xref: /linux/arch/powerpc/kvm/book3s_hv_nested.c (revision 2031f287)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright IBM Corporation, 2018
4  * Authors Suraj Jitindar Singh <sjitindarsingh@gmail.com>
5  *	   Paul Mackerras <paulus@ozlabs.org>
6  *
7  * Description: KVM functions specific to running nested KVM-HV guests
8  * on Book3S processors (specifically POWER9 and later).
9  */
10 
11 #include <linux/kernel.h>
12 #include <linux/kvm_host.h>
13 #include <linux/llist.h>
14 #include <linux/pgtable.h>
15 
16 #include <asm/kvm_ppc.h>
17 #include <asm/kvm_book3s.h>
18 #include <asm/mmu.h>
19 #include <asm/pgalloc.h>
20 #include <asm/pte-walk.h>
21 #include <asm/reg.h>
22 #include <asm/plpar_wrappers.h>
23 
24 static struct patb_entry *pseries_partition_tb;
25 
26 static void kvmhv_update_ptbl_cache(struct kvm_nested_guest *gp);
27 static void kvmhv_free_memslot_nest_rmap(struct kvm_memory_slot *free);
28 
29 void kvmhv_save_hv_regs(struct kvm_vcpu *vcpu, struct hv_guest_state *hr)
30 {
31 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
32 
33 	hr->pcr = vc->pcr | PCR_MASK;
34 	hr->dpdes = vc->dpdes;
35 	hr->hfscr = vcpu->arch.hfscr;
36 	hr->tb_offset = vc->tb_offset;
37 	hr->dawr0 = vcpu->arch.dawr0;
38 	hr->dawrx0 = vcpu->arch.dawrx0;
39 	hr->ciabr = vcpu->arch.ciabr;
40 	hr->purr = vcpu->arch.purr;
41 	hr->spurr = vcpu->arch.spurr;
42 	hr->ic = vcpu->arch.ic;
43 	hr->vtb = vc->vtb;
44 	hr->srr0 = vcpu->arch.shregs.srr0;
45 	hr->srr1 = vcpu->arch.shregs.srr1;
46 	hr->sprg[0] = vcpu->arch.shregs.sprg0;
47 	hr->sprg[1] = vcpu->arch.shregs.sprg1;
48 	hr->sprg[2] = vcpu->arch.shregs.sprg2;
49 	hr->sprg[3] = vcpu->arch.shregs.sprg3;
50 	hr->pidr = vcpu->arch.pid;
51 	hr->cfar = vcpu->arch.cfar;
52 	hr->ppr = vcpu->arch.ppr;
53 	hr->dawr1 = vcpu->arch.dawr1;
54 	hr->dawrx1 = vcpu->arch.dawrx1;
55 }
56 
57 /* Use noinline_for_stack due to https://bugs.llvm.org/show_bug.cgi?id=49610 */
58 static noinline_for_stack void byteswap_pt_regs(struct pt_regs *regs)
59 {
60 	unsigned long *addr = (unsigned long *) regs;
61 
62 	for (; addr < ((unsigned long *) (regs + 1)); addr++)
63 		*addr = swab64(*addr);
64 }
65 
66 static void byteswap_hv_regs(struct hv_guest_state *hr)
67 {
68 	hr->version = swab64(hr->version);
69 	hr->lpid = swab32(hr->lpid);
70 	hr->vcpu_token = swab32(hr->vcpu_token);
71 	hr->lpcr = swab64(hr->lpcr);
72 	hr->pcr = swab64(hr->pcr) | PCR_MASK;
73 	hr->amor = swab64(hr->amor);
74 	hr->dpdes = swab64(hr->dpdes);
75 	hr->hfscr = swab64(hr->hfscr);
76 	hr->tb_offset = swab64(hr->tb_offset);
77 	hr->dawr0 = swab64(hr->dawr0);
78 	hr->dawrx0 = swab64(hr->dawrx0);
79 	hr->ciabr = swab64(hr->ciabr);
80 	hr->hdec_expiry = swab64(hr->hdec_expiry);
81 	hr->purr = swab64(hr->purr);
82 	hr->spurr = swab64(hr->spurr);
83 	hr->ic = swab64(hr->ic);
84 	hr->vtb = swab64(hr->vtb);
85 	hr->hdar = swab64(hr->hdar);
86 	hr->hdsisr = swab64(hr->hdsisr);
87 	hr->heir = swab64(hr->heir);
88 	hr->asdr = swab64(hr->asdr);
89 	hr->srr0 = swab64(hr->srr0);
90 	hr->srr1 = swab64(hr->srr1);
91 	hr->sprg[0] = swab64(hr->sprg[0]);
92 	hr->sprg[1] = swab64(hr->sprg[1]);
93 	hr->sprg[2] = swab64(hr->sprg[2]);
94 	hr->sprg[3] = swab64(hr->sprg[3]);
95 	hr->pidr = swab64(hr->pidr);
96 	hr->cfar = swab64(hr->cfar);
97 	hr->ppr = swab64(hr->ppr);
98 	hr->dawr1 = swab64(hr->dawr1);
99 	hr->dawrx1 = swab64(hr->dawrx1);
100 }
101 
102 static void save_hv_return_state(struct kvm_vcpu *vcpu,
103 				 struct hv_guest_state *hr)
104 {
105 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
106 
107 	hr->dpdes = vc->dpdes;
108 	hr->purr = vcpu->arch.purr;
109 	hr->spurr = vcpu->arch.spurr;
110 	hr->ic = vcpu->arch.ic;
111 	hr->vtb = vc->vtb;
112 	hr->srr0 = vcpu->arch.shregs.srr0;
113 	hr->srr1 = vcpu->arch.shregs.srr1;
114 	hr->sprg[0] = vcpu->arch.shregs.sprg0;
115 	hr->sprg[1] = vcpu->arch.shregs.sprg1;
116 	hr->sprg[2] = vcpu->arch.shregs.sprg2;
117 	hr->sprg[3] = vcpu->arch.shregs.sprg3;
118 	hr->pidr = vcpu->arch.pid;
119 	hr->cfar = vcpu->arch.cfar;
120 	hr->ppr = vcpu->arch.ppr;
121 	switch (vcpu->arch.trap) {
122 	case BOOK3S_INTERRUPT_H_DATA_STORAGE:
123 		hr->hdar = vcpu->arch.fault_dar;
124 		hr->hdsisr = vcpu->arch.fault_dsisr;
125 		hr->asdr = vcpu->arch.fault_gpa;
126 		break;
127 	case BOOK3S_INTERRUPT_H_INST_STORAGE:
128 		hr->asdr = vcpu->arch.fault_gpa;
129 		break;
130 	case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
131 		hr->hfscr = ((~HFSCR_INTR_CAUSE & hr->hfscr) |
132 			     (HFSCR_INTR_CAUSE & vcpu->arch.hfscr));
133 		break;
134 	case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
135 		hr->heir = vcpu->arch.emul_inst;
136 		break;
137 	}
138 }
139 
140 static void restore_hv_regs(struct kvm_vcpu *vcpu, const struct hv_guest_state *hr)
141 {
142 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
143 
144 	vc->pcr = hr->pcr | PCR_MASK;
145 	vc->dpdes = hr->dpdes;
146 	vcpu->arch.hfscr = hr->hfscr;
147 	vcpu->arch.dawr0 = hr->dawr0;
148 	vcpu->arch.dawrx0 = hr->dawrx0;
149 	vcpu->arch.ciabr = hr->ciabr;
150 	vcpu->arch.purr = hr->purr;
151 	vcpu->arch.spurr = hr->spurr;
152 	vcpu->arch.ic = hr->ic;
153 	vc->vtb = hr->vtb;
154 	vcpu->arch.shregs.srr0 = hr->srr0;
155 	vcpu->arch.shregs.srr1 = hr->srr1;
156 	vcpu->arch.shregs.sprg0 = hr->sprg[0];
157 	vcpu->arch.shregs.sprg1 = hr->sprg[1];
158 	vcpu->arch.shregs.sprg2 = hr->sprg[2];
159 	vcpu->arch.shregs.sprg3 = hr->sprg[3];
160 	vcpu->arch.pid = hr->pidr;
161 	vcpu->arch.cfar = hr->cfar;
162 	vcpu->arch.ppr = hr->ppr;
163 	vcpu->arch.dawr1 = hr->dawr1;
164 	vcpu->arch.dawrx1 = hr->dawrx1;
165 }
166 
