits: %d\n" , |
336 | imsic->fwnode, global->hart_index_bits, global->guest_index_bits); |
337 | pr_info("%pfwP: group-index-bits: %d, group-index-shift: %d\n" , |
338 | imsic->fwnode, global->group_index_bits, global->group_index_shift); |
339 | pr_info("%pfwP: per-CPU IDs %d at base address %pa\n" , |
340 | imsic->fwnode, global->nr_ids, &global->base_addr); |
341 | pr_info("%pfwP: total %d interrupts available\n" , |
342 | imsic->fwnode, num_possible_cpus() * (global->nr_ids - 1)); |
343 | |
344 | return 0; |
345 | } |
346 | |
347 | static int imsic_platform_probe_common(struct fwnode_handle *fwnode) |
348 | { |
349 | if (imsic && imsic->fwnode != fwnode) { |
350 | pr_err("%pfwP: fwnode mismatch\n" , fwnode); |
351 | return -ENODEV; |
352 | } |
353 | |
354 | return imsic_irqdomain_init(); |
355 | } |
356 | |
357 | static int imsic_platform_dt_probe(struct platform_device *pdev) |
358 | { |
359 | return imsic_platform_probe_common(fwnode: pdev->dev.fwnode); |
360 | } |
361 | |
362 | #ifdef CONFIG_ACPI |
363 | |
364 | /* |
365 | * On ACPI based systems, PCI enumeration happens early during boot in |
366 | * acpi_scan_init(). PCI enumeration expects MSI domain setup before |
367 | * it calls pci_set_msi_domain(). Hence, unlike in DT where |
368 | * imsic-platform drive probe happens late during boot, ACPI based |
369 | * systems need to setup the MSI domain early. |
370 | */ |
371 | int imsic_platform_acpi_probe(struct fwnode_handle *fwnode) |
372 | { |
373 | return imsic_platform_probe_common(fwnode); |
374 | } |
375 | |
376 | #endif |
377 | |
378 | static const struct of_device_id imsic_platform_match[] = { |
379 | { .compatible = "riscv,imsics" }, |
380 | {} |
381 | }; |
382 | |
383 | static struct platform_driver imsic_platform_driver = { |
384 | .driver = { |
385 | .name = "riscv-imsic" , |
386 | .of_match_table = imsic_platform_match, |
387 | }, |
388 | .probe = imsic_platform_dt_probe, |
389 | }; |
390 | builtin_platform_driver(imsic_platform_driver); |
391 | |