xref: /linux/arch/riscv/kvm/aia_imsic.c (revision 0c7c237b1c35011ef0b8d30c1d5c20bc6ae7b69b)
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  * Authors:
7  *	Anup Patel <apatel@ventanamicro.com>
8  */
9 
10 #include <linux/atomic.h>
11 #include <linux/bitmap.h>
12 #include <linux/kvm_host.h>
13 #include <linux/math.h>
14 #include <linux/spinlock.h>
15 #include <linux/swab.h>
16 #include <kvm/iodev.h>
17 #include <asm/csr.h>
18 #include <asm/kvm_aia_imsic.h>
19 
20 #define IMSIC_MAX_EIX	(IMSIC_MAX_ID / BITS_PER_TYPE(u64))
21 
22 struct imsic_mrif_eix {
23 	unsigned long eip[BITS_PER_TYPE(u64) / BITS_PER_LONG];
24 	unsigned long eie[BITS_PER_TYPE(u64) / BITS_PER_LONG];
25 };
26 
27 struct imsic_mrif {
28 	struct imsic_mrif_eix eix[IMSIC_MAX_EIX];
29 	unsigned long eithreshold;
30 	unsigned long eidelivery;
31 };
32 
33 struct imsic {
34 	struct kvm_io_device iodev;
35 
36 	u32 nr_msis;
37 	u32 nr_eix;
38 	u32 nr_hw_eix;
39 
40 	/*
41 	 * At any point in time, the register state is in
42 	 * one of the following places:
43 	 *
44 	 * 1) Hardware: IMSIC VS-file (vsfile_cpu >= 0)
45 	 * 2) Software: IMSIC SW-file (vsfile_cpu < 0)
46 	 */
47 
48 	/* IMSIC VS-file */
49 	rwlock_t vsfile_lock;
50 	int vsfile_cpu;
51 	int vsfile_hgei;
52 	void __iomem *vsfile_va;
53 	phys_addr_t vsfile_pa;
54 
55 	/* IMSIC SW-file */
56 	struct imsic_mrif *swfile;
57 	phys_addr_t swfile_pa;
58 };
59 
60 #define imsic_vs_csr_read(__c)			\
61 ({						\
62 	unsigned long __r;			\
63 	csr_write(CSR_VSISELECT, __c);		\
64 	__r = csr_read(CSR_VSIREG);		\
65 	__r;					\
66 })
67 
68 #define imsic_read_switchcase(__ireg)			\
69 	case __ireg:					\
70 		return imsic_vs_csr_read(__ireg);
71 #define imsic_read_switchcase_2(__ireg)			\
72 	imsic_read_switchcase(__ireg + 0)		\
73 	imsic_read_switchcase(__ireg + 1)
74 #define imsic_read_switchcase_4(__ireg)			\
75 	imsic_read_switchcase_2(__ireg + 0)		\
76 	imsic_read_switchcase_2(__ireg + 2)
77 #define imsic_read_switchcase_8(__ireg)			\
78 	imsic_read_switchcase_4(__ireg + 0)		\
79 	imsic_read_switchcase_4(__ireg + 4)
80 #define imsic_read_switchcase_16(__ireg)		\
81 	imsic_read_switchcase_8(__ireg + 0)		\
82 	imsic_read_switchcase_8(__ireg + 8)
83 #define imsic_read_switchcase_32(__ireg)		\
84 	imsic_read_switchcase_16(__ireg + 0)		\
85 	imsic_read_switchcase_16(__ireg + 16)
86 #define imsic_read_switchcase_64(__ireg)		\
87 	imsic_read_switchcase_32(__ireg + 0)		\
88 	imsic_read_switchcase_32(__ireg + 32)
89 
90 static unsigned long imsic_eix_read(int ireg)
91 {
92 	switch (ireg) {
93 	imsic_read_switchcase_64(IMSIC_EIP0)
94 	imsic_read_switchcase_64(IMSIC_EIE0)
95 	}
96 
97 	return 0;
98 }
99 
100 #define imsic_vs_csr_swap(__c, __v)		\
101 ({						\
102 	unsigned long __r;			\
103 	csr_write(CSR_VSISELECT, __c);		\
104 	__r = csr_swap(CSR_VSIREG, __v);	\
105 	__r;					\
106 })
107 
108 #define imsic_swap_switchcase(__ireg, __v)		\
109 	case __ireg:					\
110 		return imsic_vs_csr_swap(__ireg, __v);
111 #define imsic_swap_switchcase_2(__ireg, __v)		\
112 	imsic_swap_switchcase(__ireg + 0, __v)		\
113 	imsic_swap_switchcase(__ireg + 1, __v)
114 #define imsic_swap_switchcase_4(__ireg, __v)		\
115 	imsic_swap_switchcase_2(__ireg + 0, __v)	\
116 	imsic_swap_switchcase_2(__ireg + 2, __v)
117 #define imsic_swap_switchcase_8(__ireg, __v)		\
118 	imsic_swap_switchcase_4(__ireg + 0, __v)	\
119 	imsic_swap_switchcase_4(__ireg + 4, __v)
120 #define imsic_swap_switchcase_16(__ireg, __v)		\
121 	imsic_swap_switchcase_8(__ireg + 0, __v)	\
122 	imsic_swap_switchcase_8(__ireg + 8, __v)
123 #define imsic_swap_switchcase_32(__ireg, __v)		\
124 	imsic_swap_switchcase_16(__ireg + 0, __v)	\
125 	imsic_swap_switchcase_16(__ireg + 16, __v)
126 #define imsic_swap_switchcase_64(__ireg, __v)		\
127 	imsic_swap_switchcase_32(__ireg + 0, __v)	\
128 	imsic_swap_switchcase_32(__ireg + 32, __v)
129 
130 static unsigned long imsic_eix_swap(int ireg, unsigned long val)
131 {
132 	switch (ireg) {
133 	imsic_swap_switchcase_64(IMSIC_EIP0, val)
134 	imsic_swap_switchcase_64(IMSIC_EIE0, val)
135 	}
136 
137 	return 0;
138 }
139 
140 #define imsic_vs_csr_write(__c, __v)		\
141 do {						\
142 	csr_write(CSR_VSISELECT, __c);		\
143 	csr_write(CSR_VSIREG, __v);		\
144 } while (0)
145 
146 #define imsic_write_switchcase(__ireg, __v)		\
147 	case __ireg:					\
148 		imsic_vs_csr_write(__ireg, __v);	\
149 		break;