167 void kvmhv_restore_hv_return_state(struct kvm_vcpu *vcpu,
168 				   struct hv_guest_state *hr)
169 {
170 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
171 
172 	vc->dpdes = hr->dpdes;
173 	vcpu->arch.hfscr = hr->hfscr;
174 	vcpu->arch.purr = hr->purr;
175 	vcpu->arch.spurr = hr->spurr;
176 	vcpu->arch.ic = hr->ic;
177 	vc->vtb = hr->vtb;
178 	vcpu->arch.fault_dar = hr->hdar;
179 	vcpu->arch.fault_dsisr = hr->hdsisr;
180 	vcpu->arch.fault_gpa = hr->asdr;
181 	vcpu->arch.emul_inst = hr->heir;
182 	vcpu->arch.shregs.srr0 = hr->srr0;
183 	vcpu->arch.shregs.srr1 = hr->srr1;
184 	vcpu->arch.shregs.sprg0 = hr->sprg[0];
185 	vcpu->arch.shregs.sprg1 = hr->sprg[1];
186 	vcpu->arch.shregs.sprg2 = hr->sprg[2];
187 	vcpu->arch.shregs.sprg3 = hr->sprg[3];
188 	vcpu->arch.pid = hr->pidr;
189 	vcpu->arch.cfar = hr->cfar;
190 	vcpu->arch.ppr = hr->ppr;
191 }
192 
193 static void kvmhv_nested_mmio_needed(struct kvm_vcpu *vcpu, u64 regs_ptr)
194 {
195 	/* No need to reflect the page fault to L1, we've handled it */
196 	vcpu->arch.trap = 0;
197 
198 	/*
199 	 * Since the L2 gprs have already been written back into L1 memory when
200 	 * we complete the mmio, store the L1 memory location of the L2 gpr
201 	 * being loaded into by the mmio so that the loaded value can be
202 	 * written there in kvmppc_complete_mmio_load()
203 	 */
204 	if (((vcpu->arch.io_gpr & KVM_MMIO_REG_EXT_MASK) == KVM_MMIO_REG_GPR)
205 	    && (vcpu->mmio_is_write == 0)) {
206 		vcpu->arch.nested_io_gpr = (gpa_t) regs_ptr +
207 					   offsetof(struct pt_regs,
208 						    gpr[vcpu->arch.io_gpr]);
209 		vcpu->arch.io_gpr = KVM_MMIO_REG_NESTED_GPR;
210 	}
211 }
212 
213 static int kvmhv_read_guest_state_and_regs(struct kvm_vcpu *vcpu,
214 					   struct hv_guest_state *l2_hv,
215 					   struct pt_regs *l2_regs,
216 					   u64 hv_ptr, u64 regs_ptr)
217 {
218 	int size;
219 
220 	if (kvm_vcpu_read_guest(vcpu, hv_ptr, &l2_hv->version,
221 				sizeof(l2_hv->version)))
222 		return -1;
223 
224 	if (kvmppc_need_byteswap(vcpu))
225 		l2_hv->version = swab64(l2_hv->version);
226 
227 	size = hv_guest_state_size(l2_hv->version);
228 	if (size < 0)
229 		return -1;
230 
231 	return kvm_vcpu_read_guest(vcpu, hv_ptr, l2_hv, size) ||
232 		kvm_vcpu_read_guest(vcpu, regs_ptr, l2_regs,
233 				    sizeof(struct pt_regs));
234 }
235 
236 static int kvmhv_write_guest_state_and_regs(struct kvm_vcpu *vcpu,
237 					    struct hv_guest_state *l2_hv,
238 					    struct pt_regs *l2_regs,
239 					    u64 hv_ptr, u64 regs_ptr)
240 {
241 	int size;
242 
243 	size = hv_guest_state_size(l2_hv->version);
244 	if (size < 0)
245 		return -1;
246 
247 	return kvm_vcpu_write_guest(vcpu, hv_ptr, l2_hv, size) ||
248 		kvm_vcpu_write_guest(vcpu, regs_ptr, l2_regs,
249 				     sizeof(struct pt_regs));
250 }
251 
252 static void load_l2_hv_regs(struct kvm_vcpu *vcpu,
253 			    const struct hv_guest_state *l2_hv,
254 			    const struct hv_guest_state *l1_hv, u64 *lpcr)
255 {
256 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
257 	u64 mask;
258 
259 	restore_hv_regs(vcpu, l2_hv);
260 
261 	/*
262 	 * Don't let L1 change LPCR bits for the L2 except these:
263 	 */
264 	mask = LPCR_DPFD | LPCR_ILE | LPCR_TC | LPCR_AIL | LPCR_LD |
265 		LPCR_LPES | LPCR_MER;
266 
267 	/*
268 	 * Additional filtering is required depending on hardware
269 	 * and configuration.
270 	 */
271 	*lpcr = kvmppc_filter_lpcr_hv(vcpu->kvm,
272 				      (vc->lpcr & ~mask) | (*lpcr & mask));
273 
274 	/*
275 	 * Don't let L1 enable features for L2 which we don't allow for L1,
276 	 * but preserve the interrupt cause field.
277 	 */
278 	vcpu->arch.hfscr = l2_hv->hfscr & (HFSCR_INTR_CAUSE | vcpu->arch.hfscr_permitted);
279 
280 	/* Don't let data address watchpoint match in hypervisor state */
281 	vcpu->arch.dawrx0 = l2_hv->dawrx0 & ~DAWRX_HYP;
282 	vcpu->arch.dawrx1 = l2_hv->dawrx1 & ~DAWRX_HYP;
283 
284 	/* Don't let completed instruction address breakpt match in HV state */
285 	if ((l2_hv->ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
286 		vcpu->arch.ciabr = l2_hv->ciabr & ~CIABR_PRIV;
287 }
288 
289 long kvmhv_enter_nested_guest(struct kvm_vcpu *vcpu)
290 {
291 	long int err, r;
292 	struct kvm_nested_guest *l2;
293 	struct pt_regs l2_regs, saved_l1_regs;
294 	struct hv_guest_state l2_hv = {0}, saved_l1_hv;
295 	struct kvmppc_vcore *vc = vcpu->arch.vcore;
296 	u64 hv_ptr, regs_ptr;
297 	u64 hdec_exp, lpcr;
298 	s64 delta_purr, delta_spurr, delta_ic, delta_vtb;
299 
300 	if (vcpu->kvm->arch.l1_ptcr == 0)
301 		return H_NOT_AVAILABLE;
302 
303 	if (MSR_TM_TRANSACTIONAL(vcpu->arch.shregs.msr))
304 		return H_BAD_MODE;
305 
306 	/* copy parameters in */
307 	hv_ptr = kvmppc_get_gpr(vcpu, 4);
308 	regs_ptr = kvmppc_get_gpr(vcpu, 5);
309 	kvm_vcpu_srcu_read_lock(vcpu);
310 	err = kvmhv_read_guest_state_and_regs(vcpu, &l2_hv, &l2_regs,
311 					      hv_ptr, regs_ptr);
312 	kvm_vcpu_srcu_read_unlock(vcpu);
313 	if (err)
314 		return H_PARAMETER;
315 
316 	if (kvmppc_need_byteswap(vcpu))
317 		byteswap_hv_regs(&l2_hv);
318 	if (l2_hv.version > HV_GUEST_STATE_VERSION)
319 		return H_P2;
320 
321 	if (kvmppc_need_byteswap(vcpu))
322 		byteswap_pt_regs(&l2_regs);
323 	if (l2_hv.vcpu_token >= NR_CPUS)
324 		return H_PARAMETER;
325 
326 	/*
327 	 * L1 must have set up a suspended state to enter the L2 in a
328 	 * transactional state, and only in that case. These have to be
329 	 * filtered out here to prevent causing a TM Bad Thing in the
330 	 * host HRFID. We could synthesize a TM Bad Thing back to the L1
331 	 * here but there doesn't seem like much point.
