1 // SPDX-License-Identifier: GPL-2.0-only
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>
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>
25 #include <asm/dbell.h>
26 #include <asm/cputhreads.h>
31 #define KVM_CMA_CHUNK_ORDER 18
33 #include "book3s_xics.h"
34 #include "book3s_xive.h"
37 * Hash page table alignment on newer cpus(CPU_FTR_ARCH_206)
38 * should be power of 2.
40 #define HPT_ALIGN_PAGES ((1 << 18) >> PAGE_SHIFT) /* 256k */
42 * By default we reserve 5% of memory for hash pagetable allocation.
44 static unsigned long kvm_cma_resv_ratio = 5;
46 static struct cma *kvm_cma;
48 static int __init early_parse_kvm_cma_resv(char *p)
50 pr_debug("%s(%s)\n", __func__, p);
53 return kstrtoul(p, 0, &kvm_cma_resv_ratio);
55 early_param("kvm_cma_resv_ratio", early_parse_kvm_cma_resv);
57 struct page *kvm_alloc_hpt_cma(unsigned long nr_pages)
59 VM_BUG_ON(order_base_2(nr_pages) < KVM_CMA_CHUNK_ORDER - PAGE_SHIFT);
61 return cma_alloc(kvm_cma, nr_pages, order_base_2(HPT_ALIGN_PAGES),
64 EXPORT_SYMBOL_GPL(kvm_alloc_hpt_cma);
66 void kvm_free_hpt_cma(struct page *page, unsigned long nr_pages)
68 cma_release(kvm_cma, page, nr_pages);
70 EXPORT_SYMBOL_GPL(kvm_free_hpt_cma);
73 * kvm_cma_reserve() - reserve area for kvm hash pagetable
75 * This function reserves memory from early allocator. It should be
76 * called by arch specific code once the memblock allocator
77 * has been activated and all other subsystems have already allocated/reserved
80 void __init kvm_cma_reserve(void)
82 unsigned long align_size;
83 phys_addr_t selected_size;
86 * We need CMA reservation only when we are in HV mode
88 if (!cpu_has_feature(CPU_FTR_HVMODE))
91 selected_size = PAGE_ALIGN(memblock_phys_mem_size() * kvm_cma_resv_ratio / 100);
93 pr_info("%s: reserving %ld MiB for global area\n", __func__,
94 (unsigned long)selected_size / SZ_1M);
95 align_size = HPT_ALIGN_PAGES << PAGE_SHIFT;
96 cma_declare_contiguous(0, selected_size, 0, align_size,
97 KVM_CMA_CHUNK_ORDER - PAGE_SHIFT, false, "kvm_cma",
103 * Real-mode H_CONFER implementation.
104 * We check if we are the only vcpu out of this virtual core
105 * still running in the guest and not ceded. If so, we pop up
106 * to the virtual-mode implementation; if not, just return to
109 long int kvmppc_rm_h_confer(struct kvm_vcpu *vcpu, int target,
110 unsigned int yield_count)
112 struct kvmppc_vcore *vc = local_paca->kvm_hstate.kvm_vcore;
113 int ptid = local_paca->kvm_hstate.ptid;
116 int threads_conferring;
117 u64 stop = get_tb() + 10 * tb_ticks_per_usec;
118 int rv = H_SUCCESS; /* => don't yield */
120 set_bit(ptid, &vc->conferring_threads);
121 while ((get_tb() < stop) && !VCORE_IS_EXITING(vc)) {
122 threads_running = VCORE_ENTRY_MAP(vc);
123 threads_ceded = vc->napping_threads;
124 threads_conferring = vc->conferring_threads;
125 if ((threads_ceded | threads_conferring) == threads_running) {
126 rv = H_TOO_HARD; /* => do yield */
130 clear_bit(ptid, &vc->conferring_threads);
135 * When running HV mode KVM we need to block certain operations while KVM VMs
136 * exist in the system. We use a counter of VMs to track this.
138 * One of the operations we need to block is onlining of secondaries, so we
139 * protect hv_vm_count with cpus_read_lock/unlock().
