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>
26 #include <asm/dbell.h>
27 #include <asm/cputhreads.h>
30 #define KVM_CMA_CHUNK_ORDER 18
33 * Hash page table alignment on newer cpus(CPU_FTR_ARCH_206)
34 * should be power of 2.
36 #define HPT_ALIGN_PAGES ((1 << 18) >> PAGE_SHIFT) /* 256k */
38 * By default we reserve 5% of memory for hash pagetable allocation.
40 static unsigned long kvm_cma_resv_ratio
= 5;
42 static struct cma
*kvm_cma
;
44 static int __init
early_parse_kvm_cma_resv(char *p
)
46 pr_debug("%s(%s)\n", __func__
, p
);
49 return kstrtoul(p
, 0, &kvm_cma_resv_ratio
);
51 early_param("kvm_cma_resv_ratio", early_parse_kvm_cma_resv
);
53 struct page
*kvm_alloc_hpt(unsigned long nr_pages
)
55 VM_BUG_ON(order_base_2(nr_pages
) < KVM_CMA_CHUNK_ORDER
- PAGE_SHIFT
);
57 return cma_alloc(kvm_cma
, nr_pages
, order_base_2(HPT_ALIGN_PAGES
));
59 EXPORT_SYMBOL_GPL(kvm_alloc_hpt
);
61 void kvm_release_hpt(struct page
*page
, unsigned long nr_pages
)
63 cma_release(kvm_cma
, page
, nr_pages
);
65 EXPORT_SYMBOL_GPL(kvm_release_hpt
);
68 * kvm_cma_reserve() - reserve area for kvm hash pagetable
70 * This function reserves memory from early allocator. It should be
71 * called by arch specific code once the memblock allocator
72 * has been activated and all other subsystems have already allocated/reserved
75 void __init
kvm_cma_reserve(void)
77 unsigned long align_size
;
78 struct memblock_region
*reg
;
79 phys_addr_t selected_size
= 0;
82 * We need CMA reservation only when we are in HV mode
84 if (!cpu_has_feature(CPU_FTR_HVMODE
))
87 * We cannot use memblock_phys_mem_size() here, because
88 * memblock_analyze() has not been called yet.
90 for_each_memblock(memory
, reg
)
91 selected_size
+= memblock_region_memory_end_pfn(reg
) -
92 memblock_region_memory_base_pfn(reg
);
94 selected_size
= (selected_size
* kvm_cma_resv_ratio
/ 100) << PAGE_SHIFT
;
96 pr_debug("%s: reserving %ld MiB for global area\n", __func__
,
97 (unsigned long)selected_size
/ SZ_1M
);
98 align_size
= HPT_ALIGN_PAGES
<< PAGE_SHIFT
;
99 cma_declare_contiguous(0, selected_size
, 0, align_size
,
100 KVM_CMA_CHUNK_ORDER
- PAGE_SHIFT
, false, &kvm_cma
);
105 * Real-mode H_CONFER implementation.
106 * We check if we are the only vcpu out of this virtual core
107 * still running in the guest and not ceded. If so, we pop up
108 * to the virtual-mode implementation; if not, just return to
111 long int kvmppc_rm_h_confer(struct kvm_vcpu
*vcpu
, int target
,
112 unsigned int yield_count
)
114 struct kvmppc_vcore
*vc
= local_paca
->kvm_hstate
.kvm_vcore
;
115 int ptid
= local_paca
->kvm_hstate
.ptid
;
118 int threads_conferring
;
119 u64 stop
= get_tb() + 10 * tb_ticks_per_usec
;
120 int rv
= H_SUCCESS
; /* => don't yield */
122 set_bit(ptid
, &vc
->conferring_threads
);
123 while ((get_tb() < stop
) && !VCORE_IS_EXITING(vc
)) {
124 threads_running
= VCORE_ENTRY_MAP(vc
);
125 threads_ceded
= vc
->napping_threads
;
126 threads_conferring
= vc
->conferring_threads
;
127 if ((threads_ceded
| threads_conferring
) == threads_running
) {
128 rv
= H_TOO_HARD
; /* => do yield */
132 clear_bit(ptid
, &vc
->conferring_threads
);
137 * When running HV mode KVM we need to block certain operations while KVM VMs
138 * exist in the system. We use a counter of VMs to track this.
140 * One of the operations we need to block is onlining of secondaries, so we
141 * protect hv_vm_count with get/put_online_cpus().