332 	 */
333 	if (MSR_TM_SUSPENDED(vcpu->arch.shregs.msr)) {
334 		if (!MSR_TM_ACTIVE(l2_regs.msr))
335 			return H_BAD_MODE;
336 	} else {
337 		if (l2_regs.msr & MSR_TS_MASK)
338 			return H_BAD_MODE;
339 		if (WARN_ON_ONCE(vcpu->arch.shregs.msr & MSR_TS_MASK))
340 			return H_BAD_MODE;
341 	}
342 
343 	/* translate lpid */
344 	l2 = kvmhv_get_nested(vcpu->kvm, l2_hv.lpid, true);
345 	if (!l2)
346 		return H_PARAMETER;
347 	if (!l2->l1_gr_to_hr) {
348 		mutex_lock(&l2->tlb_lock);
349 		kvmhv_update_ptbl_cache(l2);
350 		mutex_unlock(&l2->tlb_lock);
351 	}
352 
353 	/* save l1 values of things */
354 	vcpu->arch.regs.msr = vcpu->arch.shregs.msr;
355 	saved_l1_regs = vcpu->arch.regs;
356 	kvmhv_save_hv_regs(vcpu, &saved_l1_hv);
357 
358 	/* convert TB values/offsets to host (L0) values */
359 	hdec_exp = l2_hv.hdec_expiry - vc->tb_offset;
360 	vc->tb_offset += l2_hv.tb_offset;
361 	vcpu->arch.dec_expires += l2_hv.tb_offset;
362 
363 	/* set L1 state to L2 state */
364 	vcpu->arch.nested = l2;
365 	vcpu->arch.nested_vcpu_id = l2_hv.vcpu_token;
366 	vcpu->arch.nested_hfscr = l2_hv.hfscr;
367 	vcpu->arch.regs = l2_regs;
368 
369 	/* Guest must always run with ME enabled, HV disabled. */
370 	vcpu->arch.shregs.msr = (vcpu->arch.regs.msr | MSR_ME) & ~MSR_HV;
371 
372 	lpcr = l2_hv.lpcr;
373 	load_l2_hv_regs(vcpu, &l2_hv, &saved_l1_hv, &lpcr);
374 
375 	vcpu->arch.ret = RESUME_GUEST;
376 	vcpu->arch.trap = 0;
377 	do {
378 		r = kvmhv_run_single_vcpu(vcpu, hdec_exp, lpcr);
379 	} while (is_kvmppc_resume_guest(r));
380 
381 	/* save L2 state for return */
382 	l2_regs = vcpu->arch.regs;
383 	l2_regs.msr = vcpu->arch.shregs.msr;
384 	delta_purr = vcpu->arch.purr - l2_hv.purr;
385 	delta_spurr = vcpu->arch.spurr - l2_hv.spurr;
386 	delta_ic = vcpu->arch.ic - l2_hv.ic;
387 	delta_vtb = vc->vtb - l2_hv.vtb;
388 	save_hv_return_state(vcpu, &l2_hv);
389 
390 	/* restore L1 state */
391 	vcpu->arch.nested = NULL;
392 	vcpu->arch.regs = saved_l1_regs;
393 	vcpu->arch.shregs.msr = saved_l1_regs.msr & ~MSR_TS_MASK;
394 	/* set L1 MSR TS field according to L2 transaction state */
395 	if (l2_regs.msr & MSR_TS_MASK)
396 		vcpu->arch.shregs.msr |= MSR_TS_S;
397 	vc->tb_offset = saved_l1_hv.tb_offset;
398 	/* XXX: is this always the same delta as saved_l1_hv.tb_offset? */
399 	vcpu->arch.dec_expires -= l2_hv.tb_offset;
400 	restore_hv_regs(vcpu, &saved_l1_hv);
401 	vcpu->arch.purr += delta_purr;
402 	vcpu->arch.spurr += delta_spurr;
403 	vcpu->arch.ic += delta_ic;
404 	vc->vtb += delta_vtb;
405 
406 	kvmhv_put_nested(l2);
407 
408 	/* copy l2_hv_state and regs back to guest */
409 	if (kvmppc_need_byteswap(vcpu)) {
410 		byteswap_hv_regs(&l2_hv);
411 		byteswap_pt_regs(&l2_regs);
412 	}
413 	kvm_vcpu_srcu_read_lock(vcpu);
414 	err = kvmhv_write_guest_state_and_regs(vcpu, &l2_hv, &l2_regs,
415 					       hv_ptr, regs_ptr);
416 	kvm_vcpu_srcu_read_unlock(vcpu);
417 	if (err)
418 		return H_AUTHORITY;
419 
420 	if (r == -EINTR)
421 		return H_INTERRUPT;
422 
423 	if (vcpu->mmio_needed) {
424 		kvmhv_nested_mmio_needed(vcpu, regs_ptr);
425 		return H_TOO_HARD;
426 	}
427 
428 	return vcpu->arch.trap;
429 }
430 
431 long kvmhv_nested_init(void)
432 {
433 	long int ptb_order;
434 	unsigned long ptcr;
435 	long rc;
436 
437 	if (!kvmhv_on_pseries())
438 		return 0;
439 	if (!radix_enabled())
440 		return -ENODEV;
441 
442 	/* find log base 2 of KVMPPC_NR_LPIDS, rounding up */
443 	ptb_order = __ilog2(KVMPPC_NR_LPIDS - 1) + 1;
444 	if (ptb_order < 8)
445 		ptb_order = 8;
446 	pseries_partition_tb = kmalloc(sizeof(struct patb_entry) << ptb_order,
447 				       GFP_KERNEL);
448 	if (!pseries_partition_tb) {
449 		pr_err("kvm-hv: failed to allocated nested partition table\n");
450 		return -ENOMEM;
451 	}
452 
453 	ptcr = __pa(pseries_partition_tb) | (ptb_order - 8);
454 	rc = plpar_hcall_norets(H_SET_PARTITION_TABLE, ptcr);
455 	if (rc != H_SUCCESS) {
456 		pr_err("kvm-hv: Parent hypervisor does not support nesting (rc=%ld)\n",
457 		       rc);
458 		kfree(pseries_partition_tb);
459 		pseries_partition_tb = NULL;
460 		return -ENODEV;
461 	}
462 
463 	return 0;
464 }
465 
466 void kvmhv_nested_exit(void)
467 {
468 	/*
469 	 * N.B. the kvmhv_on_pseries() test is there because it enables
470 	 * the compiler to remove the call to plpar_hcall_norets()
471 	 * when CONFIG_PPC_PSERIES=n.
472 	 */
473 	if (kvmhv_on_pseries() && pseries_partition_tb) {
474 		plpar_hcall_norets(H_SET_PARTITION_TABLE, 0);
475 		kfree(pseries_partition_tb);
476 		pseries_partition_tb = NULL;
477 	}
478 }
479 
480 static void kvmhv_flush_lpid(unsigned int lpid)
481 {
482 	long rc;
483 
484 	if (!kvmhv_on_pseries()) {
485 		radix__flush_all_lpid(lpid);
486 		return;
487 	}
488 
489 	if (!firmware_has_feature(FW_FEATURE_RPT_INVALIDATE))
490 		rc = plpar_hcall_norets(H_TLB_INVALIDATE, H_TLBIE_P1_ENC(2, 0, 1),
491 					lpid, TLBIEL_INVAL_SET_LPID);
492 	else
493 		rc = pseries_rpt_invalidate(lpid, H_RPTI_TARGET_CMMU,
494 					    H_RPTI_TYPE_NESTED |
495 					    H_RPTI_TYPE_TLB | H_RPTI_TYPE_PWC |
496 					    H_RPTI_TYPE_PAT,
497 					    H_RPTI_PAGE_ALL, 0, -1UL);
498 	if (rc)
499 		pr_err("KVM: TLB LPID invalidation hcall failed, rc=%ld\n", rc);
500 }
501 
502 void kvmhv_set_ptbl_entry(unsigned int lpid, u64 dw0, u64 dw1)
503 {
504 	if (!kvmhv_on_pseries()) {
505 		mmu_partition_table_set_entry(lpid, dw0, dw1, true);
506 		return;
507 	}
508 
509 	pseries_partition_tb[lpid].patb0 = cpu_to_be64(dw0);
510 	pseries_partition_tb[lpid].patb1 = cpu_to_be64(dw1);
511 	/* L0 will do the necessary barriers */
512 	kvmhv_flush_lpid(lpid);
513 }
514 
515 static void kvmhv_set_nested_ptbl(struct kvm_nested_guest *gp)
516 {
517 	unsigned long dw0;
518 
519 	dw0 = PATB_HR | radix__get_tree_size() |
520 		__pa(gp->shadow_pgtable) | RADIX_PGD_INDEX_SIZE;
521 	kvmhv_set_ptbl_entry(gp->shadow_lpid, dw0, gp->process_table);
522 }
523 
524 void kvmhv_vm_nested_init(struct kvm *kvm)
525 {
526 	kvm->arch.max_nested_lpid = -1;
527 }
528 
529 /*
530  * Handle the H_SET_PARTITION_TABLE hcall.
531  * r4 = guest real address of partition table + log_2(size) - 12
532  * (formatted as for the PTCR).
533  */
534 long kvmhv_set_partition_table(struct kvm_vcpu *vcpu)
535 {
536 	struct kvm *kvm = vcpu->kvm;
537 	unsigned long ptcr = kvmppc_get_gpr(vcpu, 4);
538 	int srcu_idx;
539 	long ret = H_SUCCESS;
540 
541 	srcu_idx = srcu_read_lock(&kvm->srcu);
542 	/*
543 	 * Limit the partition table to 4096 entries (because that's what
544 	 * hardware supports), and check the base address.
545 	 */
546 	if ((ptcr & PRTS_MASK) > 12 - 8 ||
547 	    !kvm_is_visible_gfn(vcpu->kvm, (ptcr & PRTB_MASK) >> PAGE_SHIFT))
548 		ret = H_PARAMETER;
549 	srcu_read_unlock(&kvm->srcu, srcu_idx);
550 	if (ret == H_SUCCESS)
551 		kvm->arch.l1_ptcr = ptcr;
552 	return ret;
553 }
554 
555 /*
556  * Handle the H_COPY_TOFROM_GUEST hcall.