150 #define imsic_write_switchcase_2(__ireg, __v)		\
151 	imsic_write_switchcase(__ireg + 0, __v)		\
152 	imsic_write_switchcase(__ireg + 1, __v)
153 #define imsic_write_switchcase_4(__ireg, __v)		\
154 	imsic_write_switchcase_2(__ireg + 0, __v)	\
155 	imsic_write_switchcase_2(__ireg + 2, __v)
156 #define imsic_write_switchcase_8(__ireg, __v)		\
157 	imsic_write_switchcase_4(__ireg + 0, __v)	\
158 	imsic_write_switchcase_4(__ireg + 4, __v)
159 #define imsic_write_switchcase_16(__ireg, __v)		\
160 	imsic_write_switchcase_8(__ireg + 0, __v)	\
161 	imsic_write_switchcase_8(__ireg + 8, __v)
162 #define imsic_write_switchcase_32(__ireg, __v)		\
163 	imsic_write_switchcase_16(__ireg + 0, __v)	\
164 	imsic_write_switchcase_16(__ireg + 16, __v)
165 #define imsic_write_switchcase_64(__ireg, __v)		\
166 	imsic_write_switchcase_32(__ireg + 0, __v)	\
167 	imsic_write_switchcase_32(__ireg + 32, __v)
168 
169 static void imsic_eix_write(int ireg, unsigned long val)
170 {
171 	switch (ireg) {
172 	imsic_write_switchcase_64(IMSIC_EIP0, val)
173 	imsic_write_switchcase_64(IMSIC_EIE0, val)
174 	}
175 }
176 
177 #define imsic_vs_csr_set(__c, __v)		\
178 do {						\
179 	csr_write(CSR_VSISELECT, __c);		\
180 	csr_set(CSR_VSIREG, __v);		\
181 } while (0)
182 
183 #define imsic_set_switchcase(__ireg, __v)		\
184 	case __ireg:					\
185 		imsic_vs_csr_set(__ireg, __v);		\
186 		break;
187 #define imsic_set_switchcase_2(__ireg, __v)		\
188 	imsic_set_switchcase(__ireg + 0, __v)		\
189 	imsic_set_switchcase(__ireg + 1, __v)
190 #define imsic_set_switchcase_4(__ireg, __v)		\
191 	imsic_set_switchcase_2(__ireg + 0, __v)		\
192 	imsic_set_switchcase_2(__ireg + 2, __v)
193 #define imsic_set_switchcase_8(__ireg, __v)		\
194 	imsic_set_switchcase_4(__ireg + 0, __v)		\
195 	imsic_set_switchcase_4(__ireg + 4, __v)
196 #define imsic_set_switchcase_16(__ireg, __v)		\
197 	imsic_set_switchcase_8(__ireg + 0, __v)		\
198 	imsic_set_switchcase_8(__ireg + 8, __v)
199 #define imsic_set_switchcase_32(__ireg, __v)		\
200 	imsic_set_switchcase_16(__ireg + 0, __v)	\
201 	imsic_set_switchcase_16(__ireg + 16, __v)
202 #define imsic_set_switchcase_64(__ireg, __v)		\
203 	imsic_set_switchcase_32(__ireg + 0, __v)	\
204 	imsic_set_switchcase_32(__ireg + 32, __v)
205 
206 static void imsic_eix_set(int ireg, unsigned long val)
207 {
208 	switch (ireg) {
209 	imsic_set_switchcase_64(IMSIC_EIP0, val)
210 	imsic_set_switchcase_64(IMSIC_EIE0, val)
211 	}
212 }
213 
214 static unsigned long imsic_mrif_atomic_rmw(struct imsic_mrif *mrif,
215 					   unsigned long *ptr,
216 					   unsigned long new_val,
217 					   unsigned long wr_mask)
218 {
219 	unsigned long old_val = 0, tmp = 0;
220 
221 	__asm__ __volatile__ (
222 		"0:	lr.w.aq   %1, %0\n"
223 		"	and       %2, %1, %3\n"
224 		"	or        %2, %2, %4\n"
225 		"	sc.w.rl   %2, %2, %0\n"
226 		"	bnez      %2, 0b"
227 		: "+A" (*ptr), "+r" (old_val), "+r" (tmp)
228 		: "r" (~wr_mask), "r" (new_val & wr_mask)
229 		: "memory");
230 
231 	return old_val;
232 }
233 
234 static unsigned long imsic_mrif_atomic_or(struct imsic_mrif *mrif,
235 					  unsigned long *ptr,
236 					  unsigned long val)
237 {
238 	return atomic_long_fetch_or(val, (atomic_long_t *)ptr);
239 }
240 
241 #define imsic_mrif_atomic_write(__mrif, __ptr, __new_val)	\
242 		imsic_mrif_atomic_rmw(__mrif, __ptr, __new_val, -1UL)
243 #define imsic_mrif_atomic_read(__mrif, __ptr)			\
244 		imsic_mrif_atomic_or(__mrif, __ptr, 0)
245 
246 static u32 imsic_mrif_topei(struct imsic_mrif *mrif, u32 nr_eix, u32 nr_msis)
247 {
248 	struct imsic_mrif_eix *eix;
249 	u32 i, imin, imax, ei, max_msi;
250 	unsigned long eipend[BITS_PER_TYPE(u64) / BITS_PER_LONG];
251 	unsigned long eithreshold = imsic_mrif_atomic_read(mrif,
252 							&mrif->eithreshold);
253 
254 	max_msi = (eithreshold && (eithreshold <= nr_msis)) ?
255 		   eithreshold : nr_msis;
256 	for (ei = 0; ei < nr_eix; ei++) {
257 		eix = &mrif->eix[ei];
258 		eipend[0] = imsic_mrif_atomic_read(mrif, &eix->eie[0]) &
259 			    imsic_mrif_atomic_read(mrif, &eix->eip[0]);
260 #ifdef CONFIG_32BIT
261 		eipend[1] = imsic_mrif_atomic_read(mrif, &eix->eie[1]) &
262 			    imsic_mrif_atomic_read(mrif, &eix->eip[1]);
263 		if (!eipend[0] && !eipend[1])
264 #else
265 		if (!eipend[0])
266 #endif
267 			continue;
268 
269 		imin = ei * BITS_PER_TYPE(u64);
270 		imax = ((imin + BITS_PER_TYPE(u64)) < max_msi) ?