141 static atomic_t hv_vm_count;
143 void kvm_hv_vm_activated(void)
146 atomic_inc(&hv_vm_count);
149 EXPORT_SYMBOL_GPL(kvm_hv_vm_activated);
151 void kvm_hv_vm_deactivated(void)
154 atomic_dec(&hv_vm_count);
157 EXPORT_SYMBOL_GPL(kvm_hv_vm_deactivated);
159 bool kvm_hv_mode_active(void)
161 return atomic_read(&hv_vm_count) != 0;
164 extern int hcall_real_table[], hcall_real_table_end[];
166 int kvmppc_hcall_impl_hv_realmode(unsigned long cmd)
169 if (cmd < hcall_real_table_end - hcall_real_table &&
170 hcall_real_table[cmd])
175 EXPORT_SYMBOL_GPL(kvmppc_hcall_impl_hv_realmode);
177 int kvmppc_hwrng_present(void)
179 return ppc_md.get_random_seed != NULL;
181 EXPORT_SYMBOL_GPL(kvmppc_hwrng_present);
183 long kvmppc_rm_h_random(struct kvm_vcpu *vcpu)
185 if (ppc_md.get_random_seed &&
186 ppc_md.get_random_seed(&vcpu->arch.regs.gpr[4]))
193 * Send an interrupt or message to another CPU.
194 * The caller needs to include any barrier needed to order writes
195 * to memory vs. the IPI/message.
197 void kvmhv_rm_send_ipi(int cpu)
199 void __iomem *xics_phys;
200 unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
202 /* On POWER9 we can use msgsnd for any destination cpu. */
203 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
204 msg |= get_hard_smp_processor_id(cpu);
205 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
209 /* On POWER8 for IPIs to threads in the same core, use msgsnd. */
210 if (cpu_has_feature(CPU_FTR_ARCH_207S) &&
211 cpu_first_thread_sibling(cpu) ==
212 cpu_first_thread_sibling(raw_smp_processor_id())) {
213 msg |= cpu_thread_in_core(cpu);
214 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
218 /* We should never reach this */
219 if (WARN_ON_ONCE(xics_on_xive()))
222 /* Else poke the target with an IPI */
223 xics_phys = paca_ptrs[cpu]->kvm_hstate.xics_phys;
225 __raw_rm_writeb(IPI_PRIORITY, xics_phys + XICS_MFRR);
227 opal_int_set_mfrr(get_hard_smp_processor_id(cpu), IPI_PRIORITY);
231 * The following functions are called from the assembly code
232 * in book3s_hv_rmhandlers.S.
234 static void kvmhv_interrupt_vcore(struct kvmppc_vcore *vc, int active)
238 /* Order setting of exit map vs. msgsnd/IPI */
240 for (; active; active >>= 1, ++cpu)
242 kvmhv_rm_send_ipi(cpu);
245 void kvmhv_commence_exit(int trap)
247 struct kvmppc_vcore *vc = local_paca->kvm_hstate.kvm_vcore;
248 int ptid = local_paca->kvm_hstate.ptid;
249 struct kvm_split_mode *sip = local_paca->kvm_hstate.kvm_split_mode;
252 /* Set our bit in the threads-exiting-guest map in the 0xff00
253 bits of vcore->entry_exit_map */
256 ee = vc->entry_exit_map;
257 } while (cmpxchg(&vc->entry_exit_map, ee, ee | me) != ee);
259 /* Are we the first here? */
264 * Trigger the other threads in this vcore to exit the guest.
265 * If this is a hypervisor decrementer interrupt then they
266 * will be already on their way out of the guest.
268 if (trap != BOOK3S_INTERRUPT_HV_DECREMENTER)
269 kvmhv_interrupt_vcore(vc, ee & ~(1 << ptid));
272 * If we are doing dynamic micro-threading, interrupt the other
273 * subcores to pull them out of their guests too.
278 for (i = 0; i < MAX_SUBCORES; ++i) {
283 ee = vc->entry_exit_map;
284 /* Already asked to exit? */
287 } while (cmpxchg(&vc->entry_exit_map, ee,
288 ee | VCORE_EXIT_REQ) != ee);
290 kvmhv_interrupt_vcore(vc, ee);
294 struct kvmppc_host_rm_ops *kvmppc_host_rm_ops_hv;
295 EXPORT_SYMBOL_GPL(kvmppc_host_rm_ops_hv);
297 #ifdef CONFIG_KVM_XICS
298 static struct kvmppc_irq_map *get_irqmap(struct kvmppc_passthru_irqmap *pimap,
304 * We access the mapped array here without a lock. That
305 * is safe because we never reduce the number of entries
306 * in the array and we never change the v_hwirq field of
307 * an entry once it is set.
309 * We have also carefully ordered the stores in the writer
310 * and the loads here in the reader, so that if we find a matching
311 * hwirq here, the associated GSI and irq_desc fields are valid.
