2 * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License, version 2, as
6 * published by the Free Software Foundation.
10 #include <linux/kvm_host.h>
11 #include <linux/preempt.h>
12 #include <linux/export.h>
13 #include <linux/sched.h>
14 #include <linux/spinlock.h>
15 #include <linux/init.h>
16 #include <linux/memblock.h>
17 #include <linux/sizes.h>
18 #include <linux/cma.h>
19 #include <linux/bitops.h>
21 #include <asm/cputable.h>
22 #include <asm/kvm_ppc.h>
23 #include <asm/kvm_book3s.h>
24 #include <asm/archrandom.h>
27 #include <asm/dbell.h>
28 #include <asm/cputhreads.h>
33 #define KVM_CMA_CHUNK_ORDER 18
35 #include "book3s_xics.h"
36 #include "book3s_xive.h"
39 * The XIVE module will populate these when it loads
41 unsigned long (*__xive_vm_h_xirr
)(struct kvm_vcpu
*vcpu
);
42 unsigned long (*__xive_vm_h_ipoll
)(struct kvm_vcpu
*vcpu
, unsigned long server
);
43 int (*__xive_vm_h_ipi
)(struct kvm_vcpu
*vcpu
, unsigned long server
,
45 int (*__xive_vm_h_cppr
)(struct kvm_vcpu
*vcpu
, unsigned long cppr
);
46 int (*__xive_vm_h_eoi
)(struct kvm_vcpu
*vcpu
, unsigned long xirr
);
47 EXPORT_SYMBOL_GPL(__xive_vm_h_xirr
);
48 EXPORT_SYMBOL_GPL(__xive_vm_h_ipoll
);
49 EXPORT_SYMBOL_GPL(__xive_vm_h_ipi
);
50 EXPORT_SYMBOL_GPL(__xive_vm_h_cppr
);
51 EXPORT_SYMBOL_GPL(__xive_vm_h_eoi
);
54 * Hash page table alignment on newer cpus(CPU_FTR_ARCH_206)
55 * should be power of 2.
57 #define HPT_ALIGN_PAGES ((1 << 18) >> PAGE_SHIFT) /* 256k */
59 * By default we reserve 5% of memory for hash pagetable allocation.
61 static unsigned long kvm_cma_resv_ratio
= 5;
63 static struct cma
*kvm_cma
;
65 static int __init
early_parse_kvm_cma_resv(char *p
)
67 pr_debug("%s(%s)\n", __func__
, p
);
70 return kstrtoul(p
, 0, &kvm_cma_resv_ratio
);
72 early_param("kvm_cma_resv_ratio", early_parse_kvm_cma_resv
);
74 struct page
*kvm_alloc_hpt_cma(unsigned long nr_pages
)
76 VM_BUG_ON(order_base_2(nr_pages
) < KVM_CMA_CHUNK_ORDER
- PAGE_SHIFT
);
78 return cma_alloc(kvm_cma
, nr_pages
, order_base_2(HPT_ALIGN_PAGES
),
81 EXPORT_SYMBOL_GPL(kvm_alloc_hpt_cma
);
83 void kvm_free_hpt_cma(struct page
*page
, unsigned long nr_pages
)
85 cma_release(kvm_cma
, page
, nr_pages
);
87 EXPORT_SYMBOL_GPL(kvm_free_hpt_cma
);
90 * kvm_cma_reserve() - reserve area for kvm hash pagetable
92 * This function reserves memory from early allocator. It should be
93 * called by arch specific code once the memblock allocator
94 * has been activated and all other subsystems have already allocated/reserved
97 void __init
kvm_cma_reserve(void)
99 unsigned long align_size
;
100 struct memblock_region
*reg
;
101 phys_addr_t selected_size
= 0;
104 * We need CMA reservation only when we are in HV mode
106 if (!cpu_has_feature(CPU_FTR_HVMODE
))
109 * We cannot use memblock_phys_mem_size() here, because
110 * memblock_analyze() has not been called yet.