271 			imin + BITS_PER_TYPE(u64) : max_msi;
272 		for (i = (!imin) ? 1 : imin; i < imax; i++) {
273 			if (test_bit(i - imin, eipend))
274 				return (i << TOPEI_ID_SHIFT) | i;
275 		}
276 	}
277 
278 	return 0;
279 }
280 
281 static int imsic_mrif_isel_check(u32 nr_eix, unsigned long isel)
282 {
283 	u32 num = 0;
284 
285 	switch (isel) {
286 	case IMSIC_EIDELIVERY:
287 	case IMSIC_EITHRESHOLD:
288 		break;
289 	case IMSIC_EIP0 ... IMSIC_EIP63:
290 		num = isel - IMSIC_EIP0;
291 		break;
292 	case IMSIC_EIE0 ... IMSIC_EIE63:
293 		num = isel - IMSIC_EIE0;
294 		break;
295 	default:
296 		return -ENOENT;
297 	}
298 #ifndef CONFIG_32BIT
299 	if (num & 0x1)
300 		return -EINVAL;
301 #endif
302 	if ((num / 2) >= nr_eix)
303 		return -EINVAL;
304 
305 	return 0;
306 }
307 
308 static int imsic_mrif_rmw(struct imsic_mrif *mrif, u32 nr_eix,
309 			  unsigned long isel, unsigned long *val,
310 			  unsigned long new_val, unsigned long wr_mask)
311 {
312 	bool pend;
313 	struct imsic_mrif_eix *eix;
314 	unsigned long *ei, num, old_val = 0;
315 
316 	switch (isel) {
317 	case IMSIC_EIDELIVERY:
318 		old_val = imsic_mrif_atomic_rmw(mrif, &mrif->eidelivery,
319 						new_val, wr_mask & 0x1);
320 		break;
321 	case IMSIC_EITHRESHOLD:
322 		old_val = imsic_mrif_atomic_rmw(mrif, &mrif->eithreshold,
323 				new_val, wr_mask & (IMSIC_MAX_ID - 1));
324 		break;
325 	case IMSIC_EIP0 ... IMSIC_EIP63:
326 	case IMSIC_EIE0 ... IMSIC_EIE63:
327 		if (isel >= IMSIC_EIP0 && isel <= IMSIC_EIP63) {
328 			pend = true;
329 			num = isel - IMSIC_EIP0;
330 		} else {
331 			pend = false;
332 			num = isel - IMSIC_EIE0;
333 		}
334 
335 		if ((num / 2) >= nr_eix)
336 			return -EINVAL;
337 		eix = &mrif->eix[num / 2];
338 
339 #ifndef CONFIG_32BIT
340 		if (num & 0x1)
341 			return -EINVAL;
342 		ei = (pend) ? &eix->eip[0] : &eix->eie[0];
343 #else
344 		ei = (pend) ? &eix->eip[num & 0x1] : &eix->eie[num & 0x1];
345 #endif
346 
347 		/* Bit0 of EIP0 or EIE0 is read-only */
348 		if (!num)
349 			wr_mask &= ~BIT(0);
350 
351 		old_val = imsic_mrif_atomic_rmw(mrif, ei, new_val, wr_mask);
352 		break;
353 	default:
354 		return -ENOENT;
355 	}
356 
357 	if (val)
358 		*val = old_val;
359 
360 	return 0;
361 }
362 
363 struct imsic_vsfile_read_data {
364 	int hgei;
365 	u32 nr_eix;
366 	bool clear;
367 	struct imsic_mrif *mrif;
368 };
369 
370 static void imsic_vsfile_local_read(void *data)
371 {
372 	u32 i;
373 	struct imsic_mrif_eix *eix;
374 	struct imsic_vsfile_read_data *idata = data;
375 	struct imsic_mrif *mrif = idata->mrif;
376 	unsigned long new_hstatus, old_hstatus, old_vsiselect;
377 
378 	old_vsiselect = csr_read(CSR_VSISELECT);
379 	old_hstatus = csr_read(CSR_HSTATUS);
380 	new_hstatus = old_hstatus & ~HSTATUS_VGEIN;
381 	new_hstatus |= ((unsigned long)idata->hgei) << HSTATUS_VGEIN_SHIFT;
382 	csr_write(CSR_HSTATUS, new_hstatus);
383 
384 	/*
385 	 * We don't use imsic_mrif_atomic_xyz() functions to store
386 	 * values in MRIF because imsic_vsfile_read() is always called
387 	 * with pointer to temporary MRIF on stack.