313 for (i = 0; i < pimap->n_mapped; i++) {
314 if (xisr == pimap->mapped[i].r_hwirq) {
316 * Order subsequent reads in the caller to serialize
320 return &pimap->mapped[i];
327 * If we have an interrupt that's not an IPI, check if we have a
328 * passthrough adapter and if so, check if this external interrupt
329 * is for the adapter.
330 * We will attempt to deliver the IRQ directly to the target VCPU's
331 * ICP, the virtual ICP (based on affinity - the xive value in ICS).
333 * If the delivery fails or if this is not for a passthrough adapter,
334 * return to the host to handle this interrupt. We earlier
335 * saved a copy of the XIRR in the PACA, it will be picked up by
336 * the host ICP driver.
338 static int kvmppc_check_passthru(u32 xisr, __be32 xirr, bool *again)
340 struct kvmppc_passthru_irqmap *pimap;
341 struct kvmppc_irq_map *irq_map;
342 struct kvm_vcpu *vcpu;
344 vcpu = local_paca->kvm_hstate.kvm_vcpu;
347 pimap = kvmppc_get_passthru_irqmap(vcpu->kvm);
350 irq_map = get_irqmap(pimap, xisr);
354 /* We're handling this interrupt, generic code doesn't need to */
355 local_paca->kvm_hstate.saved_xirr = 0;
357 return kvmppc_deliver_irq_passthru(vcpu, xirr, irq_map, pimap, again);
361 static inline int kvmppc_check_passthru(u32 xisr, __be32 xirr, bool *again)
368 * Determine what sort of external interrupt is pending (if any).
370 * 0 if no interrupt is pending
371 * 1 if an interrupt is pending that needs to be handled by the host
372 * 2 Passthrough that needs completion in the host
373 * -1 if there was a guest wakeup IPI (which has now been cleared)
374 * -2 if there is PCI passthrough external interrupt that was handled
376 static long kvmppc_read_one_intr(bool *again);
378 long kvmppc_read_intr(void)
389 rc = kvmppc_read_one_intr(&again);
390 if (rc && (ret == 0 || rc > ret))
396 static long kvmppc_read_one_intr(bool *again)
398 void __iomem *xics_phys;
408 /* see if a host IPI is pending */
409 host_ipi = local_paca->kvm_hstate.host_ipi;
413 /* Now read the interrupt from the ICP */
414 xics_phys = local_paca->kvm_hstate.xics_phys;
417 rc = opal_int_get_xirr(&xirr, false);
419 xirr = __raw_rm_readl(xics_phys + XICS_XIRR);
424 * Save XIRR for later. Since we get control in reverse endian
425 * on LE systems, save it byte reversed and fetch it back in
426 * host endian. Note that xirr is the value read from the
427 * XIRR register, while h_xirr is the host endian version.
429 h_xirr = be32_to_cpu(xirr);
430 local_paca->kvm_hstate.saved_xirr = h_xirr;
431 xisr = h_xirr & 0xffffff;
433 * Ensure that the store/load complete to guarantee all side
434 * effects of loading from XIRR has completed
438 /* if nothing pending in the ICP */
442 /* We found something in the ICP...
444 * If it is an IPI, clear the MFRR and EOI it.
446 if (xisr == XICS_IPI) {
449 __raw_rm_writeb(0xff, xics_phys + XICS_MFRR);
450 __raw_rm_writel(xirr, xics_phys + XICS_XIRR);
452 opal_int_set_mfrr(hard_smp_processor_id(), 0xff);
453 rc = opal_int_eoi(h_xirr);
455 /* If rc > 0, there is another interrupt pending */
459 * Need to ensure side effects of above stores
460 * complete before proceeding.
465 * We need to re-check host IPI now in case it got set in the
466 * meantime. If it's clear, we bounce the interrupt to the
469 host_ipi = local_paca->kvm_hstate.host_ipi;
470 if (unlikely(host_ipi != 0)) {
471 /* We raced with the host,
472 * we need to resend that IPI, bummer
475 __raw_rm_writeb(IPI_PRIORITY,
476 xics_phys + XICS_MFRR);
478 opal_int_set_mfrr(hard_smp_processor_id(),
480 /* Let side effects complete */
485 /* OK, it's an IPI for us */
486 local_paca->kvm_hstate.saved_xirr = 0;
490 return kvmppc_check_passthru(xisr, xirr, again);
493 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
495 vcpu->arch.ceded = 0;
496 if (vcpu->arch.timer_running) {
497 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
498 vcpu->arch.timer_running = 0;
502 void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
504 /* Guest must always run with ME enabled, HV disabled. */
505 msr = (msr | MSR_ME) & ~MSR_HV;
508 * Check for illegal transactional state bit combination
509 * and if we find it, force the TS field to a safe state.