112 for_each_memblock(memory
, reg
)
113 selected_size
+= memblock_region_memory_end_pfn(reg
) -
114 memblock_region_memory_base_pfn(reg
);
116 selected_size
= (selected_size
* kvm_cma_resv_ratio
/ 100) << PAGE_SHIFT
;
118 pr_debug("%s: reserving %ld MiB for global area\n", __func__
,
119 (unsigned long)selected_size
/ SZ_1M
);
120 align_size
= HPT_ALIGN_PAGES
<< PAGE_SHIFT
;
121 cma_declare_contiguous(0, selected_size
, 0, align_size
,
122 KVM_CMA_CHUNK_ORDER
- PAGE_SHIFT
, false, "kvm_cma",
128 * Real-mode H_CONFER implementation.
129 * We check if we are the only vcpu out of this virtual core
130 * still running in the guest and not ceded. If so, we pop up
131 * to the virtual-mode implementation; if not, just return to
134 long int kvmppc_rm_h_confer(struct kvm_vcpu
*vcpu
, int target
,
135 unsigned int yield_count
)
137 struct kvmppc_vcore
*vc
= local_paca
->kvm_hstate
.kvm_vcore
;
138 int ptid
= local_paca
->kvm_hstate
.ptid
;
141 int threads_conferring
;
142 u64 stop
= get_tb() + 10 * tb_ticks_per_usec
;
143 int rv
= H_SUCCESS
; /* => don't yield */
145 set_bit(ptid
, &vc
->conferring_threads
);
146 while ((get_tb() < stop
) && !VCORE_IS_EXITING(vc
)) {
147 threads_running
= VCORE_ENTRY_MAP(vc
);
148 threads_ceded
= vc
->napping_threads
;
149 threads_conferring
= vc
->conferring_threads
;
150 if ((threads_ceded
| threads_conferring
) == threads_running
) {
151 rv
= H_TOO_HARD
; /* => do yield */
155 clear_bit(ptid
, &vc
->conferring_threads
);
160 * When running HV mode KVM we need to block certain operations while KVM VMs
161 * exist in the system. We use a counter of VMs to track this.
163 * One of the operations we need to block is onlining of secondaries, so we
164 * protect hv_vm_count with get/put_online_cpus().
166 static atomic_t hv_vm_count
;
168 void kvm_hv_vm_activated(void)
171 atomic_inc(&hv_vm_count
);
174 EXPORT_SYMBOL_GPL(kvm_hv_vm_activated
);
176 void kvm_hv_vm_deactivated(void)
179 atomic_dec(&hv_vm_count
);
182 EXPORT_SYMBOL_GPL(kvm_hv_vm_deactivated
);
184 bool kvm_hv_mode_active(void)
186 return atomic_read(&hv_vm_count
) != 0;
189 extern int hcall_real_table
[], hcall_real_table_end
[];
191 int kvmppc_hcall_impl_hv_realmode(unsigned long cmd
)
194 if (cmd
< hcall_real_table_end
- hcall_real_table
&&
195 hcall_real_table
[cmd
])
200 EXPORT_SYMBOL_GPL(kvmppc_hcall_impl_hv_realmode
);
202 int kvmppc_hwrng_present(void)
204 return powernv_hwrng_present();
206 EXPORT_SYMBOL_GPL(kvmppc_hwrng_present
);
208 long kvmppc_h_random(struct kvm_vcpu
*vcpu
)
212 /* Only need to do the expensive mfmsr() on radix */
213 if (kvm_is_radix(vcpu
->kvm
) && (mfmsr() & MSR_IR
))
214 r
= powernv_get_random_long(&vcpu
->arch
.gpr
[4]);
216 r
= powernv_get_random_real_mode(&vcpu
->arch
.gpr
[4]);
224 * Send an interrupt or message to another CPU.