388 	 */
389 
390 	if (idata->clear) {
391 		mrif->eidelivery = imsic_vs_csr_swap(IMSIC_EIDELIVERY, 0);
392 		mrif->eithreshold = imsic_vs_csr_swap(IMSIC_EITHRESHOLD, 0);
393 		for (i = 0; i < idata->nr_eix; i++) {
394 			eix = &mrif->eix[i];
395 			eix->eip[0] = imsic_eix_swap(IMSIC_EIP0 + i * 2, 0);
396 			eix->eie[0] = imsic_eix_swap(IMSIC_EIE0 + i * 2, 0);
397 #ifdef CONFIG_32BIT
398 			eix->eip[1] = imsic_eix_swap(IMSIC_EIP0 + i * 2 + 1, 0);
399 			eix->eie[1] = imsic_eix_swap(IMSIC_EIE0 + i * 2 + 1, 0);
400 #endif
401 		}
402 	} else {
403 		mrif->eidelivery = imsic_vs_csr_read(IMSIC_EIDELIVERY);
404 		mrif->eithreshold = imsic_vs_csr_read(IMSIC_EITHRESHOLD);
405 		for (i = 0; i < idata->nr_eix; i++) {
406 			eix = &mrif->eix[i];
407 			eix->eip[0] = imsic_eix_read(IMSIC_EIP0 + i * 2);
408 			eix->eie[0] = imsic_eix_read(IMSIC_EIE0 + i * 2);
409 #ifdef CONFIG_32BIT
410 			eix->eip[1] = imsic_eix_read(IMSIC_EIP0 + i * 2 + 1);
411 			eix->eie[1] = imsic_eix_read(IMSIC_EIE0 + i * 2 + 1);
412 #endif
413 		}
414 	}
415 
416 	csr_write(CSR_HSTATUS, old_hstatus);
417 	csr_write(CSR_VSISELECT, old_vsiselect);
418 }
419 
420 static void imsic_vsfile_read(int vsfile_hgei, int vsfile_cpu, u32 nr_eix,
421 			      bool clear, struct imsic_mrif *mrif)
422 {
423 	struct imsic_vsfile_read_data idata;
424 
425 	/* We can only read clear if we have a IMSIC VS-file */
426 	if (vsfile_cpu < 0 || vsfile_hgei <= 0)
427 		return;
428 
429 	/* We can only read clear on local CPU */
430 	idata.hgei = vsfile_hgei;
431 	idata.nr_eix = nr_eix;
432 	idata.clear = clear;
433 	idata.mrif = mrif;
434 	on_each_cpu_mask(cpumask_of(vsfile_cpu),
435 			 imsic_vsfile_local_read, &idata, 1);
436 }
437 
438 struct imsic_vsfile_rw_data {
439 	int hgei;
440 	int isel;
441 	bool write;
442 	unsigned long val;
443 };
444 
445 static void imsic_vsfile_local_rw(void *data)
446 {
447 	struct imsic_vsfile_rw_data *idata = data;
448 	unsigned long new_hstatus, old_hstatus, old_vsiselect;
449 
450 	old_vsiselect = csr_read(CSR_VSISELECT);
451 	old_hstatus = csr_read(CSR_HSTATUS);
452 	new_hstatus = old_hstatus & ~HSTATUS_VGEIN;
453 	new_hstatus |= ((unsigned long)idata->hgei) << HSTATUS_VGEIN_SHIFT;
454 	csr_write(CSR_HSTATUS, new_hstatus);
455 
456 	switch (idata->isel) {
457 	case IMSIC_EIDELIVERY:
458 		if (idata->write)
459 			imsic_vs_csr_write(IMSIC_EIDELIVERY, idata->val);
460 		else
461 			idata->val = imsic_vs_csr_read(IMSIC_EIDELIVERY);
462 		break;
463 	case IMSIC_EITHRESHOLD:
464 		if (idata->write)
465 			imsic_vs_csr_write(IMSIC_EITHRESHOLD, idata->val);
466 		else
467 			idata->val = imsic_vs_csr_read(IMSIC_EITHRESHOLD);
468 		break;
469 	case IMSIC_EIP0 ... IMSIC_EIP63:
470 	case IMSIC_EIE0 ... IMSIC_EIE63:
471 #ifndef CONFIG_32BIT
472 		if (idata->isel & 0x1)
473 			break;
474 #endif
475 		if (idata->write)
476 			imsic_eix_write(idata->isel, idata->val);
477 		else
478 			idata->val = imsic_eix_read(idata->isel);
479 		break;
480 	default:
481 		break;
482 	}
483 
484 	csr_write(CSR_HSTATUS, old_hstatus);
485 	csr_write(CSR_VSISELECT, old_vsiselect);
486 }
487 
488 static int imsic_vsfile_rw(int vsfile_hgei, int vsfile_cpu, u32 nr_eix,
489 			   unsigned long isel, bool write,
490 			   unsigned long *val)
491 {
492 	int rc;
493 	struct imsic_vsfile_rw_data rdata;
494 
495 	/* We can only access register if we have a IMSIC VS-file */
496 	if (vsfile_cpu < 0 || vsfile_hgei <= 0)
497 		return -EINVAL;
498 
499 	/* Check IMSIC register iselect */
500 	rc = imsic_mrif_isel_check(nr_eix, isel);
501 	if (rc)
502 		return rc;
503 
504 	/* We can only access register on local CPU */
505 	rdata.hgei = vsfile_hgei;
506 	rdata.isel = isel;
507 	rdata.write = write;
508 	rdata.val = (write) ? *val : 0;
509 	on_each_cpu_mask(cpumask_of(vsfile_cpu),
510 			 imsic_vsfile_local_rw, &rdata, 1);
511 
512 	if (!write)
513 		*val = rdata.val;
514 
515 	return 0;
516 }
517 
518 static void imsic_vsfile_local_clear(int vsfile_hgei, u32 nr_eix)
519 {
520 	u32 i;
521 	unsigned long new_hstatus, old_hstatus, old_vsiselect;
522 
523 	/* We can only zero-out if we have a IMSIC VS-file */
524 	if (vsfile_hgei <= 0)
525 		return;
526 
527 	old_vsiselect = csr_read(CSR_VSISELECT);
528 	old_hstatus = csr_read(CSR_HSTATUS);
529 	new_hstatus = old_hstatus & ~HSTATUS_VGEIN;
530 	new_hstatus |= ((unsigned long)vsfile_hgei) << HSTATUS_VGEIN_SHIFT;
531 	csr_write(CSR_HSTATUS, new_hstatus);
532 
533 	imsic_vs_csr_write(IMSIC_EIDELIVERY, 0);
534 	imsic_vs_csr_write(IMSIC_EITHRESHOLD, 0);
535 	for (i = 0; i < nr_eix; i++) {
536 		imsic_eix_write(IMSIC_EIP0 + i * 2, 0);
537 		imsic_eix_write(IMSIC_EIE0 + i * 2, 0);
538 #ifdef CONFIG_32BIT
539 		imsic_eix_write(IMSIC_EIP0 + i * 2 + 1, 0);
540 		imsic_eix_write(IMSIC_EIE0 + i * 2 + 1, 0);
541 #endif
542 	}
543 
544 	csr_write(CSR_HSTATUS, old_hstatus);
545 	csr_write(CSR_VSISELECT, old_vsiselect);
546 }
547 
548 static void imsic_vsfile_local_update(int vsfile_hgei, u32 nr_eix,
549 				      struct imsic_mrif *mrif)
550 {
551 	u32 i;
552 	struct imsic_mrif_eix *eix;
553 	unsigned long new_hstatus, old_hstatus, old_vsiselect;
554 
555 	/* We can only update if we have a HW IMSIC context */
556 	if (vsfile_hgei <= 0)
557 		return;
558 
559 	/*
560 	 * We don't use imsic_mrif_atomic_xyz() functions to read values
561 	 * from MRIF in this function because it is always called with
562 	 * pointer to temporary MRIF on stack.