557  * r4 = L1 lpid of nested guest
558  * r5 = pid
559  * r6 = eaddr to access
560  * r7 = to buffer (L1 gpa)
561  * r8 = from buffer (L1 gpa)
562  * r9 = n bytes to copy
563  */
564 long kvmhv_copy_tofrom_guest_nested(struct kvm_vcpu *vcpu)
565 {
566 	struct kvm_nested_guest *gp;
567 	int l1_lpid = kvmppc_get_gpr(vcpu, 4);
568 	int pid = kvmppc_get_gpr(vcpu, 5);
569 	gva_t eaddr = kvmppc_get_gpr(vcpu, 6);
570 	gpa_t gp_to = (gpa_t) kvmppc_get_gpr(vcpu, 7);
571 	gpa_t gp_from = (gpa_t) kvmppc_get_gpr(vcpu, 8);
572 	void *buf;
573 	unsigned long n = kvmppc_get_gpr(vcpu, 9);
574 	bool is_load = !!gp_to;
575 	long rc;
576 
577 	if (gp_to && gp_from) /* One must be NULL to determine the direction */
578 		return H_PARAMETER;
579 
580 	if (eaddr & (0xFFFUL << 52))
581 		return H_PARAMETER;
582 
583 	buf = kzalloc(n, GFP_KERNEL | __GFP_NOWARN);
584 	if (!buf)
585 		return H_NO_MEM;
586 
587 	gp = kvmhv_get_nested(vcpu->kvm, l1_lpid, false);
588 	if (!gp) {
589 		rc = H_PARAMETER;
590 		goto out_free;
591 	}
592 
593 	mutex_lock(&gp->tlb_lock);
594 
595 	if (is_load) {
596 		/* Load from the nested guest into our buffer */
597 		rc = __kvmhv_copy_tofrom_guest_radix(gp->shadow_lpid, pid,
598 						     eaddr, buf, NULL, n);
599 		if (rc)
600 			goto not_found;
601 
602 		/* Write what was loaded into our buffer back to the L1 guest */
603 		kvm_vcpu_srcu_read_lock(vcpu);
604 		rc = kvm_vcpu_write_guest(vcpu, gp_to, buf, n);
605 		kvm_vcpu_srcu_read_unlock(vcpu);
606 		if (rc)
607 			goto not_found;
608 	} else {
609 		/* Load the data to be stored from the L1 guest into our buf */
610 		kvm_vcpu_srcu_read_lock(vcpu);
611 		rc = kvm_vcpu_read_guest(vcpu, gp_from, buf, n);
612 		kvm_vcpu_srcu_read_unlock(vcpu);
613 		if (rc)
614 			goto not_found;
615 
616 		/* Store from our buffer into the nested guest */
617 		rc = __kvmhv_copy_tofrom_guest_radix(gp->shadow_lpid, pid,
618 						     eaddr, NULL, buf, n);
619 		if (rc)
620 			goto not_found;
621 	}
622 
623 out_unlock:
624 	mutex_unlock(&gp->tlb_lock);
625 	kvmhv_put_nested(gp);
626 out_free:
627 	kfree(buf);
628 	return rc;
629 not_found:
630 	rc = H_NOT_FOUND;
631 	goto out_unlock;
632 }
633 
634 /*
635  * Reload the partition table entry for a guest.
636  * Caller must hold gp->tlb_lock.
637  */
638 static void kvmhv_update_ptbl_cache(struct kvm_nested_guest *gp)
639 {
640 	int ret;
641 	struct patb_entry ptbl_entry;
642 	unsigned long ptbl_addr;
643 	struct kvm *kvm = gp->l1_host;
644 
645 	ret = -EFAULT;
646 	ptbl_addr = (kvm->arch.l1_ptcr & PRTB_MASK) + (gp->l1_lpid << 4);
647 	if (gp->l1_lpid < (1ul << ((kvm->arch.l1_ptcr & PRTS_MASK) + 8))) {
648 		int srcu_idx = srcu_read_lock(&kvm->srcu);
649 		ret = kvm_read_guest(kvm, ptbl_addr,
650 				     &ptbl_entry, sizeof(ptbl_entry));
651 		srcu_read_unlock(&kvm->srcu, srcu_idx);
652 	}
653 	if (ret) {
654 		gp->l1_gr_to_hr = 0;
655 		gp->process_table = 0;
656 	} else {
657 		gp->l1_gr_to_hr = be64_to_cpu(ptbl_entry.patb0);
658 		gp->process_table = be64_to_cpu(ptbl_entry.patb1);
659 	}
660 	kvmhv_set_nested_ptbl(gp);
661 }
662 
663 static struct kvm_nested_guest *kvmhv_alloc_nested(struct kvm *kvm, unsigned int lpid)
664 {
665 	struct kvm_nested_guest *gp;
666 	long shadow_lpid;
667 
668 	gp = kzalloc(sizeof(*gp), GFP_KERNEL);
669 	if (!gp)
670 		return NULL;
671 	gp->l1_host = kvm;
672 	gp->l1_lpid = lpid;
673 	mutex_init(&gp->tlb_lock);
674 	gp->shadow_pgtable = pgd_alloc(kvm->mm);
675 	if (!gp->shadow_pgtable)
676 		goto out_free;
677 	shadow_lpid = kvmppc_alloc_lpid();
678 	if (shadow_lpid < 0)
679 		goto out_free2;
680 	gp->shadow_lpid = shadow_lpid;
681 	gp->radix = 1;
682 
683 	memset(gp->prev_cpu, -1, sizeof(gp->prev_cpu));
684 
685 	return gp;
686 
687  out_free2:
688 	pgd_free(kvm->mm, gp->shadow_pgtable);
689  out_free:
690 	kfree(gp);
691 	return NULL;
692 }
693 
694 /*
695  * Free up any resources allocated for a nested guest.
696  */
697 static void kvmhv_release_nested(struct kvm_nested_guest *gp)
698 {
699 	struct kvm *kvm = gp->l1_host;
700 
701 	if (gp->shadow_pgtable) {
702 		/*
703 		 * No vcpu is using this struct and no call to
704 		 * kvmhv_get_nested can find this struct,
705 		 * so we don't need to hold kvm->mmu_lock.
706 		 */
707 		kvmppc_free_pgtable_radix(kvm, gp->shadow_pgtable,
708 					  gp->shadow_lpid);
709 		pgd_free(kvm->mm, gp->shadow_pgtable);
710 	}
711 	kvmhv_set_ptbl_entry(gp->shadow_lpid, 0, 0);
712 	kvmppc_free_lpid(gp->shadow_lpid);
713 	kfree(gp);
714 }
715 
716 static void kvmhv_remove_nested(struct kvm_nested_guest *gp)
717 {
718 	struct kvm *kvm = gp->l1_host;
719 	int lpid = gp->l1_lpid;
720 	long ref;
721 
722 	spin_lock(&kvm->mmu_lock);
723 	if (gp == kvm->arch.nested_guests[lpid]) {
724 		kvm->arch.nested_guests[lpid] = NULL;
725 		if (lpid == kvm->arch.max_nested_lpid) {
726 			while (--lpid >= 0 && !kvm->arch.nested_guests[lpid])
727 				;
728 			kvm->arch.max_nested_lpid = lpid;
729 		}
730 		--gp->refcnt;
731 	}
732 	ref = gp->refcnt;
733 	spin_unlock(&kvm->mmu_lock);
734 	if (ref == 0)
735 		kvmhv_release_nested(gp);
736 }
737 
738 /*
739  * Free up all nested resources allocated for this guest.
740  * This is called with no vcpus of the guest running, when
741  * switching the guest to HPT mode or when destroying the
742  * guest.
743  */
744 void kvmhv_release_all_nested(struct kvm *kvm)
745 {
746 	int i;
747 	struct kvm_nested_guest *gp;
748 	struct kvm_nested_guest *freelist = NULL;
749 	struct kvm_memory_slot *memslot;
750 	int srcu_idx, bkt;
751 
752 	spin_lock(&kvm->mmu_lock);
753 	for (i = 0; i <= kvm->arch.max_nested_lpid; i++) {
754 		gp = kvm->arch.nested_guests[i];
755 		if (!gp)
756 			continue;
757 		kvm->arch.nested_guests[i] = NULL;
758 		if (--gp->refcnt == 0) {
759 			gp->next = freelist;
760 			freelist = gp;
761 		}
762 	}
763 	kvm->arch.max_nested_lpid = -1;
764 	spin_unlock(&kvm->mmu_lock);
765 	while ((gp = freelist) != NULL) {
766 		freelist = gp->next;
767 		kvmhv_release_nested(gp);
768 	}
769 
770 	srcu_idx = srcu_read_lock(&kvm->srcu);
771 	kvm_for_each_memslot(memslot, bkt, kvm_memslots(kvm))
772 		kvmhv_free_memslot_nest_rmap(memslot);
773 	srcu_read_unlock(&kvm->srcu, srcu_idx);
774 }
775 
776 /* caller must hold gp->tlb_lock */
777 static void kvmhv_flush_nested(struct kvm_nested_guest *gp)
778 {
779 	struct kvm *kvm = gp->l1_host;
780 
781 	spin_lock(&kvm->mmu_lock);
782 	kvmppc_free_pgtable_radix(kvm, gp->shadow_pgtable, gp->shadow_lpid);
783 	spin_unlock(&kvm->mmu_lock);
784 	kvmhv_flush_lpid(gp->shadow_lpid);
785 	kvmhv_update_ptbl_cache(gp);
786 	if (gp->l1_gr_to_hr == 0)
787 		kvmhv_remove_nested(gp);
788 }
789 
790 struct kvm_nested_guest *kvmhv_get_nested(struct kvm *kvm, int l1_lpid,
791 					  bool create)
792 {
793 	struct kvm_nested_guest *gp, *newgp;