225 * The caller needs to include any barrier needed to order writes
226 * to memory vs. the IPI/message.
228 void kvmhv_rm_send_ipi(int cpu
)
230 void __iomem
*xics_phys
;
231 unsigned long msg
= PPC_DBELL_TYPE(PPC_DBELL_SERVER
);
233 /* On POWER9 we can use msgsnd for any destination cpu. */
234 if (cpu_has_feature(CPU_FTR_ARCH_300
)) {
235 msg
|= get_hard_smp_processor_id(cpu
);
236 __asm__
__volatile__ (PPC_MSGSND(%0) : : "r" (msg
));
240 /* On POWER8 for IPIs to threads in the same core, use msgsnd. */
241 if (cpu_has_feature(CPU_FTR_ARCH_207S
) &&
242 cpu_first_thread_sibling(cpu
) ==
243 cpu_first_thread_sibling(raw_smp_processor_id())) {
244 msg
|= cpu_thread_in_core(cpu
);
245 __asm__
__volatile__ (PPC_MSGSND(%0) : : "r" (msg
));
249 /* We should never reach this */
250 if (WARN_ON_ONCE(xive_enabled()))
253 /* Else poke the target with an IPI */
254 xics_phys
= paca
[cpu
].kvm_hstate
.xics_phys
;
256 __raw_rm_writeb(IPI_PRIORITY
, xics_phys
+ XICS_MFRR
);
258 opal_int_set_mfrr(get_hard_smp_processor_id(cpu
), IPI_PRIORITY
);
262 * The following functions are called from the assembly code
263 * in book3s_hv_rmhandlers.S.
265 static void kvmhv_interrupt_vcore(struct kvmppc_vcore
*vc
, int active
)
269 /* Order setting of exit map vs. msgsnd/IPI */
271 for (; active
; active
>>= 1, ++cpu
)
273 kvmhv_rm_send_ipi(cpu
);
276 void kvmhv_commence_exit(int trap
)
278 struct kvmppc_vcore
*vc
= local_paca
->kvm_hstate
.kvm_vcore
;
279 int ptid
= local_paca
->kvm_hstate
.ptid
;
280 struct kvm_split_mode
*sip
= local_paca
->kvm_hstate
.kvm_split_mode
;
283 /* Set our bit in the threads-exiting-guest map in the 0xff00
284 bits of vcore->entry_exit_map */
287 ee
= vc
->entry_exit_map
;
288 } while (cmpxchg(&vc
->entry_exit_map
, ee
, ee
| me
) != ee
);
290 /* Are we the first here? */
295 * Trigger the other threads in this vcore to exit the guest.
296 * If this is a hypervisor decrementer interrupt then they
297 * will be already on their way out of the guest.
299 if (trap
!= BOOK3S_INTERRUPT_HV_DECREMENTER
)
300 kvmhv_interrupt_vcore(vc
, ee
& ~(1 << ptid
));
303 * If we are doing dynamic micro-threading, interrupt the other
304 * subcores to pull them out of their guests too.
309 for (i
= 0; i
< MAX_SUBCORES
; ++i
) {
310 vc
= sip
->master_vcs
[i
];
314 ee
= vc
->entry_exit_map
;
315 /* Already asked to exit? */
318 } while (cmpxchg(&vc
->entry_exit_map
, ee
,
319 ee
| VCORE_EXIT_REQ
) != ee
);
321 kvmhv_interrupt_vcore(vc
, ee
);
325 struct kvmppc_host_rm_ops
*kvmppc_host_rm_ops_hv
;
326 EXPORT_SYMBOL_GPL(kvmppc_host_rm_ops_hv
);
328 #ifdef CONFIG_KVM_XICS
329 static struct kvmppc_irq_map
*get_irqmap(struct kvmppc_passthru_irqmap
*pimap
,
335 * We access the mapped array here without a lock. That
336 * is safe because we never reduce the number of entries
337 * in the array and we never change the v_hwirq field of
338 * an entry once it is set.
