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
)
210 if (powernv_get_random_real_mode(&vcpu
->arch
.gpr
[4]))
217 * Send an interrupt or message to another CPU.
218 * The caller needs to include any barrier needed to order writes
219 * to memory vs. the IPI/message.
221 void kvmhv_rm_send_ipi(int cpu
)
223 void __iomem
*xics_phys
;
224 unsigned long msg
= PPC_DBELL_TYPE(PPC_DBELL_SERVER
);
226 /* On POWER9 we can use msgsnd for any destination cpu. */
227 if (cpu_has_feature(CPU_FTR_ARCH_300
)) {
228 msg
|= get_hard_smp_processor_id(cpu
);
229 __asm__
__volatile__ (PPC_MSGSND(%0) : : "r" (msg
));
233 /* On POWER8 for IPIs to threads in the same core, use msgsnd. */
234 if (cpu_has_feature(CPU_FTR_ARCH_207S
) &&
235 cpu_first_thread_sibling(cpu
) ==
236 cpu_first_thread_sibling(raw_smp_processor_id())) {
237 msg
|= cpu_thread_in_core(cpu
);
238 __asm__
__volatile__ (PPC_MSGSND(%0) : : "r" (msg
));
242 /* We should never reach this */
243 if (WARN_ON_ONCE(xive_enabled()))
246 /* Else poke the target with an IPI */
247 xics_phys
= paca
[cpu
].kvm_hstate
.xics_phys
;
249 __raw_rm_writeb(IPI_PRIORITY
, xics_phys
+ XICS_MFRR
);
251 opal_int_set_mfrr(get_hard_smp_processor_id(cpu
), IPI_PRIORITY
);
255 * The following functions are called from the assembly code
256 * in book3s_hv_rmhandlers.S.
258 static void kvmhv_interrupt_vcore(struct kvmppc_vcore
*vc
, int active
)
262 /* Order setting of exit map vs. msgsnd/IPI */
264 for (; active
; active
>>= 1, ++cpu
)
266 kvmhv_rm_send_ipi(cpu
);
269 void kvmhv_commence_exit(int trap
)
271 struct kvmppc_vcore
*vc
= local_paca
->kvm_hstate
.kvm_vcore
;
272 int ptid
= local_paca
->kvm_hstate
.ptid
;
273 struct kvm_split_mode
*sip
= local_paca
->kvm_hstate
.kvm_split_mode
;
276 /* Set our bit in the threads-exiting-guest map in the 0xff00
277 bits of vcore->entry_exit_map */
280 ee
= vc
->entry_exit_map
;
281 } while (cmpxchg(&vc
->entry_exit_map
, ee
, ee
| me
) != ee
);
283 /* Are we the first here? */
288 * Trigger the other threads in this vcore to exit the guest.
289 * If this is a hypervisor decrementer interrupt then they
290 * will be already on their way out of the guest.
292 if (trap
!= BOOK3S_INTERRUPT_HV_DECREMENTER
)
293 kvmhv_interrupt_vcore(vc
, ee
& ~(1 << ptid
));
296 * If we are doing dynamic micro-threading, interrupt the other
297 * subcores to pull them out of their guests too.
302 for (i
= 0; i
< MAX_SUBCORES
; ++i
) {
303 vc
= sip
->master_vcs
[i
];
307 ee
= vc
->entry_exit_map
;
308 /* Already asked to exit? */
311 } while (cmpxchg(&vc
->entry_exit_map
, ee
,
312 ee
| VCORE_EXIT_REQ
) != ee
);
314 kvmhv_interrupt_vcore(vc
, ee
);
318 struct kvmppc_host_rm_ops
*kvmppc_host_rm_ops_hv
;
319 EXPORT_SYMBOL_GPL(kvmppc_host_rm_ops_hv
);
321 #ifdef CONFIG_KVM_XICS
322 static struct kvmppc_irq_map
*get_irqmap(struct kvmppc_passthru_irqmap
*pimap
,
328 * We access the mapped array here without a lock. That
329 * is safe because we never reduce the number of entries
330 * in the array and we never change the v_hwirq field of
331 * an entry once it is set.