794 
795 	if (l1_lpid >= KVM_MAX_NESTED_GUESTS ||
796 	    l1_lpid >= (1ul << ((kvm->arch.l1_ptcr & PRTS_MASK) + 12 - 4)))
797 		return NULL;
798 
799 	spin_lock(&kvm->mmu_lock);
800 	gp = kvm->arch.nested_guests[l1_lpid];
801 	if (gp)
802 		++gp->refcnt;
803 	spin_unlock(&kvm->mmu_lock);
804 
805 	if (gp || !create)
806 		return gp;
807 
808 	newgp = kvmhv_alloc_nested(kvm, l1_lpid);
809 	if (!newgp)
810 		return NULL;
811 	spin_lock(&kvm->mmu_lock);
812 	if (kvm->arch.nested_guests[l1_lpid]) {
813 		/* someone else beat us to it */
814 		gp = kvm->arch.nested_guests[l1_lpid];
815 	} else {
816 		kvm->arch.nested_guests[l1_lpid] = newgp;
817 		++newgp->refcnt;
818 		gp = newgp;
819 		newgp = NULL;
820 		if (l1_lpid > kvm->arch.max_nested_lpid)
821 			kvm->arch.max_nested_lpid = l1_lpid;
822 	}
823 	++gp->refcnt;
824 	spin_unlock(&kvm->mmu_lock);
825 
826 	if (newgp)
827 		kvmhv_release_nested(newgp);
828 
829 	return gp;
830 }
831 
832 void kvmhv_put_nested(struct kvm_nested_guest *gp)
833 {
834 	struct kvm *kvm = gp->l1_host;
835 	long ref;
836 
837 	spin_lock(&kvm->mmu_lock);
838 	ref = --gp->refcnt;
839 	spin_unlock(&kvm->mmu_lock);
840 	if (ref == 0)
841 		kvmhv_release_nested(gp);
842 }
843 
844 static struct kvm_nested_guest *kvmhv_find_nested(struct kvm *kvm, int lpid)
845 {
846 	if (lpid > kvm->arch.max_nested_lpid)
847 		return NULL;
848 	return kvm->arch.nested_guests[lpid];
849 }
850 
851 pte_t *find_kvm_nested_guest_pte(struct kvm *kvm, unsigned long lpid,
852 				 unsigned long ea, unsigned *hshift)
853 {
854 	struct kvm_nested_guest *gp;
855 	pte_t *pte;
856 
857 	gp = kvmhv_find_nested(kvm, lpid);
858 	if (!gp)
859 		return NULL;
860 
861 	VM_WARN(!spin_is_locked(&kvm->mmu_lock),
862 		"%s called with kvm mmu_lock not held \n", __func__);
863 	pte = __find_linux_pte(gp->shadow_pgtable, ea, NULL, hshift);
864 
865 	return pte;
866 }
867 
868 static inline bool kvmhv_n_rmap_is_equal(u64 rmap_1, u64 rmap_2)
869 {
870 	return !((rmap_1 ^ rmap_2) & (RMAP_NESTED_LPID_MASK |
871 				       RMAP_NESTED_GPA_MASK));
872 }
873 
874 void kvmhv_insert_nest_rmap(struct kvm *kvm, unsigned long *rmapp,
875 			    struct rmap_nested **n_rmap)
876 {
877 	struct llist_node *entry = ((struct llist_head *) rmapp)->first;
878 	struct rmap_nested *cursor;
879 	u64 rmap, new_rmap = (*n_rmap)->rmap;
880 
881 	/* Are there any existing entries? */
882 	if (!(*rmapp)) {
883 		/* No -> use the rmap as a single entry */
884 		*rmapp = new_rmap | RMAP_NESTED_IS_SINGLE_ENTRY;
885 		return;
886 	}
887 
888 	/* Do any entries match what we're trying to insert? */
889 	for_each_nest_rmap_safe(cursor, entry, &rmap) {
890 		if (kvmhv_n_rmap_is_equal(rmap, new_rmap))
891 			return;
892 	}
893 
894 	/* Do we need to create a list or just add the new entry? */
895 	rmap = *rmapp;
896 	if (rmap & RMAP_NESTED_IS_SINGLE_ENTRY) /* Not previously a list */
897 		*rmapp = 0UL;
898 	llist_add(&((*n_rmap)->list), (struct llist_head *) rmapp);
899 	if (rmap & RMAP_NESTED_IS_SINGLE_ENTRY) /* Not previously a list */
900 		(*n_rmap)->list.next = (struct llist_node *) rmap;
901 
902 	/* Set NULL so not freed by caller */
903 	*n_rmap = NULL;
904 }
905 
906 static void kvmhv_update_nest_rmap_rc(struct kvm *kvm, u64 n_rmap,
907 				      unsigned long clr, unsigned long set,
908 				      unsigned long hpa, unsigned long mask)
909 {
910 	unsigned long gpa;
911 	unsigned int shift, lpid;
912 	pte_t *ptep;
913 
914 	gpa = n_rmap & RMAP_NESTED_GPA_MASK;
915 	lpid = (n_rmap & RMAP_NESTED_LPID_MASK) >> RMAP_NESTED_LPID_SHIFT;
916 
917 	/* Find the pte */
918 	ptep = find_kvm_nested_guest_pte(kvm, lpid, gpa, &shift);
919 	/*
920 	 * If the pte is present and the pfn is still the same, update the pte.
921 	 * If the pfn has changed then this is a stale rmap entry, the nested
922 	 * gpa actually points somewhere else now, and there is nothing to do.
923 	 * XXX A future optimisation would be to remove the rmap entry here.
924 	 */
925 	if (ptep && pte_present(*ptep) && ((pte_val(*ptep) & mask) == hpa)) {
926 		__radix_pte_update(ptep, clr, set);
927 		kvmppc_radix_tlbie_page(kvm, gpa, shift, lpid);
928 	}
929 }
930 
931 /*
932  * For a given list of rmap entries, update the rc bits in all ptes in shadow
933  * page tables for nested guests which are referenced by the rmap list.
934  */
935 void kvmhv_update_nest_rmap_rc_list(struct kvm *kvm, unsigned long *rmapp,
936 				    unsigned long clr, unsigned long set,
937 				    unsigned long hpa, unsigned long nbytes)
938 {
939 	struct llist_node *entry = ((struct llist_head *) rmapp)->first;
940 	struct rmap_nested *cursor;
941 	unsigned long rmap, mask;
942 
943 	if ((clr | set) & ~(_PAGE_DIRTY | _PAGE_ACCESSED))
944 		return;
945 
946 	mask = PTE_RPN_MASK & ~(nbytes - 1);
947 	hpa &= mask;
948 
949 	for_each_nest_rmap_safe(cursor, entry, &rmap)
950 		kvmhv_update_nest_rmap_rc(kvm, rmap, clr, set, hpa, mask);
951 }
952 
953 static void kvmhv_remove_nest_rmap(struct kvm *kvm, u64 n_rmap,
954 				   unsigned long hpa, unsigned long mask)
955 {
956 	struct kvm_nested_guest *gp;
957 	unsigned long gpa;
958 	unsigned int shift, lpid;
959 	pte_t *ptep;
960 
961 	gpa = n_rmap & RMAP_NESTED_GPA_MASK;
962 	lpid = (n_rmap & RMAP_NESTED_LPID_MASK) >> RMAP_NESTED_LPID_SHIFT;
963 	gp = kvmhv_find_nested(kvm, lpid);
964 	if (!gp)
965 		return;
966 
967 	/* Find and invalidate the pte */
968 	ptep = find_kvm_nested_guest_pte(kvm, lpid, gpa, &shift);
969 	/* Don't spuriously invalidate ptes if the pfn has changed */
970 	if (ptep && pte_present(*ptep) && ((pte_val(*ptep) & mask) == hpa))
971 		kvmppc_unmap_pte(kvm, ptep, gpa, shift, NULL, gp->shadow_lpid);
972 }
973 
974 static void kvmhv_remove_nest_rmap_list(struct kvm *kvm, unsigned long *rmapp,
975 					unsigned long hpa, unsigned long mask)
976 {
977 	struct llist_node *entry = llist_del_all((struct llist_head *) rmapp);
978 	struct rmap_nested *cursor;
979 	unsigned long rmap;
980 
981 	for_each_nest_rmap_safe(cursor, entry, &rmap) {
982 		kvmhv_remove_nest_rmap(kvm, rmap, hpa, mask);
983 		kfree(cursor);
984 	}
985 }
986 
987 /* called with kvm->mmu_lock held */
988 void kvmhv_remove_nest_rmap_range(struct kvm *kvm,
989 				  const struct kvm_memory_slot *memslot,
990 				  unsigned long gpa, unsigned long hpa,
991 				  unsigned long nbytes)
992 {
993 	unsigned long gfn, end_gfn;
994 	unsigned long addr_mask;
995 
996 	if (!memslot)
997 		return;
998 	gfn = (gpa >> PAGE_SHIFT) - memslot->base_gfn;
999 	end_gfn = gfn + (nbytes >> PAGE_SHIFT);
1000 
1001 	addr_mask = PTE_RPN_MASK & ~(nbytes - 1);
1002 	hpa &= addr_mask;
1003 
1004 	for (; gfn < end_gfn; gfn++) {
1005 		unsigned long *rmap = &memslot->arch.rmap[gfn];
1006 		kvmhv_remove_nest_rmap_list(kvm, rmap, hpa, addr_mask);
