1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
4 */
5
6#include <linux/cpu.h>
7#include <linux/kvm_host.h>
8#include <linux/preempt.h>
9#include <linux/export.h>
10#include <linux/sched.h>
11#include <linux/spinlock.h>
12#include <linux/init.h>
13#include <linux/memblock.h>
14#include <linux/sizes.h>
15#include <linux/cma.h>
16#include <linux/bitops.h>
17
18#include <asm/cputable.h>
19#include <asm/interrupt.h>
20#include <asm/kvm_ppc.h>
21#include <asm/kvm_book3s.h>
22#include <asm/machdep.h>
23#include <asm/xics.h>
24#include <asm/xive.h>
25#include <asm/dbell.h>
26#include <asm/cputhreads.h>
27#include <asm/io.h>
28#include <asm/opal.h>
29#include <asm/smp.h>
30
31#define KVM_CMA_CHUNK_ORDER 18
32
33#include "book3s_xics.h"
34#include "book3s_xive.h"
35#include "book3s_hv.h"
36
37/*
38 * Hash page table alignment on newer cpus(CPU_FTR_ARCH_206)
39 * should be power of 2.
40 */
41#define HPT_ALIGN_PAGES ((1 << 18) >> PAGE_SHIFT) /* 256k */
42/*
43 * By default we reserve 5% of memory for hash pagetable allocation.
44 */
45static unsigned long kvm_cma_resv_ratio = 5;
46
47static struct cma *kvm_cma;
48
49static int __init early_parse_kvm_cma_resv(char *p)
50{
51 pr_debug("%s(%s)\n", __func__, p);
52 if (!p)
53 return -EINVAL;
54 return kstrtoul(s: p, base: 0, res: &kvm_cma_resv_ratio);
55}
56early_param("kvm_cma_resv_ratio", early_parse_kvm_cma_resv);
57
58struct page *kvm_alloc_hpt_cma(unsigned long nr_pages)
59{
60 VM_BUG_ON(order_base_2(nr_pages) < KVM_CMA_CHUNK_ORDER - PAGE_SHIFT);
61
62 return cma_alloc(cma: kvm_cma, count: nr_pages, order_base_2(HPT_ALIGN_PAGES),
63 no_warn: false);
64}
65EXPORT_SYMBOL_GPL(kvm_alloc_hpt_cma);
66
67void kvm_free_hpt_cma(struct page *page, unsigned long nr_pages)
68{
69 cma_release(cma: kvm_cma, pages: page, count: nr_pages);
70}
71EXPORT_SYMBOL_GPL(kvm_free_hpt_cma);
72
73/**
74 * kvm_cma_reserve() - reserve area for kvm hash pagetable
75 *
76 * This function reserves memory from early allocator. It should be
77 * called by arch specific code once the memblock allocator
78 * has been activated and all other subsystems have already allocated/reserved
79 * memory.
80 */
81void __init kvm_cma_reserve(void)
82{
83 unsigned long align_size;
84 phys_addr_t selected_size;
85
86 /*
87 * We need CMA reservation only when we are in HV mode
88 */
89 if (!cpu_has_feature(CPU_FTR_HVMODE))
90 return;
91
92 selected_size = PAGE_ALIGN(memblock_phys_mem_size() * kvm_cma_resv_ratio / 100);
93 if (selected_size) {
94 pr_info("%s: reserving %ld MiB for global area\n", __func__,
95 (unsigned long)selected_size / SZ_1M);
96 align_size = HPT_ALIGN_PAGES << PAGE_SHIFT;
97 cma_declare_contiguous(base: 0, size: selected_size, limit: 0, alignment: align_size,
98 KVM_CMA_CHUNK_ORDER - PAGE_SHIFT, fixed: false, name: "kvm_cma",
99 res_cma: &kvm_cma);
100 }
101}
102
103/*
104 * Real-mode H_CONFER implementation.