333 * We have also carefully ordered the stores in the writer
334 * and the loads here in the reader, so that if we find a matching
335 * hwirq here, the associated GSI and irq_desc fields are valid.
337 for (i
= 0; i
< pimap
->n_mapped
; i
++) {
338 if (xisr
== pimap
->mapped
[i
].r_hwirq
) {
340 * Order subsequent reads in the caller to serialize
344 return &pimap
->mapped
[i
];
351 * If we have an interrupt that's not an IPI, check if we have a
352 * passthrough adapter and if so, check if this external interrupt
353 * is for the adapter.
354 * We will attempt to deliver the IRQ directly to the target VCPU's
355 * ICP, the virtual ICP (based on affinity - the xive value in ICS).
357 * If the delivery fails or if this is not for a passthrough adapter,
358 * return to the host to handle this interrupt. We earlier
359 * saved a copy of the XIRR in the PACA, it will be picked up by
360 * the host ICP driver.
362 static int kvmppc_check_passthru(u32 xisr
, __be32 xirr
, bool *again
)
364 struct kvmppc_passthru_irqmap
*pimap
;
365 struct kvmppc_irq_map
*irq_map
;
366 struct kvm_vcpu
*vcpu
;
368 vcpu
= local_paca
->kvm_hstate
.kvm_vcpu
;
371 pimap
= kvmppc_get_passthru_irqmap(vcpu
->kvm
);
374 irq_map
= get_irqmap(pimap
, xisr
);
378 /* We're handling this interrupt, generic code doesn't need to */
379 local_paca
->kvm_hstate
.saved_xirr
= 0;
381 return kvmppc_deliver_irq_passthru(vcpu
, xirr
, irq_map
, pimap
, again
);
385 static inline int kvmppc_check_passthru(u32 xisr
, __be32 xirr
, bool *again
)
392 * Determine what sort of external interrupt is pending (if any).
394 * 0 if no interrupt is pending
395 * 1 if an interrupt is pending that needs to be handled by the host
396 * 2 Passthrough that needs completion in the host
397 * -1 if there was a guest wakeup IPI (which has now been cleared)
398 * -2 if there is PCI passthrough external interrupt that was handled
400 static long kvmppc_read_one_intr(bool *again
);
402 long kvmppc_read_intr(void)
413 rc
= kvmppc_read_one_intr(&again
);
414 if (rc
&& (ret
== 0 || rc
> ret
))
420 static long kvmppc_read_one_intr(bool *again
)
422 void __iomem
*xics_phys
;
432 /* see if a host IPI is pending */
433 host_ipi
= local_paca
->kvm_hstate
.host_ipi
;
437 /* Now read the interrupt from the ICP */
438 xics_phys
= local_paca
->kvm_hstate
.xics_phys
;
441 rc
= opal_int_get_xirr(&xirr
, false);
443 xirr
= __raw_rm_readl(xics_phys
+ XICS_XIRR
);
448 * Save XIRR for later. Since we get control in reverse endian
449 * on LE systems, save it byte reversed and fetch it back in
450 * host endian. Note that xirr is the value read from the
451 * XIRR register, while h_xirr is the host endian version.
453 h_xirr
= be32_to_cpu(xirr
);
454 local_paca
->kvm_hstate
.saved_xirr
= h_xirr
;
455 xisr
= h_xirr
& 0xffffff;
457 * Ensure that the store/load complete to guarantee all side
458 * effects of loading from XIRR has completed
462 /* if nothing pending in the ICP */
466 /* We found something in the ICP...
468 * If it is an IPI, clear the MFRR and EOI it.