1007 	}
1008 }
1009 
1010 static void kvmhv_free_memslot_nest_rmap(struct kvm_memory_slot *free)
1011 {
1012 	unsigned long page;
1013 
1014 	for (page = 0; page < free->npages; page++) {
1015 		unsigned long rmap, *rmapp = &free->arch.rmap[page];
1016 		struct rmap_nested *cursor;
1017 		struct llist_node *entry;
1018 
1019 		entry = llist_del_all((struct llist_head *) rmapp);
1020 		for_each_nest_rmap_safe(cursor, entry, &rmap)
1021 			kfree(cursor);
1022 	}
1023 }
1024 
1025 static bool kvmhv_invalidate_shadow_pte(struct kvm_vcpu *vcpu,
1026 					struct kvm_nested_guest *gp,
1027 					long gpa, int *shift_ret)
1028 {
1029 	struct kvm *kvm = vcpu->kvm;
1030 	bool ret = false;
1031 	pte_t *ptep;
1032 	int shift;
1033 
1034 	spin_lock(&kvm->mmu_lock);
1035 	ptep = find_kvm_nested_guest_pte(kvm, gp->l1_lpid, gpa, &shift);
1036 	if (!shift)
1037 		shift = PAGE_SHIFT;
1038 	if (ptep && pte_present(*ptep)) {
1039 		kvmppc_unmap_pte(kvm, ptep, gpa, shift, NULL, gp->shadow_lpid);
1040 		ret = true;
1041 	}
1042 	spin_unlock(&kvm->mmu_lock);
1043 
1044 	if (shift_ret)
1045 		*shift_ret = shift;
1046 	return ret;
1047 }
1048 
1049 static inline int get_ric(unsigned int instr)
1050 {
1051 	return (instr >> 18) & 0x3;
1052 }
1053 
1054 static inline int get_prs(unsigned int instr)
1055 {
1056 	return (instr >> 17) & 0x1;
1057 }
1058 
1059 static inline int get_r(unsigned int instr)
1060 {
1061 	return (instr >> 16) & 0x1;
1062 }
1063 
1064 static inline int get_lpid(unsigned long r_val)
1065 {
1066 	return r_val & 0xffffffff;
1067 }
1068 
1069 static inline int get_is(unsigned long r_val)
1070 {
1071 	return (r_val >> 10) & 0x3;
1072 }
1073 
1074 static inline int get_ap(unsigned long r_val)
1075 {
1076 	return (r_val >> 5) & 0x7;
1077 }
1078 
1079 static inline long get_epn(unsigned long r_val)
1080 {
1081 	return r_val >> 12;
1082 }
1083 
1084 static int kvmhv_emulate_tlbie_tlb_addr(struct kvm_vcpu *vcpu, int lpid,
1085 					int ap, long epn)
1086 {
1087 	struct kvm *kvm = vcpu->kvm;
1088 	struct kvm_nested_guest *gp;
1089 	long npages;
1090 	int shift, shadow_shift;
1091 	unsigned long addr;
1092 
1093 	shift = ap_to_shift(ap);
1094 	addr = epn << 12;
1095 	if (shift < 0)
1096 		/* Invalid ap encoding */
1097 		return -EINVAL;
1098 
1099 	addr &= ~((1UL << shift) - 1);
1100 	npages = 1UL << (shift - PAGE_SHIFT);
1101 
1102 	gp = kvmhv_get_nested(kvm, lpid, false);
1103 	if (!gp) /* No such guest -> nothing to do */
1104 		return 0;
1105 	mutex_lock(&gp->tlb_lock);
1106 
1107 	/* There may be more than one host page backing this single guest pte */
1108 	do {
1109 		kvmhv_invalidate_shadow_pte(vcpu, gp, addr, &shadow_shift);
1110 
1111 		npages -= 1UL << (shadow_shift - PAGE_SHIFT);
1112 		addr += 1UL << shadow_shift;
1113 	} while (npages > 0);
1114 
1115 	mutex_unlock(&gp->tlb_lock);
1116 	kvmhv_put_nested(gp);
1117 	return 0;
1118 }
1119 
1120 static void kvmhv_emulate_tlbie_lpid(struct kvm_vcpu *vcpu,
1121 				     struct kvm_nested_guest *gp, int ric)
1122 {
1123 	struct kvm *kvm = vcpu->kvm;
1124 
1125 	mutex_lock(&gp->tlb_lock);
1126 	switch (ric) {
1127 	case 0:
1128 		/* Invalidate TLB */
1129 		spin_lock(&kvm->mmu_lock);
1130 		kvmppc_free_pgtable_radix(kvm, gp->shadow_pgtable,
1131 					  gp->shadow_lpid);
1132 		kvmhv_flush_lpid(gp->shadow_lpid);
1133 		spin_unlock(&kvm->mmu_lock);
1134 		break;
1135 	case 1:
1136 		/*
1137 		 * Invalidate PWC
1138 		 * We don't cache this -> nothing to do
1139 		 */
1140 		break;
1141 	case 2:
1142 		/* Invalidate TLB, PWC and caching of partition table entries */
1143 		kvmhv_flush_nested(gp);
1144 		break;
1145 	default:
1146 		break;
1147 	}
1148 	mutex_unlock(&gp->tlb_lock);
1149 }
1150 
1151 static void kvmhv_emulate_tlbie_all_lpid(struct kvm_vcpu *vcpu, int ric)
1152 {
1153 	struct kvm *kvm = vcpu->kvm;
1154 	struct kvm_nested_guest *gp;
1155 	int i;
1156 
1157 	spin_lock(&kvm->mmu_lock);
1158 	for (i = 0; i <= kvm->arch.max_nested_lpid; i++) {
1159 		gp = kvm->arch.nested_guests[i];
1160 		if (gp) {
1161 			spin_unlock(&kvm->mmu_lock);
1162 			kvmhv_emulate_tlbie_lpid(vcpu, gp, ric);
1163 			spin_lock(&kvm->mmu_lock);
1164 		}
1165 	}
1166 	spin_unlock(&kvm->mmu_lock);
1167 }
1168 
1169 static int kvmhv_emulate_priv_tlbie(struct kvm_vcpu *vcpu, unsigned int instr,
1170 				    unsigned long rsval, unsigned long rbval)
1171 {
1172 	struct kvm *kvm = vcpu->kvm;
1173 	struct kvm_nested_guest *gp;
1174 	int r, ric, prs, is, ap;
1175 	int lpid;
1176 	long epn;
1177 	int ret = 0;
1178 
1179 	ric = get_ric(instr);
1180 	prs = get_prs(instr);
1181 	r = get_r(instr);
1182 	lpid = get_lpid(rsval);
1183 	is = get_is(rbval);
1184 
1185 	/*
1186 	 * These cases are invalid and are not handled:
1187 	 * r   != 1 -> Only radix supported
1188 	 * prs == 1 -> Not HV privileged
1189 	 * ric == 3 -> No cluster bombs for radix
1190 	 * is  == 1 -> Partition scoped translations not associated with pid
1191 	 * (!is) && (ric == 1 || ric == 2) -> Not supported by ISA
1192 	 */
1193 	if ((!r) || (prs) || (ric == 3) || (is == 1) ||
1194 	    ((!is) && (ric == 1 || ric == 2)))
1195 		return -EINVAL;
1196 
1197 	switch (is) {
1198 	case 0:
1199 		/*
1200 		 * We know ric == 0
1201 		 * Invalidate TLB for a given target address
1202 		 */
1203 		epn = get_epn(rbval);
1204 		ap = get_ap(rbval);
1205 		ret = kvmhv_emulate_tlbie_tlb_addr(vcpu, lpid, ap, epn);
1206 		break;
1207 	case 2:
1208 		/* Invalidate matching LPID */
1209 		gp = kvmhv_get_nested(kvm, lpid, false);
1210 		if (gp) {
1211 			kvmhv_emulate_tlbie_lpid(vcpu, gp, ric);
1212 			kvmhv_put_nested(gp);
1213 		}
1214 		break;
1215 	case 3:
1216 		/* Invalidate ALL LPIDs */
1217 		kvmhv_emulate_tlbie_all_lpid(vcpu, ric);
1218 		break;
1219 	default:
1220 		ret = -EINVAL;
1221 		break;
1222 	}
1223 
1224 	return ret;
1225 }
1226 
1227 /*
1228  * This handles the H_TLB_INVALIDATE hcall.
1229  * Parameters are (r4) tlbie instruction code, (r5) rS contents,
1230  * (r6) rB contents.
1231  */
1232 long kvmhv_do_nested_tlbie(struct kvm_vcpu *vcpu)
1233 {
1234 	int ret;
1235 
1236 	ret = kvmhv_emulate_priv_tlbie(vcpu, kvmppc_get_gpr(vcpu, 4),
1237 			kvmppc_get_gpr(vcpu, 5), kvmppc_get_gpr(vcpu, 6));
1238 	if (ret)
1239 		return H_PARAMETER;
1240 	return H_SUCCESS;
1241 }
1242 
1243 static long do_tlb_invalidate_nested_all(struct kvm_vcpu *vcpu,
1244 					 unsigned long lpid, unsigned long ric)
1245 {
1246 	struct kvm *kvm = vcpu->kvm;
1247 	struct kvm_nested_guest *gp;
1248 
1249 	gp = kvmhv_get_nested(kvm, lpid, false);
1250 	if (gp) {
1251 		kvmhv_emulate_tlbie_lpid(vcpu, gp, ric);
1252 		kvmhv_put_nested(gp);
1253 	}
1254 	return H_SUCCESS;
1255 }
1256 
1257 /*
1258  * Number of pages above which we invalidate the entire LPID rather than
1259  * flush individual pages.