105 * We check if we are the only vcpu out of this virtual core
106 * still running in the guest and not ceded. If so, we pop up
107 * to the virtual-mode implementation; if not, just return to
108 * the guest.
109 */
110long int kvmppc_rm_h_confer(struct kvm_vcpu *vcpu, int target,
111 unsigned int yield_count)
112{
113 struct kvmppc_vcore *vc = local_paca->kvm_hstate.kvm_vcore;
114 int ptid = local_paca->kvm_hstate.ptid;
115 int threads_running;
116 int threads_ceded;
117 int threads_conferring;
118 u64 stop = get_tb() + 10 * tb_ticks_per_usec;
119 int rv = H_SUCCESS; /* => don't yield */
120
121 set_bit(nr: ptid, addr: &vc->conferring_threads);
122 while ((get_tb() < stop) && !VCORE_IS_EXITING(vc)) {
123 threads_running = VCORE_ENTRY_MAP(vc);
124 threads_ceded = vc->napping_threads;
125 threads_conferring = vc->conferring_threads;
126 if ((threads_ceded | threads_conferring) == threads_running) {
127 rv = H_TOO_HARD; /* => do yield */
128 break;
129 }
130 }
131 clear_bit(nr: ptid, addr: &vc->conferring_threads);
132 return rv;
133}
134
135/*
136 * When running HV mode KVM we need to block certain operations while KVM VMs
137 * exist in the system. We use a counter of VMs to track this.
138 *
139 * One of the operations we need to block is onlining of secondaries, so we
140 * protect hv_vm_count with cpus_read_lock/unlock().
141 */
142static atomic_t hv_vm_count;
143
144void kvm_hv_vm_activated(void)
145{
146 cpus_read_lock();
147 atomic_inc(v: &hv_vm_count);
148 cpus_read_unlock();
149}
150EXPORT_SYMBOL_GPL(kvm_hv_vm_activated);
151
152void kvm_hv_vm_deactivated(void)
153{
154 cpus_read_lock();
155 atomic_dec(v: &hv_vm_count);
156 cpus_read_unlock();
157}
158EXPORT_SYMBOL_GPL(kvm_hv_vm_deactivated);
159
160bool kvm_hv_mode_active(void)
161{
162 return atomic_read(v: &hv_vm_count) != 0;
163}
164
165extern int hcall_real_table[], hcall_real_table_end[];
166
167int kvmppc_hcall_impl_hv_realmode(unsigned long cmd)
168{
169 cmd /= 4;
170 if (cmd < hcall_real_table_end - hcall_real_table &&
171 hcall_real_table[cmd])
172 return 1;
173
174 return 0;
175}
176EXPORT_SYMBOL_GPL(kvmppc_hcall_impl_hv_realmode);
177
178int kvmppc_hwrng_present(void)
179{
180 return ppc_md.get_random_seed != NULL;
181}
182EXPORT_SYMBOL_GPL(kvmppc_hwrng_present);
183
184long kvmppc_rm_h_random(struct kvm_vcpu *vcpu)
185{
186 unsigned long rand;
187
188 if (ppc_md.get_random_seed &&
189 ppc_md.get_random_seed(&rand)) {
190 kvmppc_set_gpr(vcpu, 4, rand);
191 return H_SUCCESS;
192 }
193
194 return H_HARDWARE;
195}
196
197/*
198 * Send an interrupt or message to another CPU.
199 * The caller needs to include any barrier needed to order writes
200 * to memory vs. the IPI/message.