563 	 */
564 
565 	old_vsiselect = csr_read(CSR_VSISELECT);
566 	old_hstatus = csr_read(CSR_HSTATUS);
567 	new_hstatus = old_hstatus & ~HSTATUS_VGEIN;
568 	new_hstatus |= ((unsigned long)vsfile_hgei) << HSTATUS_VGEIN_SHIFT;
569 	csr_write(CSR_HSTATUS, new_hstatus);
570 
571 	for (i = 0; i < nr_eix; i++) {
572 		eix = &mrif->eix[i];
573 		imsic_eix_set(IMSIC_EIP0 + i * 2, eix->eip[0]);
574 		imsic_eix_set(IMSIC_EIE0 + i * 2, eix->eie[0]);
575 #ifdef CONFIG_32BIT
576 		imsic_eix_set(IMSIC_EIP0 + i * 2 + 1, eix->eip[1]);
577 		imsic_eix_set(IMSIC_EIE0 + i * 2 + 1, eix->eie[1]);
578 #endif
579 	}
580 	imsic_vs_csr_write(IMSIC_EITHRESHOLD, mrif->eithreshold);
581 	imsic_vs_csr_write(IMSIC_EIDELIVERY, mrif->eidelivery);
582 
583 	csr_write(CSR_HSTATUS, old_hstatus);
584 	csr_write(CSR_VSISELECT, old_vsiselect);
585 }
586 
587 static void imsic_vsfile_cleanup(struct imsic *imsic)
588 {
589 	int old_vsfile_hgei, old_vsfile_cpu;
590 	unsigned long flags;
591 
592 	/*
593 	 * We don't use imsic_mrif_atomic_xyz() functions to clear the
594 	 * SW-file in this function because it is always called when the
595 	 * VCPU is being destroyed.
596 	 */
597 
598 	write_lock_irqsave(&imsic->vsfile_lock, flags);
599 	old_vsfile_hgei = imsic->vsfile_hgei;
600 	old_vsfile_cpu = imsic->vsfile_cpu;
601 	imsic->vsfile_cpu = imsic->vsfile_hgei = -1;
602 	imsic->vsfile_va = NULL;
603 	imsic->vsfile_pa = 0;
604 	write_unlock_irqrestore(&imsic->vsfile_lock, flags);
605 
606 	memset(imsic->swfile, 0, sizeof(*imsic->swfile));
607 
608 	if (old_vsfile_cpu >= 0)
609 		kvm_riscv_aia_free_hgei(old_vsfile_cpu, old_vsfile_hgei);
610 }
611 
612 static void imsic_swfile_extirq_update(struct kvm_vcpu *vcpu)
613 {
614 	struct imsic *imsic = vcpu->arch.aia_context.imsic_state;
615 	struct imsic_mrif *mrif = imsic->swfile;
616 
617 	if (imsic_mrif_atomic_read(mrif, &mrif->eidelivery) &&
618 	    imsic_mrif_topei(mrif, imsic->nr_eix, imsic->nr_msis))
619 		kvm_riscv_vcpu_set_interrupt(vcpu, IRQ_VS_EXT);
620 	else
621 		kvm_riscv_vcpu_unset_interrupt(vcpu, IRQ_VS_EXT);
622 }
623 
624 static void imsic_swfile_read(struct kvm_vcpu *vcpu, bool clear,
625 			      struct imsic_mrif *mrif)
626 {
627 	struct imsic *imsic = vcpu->arch.aia_context.imsic_state;
628 
629 	/*
630 	 * We don't use imsic_mrif_atomic_xyz() functions to read and
631 	 * write SW-file and MRIF in this function because it is always
632 	 * called when VCPU is not using SW-file and the MRIF points to
633 	 * a temporary MRIF on stack.
634 	 */
635 
636 	memcpy(mrif, imsic->swfile, sizeof(*mrif));
637 	if (clear) {
638 		memset(imsic->swfile, 0, sizeof(*imsic->swfile));
639 		kvm_riscv_vcpu_unset_interrupt(vcpu, IRQ_VS_EXT);
640 	}
641 }
642 
643 static void imsic_swfile_update(struct kvm_vcpu *vcpu,
644 				struct imsic_mrif *mrif)
645 {
646 	u32 i;
647 	struct imsic_mrif_eix *seix, *eix;
648 	struct imsic *imsic = vcpu->arch.aia_context.imsic_state;
649 	struct imsic_mrif *smrif = imsic->swfile;
650 
651 	imsic_mrif_atomic_write(smrif, &smrif->eidelivery, mrif->eidelivery);
652 	imsic_mrif_atomic_write(smrif, &smrif->eithreshold, mrif->eithreshold);
653 	for (i = 0; i < imsic->nr_eix; i++) {
654 		seix = &smrif->eix[i];
655 		eix = &mrif->eix[i];
656 		imsic_mrif_atomic_or(smrif, &seix->eip[0], eix->eip[0]);
657 		imsic_mrif_atomic_or(smrif, &seix->eie[0], eix->eie[0]);
658 #ifdef CONFIG_32BIT
659 		imsic_mrif_atomic_or(smrif, &seix->eip[1], eix->eip[1]);
660 		imsic_mrif_atomic_or(smrif, &seix->eie[1], eix->eie[1]);
661 #endif
662 	}
663 
664 	imsic_swfile_extirq_update(vcpu);
665 }
666 
667 void kvm_riscv_vcpu_aia_imsic_release(struct kvm_vcpu *vcpu)
668 {
669 	unsigned long flags;
670 	struct imsic_mrif tmrif;
671 	int old_vsfile_hgei, old_vsfile_cpu;
672 	struct imsic *imsic = vcpu->arch.aia_context.imsic_state;
673 
674 	/* Read and clear IMSIC VS-file details */
675 	write_lock_irqsave(&imsic->vsfile_lock, flags);
676 	old_vsfile_hgei = imsic->vsfile_hgei;
677 	old_vsfile_cpu = imsic->vsfile_cpu;
678 	imsic->vsfile_cpu = imsic->vsfile_hgei = -1;
679 	imsic->vsfile_va = NULL;
680 	imsic->vsfile_pa = 0;
681 	write_unlock_irqrestore(&imsic->vsfile_lock, flags);
682 
683 	/* Do nothing, if no IMSIC VS-file to release */
684 	if (old_vsfile_cpu < 0)
685 		return;
686 
687 	/*
688 	 * At this point, all interrupt producers are still using
689 	 * the old IMSIC VS-file so we first re-direct all interrupt
690 	 * producers.