340 * We have also carefully ordered the stores in the writer
341 * and the loads here in the reader, so that if we find a matching
342 * hwirq here, the associated GSI and irq_desc fields are valid.
344 for (i
= 0; i
< pimap
->n_mapped
; i
++) {
345 if (xisr
== pimap
->mapped
[i
].r_hwirq
) {
347 * Order subsequent reads in the caller to serialize
351 return &pimap
->mapped
[i
];
358 * If we have an interrupt that's not an IPI, check if we have a
359 * passthrough adapter and if so, check if this external interrupt
360 * is for the adapter.
361 * We will attempt to deliver the IRQ directly to the target VCPU's
362 * ICP, the virtual ICP (based on affinity - the xive value in ICS).
364 * If the delivery fails or if this is not for a passthrough adapter,
365 * return to the host to handle this interrupt. We earlier
366 * saved a copy of the XIRR in the PACA, it will be picked up by
367 * the host ICP driver.
369 static int kvmppc_check_passthru(u32 xisr
, __be32 xirr
, bool *again
)
371 struct kvmppc_passthru_irqmap
*pimap
;
372 struct kvmppc_irq_map
*irq_map
;
373 struct kvm_vcpu
*vcpu
;
375 vcpu
= local_paca
->kvm_hstate
.kvm_vcpu
;
378 pimap
= kvmppc_get_passthru_irqmap(vcpu
->kvm
);
381 irq_map
= get_irqmap(pimap
, xisr
);
385 /* We're handling this interrupt, generic code doesn't need to */
386 local_paca
->kvm_hstate
.saved_xirr
= 0;
388 return kvmppc_deliver_irq_passthru(vcpu
, xirr
, irq_map
, pimap
, again
);
392 static inline int kvmppc_check_passthru(u32 xisr
, __be32 xirr
, bool *again
)
399 * Determine what sort of external interrupt is pending (if any).
401 * 0 if no interrupt is pending
402 * 1 if an interrupt is pending that needs to be handled by the host
403 * 2 Passthrough that needs completion in the host
404 * -1 if there was a guest wakeup IPI (which has now been cleared)
405 * -2 if there is PCI passthrough external interrupt that was handled
407 static long kvmppc_read_one_intr(bool *again
);
409 long kvmppc_read_intr(void)
420 rc
= kvmppc_read_one_intr(&again
);
421 if (rc
&& (ret
== 0 || rc
> ret
))
427 static long kvmppc_read_one_intr(bool *again
)
429 void __iomem
*xics_phys
;
439 /* see if a host IPI is pending */
440 host_ipi
= local_paca
->kvm_hstate
.host_ipi
;
444 /* Now read the interrupt from the ICP */
445 xics_phys
= local_paca
->kvm_hstate
.xics_phys
;
448 rc
= opal_int_get_xirr(&xirr
, false);
450 xirr
= __raw_rm_readl(xics_phys
+ XICS_XIRR
);
455 * Save XIRR for later. Since we get control in reverse endian
456 * on LE systems, save it byte reversed and fetch it back in
457 * host endian. Note that xirr is the value read from the
458 * XIRR register, while h_xirr is the host endian version.
460 h_xirr
= be32_to_cpu(xirr
);
461 local_paca
->kvm_hstate
.saved_xirr
= h_xirr
;
462 xisr
= h_xirr
& 0xffffff;
464 * Ensure that the store/load complete to guarantee all side
465 * effects of loading from XIRR has completed
469 /* if nothing pending in the ICP */
473 /* We found something in the ICP...
475 * If it is an IPI, clear the MFRR and EOI it.
477 if (xisr
== XICS_IPI
) {
480 __raw_rm_writeb(0xff, xics_phys
+ XICS_MFRR
);
481 __raw_rm_writel(xirr
, xics_phys
+ XICS_XIRR
);
483 opal_int_set_mfrr(hard_smp_processor_id(), 0xff);
484 rc
= opal_int_eoi(h_xirr
);
486 /* If rc > 0, there is another interrupt pending */
490 * Need to ensure side effects of above stores
491 * complete before proceeding.