511 if ((msr & MSR_TS_MASK) == MSR_TS_MASK)
513 vcpu->arch.shregs.msr = msr;
514 kvmppc_end_cede(vcpu);
516 EXPORT_SYMBOL_GPL(kvmppc_set_msr_hv);
518 static void inject_interrupt(struct kvm_vcpu *vcpu, int vec, u64 srr1_flags)
520 unsigned long msr, pc, new_msr, new_pc;
522 msr = kvmppc_get_msr(vcpu);
523 pc = kvmppc_get_pc(vcpu);
524 new_msr = vcpu->arch.intr_msr;
527 /* If transactional, change to suspend mode on IRQ delivery */
528 if (MSR_TM_TRANSACTIONAL(msr))
531 new_msr |= msr & MSR_TS_MASK;
534 * Perform MSR and PC adjustment for LPCR[AIL]=3 if it is set and
535 * applicable. AIL=2 is not supported.
537 * AIL does not apply to SRESET, MCE, or HMI (which is never
538 * delivered to the guest), and does not apply if IR=0 or DR=0.
540 if (vec != BOOK3S_INTERRUPT_SYSTEM_RESET &&
541 vec != BOOK3S_INTERRUPT_MACHINE_CHECK &&
542 (vcpu->arch.vcore->lpcr & LPCR_AIL) == LPCR_AIL_3 &&
543 (msr & (MSR_IR|MSR_DR)) == (MSR_IR|MSR_DR) ) {
544 new_msr |= MSR_IR | MSR_DR;
545 new_pc += 0xC000000000004000ULL;
548 kvmppc_set_srr0(vcpu, pc);
549 kvmppc_set_srr1(vcpu, (msr & SRR1_MSR_BITS) | srr1_flags);
550 kvmppc_set_pc(vcpu, new_pc);
551 vcpu->arch.shregs.msr = new_msr;
554 void kvmppc_inject_interrupt_hv(struct kvm_vcpu *vcpu, int vec, u64 srr1_flags)
556 inject_interrupt(vcpu, vec, srr1_flags);
557 kvmppc_end_cede(vcpu);
559 EXPORT_SYMBOL_GPL(kvmppc_inject_interrupt_hv);
562 * Is there a PRIV_DOORBELL pending for the guest (on POWER9)?
563 * Can we inject a Decrementer or a External interrupt?
565 void kvmppc_guest_entry_inject_int(struct kvm_vcpu *vcpu)
570 WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300));
572 /* Insert EXTERNAL bit into LPCR at the MER bit position */
573 ext = (vcpu->arch.pending_exceptions >> BOOK3S_IRQPRIO_EXTERNAL) & 1;
574 lpcr = mfspr(SPRN_LPCR);
575 lpcr |= ext << LPCR_MER_SH;
576 mtspr(SPRN_LPCR, lpcr);
579 if (vcpu->arch.shregs.msr & MSR_EE) {
581 inject_interrupt(vcpu, BOOK3S_INTERRUPT_EXTERNAL, 0);
583 long int dec = mfspr(SPRN_DEC);
584 if (!(lpcr & LPCR_LD))
587 inject_interrupt(vcpu,
588 BOOK3S_INTERRUPT_DECREMENTER, 0);
592 if (vcpu->arch.doorbell_request) {
593 mtspr(SPRN_DPDES, 1);
594 vcpu->arch.vcore->dpdes = 1;
596 vcpu->arch.doorbell_request = 0;
600 static void flush_guest_tlb(struct kvm *kvm)
602 unsigned long rb, set;
604 rb = PPC_BIT(52); /* IS = 2 */
605 for (set = 0; set < kvm->arch.tlb_sets; ++set) {
606 /* R=0 PRS=0 RIC=0 */
607 asm volatile(PPC_TLBIEL(%0, %4, %3, %2, %1)
608 : : "r" (rb), "i" (0), "i" (0), "i" (0),
610 rb += PPC_BIT(51); /* increment set number */
612 asm volatile("ptesync": : :"memory");
615 void kvmppc_check_need_tlb_flush(struct kvm *kvm, int pcpu)
617 if (cpumask_test_cpu(pcpu, &kvm->arch.need_tlb_flush)) {
618 flush_guest_tlb(kvm);
620 /* Clear the bit after the TLB flush */
621 cpumask_clear_cpu(pcpu, &kvm->arch.need_tlb_flush);
624 EXPORT_SYMBOL_GPL(kvmppc_check_need_tlb_flush);