691 	 */
692 
693 	/* Purge the G-stage mapping */
694 	kvm_riscv_gstage_iounmap(vcpu->kvm,
695 				 vcpu->arch.aia_context.imsic_addr,
696 				 IMSIC_MMIO_PAGE_SZ);
697 
698 	/* TODO: Purge the IOMMU mapping ??? */
699 
700 	/*
701 	 * At this point, all interrupt producers have been re-directed
702 	 * to somewhere else so we move register state from the old IMSIC
703 	 * VS-file to the IMSIC SW-file.
704 	 */
705 
706 	/* Read and clear register state from old IMSIC VS-file */
707 	memset(&tmrif, 0, sizeof(tmrif));
708 	imsic_vsfile_read(old_vsfile_hgei, old_vsfile_cpu, imsic->nr_hw_eix,
709 			  true, &tmrif);
710 
711 	/* Update register state in IMSIC SW-file */
712 	imsic_swfile_update(vcpu, &tmrif);
713 
714 	/* Free-up old IMSIC VS-file */
715 	kvm_riscv_aia_free_hgei(old_vsfile_cpu, old_vsfile_hgei);
716 }
717 
718 int kvm_riscv_vcpu_aia_imsic_update(struct kvm_vcpu *vcpu)
719 {
720 	unsigned long flags;
721 	phys_addr_t new_vsfile_pa;
722 	struct imsic_mrif tmrif;
723 	void __iomem *new_vsfile_va;
724 	struct kvm *kvm = vcpu->kvm;
725 	struct kvm_run *run = vcpu->run;
726 	struct kvm_vcpu_aia *vaia = &vcpu->arch.aia_context;
727 	struct imsic *imsic = vaia->imsic_state;
728 	int ret = 0, new_vsfile_hgei = -1, old_vsfile_hgei, old_vsfile_cpu;
729 
730 	/* Do nothing for emulation mode */
731 	if (kvm->arch.aia.mode == KVM_DEV_RISCV_AIA_MODE_EMUL)
732 		return 1;
733 
734 	/* Read old IMSIC VS-file details */
735 	read_lock_irqsave(&imsic->vsfile_lock, flags);
736 	old_vsfile_hgei = imsic->vsfile_hgei;
737 	old_vsfile_cpu = imsic->vsfile_cpu;
738 	read_unlock_irqrestore(&imsic->vsfile_lock, flags);
739 
740 	/* Do nothing if we are continuing on same CPU */
741 	if (old_vsfile_cpu == vcpu->cpu)
742 		return 1;
743 
744 	/* Allocate new IMSIC VS-file */
745 	ret = kvm_riscv_aia_alloc_hgei(vcpu->cpu, vcpu,
746 				       &new_vsfile_va, &new_vsfile_pa);
747 	if (ret <= 0) {
748 		/* For HW acceleration mode, we can't continue */
749 		if (kvm->arch.aia.mode == KVM_DEV_RISCV_AIA_MODE_HWACCEL) {
750 			run->fail_entry.hardware_entry_failure_reason =
751 								CSR_HSTATUS;
752 			run->fail_entry.cpu = vcpu->cpu;
753 			run->exit_reason = KVM_EXIT_FAIL_ENTRY;
754 			return 0;
755 		}
756 
757 		/* Release old IMSIC VS-file */
758 		if (old_vsfile_cpu >= 0)
759 			kvm_riscv_vcpu_aia_imsic_release(vcpu);
760 
761 		/* For automatic mode, we continue */
762 		goto done;
763 	}
764 	new_vsfile_hgei = ret;
765 
766 	/*
767 	 * At this point, all interrupt producers are still using
768 	 * to the old IMSIC VS-file so we first move all interrupt
769 	 * producers to the new IMSIC VS-file.
770 	 */
771 
772 	/* Zero-out new IMSIC VS-file */
773 	imsic_vsfile_local_clear(new_vsfile_hgei, imsic->nr_hw_eix);
774 
775 	/* Update G-stage mapping for the new IMSIC VS-file */
776 	ret = kvm_riscv_gstage_ioremap(kvm, vcpu->arch.aia_context.imsic_addr,
777 				       new_vsfile_pa, IMSIC_MMIO_PAGE_SZ,
778 				       true, true);
779 	if (ret)
780 		goto fail_free_vsfile_hgei;
781 
782 	/* TODO: Update the IOMMU mapping ??? */
783 
784 	/* Update new IMSIC VS-file details in IMSIC context */
785 	write_lock_irqsave(&imsic->vsfile_lock, flags);
786 	imsic->vsfile_hgei = new_vsfile_hgei;
787 	imsic->vsfile_cpu = vcpu->cpu;
788 	imsic->vsfile_va = new_vsfile_va;
789 	imsic->vsfile_pa = new_vsfile_pa;
790 	write_unlock_irqrestore(&imsic->vsfile_lock, flags);
791 
792 	/*
793 	 * At this point, all interrupt producers have been moved
794 	 * to the new IMSIC VS-file so we move register state from
795 	 * the old IMSIC VS/SW-file to the new IMSIC VS-file.