201 */
202void kvmhv_rm_send_ipi(int cpu)
203{
204 void __iomem *xics_phys;
205 unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
206
207 /* On POWER9 we can use msgsnd for any destination cpu. */
208 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
209 msg |= get_hard_smp_processor_id(cpu);
210 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
211 return;
212 }
213
214 /* On POWER8 for IPIs to threads in the same core, use msgsnd. */
215 if (cpu_has_feature(CPU_FTR_ARCH_207S) &&
216 cpu_first_thread_sibling(cpu) ==
217 cpu_first_thread_sibling(raw_smp_processor_id())) {
218 msg |= cpu_thread_in_core(cpu);
219 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
220 return;
221 }
222
223 /* We should never reach this */
224 if (WARN_ON_ONCE(xics_on_xive()))
225 return;
226
227 /* Else poke the target with an IPI */
228 xics_phys = paca_ptrs[cpu]->kvm_hstate.xics_phys;
229 if (xics_phys)
230 __raw_rm_writeb(IPI_PRIORITY, xics_phys + XICS_MFRR);
231 else
232 opal_int_set_mfrr(get_hard_smp_processor_id(cpu), IPI_PRIORITY);
233}
234
235/*
236 * The following functions are called from the assembly code
237 * in book3s_hv_rmhandlers.S.
238 */
239static void kvmhv_interrupt_vcore(struct kvmppc_vcore *vc, int active)
240{
241 int cpu = vc->pcpu;
242
243 /* Order setting of exit map vs. msgsnd/IPI */
244 smp_mb();
245 for (; active; active >>= 1, ++cpu)
246 if (active & 1)
247 kvmhv_rm_send_ipi(cpu);
248}
249
250void kvmhv_commence_exit(int trap)
251{
252 struct kvmppc_vcore *vc = local_paca->kvm_hstate.kvm_vcore;
253 int ptid = local_paca->kvm_hstate.ptid;
254 struct kvm_split_mode *sip = local_paca->kvm_hstate.kvm_split_mode;
255 int me, ee, i;
256
257 /* Set our bit in the threads-exiting-guest map in the 0xff00
258 bits of vcore->entry_exit_map */
259 me = 0x100 << ptid;
260 do {
261 ee = vc->entry_exit_map;
262 } while (cmpxchg(&vc->entry_exit_map, ee, ee | me) != ee);
263
264 /* Are we the first here? */
265 if ((ee >> 8) != 0)
266 return;
267
268 /*
269 * Trigger the other threads in this vcore to exit the guest.
270 * If this is a hypervisor decrementer interrupt then they
271 * will be already on their way out of the guest.
272 */
273 if (trap != BOOK3S_INTERRUPT_HV_DECREMENTER)
274 kvmhv_interrupt_vcore(vc, active: ee & ~(1 << ptid));
275
276 /*
277 * If we are doing dynamic micro-threading, interrupt the other
278 * subcores to pull them out of their guests too.
279 */
280 if (!sip)
281 return;
282
283 for (i = 0; i < MAX_SUBCORES; ++i) {
284 vc = sip->vc[i];
285 if (!vc)
286 break;
287 do {
288 ee = vc->entry_exit_map;
289 /* Already asked to exit? */
290 if ((ee >> 8) != 0)
291 break;
292 } while (cmpxchg(&vc->entry_exit_map, ee,
293 ee | VCORE_EXIT_REQ) != ee);
294 if ((ee >> 8) == 0)
295 kvmhv_interrupt_vcore(vc, active: ee);
296 }
297}
298
299struct kvmppc_host_rm_ops *kvmppc_host_rm_ops_hv;
300EXPORT_SYMBOL_GPL(kvmppc_host_rm_ops_hv);
301
302#ifdef CONFIG_KVM_XICS
303static struct kvmppc_irq_map *get_irqmap(struct kvmppc_passthru_irqmap *pimap,
304 u32 xisr)
305{
306 int i;
307
308 /*
309 * We access the mapped array here without a lock. That
310 * is safe because we never reduce the number of entries
311 * in the array and we never change the v_hwirq field of
312 * an entry once it is set.
313 *
314 * We have also carefully ordered the stores in the writer
315 * and the loads here in the reader, so that if we find a matching
316 * hwirq here, the associated GSI and irq_desc fields are valid.