496 * We need to re-check host IPI now in case it got set in the
497 * meantime. If it's clear, we bounce the interrupt to the
500 host_ipi
= local_paca
->kvm_hstate
.host_ipi
;
501 if (unlikely(host_ipi
!= 0)) {
502 /* We raced with the host,
503 * we need to resend that IPI, bummer
506 __raw_rm_writeb(IPI_PRIORITY
,
507 xics_phys
+ XICS_MFRR
);
509 opal_int_set_mfrr(hard_smp_processor_id(),
511 /* Let side effects complete */
516 /* OK, it's an IPI for us */
517 local_paca
->kvm_hstate
.saved_xirr
= 0;
521 return kvmppc_check_passthru(xisr
, xirr
, again
);
524 #ifdef CONFIG_KVM_XICS
525 static inline bool is_rm(void)
527 return !(mfmsr() & MSR_DR
);
530 unsigned long kvmppc_rm_h_xirr(struct kvm_vcpu
*vcpu
)
532 if (xive_enabled()) {
534 return xive_rm_h_xirr(vcpu
);
535 if (unlikely(!__xive_vm_h_xirr
))
536 return H_NOT_AVAILABLE
;
537 return __xive_vm_h_xirr(vcpu
);
539 return xics_rm_h_xirr(vcpu
);
542 unsigned long kvmppc_rm_h_xirr_x(struct kvm_vcpu
*vcpu
)
544 vcpu
->arch
.gpr
[5] = get_tb();
545 if (xive_enabled()) {
547 return xive_rm_h_xirr(vcpu
);
548 if (unlikely(!__xive_vm_h_xirr
))
549 return H_NOT_AVAILABLE
;
550 return __xive_vm_h_xirr(vcpu
);
552 return xics_rm_h_xirr(vcpu
);
555 unsigned long kvmppc_rm_h_ipoll(struct kvm_vcpu
*vcpu
, unsigned long server
)
557 if (xive_enabled()) {
559 return xive_rm_h_ipoll(vcpu
, server
);
560 if (unlikely(!__xive_vm_h_ipoll
))
561 return H_NOT_AVAILABLE
;
562 return __xive_vm_h_ipoll(vcpu
, server
);
567 int kvmppc_rm_h_ipi(struct kvm_vcpu
*vcpu
, unsigned long server
,
570 if (xive_enabled()) {
572 return xive_rm_h_ipi(vcpu
, server
, mfrr
);
573 if (unlikely(!__xive_vm_h_ipi
))
574 return H_NOT_AVAILABLE
;
575 return __xive_vm_h_ipi(vcpu
, server
, mfrr
);
577 return xics_rm_h_ipi(vcpu
, server
, mfrr
);
580 int kvmppc_rm_h_cppr(struct kvm_vcpu
*vcpu
, unsigned long cppr
)
582 if (xive_enabled()) {
584 return xive_rm_h_cppr(vcpu
, cppr
);
585 if (unlikely(!__xive_vm_h_cppr
))
586 return H_NOT_AVAILABLE
;
587 return __xive_vm_h_cppr(vcpu
, cppr
);
589 return xics_rm_h_cppr(vcpu
, cppr
);
592 int kvmppc_rm_h_eoi(struct kvm_vcpu
*vcpu
, unsigned long xirr
)
594 if (xive_enabled()) {
596 return xive_rm_h_eoi(vcpu
, xirr
);
597 if (unlikely(!__xive_vm_h_eoi
))
598 return H_NOT_AVAILABLE
;
599 return __xive_vm_h_eoi(vcpu
, xirr
);
601 return xics_rm_h_eoi(vcpu
, xirr
);
603 #endif /* CONFIG_KVM_XICS */