796 	 */
797 
798 	memset(&tmrif, 0, sizeof(tmrif));
799 	if (old_vsfile_cpu >= 0) {
800 		/* Read and clear register state from old IMSIC VS-file */
801 		imsic_vsfile_read(old_vsfile_hgei, old_vsfile_cpu,
802 				  imsic->nr_hw_eix, true, &tmrif);
803 
804 		/* Free-up old IMSIC VS-file */
805 		kvm_riscv_aia_free_hgei(old_vsfile_cpu, old_vsfile_hgei);
806 	} else {
807 		/* Read and clear register state from IMSIC SW-file */
808 		imsic_swfile_read(vcpu, true, &tmrif);
809 	}
810 
811 	/* Restore register state in the new IMSIC VS-file */
812 	imsic_vsfile_local_update(new_vsfile_hgei, imsic->nr_hw_eix, &tmrif);
813 
814 done:
815 	/* Set VCPU HSTATUS.VGEIN to new IMSIC VS-file */
816 	vcpu->arch.guest_context.hstatus &= ~HSTATUS_VGEIN;
817 	if (new_vsfile_hgei > 0)
818 		vcpu->arch.guest_context.hstatus |=
819 			((unsigned long)new_vsfile_hgei) << HSTATUS_VGEIN_SHIFT;
820 
821 	/* Continue run-loop */
822 	return 1;
823 
824 fail_free_vsfile_hgei:
825 	kvm_riscv_aia_free_hgei(vcpu->cpu, new_vsfile_hgei);
826 	return ret;
827 }
828 
829 int kvm_riscv_vcpu_aia_imsic_rmw(struct kvm_vcpu *vcpu, unsigned long isel,
830 				 unsigned long *val, unsigned long new_val,
831 				 unsigned long wr_mask)
832 {
833 	u32 topei;
834 	struct imsic_mrif_eix *eix;
835 	int r, rc = KVM_INSN_CONTINUE_NEXT_SEPC;
836 	struct imsic *imsic = vcpu->arch.aia_context.imsic_state;
837 
838 	if (isel == KVM_RISCV_AIA_IMSIC_TOPEI) {
839 		/* Read pending and enabled interrupt with highest priority */
840 		topei = imsic_mrif_topei(imsic->swfile, imsic->nr_eix,
841 					 imsic->nr_msis);
842 		if (val)
843 			*val = topei;
844 
845 		/* Writes ignore value and clear top pending interrupt */
846 		if (topei && wr_mask) {
847 			topei >>= TOPEI_ID_SHIFT;
848 			if (topei) {
849 				eix = &imsic->swfile->eix[topei /
850 							  BITS_PER_TYPE(u64)];
851 				clear_bit(topei & (BITS_PER_TYPE(u64) - 1),
852 					  eix->eip);
853 			}
854 		}
855 	} else {
856 		r = imsic_mrif_rmw(imsic->swfile, imsic->nr_eix, isel,
857 				   val, new_val, wr_mask);
858 		/* Forward unknown IMSIC register to user-space */
859 		if (r)
860 			rc = (r == -ENOENT) ? 0 : KVM_INSN_ILLEGAL_TRAP;
861 	}
862 
863 	if (wr_mask)
864 		imsic_swfile_extirq_update(vcpu);
865 
866 	return rc;
867 }
868 
869 int kvm_riscv_aia_imsic_rw_attr(struct kvm *kvm, unsigned long type,
870 				bool write, unsigned long *val)
871 {
872 	u32 isel, vcpu_id;
873 	unsigned long flags;
874 	struct imsic *imsic;
875 	struct kvm_vcpu *vcpu;
876 	int rc, vsfile_hgei, vsfile_cpu;
877 
878 	if (!kvm_riscv_aia_initialized(kvm))
879 		return -ENODEV;
880 
881 	vcpu_id = KVM_DEV_RISCV_AIA_IMSIC_GET_VCPU(type);
882 	vcpu = kvm_get_vcpu_by_id(kvm, vcpu_id);
883 	if (!vcpu)
884 		return -ENODEV;
885 
886 	isel = KVM_DEV_RISCV_AIA_IMSIC_GET_ISEL(type);
887 	imsic = vcpu->arch.aia_context.imsic_state;
888 
889 	read_lock_irqsave(&imsic->vsfile_lock, flags);
890 
891 	rc = 0;
892 	vsfile_hgei = imsic->vsfile_hgei;
893 	vsfile_cpu = imsic->vsfile_cpu;
894 	if (vsfile_cpu < 0) {
895 		if (write) {
896 			rc = imsic_mrif_rmw(imsic->swfile, imsic->nr_eix,
897 					    isel, NULL, *val, -1UL);
898 			imsic_swfile_extirq_update(vcpu);
899 		} else
900 			rc = imsic_mrif_rmw(imsic->swfile, imsic->nr_eix,
901 					    isel, val, 0, 0);
902 	}
903 
904 	read_unlock_irqrestore(&imsic->vsfile_lock, flags);
905 
906 	if (!rc && vsfile_cpu >= 0)
907 		rc = imsic_vsfile_rw(vsfile_hgei, vsfile_cpu, imsic->nr_eix,
908 				     isel, write, val);
909 
910 	return rc;
911 }
912 
913 int kvm_riscv_aia_imsic_has_attr(struct kvm *kvm, unsigned long type)
914 {
915 	u32 isel, vcpu_id;
916 	struct imsic *imsic;
917 	struct kvm_vcpu *vcpu;
918 
919 	if (!kvm_riscv_aia_initialized(kvm))
920 		return -ENODEV;
921 
922 	vcpu_id = KVM_DEV_RISCV_AIA_IMSIC_GET_VCPU(type);
923 	vcpu = kvm_get_vcpu_by_id(kvm, vcpu_id);
924 	if (!vcpu)
925 		return -ENODEV;
926 
927 	isel = KVM_DEV_RISCV_AIA_IMSIC_GET_ISEL(type);
928 	imsic = vcpu->arch.aia_context.imsic_state;
929 	return imsic_mrif_isel_check(imsic->nr_eix, isel);
930 }
931 
932 void kvm_riscv_vcpu_aia_imsic_reset(struct kvm_vcpu *vcpu)
933 {
934 	struct imsic *imsic = vcpu->arch.aia_context.imsic_state;
935 
936 	if (!imsic)
937 		return;
938 
939 	kvm_riscv_vcpu_aia_imsic_release(vcpu);