317 */
318 for (i = 0; i < pimap->n_mapped; i++) {
319 if (xisr == pimap->mapped[i].r_hwirq) {
320 /*
321 * Order subsequent reads in the caller to serialize
322 * with the writer.
323 */
324 smp_rmb();
325 return &pimap->mapped[i];
326 }
327 }
328 return NULL;
329}
330
331/*
332 * If we have an interrupt that's not an IPI, check if we have a
333 * passthrough adapter and if so, check if this external interrupt
334 * is for the adapter.
335 * We will attempt to deliver the IRQ directly to the target VCPU's
336 * ICP, the virtual ICP (based on affinity - the xive value in ICS).
337 *
338 * If the delivery fails or if this is not for a passthrough adapter,
339 * return to the host to handle this interrupt. We earlier
340 * saved a copy of the XIRR in the PACA, it will be picked up by
341 * the host ICP driver.
342 */
343static int kvmppc_check_passthru(u32 xisr, __be32 xirr, bool *again)
344{
345 struct kvmppc_passthru_irqmap *pimap;
346 struct kvmppc_irq_map *irq_map;
347 struct kvm_vcpu *vcpu;
348
349 vcpu = local_paca->kvm_hstate.kvm_vcpu;
350 if (!vcpu)
351 return 1;
352 pimap = kvmppc_get_passthru_irqmap(vcpu->kvm);
353 if (!pimap)
354 return 1;
355 irq_map = get_irqmap(pimap, xisr);
356 if (!irq_map)
357 return 1;
358
359 /* We're handling this interrupt, generic code doesn't need to */
360 local_paca->kvm_hstate.saved_xirr = 0;
361
362 return kvmppc_deliver_irq_passthru(vcpu, xirr, irq_map, pimap, again);
363}
364
365#else
366static inline int kvmppc_check_passthru(u32 xisr, __be32 xirr, bool *again)
367{
368 return 1;
369}
370#endif
371
372/*
373 * Determine what sort of external interrupt is pending (if any).
374 * Returns:
375 * 0 if no interrupt is pending
376 * 1 if an interrupt is pending that needs to be handled by the host
377 * 2 Passthrough that needs completion in the host
378 * -1 if there was a guest wakeup IPI (which has now been cleared)
379 * -2 if there is PCI passthrough external interrupt that was handled
380 */
381static long kvmppc_read_one_intr(bool *again);
382
383long kvmppc_read_intr(void)
384{
385 long ret = 0;
386 long rc;
387 bool again;
388
389 if (xive_enabled())
390 return 1;
391
392 do {
393 again = false;
394 rc = kvmppc_read_one_intr(again: &again);
395 if (rc && (ret == 0 || rc > ret))
396 ret = rc;
397 } while (again);
398 return ret;
399}
400
401static long kvmppc_read_one_intr(bool *again)
402{
403 void __iomem *xics_phys;
404 u32 h_xirr;
405 __be32 xirr;
406 u32 xisr;
407 u8 host_ipi;
408 int64_t rc;
409
410 if (xive_enabled())
411 return 1;
412
413 /* see if a host IPI is pending */
414 host_ipi = READ_ONCE(local_paca->kvm_hstate.host_ipi);
415 if (host_ipi)
416 return 1;
417
418 /* Now read the interrupt from the ICP */
419 xics_phys = local_paca->kvm_hstate.xics_phys;
420 rc = 0;
421 if (!xics_phys)
422 rc = opal_int_get_xirr(&xirr, false);
423 else
424 xirr = __raw_rm_readl(xics_phys + XICS_XIRR);
425 if (rc < 0)
426 return 1;
427
428 /*
429 * Save XIRR for later. Since we get control in reverse endian
430 * on LE systems, save it byte reversed and fetch it back in
431 * host endian. Note that xirr is the value read from the
432 * XIRR register, while h_xirr is the host endian version.