1260  */
1261 static unsigned long tlb_range_flush_page_ceiling __read_mostly = 33;
1262 
1263 static long do_tlb_invalidate_nested_tlb(struct kvm_vcpu *vcpu,
1264 					 unsigned long lpid,
1265 					 unsigned long pg_sizes,
1266 					 unsigned long start,
1267 					 unsigned long end)
1268 {
1269 	int ret = H_P4;
1270 	unsigned long addr, nr_pages;
1271 	struct mmu_psize_def *def;
1272 	unsigned long psize, ap, page_size;
1273 	bool flush_lpid;
1274 
1275 	for (psize = 0; psize < MMU_PAGE_COUNT; psize++) {
1276 		def = &mmu_psize_defs[psize];
1277 		if (!(pg_sizes & def->h_rpt_pgsize))
1278 			continue;
1279 
1280 		nr_pages = (end - start) >> def->shift;
1281 		flush_lpid = nr_pages > tlb_range_flush_page_ceiling;
1282 		if (flush_lpid)
1283 			return do_tlb_invalidate_nested_all(vcpu, lpid,
1284 							RIC_FLUSH_TLB);
1285 		addr = start;
1286 		ap = mmu_get_ap(psize);
1287 		page_size = 1UL << def->shift;
1288 		do {
1289 			ret = kvmhv_emulate_tlbie_tlb_addr(vcpu, lpid, ap,
1290 						   get_epn(addr));
1291 			if (ret)
1292 				return H_P4;
1293 			addr += page_size;
1294 		} while (addr < end);
1295 	}
1296 	return ret;
1297 }
1298 
1299 /*
1300  * Performs partition-scoped invalidations for nested guests
1301  * as part of H_RPT_INVALIDATE hcall.
1302  */
1303 long do_h_rpt_invalidate_pat(struct kvm_vcpu *vcpu, unsigned long lpid,
1304 			     unsigned long type, unsigned long pg_sizes,
1305 			     unsigned long start, unsigned long end)
1306 {
1307 	/*
1308 	 * If L2 lpid isn't valid, we need to return H_PARAMETER.
1309 	 *
1310 	 * However, nested KVM issues a L2 lpid flush call when creating
1311 	 * partition table entries for L2. This happens even before the
1312 	 * corresponding shadow lpid is created in HV which happens in
1313 	 * H_ENTER_NESTED call. Since we can't differentiate this case from
1314 	 * the invalid case, we ignore such flush requests and return success.
1315 	 */
1316 	if (!kvmhv_find_nested(vcpu->kvm, lpid))
1317 		return H_SUCCESS;
1318 
1319 	/*
1320 	 * A flush all request can be handled by a full lpid flush only.
1321 	 */
1322 	if ((type & H_RPTI_TYPE_NESTED_ALL) == H_RPTI_TYPE_NESTED_ALL)
1323 		return do_tlb_invalidate_nested_all(vcpu, lpid, RIC_FLUSH_ALL);
1324 
1325 	/*
1326 	 * We don't need to handle a PWC flush like process table here,
1327 	 * because intermediate partition scoped table in nested guest doesn't
1328 	 * really have PWC. Only level we have PWC is in L0 and for nested
1329 	 * invalidate at L0 we always do kvm_flush_lpid() which does
1330 	 * radix__flush_all_lpid(). For range invalidate at any level, we
1331 	 * are not removing the higher level page tables and hence there is
1332 	 * no PWC invalidate needed.
1333 	 *
1334 	 * if (type & H_RPTI_TYPE_PWC) {
1335 	 *	ret = do_tlb_invalidate_nested_all(vcpu, lpid, RIC_FLUSH_PWC);
1336 	 *	if (ret)
1337 	 *		return H_P4;
1338 	 * }
1339 	 */
1340 
1341 	if (start == 0 && end == -1)
1342 		return do_tlb_invalidate_nested_all(vcpu, lpid, RIC_FLUSH_TLB);
1343 
1344 	if (type & H_RPTI_TYPE_TLB)
1345 		return do_tlb_invalidate_nested_tlb(vcpu, lpid, pg_sizes,
1346 						    start, end);
1347 	return H_SUCCESS;
1348 }
1349 
1350 /* Used to convert a nested guest real address to a L1 guest real address */
1351 static int kvmhv_translate_addr_nested(struct kvm_vcpu *vcpu,
1352 				       struct kvm_nested_guest *gp,
1353 				       unsigned long n_gpa, unsigned long dsisr,
1354 				       struct kvmppc_pte *gpte_p)
1355 {
1356 	u64 fault_addr, flags = dsisr & DSISR_ISSTORE;
1357 	int ret;
1358 
1359 	ret = kvmppc_mmu_walk_radix_tree(vcpu, n_gpa, gpte_p, gp->l1_gr_to_hr,
1360 					 &fault_addr);
1361 
1362 	if (ret) {
1363 		/* We didn't find a pte */
1364 		if (ret == -EINVAL) {
1365 			/* Unsupported mmu config */
1366 			flags |= DSISR_UNSUPP_MMU;
1367 		} else if (ret == -ENOENT) {
1368 			/* No translation found */
1369 			flags |= DSISR_NOHPTE;
1370 		} else if (ret == -EFAULT) {
1371 			/* Couldn't access L1 real address */
1372 			flags |= DSISR_PRTABLE_FAULT;
1373 			vcpu->arch.fault_gpa = fault_addr;
1374 		} else {
1375 			/* Unknown error */
1376 			return ret;
1377 		}
1378 		goto forward_to_l1;
1379 	} else {
1380 		/* We found a pte -> check permissions */
1381 		if (dsisr & DSISR_ISSTORE) {
1382 			/* Can we write? */
1383 			if (!gpte_p->may_write) {
1384 				flags |= DSISR_PROTFAULT;
1385 				goto forward_to_l1;
1386 			}
1387 		} else if (vcpu->arch.trap == BOOK3S_INTERRUPT_H_INST_STORAGE) {
1388 			/* Can we execute? */
1389 			if (!gpte_p->may_execute) {
1390 				flags |= SRR1_ISI_N_G_OR_CIP;
1391 				goto forward_to_l1;
1392 			}
1393 		} else {
1394 			/* Can we read? */
1395 			if (!gpte_p->may_read && !gpte_p->may_write) {
1396 				flags |= DSISR_PROTFAULT;
1397 				goto forward_to_l1;
1398 			}
1399 		}
1400 	}
1401 
1402 	return 0;
1403 
1404 forward_to_l1:
1405 	vcpu->arch.fault_dsisr = flags;
1406 	if (vcpu->arch.trap == BOOK3S_INTERRUPT_H_INST_STORAGE) {
1407 		vcpu->arch.shregs.msr &= SRR1_MSR_BITS;
1408 		vcpu->arch.shregs.msr |= flags;
1409 	}
1410 	return RESUME_HOST;
1411 }
1412 
1413 static long kvmhv_handle_nested_set_rc(struct kvm_vcpu *vcpu,
1414 				       struct kvm_nested_guest *gp,
1415 				       unsigned long n_gpa,
1416 				       struct kvmppc_pte gpte,
1417 				       unsigned long dsisr)
1418 {
1419 	struct kvm *kvm = vcpu->kvm;
1420 	bool writing = !!(dsisr & DSISR_ISSTORE);
1421 	u64 pgflags;
1422 	long ret;
1423 
1424 	/* Are the rc bits set in the L1 partition scoped pte? */
1425 	pgflags = _PAGE_ACCESSED;
1426 	if (writing)
1427 		pgflags |= _PAGE_DIRTY;
1428 	if (pgflags & ~gpte.rc)
1429 		return RESUME_HOST;
1430 
1431 	spin_lock(&kvm->mmu_lock);
1432 	/* Set the rc bit in the pte of our (L0) pgtable for the L1 guest */
1433 	ret = kvmppc_hv_handle_set_rc(kvm, false, writing,
1434 				      gpte.raddr, kvm->arch.lpid);
1435 	if (!ret) {
1436 		ret = -EINVAL;
1437 		goto out_unlock;
1438 	}
1439 
1440 	/* Set the rc bit in the pte of the shadow_pgtable for the nest guest */
1441 	ret = kvmppc_hv_handle_set_rc(kvm, true, writing,
1442 				      n_gpa, gp->l1_lpid);
1443 	if (!ret)
1444 		ret = -EINVAL;
1445 	else
1446 		ret = 0;
1447 
1448 out_unlock:
1449 	spin_unlock(&kvm->mmu_lock);
1450 	return ret;
1451 }
1452 
1453 static inline int kvmppc_radix_level_to_shift(int level)
1454 {
1455 	switch (level) {
1456 	case 2:
1457 		return PUD_SHIFT;
1458 	case 1:
1459 		return PMD_SHIFT;
1460 	default:
1461 		return PAGE_SHIFT;
1462 	}
1463 }
1464 
1465 static inline int kvmppc_radix_shift_to_level(int shift)
1466 {
1467 	if (shift == PUD_SHIFT)
1468 		return 2;
1469 	if (shift == PMD_SHIFT)
1470 		return 1;
1471 	if (shift == PAGE_SHIFT)
1472 		return 0;
1473 	WARN_ON_ONCE(1);
1474 	return 0;
1475 }
1476 
1477 /* called with gp->tlb_lock held */
1478 static long int __kvmhv_nested_page_fault(struct kvm_vcpu *vcpu,
1479 					  struct kvm_nested_guest *gp)
1480 {
1481 	struct kvm *kvm = vcpu->kvm;
1482 	struct kvm_memory_slot *memslot;
1483 	struct rmap_nested *n_rmap;
1484 	struct kvmppc_pte gpte;
1485 	pte_t pte, *pte_p;
1486 	unsigned long mmu_seq;
1487 	unsigned long dsisr = vcpu->arch.fault_dsisr;
1488 	unsigned long ea = vcpu->arch.fault_dar;
1489 	unsigned long *rmapp;
1490 	unsigned long n_gpa, gpa, gfn, perm = 0UL;
1491 	unsigned int shift, l1_shift, level;
1492 	bool writing = !!(dsisr & DSISR_ISSTORE);
1493 	bool kvm_ro = false;
1494 	long int ret;
1495 
1496 	if (!gp->l1_gr_to_hr) {
1497 		kvmhv_update_ptbl_cache(gp);
1498 		if (!gp->l1_gr_to_hr)
1499 			return RESUME_HOST;
1500 	}
1501 
1502 	/* Convert the nested guest real address into a L1 guest real address */
1503 
1504 	n_gpa = vcpu->arch.fault_gpa & ~0xF000000000000FFFULL;
1505 	if (!(dsisr & DSISR_PRTABLE_FAULT))
1506 		n_gpa |= ea & 0xFFF;
1507 	ret = kvmhv_translate_addr_nested(vcpu, gp, n_gpa, dsisr, &gpte);
1508 
1509 	/*
1510 	 * If the hardware found a translation but we don't now have a usable
1511 	 * translation in the l1 partition-scoped tree, remove the shadow pte
1512 	 * and let the guest retry.