940 
941 	memset(imsic->swfile, 0, sizeof(*imsic->swfile));
942 }
943 
944 int kvm_riscv_vcpu_aia_imsic_inject(struct kvm_vcpu *vcpu,
945 				    u32 guest_index, u32 offset, u32 iid)
946 {
947 	unsigned long flags;
948 	struct imsic_mrif_eix *eix;
949 	struct imsic *imsic = vcpu->arch.aia_context.imsic_state;
950 
951 	/* We only emulate one IMSIC MMIO page for each Guest VCPU */
952 	if (!imsic || !iid || guest_index ||
953 	    (offset != IMSIC_MMIO_SETIPNUM_LE &&
954 	     offset != IMSIC_MMIO_SETIPNUM_BE))
955 		return -ENODEV;
956 
957 	iid = (offset == IMSIC_MMIO_SETIPNUM_BE) ? __swab32(iid) : iid;
958 	if (imsic->nr_msis <= iid)
959 		return -EINVAL;
960 
961 	read_lock_irqsave(&imsic->vsfile_lock, flags);
962 
963 	if (imsic->vsfile_cpu >= 0) {
964 		writel(iid, imsic->vsfile_va + IMSIC_MMIO_SETIPNUM_LE);
965 		kvm_vcpu_kick(vcpu);
966 	} else {
967 		eix = &imsic->swfile->eix[iid / BITS_PER_TYPE(u64)];
968 		set_bit(iid & (BITS_PER_TYPE(u64) - 1), eix->eip);
969 		imsic_swfile_extirq_update(vcpu);
970 	}
971 
972 	read_unlock_irqrestore(&imsic->vsfile_lock, flags);
973 
974 	return 0;
975 }
976 
977 static int imsic_mmio_read(struct kvm_vcpu *vcpu, struct kvm_io_device *dev,
978 			   gpa_t addr, int len, void *val)
979 {
980 	if (len != 4 || (addr & 0x3) != 0)
981 		return -EOPNOTSUPP;
982 
983 	*((u32 *)val) = 0;
984 
985 	return 0;
986 }
987 
988 static int imsic_mmio_write(struct kvm_vcpu *vcpu, struct kvm_io_device *dev,
989 			    gpa_t addr, int len, const void *val)
990 {
991 	struct kvm_msi msi = { 0 };
992 
993 	if (len != 4 || (addr & 0x3) != 0)
994 		return -EOPNOTSUPP;
995 
996 	msi.address_hi = addr >> 32;
997 	msi.address_lo = (u32)addr;
998 	msi.data = *((const u32 *)val);
999 	kvm_riscv_aia_inject_msi(vcpu->kvm, &msi);
1000 
1001 	return 0;
1002 };
1003 
1004 static struct kvm_io_device_ops imsic_iodoev_ops = {
1005 	.read = imsic_mmio_read,
1006 	.write = imsic_mmio_write,
1007 };
1008 
1009 int kvm_riscv_vcpu_aia_imsic_init(struct kvm_vcpu *vcpu)
1010 {
1011 	int ret = 0;
1012 	struct imsic *imsic;
1013 	struct page *swfile_page;
1014 	struct kvm *kvm = vcpu->kvm;
1015 
1016 	/* Fail if we have zero IDs */
1017 	if (!kvm->arch.aia.nr_ids)
1018 		return -EINVAL;
1019 
1020 	/* Allocate IMSIC context */
1021 	imsic = kzalloc(sizeof(*imsic), GFP_KERNEL);
1022 	if (!imsic)
1023 		return -ENOMEM;
1024 	vcpu->arch.aia_context.imsic_state = imsic;
1025 
1026 	/* Setup IMSIC context  */
1027 	imsic->nr_msis = kvm->arch.aia.nr_ids + 1;
1028 	rwlock_init(&imsic->vsfile_lock);
1029 	imsic->nr_eix = BITS_TO_U64(imsic->nr_msis);
1030 	imsic->nr_hw_eix = BITS_TO_U64(kvm_riscv_aia_max_ids);
1031 	imsic->vsfile_hgei = imsic->vsfile_cpu = -1;
1032 
1033 	/* Setup IMSIC SW-file */
1034 	swfile_page = alloc_pages(GFP_KERNEL | __GFP_ZERO,
1035 				  get_order(sizeof(*imsic->swfile)));
1036 	if (!swfile_page) {
1037 		ret = -ENOMEM;
1038 		goto fail_free_imsic;
1039 	}
1040 	imsic->swfile = page_to_virt(swfile_page);
1041 	imsic->swfile_pa = page_to_phys(swfile_page);
1042 
1043 	/* Setup IO device */
1044 	kvm_iodevice_init(&imsic->iodev, &imsic_iodoev_ops);
1045 	mutex_lock(&kvm->slots_lock);
1046 	ret = kvm_io_bus_register_dev(kvm, KVM_MMIO_BUS,
1047 				      vcpu->arch.aia_context.imsic_addr,
1048 				      KVM_DEV_RISCV_IMSIC_SIZE,
1049 				      &imsic->iodev);
1050 	mutex_unlock(&kvm->slots_lock);
1051 	if (ret)
1052 		goto fail_free_swfile;
1053 
1054 	return 0;
1055 
1056 fail_free_swfile:
1057 	free_pages((unsigned long)imsic->swfile,
1058 		   get_order(sizeof(*imsic->swfile)));
1059 fail_free_imsic:
1060 	vcpu->arch.aia_context.imsic_state = NULL;
1061 	kfree(imsic);
1062 	return ret;
1063 }
1064 
1065 void kvm_riscv_vcpu_aia_imsic_cleanup(struct kvm_vcpu *vcpu)
1066 {
1067 	struct kvm *kvm = vcpu->kvm;
1068 	struct imsic *imsic = vcpu->arch.aia_context.imsic_state;
1069 
1070 	if (!imsic)
1071 		return;
1072 
1073 	imsic_vsfile_cleanup(imsic);
1074 
1075 	mutex_lock(&kvm->slots_lock);
1076 	kvm_io_bus_unregister_dev(kvm, KVM_MMIO_BUS, &imsic->iodev);
1077 	mutex_unlock(&kvm->slots_lock);
1078 
1079 	free_pages((unsigned long)imsic->swfile,
1080 		   get_order(sizeof(*imsic->swfile)));
1081 
1082 	vcpu->arch.aia_context.imsic_state = NULL;
1083 	kfree(imsic);
1084 }
1085