433 */
434 h_xirr = be32_to_cpu(xirr);
435 local_paca->kvm_hstate.saved_xirr = h_xirr;
436 xisr = h_xirr & 0xffffff;
437 /*
438 * Ensure that the store/load complete to guarantee all side
439 * effects of loading from XIRR has completed
440 */
441 smp_mb();
442
443 /* if nothing pending in the ICP */
444 if (!xisr)
445 return 0;
446
447 /* We found something in the ICP...
448 *
449 * If it is an IPI, clear the MFRR and EOI it.
450 */
451 if (xisr == XICS_IPI) {
452 rc = 0;
453 if (xics_phys) {
454 __raw_rm_writeb(0xff, xics_phys + XICS_MFRR);
455 __raw_rm_writel(xirr, xics_phys + XICS_XIRR);
456 } else {
457 opal_int_set_mfrr(hard_smp_processor_id(), 0xff);
458 rc = opal_int_eoi(h_xirr);
459 }
460 /* If rc > 0, there is another interrupt pending */
461 *again = rc > 0;
462
463 /*
464 * Need to ensure side effects of above stores
465 * complete before proceeding.
466 */
467 smp_mb();
468
469 /*
470 * We need to re-check host IPI now in case it got set in the
471 * meantime. If it's clear, we bounce the interrupt to the
472 * guest
473 */
474 host_ipi = READ_ONCE(local_paca->kvm_hstate.host_ipi);
475 if (unlikely(host_ipi != 0)) {
476 /* We raced with the host,
477 * we need to resend that IPI, bummer
478 */
479 if (xics_phys)
480 __raw_rm_writeb(IPI_PRIORITY,
481 xics_phys + XICS_MFRR);
482 else
483 opal_int_set_mfrr(hard_smp_processor_id(),
484 IPI_PRIORITY);
485 /* Let side effects complete */
486 smp_mb();
487 return 1;
488 }
489
490 /* OK, it's an IPI for us */
491 local_paca->kvm_hstate.saved_xirr = 0;
492 return -1;
493 }
494
495 return kvmppc_check_passthru(xisr, xirr, again);
496}
497
498static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
499{
500 vcpu->arch.ceded = 0;
501 if (vcpu->arch.timer_running) {
502 hrtimer_try_to_cancel(timer: &vcpu->arch.dec_timer);
503 vcpu->arch.timer_running = 0;
504 }
505}
506
507void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
508{
509 /* Guest must always run with ME enabled, HV disabled. */
510 msr = (msr | MSR_ME) & ~MSR_HV;
511
512 /*
513 * Check for illegal transactional state bit combination
514 * and if we find it, force the TS field to a safe state.
515 */
516 if ((msr & MSR_TS_MASK) == MSR_TS_MASK)
517 msr &= ~MSR_TS_MASK;
518 __kvmppc_set_msr_hv(vcpu, val: msr);
519 kvmppc_end_cede(vcpu);
520}
521EXPORT_SYMBOL_GPL(kvmppc_set_msr_hv);
522
523static void inject_interrupt(struct kvm_vcpu *vcpu, int vec, u64 srr1_flags)
524{
525 unsigned long msr, pc, new_msr, new_pc;
526
527 msr = kvmppc_get_msr(vcpu);
528 pc = kvmppc_get_pc(vcpu);
529 new_msr = vcpu->arch.intr_msr;
530 new_pc = vec;
531
532 /* If transactional, change to suspend mode on IRQ delivery */
533 if (MSR_TM_TRANSACTIONAL(msr))
534 new_msr |= MSR_TS_S;
535 else
536 new_msr |= msr & MSR_TS_MASK;
537
538 /*
539 * Perform MSR and PC adjustment for LPCR[AIL]=3 if it is set and
540 * applicable. AIL=2 is not supported.
541 *
542 * AIL does not apply to SRESET, MCE, or HMI (which is never
543 * delivered to the guest), and does not apply if IR=0 or DR=0.