1513 	 */
1514 	if (ret == RESUME_HOST &&
1515 	    (dsisr & (DSISR_PROTFAULT | DSISR_BADACCESS | DSISR_NOEXEC_OR_G |
1516 		      DSISR_BAD_COPYPASTE)))
1517 		goto inval;
1518 	if (ret)
1519 		return ret;
1520 
1521 	/* Failed to set the reference/change bits */
1522 	if (dsisr & DSISR_SET_RC) {
1523 		ret = kvmhv_handle_nested_set_rc(vcpu, gp, n_gpa, gpte, dsisr);
1524 		if (ret == RESUME_HOST)
1525 			return ret;
1526 		if (ret)
1527 			goto inval;
1528 		dsisr &= ~DSISR_SET_RC;
1529 		if (!(dsisr & (DSISR_BAD_FAULT_64S | DSISR_NOHPTE |
1530 			       DSISR_PROTFAULT)))
1531 			return RESUME_GUEST;
1532 	}
1533 
1534 	/*
1535 	 * We took an HISI or HDSI while we were running a nested guest which
1536 	 * means we have no partition scoped translation for that. This means
1537 	 * we need to insert a pte for the mapping into our shadow_pgtable.
1538 	 */
1539 
1540 	l1_shift = gpte.page_shift;
1541 	if (l1_shift < PAGE_SHIFT) {
1542 		/* We don't support l1 using a page size smaller than our own */
1543 		pr_err("KVM: L1 guest page shift (%d) less than our own (%d)\n",
1544 			l1_shift, PAGE_SHIFT);
1545 		return -EINVAL;
1546 	}
1547 	gpa = gpte.raddr;
1548 	gfn = gpa >> PAGE_SHIFT;
1549 
1550 	/* 1. Get the corresponding host memslot */
1551 
1552 	memslot = gfn_to_memslot(kvm, gfn);
1553 	if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID)) {
1554 		if (dsisr & (DSISR_PRTABLE_FAULT | DSISR_BADACCESS)) {
1555 			/* unusual error -> reflect to the guest as a DSI */
1556 			kvmppc_core_queue_data_storage(vcpu, ea, dsisr);
1557 			return RESUME_GUEST;
1558 		}
1559 
1560 		/* passthrough of emulated MMIO case */
1561 		return kvmppc_hv_emulate_mmio(vcpu, gpa, ea, writing);
1562 	}
1563 	if (memslot->flags & KVM_MEM_READONLY) {
1564 		if (writing) {
1565 			/* Give the guest a DSI */
1566 			kvmppc_core_queue_data_storage(vcpu, ea,
1567 					DSISR_ISSTORE | DSISR_PROTFAULT);
1568 			return RESUME_GUEST;
1569 		}
1570 		kvm_ro = true;
1571 	}
1572 
1573 	/* 2. Find the host pte for this L1 guest real address */
1574 
1575 	/* Used to check for invalidations in progress */
1576 	mmu_seq = kvm->mmu_notifier_seq;
1577 	smp_rmb();
1578 
1579 	/* See if can find translation in our partition scoped tables for L1 */
1580 	pte = __pte(0);
1581 	spin_lock(&kvm->mmu_lock);
1582 	pte_p = find_kvm_secondary_pte(kvm, gpa, &shift);
1583 	if (!shift)
1584 		shift = PAGE_SHIFT;
1585 	if (pte_p)
1586 		pte = *pte_p;
1587 	spin_unlock(&kvm->mmu_lock);
1588 
1589 	if (!pte_present(pte) || (writing && !(pte_val(pte) & _PAGE_WRITE))) {
1590 		/* No suitable pte found -> try to insert a mapping */
1591 		ret = kvmppc_book3s_instantiate_page(vcpu, gpa, memslot,
1592 					writing, kvm_ro, &pte, &level);
1593 		if (ret == -EAGAIN)
1594 			return RESUME_GUEST;
1595 		else if (ret)
1596 			return ret;
1597 		shift = kvmppc_radix_level_to_shift(level);
1598 	}
1599 	/* Align gfn to the start of the page */
1600 	gfn = (gpa & ~((1UL << shift) - 1)) >> PAGE_SHIFT;
1601 
1602 	/* 3. Compute the pte we need to insert for nest_gpa -> host r_addr */
1603 
1604 	/* The permissions is the combination of the host and l1 guest ptes */
1605 	perm |= gpte.may_read ? 0UL : _PAGE_READ;
1606 	perm |= gpte.may_write ? 0UL : _PAGE_WRITE;
1607 	perm |= gpte.may_execute ? 0UL : _PAGE_EXEC;
1608 	/* Only set accessed/dirty (rc) bits if set in host and l1 guest ptes */
1609 	perm |= (gpte.rc & _PAGE_ACCESSED) ? 0UL : _PAGE_ACCESSED;
1610 	perm |= ((gpte.rc & _PAGE_DIRTY) && writing) ? 0UL : _PAGE_DIRTY;
1611 	pte = __pte(pte_val(pte) & ~perm);
1612 
1613 	/* What size pte can we insert? */
1614 	if (shift > l1_shift) {
1615 		u64 mask;
1616 		unsigned int actual_shift = PAGE_SHIFT;
1617 		if (PMD_SHIFT < l1_shift)
1618 			actual_shift = PMD_SHIFT;
1619 		mask = (1UL << shift) - (1UL << actual_shift);
1620 		pte = __pte(pte_val(pte) | (gpa & mask));
1621 		shift = actual_shift;
1622 	}
1623 	level = kvmppc_radix_shift_to_level(shift);
1624 	n_gpa &= ~((1UL << shift) - 1);
1625 
1626 	/* 4. Insert the pte into our shadow_pgtable */
1627 
1628 	n_rmap = kzalloc(sizeof(*n_rmap), GFP_KERNEL);
1629 	if (!n_rmap)
1630 		return RESUME_GUEST; /* Let the guest try again */
1631 	n_rmap->rmap = (n_gpa & RMAP_NESTED_GPA_MASK) |
1632 		(((unsigned long) gp->l1_lpid) << RMAP_NESTED_LPID_SHIFT);
1633 	rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn];
1634 	ret = kvmppc_create_pte(kvm, gp->shadow_pgtable, pte, n_gpa, level,
1635 				mmu_seq, gp->shadow_lpid, rmapp, &n_rmap);
1636 	kfree(n_rmap);
1637 	if (ret == -EAGAIN)
1638 		ret = RESUME_GUEST;	/* Let the guest try again */
1639 
1640 	return ret;
1641 
1642  inval:
1643 	kvmhv_invalidate_shadow_pte(vcpu, gp, n_gpa, NULL);
1644 	return RESUME_GUEST;
1645 }
1646 
1647 long int kvmhv_nested_page_fault(struct kvm_vcpu *vcpu)
1648 {
1649 	struct kvm_nested_guest *gp = vcpu->arch.nested;
1650 	long int ret;
1651 
1652 	mutex_lock(&gp->tlb_lock);
1653 	ret = __kvmhv_nested_page_fault(vcpu, gp);
1654 	mutex_unlock(&gp->tlb_lock);
1655 	return ret;
1656 }
1657 
1658 int kvmhv_nested_next_lpid(struct kvm *kvm, int lpid)
1659 {
1660 	int ret = -1;
1661 
1662 	spin_lock(&kvm->mmu_lock);
1663 	while (++lpid <= kvm->arch.max_nested_lpid) {
1664 		if (kvm->arch.nested_guests[lpid]) {
1665 			ret = lpid;
1666 			break;
1667 		}
1668 	}
1669 	spin_unlock(&kvm->mmu_lock);
1670 	return ret;
1671 }
1672