143 static atomic_t hv_vm_count
;
145 void kvm_hv_vm_activated(void)
148 atomic_inc(&hv_vm_count
);
151 EXPORT_SYMBOL_GPL(kvm_hv_vm_activated
);
153 void kvm_hv_vm_deactivated(void)
156 atomic_dec(&hv_vm_count
);
159 EXPORT_SYMBOL_GPL(kvm_hv_vm_deactivated
);
161 bool kvm_hv_mode_active(void)
163 return atomic_read(&hv_vm_count
) != 0;
166 extern int hcall_real_table
[], hcall_real_table_end
[];
168 int kvmppc_hcall_impl_hv_realmode(unsigned long cmd
)
171 if (cmd
< hcall_real_table_end
- hcall_real_table
&&
172 hcall_real_table
[cmd
])
177 EXPORT_SYMBOL_GPL(kvmppc_hcall_impl_hv_realmode
);
179 int kvmppc_hwrng_present(void)
181 return powernv_hwrng_present();
183 EXPORT_SYMBOL_GPL(kvmppc_hwrng_present
);
185 long kvmppc_h_random(struct kvm_vcpu
*vcpu
)
187 if (powernv_get_random_real_mode(&vcpu
->arch
.gpr
[4]))
193 static inline void rm_writeb(unsigned long paddr
, u8 val
)
195 __asm__
__volatile__("stbcix %0,0,%1"
196 : : "r" (val
), "r" (paddr
) : "memory");
200 * Send an interrupt or message to another CPU.
201 * This can only be called in real mode.
202 * The caller needs to include any barrier needed to order writes
203 * to memory vs. the IPI/message.
205 void kvmhv_rm_send_ipi(int cpu
)
207 unsigned long xics_phys
;
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 unsigned long msg
= PPC_DBELL_TYPE(PPC_DBELL_SERVER
);
214 msg
|= cpu_thread_in_core(cpu
);
215 __asm__
__volatile__ (PPC_MSGSND(%0) : : "r" (msg
));
219 /* Else poke the target with an IPI */
220 xics_phys
= paca
[cpu
].kvm_hstate
.xics_phys
;
221 rm_writeb(xics_phys
+ XICS_MFRR
, IPI_PRIORITY
);
225 * The following functions are called from the assembly code
226 * in book3s_hv_rmhandlers.S.
228 static void kvmhv_interrupt_vcore(struct kvmppc_vcore
*vc
, int active
)
232 /* Order setting of exit map vs. msgsnd/IPI */
234 for (; active
; active
>>= 1, ++cpu
)
236 kvmhv_rm_send_ipi(cpu
);
239 void kvmhv_commence_exit(int trap
)
241 struct kvmppc_vcore
*vc
= local_paca
->kvm_hstate
.kvm_vcore
;
242 int ptid
= local_paca
->kvm_hstate
.ptid
;
243 struct kvm_split_mode
*sip
= local_paca
->kvm_hstate
.kvm_split_mode
;
246 /* Set our bit in the threads-exiting-guest map in the 0xff00
247 bits of vcore->entry_exit_map */
250 ee
= vc
->entry_exit_map
;
251 } while (cmpxchg(&vc
->entry_exit_map
, ee
, ee
| me
) != ee
);
253 /* Are we the first here? */
258 * Trigger the other threads in this vcore to exit the guest.
259 * If this is a hypervisor decrementer interrupt then they
260 * will be already on their way out of the guest.
262 if (trap
!= BOOK3S_INTERRUPT_HV_DECREMENTER
)
263 kvmhv_interrupt_vcore(vc
, ee
& ~(1 << ptid
));
266 * If we are doing dynamic micro-threading, interrupt the other
267 * subcores to pull them out of their guests too.
272 for (i
= 0; i
< MAX_SUBCORES
; ++i
) {
273 vc
= sip
->master_vcs
[i
];
277 ee
= vc
->entry_exit_map
;
278 /* Already asked to exit? */
281 } while (cmpxchg(&vc
->entry_exit_map
, ee
,
282 ee
| VCORE_EXIT_REQ
) != ee
);
284 kvmhv_interrupt_vcore(vc
, ee
);
288 struct kvmppc_host_rm_ops
*kvmppc_host_rm_ops_hv
;
289 EXPORT_SYMBOL_GPL(kvmppc_host_rm_ops_hv
);
291 #ifdef CONFIG_KVM_XICS
292 static struct kvmppc_irq_map
*get_irqmap(struct kvmppc_passthru_irqmap
*pimap
,
298 * We access the mapped array here without a lock. That
299 * is safe because we never reduce the number of entries
300 * in the array and we never change the v_hwirq field of
301 * an entry once it is set.