544 */
545 if (vec != BOOK3S_INTERRUPT_SYSTEM_RESET &&
546 vec != BOOK3S_INTERRUPT_MACHINE_CHECK &&
547 (vcpu->arch.vcore->lpcr & LPCR_AIL) == LPCR_AIL_3 &&
548 (msr & (MSR_IR|MSR_DR)) == (MSR_IR|MSR_DR) ) {
549 new_msr |= MSR_IR | MSR_DR;
550 new_pc += 0xC000000000004000ULL;
551 }
552
553 kvmppc_set_srr0(vcpu, pc);
554 kvmppc_set_srr1(vcpu, (msr & SRR1_MSR_BITS) | srr1_flags);
555 kvmppc_set_pc(vcpu, new_pc);
556 __kvmppc_set_msr_hv(vcpu, val: new_msr);
557}
558
559void kvmppc_inject_interrupt_hv(struct kvm_vcpu *vcpu, int vec, u64 srr1_flags)
560{
561 inject_interrupt(vcpu, vec, srr1_flags);
562 kvmppc_end_cede(vcpu);
563}
564EXPORT_SYMBOL_GPL(kvmppc_inject_interrupt_hv);
565
566/*
567 * Is there a PRIV_DOORBELL pending for the guest (on POWER9)?
568 * Can we inject a Decrementer or a External interrupt?
569 */
570void kvmppc_guest_entry_inject_int(struct kvm_vcpu *vcpu)
571{
572 int ext;
573 unsigned long lpcr;
574
575 WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
576
577 /* Insert EXTERNAL bit into LPCR at the MER bit position */
578 ext = (vcpu->arch.pending_exceptions >> BOOK3S_IRQPRIO_EXTERNAL) & 1;
579 lpcr = mfspr(SPRN_LPCR);
580 lpcr |= ext << LPCR_MER_SH;
581 mtspr(SPRN_LPCR, lpcr);
582 isync();
583
584 if (vcpu->arch.shregs.msr & MSR_EE) {
585 if (ext) {
586 inject_interrupt(vcpu, BOOK3S_INTERRUPT_EXTERNAL, 0);
587 } else {
588 long int dec = mfspr(SPRN_DEC);
589 if (!(lpcr & LPCR_LD))
590 dec = (int) dec;
591 if (dec < 0)
592 inject_interrupt(vcpu,
593 BOOK3S_INTERRUPT_DECREMENTER, 0);
594 }
595 }
596
597 if (vcpu->arch.doorbell_request) {
598 mtspr(SPRN_DPDES, 1);
599 vcpu->arch.vcore->dpdes = 1;
600 smp_wmb();
601 vcpu->arch.doorbell_request = 0;
602 }
603}
604
605static void flush_guest_tlb(struct kvm *kvm)
606{
607 unsigned long rb, set;
608
609 rb = PPC_BIT(52); /* IS = 2 */
610 for (set = 0; set < kvm->arch.tlb_sets; ++set) {
611 /* R=0 PRS=0 RIC=0 */
612 asm volatile(PPC_TLBIEL(%0, %4, %3, %2, %1)
613 : : "r" (rb), "i" (0), "i" (0), "i" (0),
614 "r" (0) : "memory");
615 rb += PPC_BIT(51); /* increment set number */
616 }
617 asm volatile("ptesync": : :"memory");
618}
619
620void kvmppc_check_need_tlb_flush(struct kvm *kvm, int pcpu)
621{
622 if (cpumask_test_cpu(cpu: pcpu, cpumask: &kvm->arch.need_tlb_flush)) {
623 flush_guest_tlb(kvm);
624
625 /* Clear the bit after the TLB flush */
626 cpumask_clear_cpu(cpu: pcpu, dstp: &kvm->arch.need_tlb_flush);
627 }
628}
629EXPORT_SYMBOL_GPL(kvmppc_check_need_tlb_flush);
630

source code of linux/arch/powerpc/kvm/book3s_hv_builtin.c