470 if (xisr
== XICS_IPI
) {
473 __raw_rm_writeb(0xff, xics_phys
+ XICS_MFRR
);
474 __raw_rm_writel(xirr
, xics_phys
+ XICS_XIRR
);
476 opal_int_set_mfrr(hard_smp_processor_id(), 0xff);
477 rc
= opal_int_eoi(h_xirr
);
479 /* If rc > 0, there is another interrupt pending */
483 * Need to ensure side effects of above stores
484 * complete before proceeding.
489 * We need to re-check host IPI now in case it got set in the
490 * meantime. If it's clear, we bounce the interrupt to the
493 host_ipi
= local_paca
->kvm_hstate
.host_ipi
;
494 if (unlikely(host_ipi
!= 0)) {
495 /* We raced with the host,
496 * we need to resend that IPI, bummer
499 __raw_rm_writeb(IPI_PRIORITY
,
500 xics_phys
+ XICS_MFRR
);
502 opal_int_set_mfrr(hard_smp_processor_id(),
504 /* Let side effects complete */
509 /* OK, it's an IPI for us */
510 local_paca
->kvm_hstate
.saved_xirr
= 0;
514 return kvmppc_check_passthru(xisr
, xirr
, again
);
517 #ifdef CONFIG_KVM_XICS
518 static inline bool is_rm(void)
520 return !(mfmsr() & MSR_DR
);
523 unsigned long kvmppc_rm_h_xirr(struct kvm_vcpu
*vcpu
)
525 if (xive_enabled()) {
527 return xive_rm_h_xirr(vcpu
);
528 if (unlikely(!__xive_vm_h_xirr
))
529 return H_NOT_AVAILABLE
;
530 return __xive_vm_h_xirr(vcpu
);
532 return xics_rm_h_xirr(vcpu
);
535 unsigned long kvmppc_rm_h_xirr_x(struct kvm_vcpu
*vcpu
)
537 vcpu
->arch
.gpr
[5] = get_tb();
538 if (xive_enabled()) {
540 return xive_rm_h_xirr(vcpu
);
541 if (unlikely(!__xive_vm_h_xirr
))
542 return H_NOT_AVAILABLE
;
543 return __xive_vm_h_xirr(vcpu
);
545 return xics_rm_h_xirr(vcpu
);
548 unsigned long kvmppc_rm_h_ipoll(struct kvm_vcpu
*vcpu
, unsigned long server
)
550 if (xive_enabled()) {
552 return xive_rm_h_ipoll(vcpu
, server
);
553 if (unlikely(!__xive_vm_h_ipoll
))
554 return H_NOT_AVAILABLE
;
555 return __xive_vm_h_ipoll(vcpu
, server
);
560 int kvmppc_rm_h_ipi(struct kvm_vcpu
*vcpu
, unsigned long server
,
563 if (xive_enabled()) {
565 return xive_rm_h_ipi(vcpu
, server
, mfrr
);
566 if (unlikely(!__xive_vm_h_ipi
))
567 return H_NOT_AVAILABLE
;
568 return __xive_vm_h_ipi(vcpu
, server
, mfrr
);
570 return xics_rm_h_ipi(vcpu
, server
, mfrr
);
573 int kvmppc_rm_h_cppr(struct kvm_vcpu
*vcpu
, unsigned long cppr
)
575 if (xive_enabled()) {
577 return xive_rm_h_cppr(vcpu
, cppr
);
578 if (unlikely(!__xive_vm_h_cppr
))
579 return H_NOT_AVAILABLE
;
580 return __xive_vm_h_cppr(vcpu
, cppr
);
582 return xics_rm_h_cppr(vcpu
, cppr
);
585 int kvmppc_rm_h_eoi(struct kvm_vcpu
*vcpu
, unsigned long xirr
)
587 if (xive_enabled()) {
589 return xive_rm_h_eoi(vcpu
, xirr
);
590 if (unlikely(!__xive_vm_h_eoi
))
591 return H_NOT_AVAILABLE
;
592 return __xive_vm_h_eoi(vcpu
, xirr
);
594 return xics_rm_h_eoi(vcpu
, xirr
);
596 #endif /* CONFIG_KVM_XICS */