303 * We have also carefully ordered the stores in the writer
304 * and the loads here in the reader, so that if we find a matching
305 * hwirq here, the associated GSI and irq_desc fields are valid.
307 for (i
= 0; i
< pimap
->n_mapped
; i
++) {
308 if (xisr
== pimap
->mapped
[i
].r_hwirq
) {
310 * Order subsequent reads in the caller to serialize
314 return &pimap
->mapped
[i
];
321 * If we have an interrupt that's not an IPI, check if we have a
322 * passthrough adapter and if so, check if this external interrupt
323 * is for the adapter.
324 * We will attempt to deliver the IRQ directly to the target VCPU's
325 * ICP, the virtual ICP (based on affinity - the xive value in ICS).
327 * If the delivery fails or if this is not for a passthrough adapter,
328 * return to the host to handle this interrupt. We earlier
329 * saved a copy of the XIRR in the PACA, it will be picked up by
330 * the host ICP driver.
332 static int kvmppc_check_passthru(u32 xisr
, __be32 xirr
)
334 struct kvmppc_passthru_irqmap
*pimap
;
335 struct kvmppc_irq_map
*irq_map
;
336 struct kvm_vcpu
*vcpu
;
338 vcpu
= local_paca
->kvm_hstate
.kvm_vcpu
;
341 pimap
= kvmppc_get_passthru_irqmap(vcpu
->kvm
);
344 irq_map
= get_irqmap(pimap
, xisr
);
348 /* We're handling this interrupt, generic code doesn't need to */
349 local_paca
->kvm_hstate
.saved_xirr
= 0;
351 return kvmppc_deliver_irq_passthru(vcpu
, xirr
, irq_map
, pimap
);
355 static inline int kvmppc_check_passthru(u32 xisr
, __be32 xirr
)
362 * Determine what sort of external interrupt is pending (if any).
364 * 0 if no interrupt is pending
365 * 1 if an interrupt is pending that needs to be handled by the host
366 * 2 Passthrough that needs completion in the host
367 * -1 if there was a guest wakeup IPI (which has now been cleared)
368 * -2 if there is PCI passthrough external interrupt that was handled
371 long kvmppc_read_intr(void)
373 unsigned long xics_phys
;
379 /* see if a host IPI is pending */
380 host_ipi
= local_paca
->kvm_hstate
.host_ipi
;
384 /* Now read the interrupt from the ICP */
385 xics_phys
= local_paca
->kvm_hstate
.xics_phys
;
386 if (unlikely(!xics_phys
))
390 * Save XIRR for later. Since we get control in reverse endian
391 * on LE systems, save it byte reversed and fetch it back in
392 * host endian. Note that xirr is the value read from the
393 * XIRR register, while h_xirr is the host endian version.
395 xirr
= _lwzcix(xics_phys
+ XICS_XIRR
);
396 h_xirr
= be32_to_cpu(xirr
);
397 local_paca
->kvm_hstate
.saved_xirr
= h_xirr
;
398 xisr
= h_xirr
& 0xffffff;
400 * Ensure that the store/load complete to guarantee all side
401 * effects of loading from XIRR has completed
405 /* if nothing pending in the ICP */
409 /* We found something in the ICP...
411 * If it is an IPI, clear the MFRR and EOI it.
413 if (xisr
== XICS_IPI
) {
414 _stbcix(xics_phys
+ XICS_MFRR
, 0xff);
415 _stwcix(xics_phys
+ XICS_XIRR
, xirr
);
417 * Need to ensure side effects of above stores
418 * complete before proceeding.
423 * We need to re-check host IPI now in case it got set in the
424 * meantime. If it's clear, we bounce the interrupt to the
427 host_ipi
= local_paca
->kvm_hstate
.host_ipi
;
428 if (unlikely(host_ipi
!= 0)) {
429 /* We raced with the host,
430 * we need to resend that IPI, bummer
432 _stbcix(xics_phys
+ XICS_MFRR
, IPI_PRIORITY
);
433 /* Let side effects complete */
438 /* OK, it's an IPI for us */
439 local_paca
->kvm_hstate
.saved_xirr
= 0;
443 return kvmppc_check_passthru(xisr
, xirr
);