]>
Commit | Line | Data |
---|---|---|
aa04b4cc PM |
1 | /* |
2 | * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com> | |
3 | * | |
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. | |
7 | */ | |
8 | ||
441c19c8 | 9 | #include <linux/cpu.h> |
aa04b4cc PM |
10 | #include <linux/kvm_host.h> |
11 | #include <linux/preempt.h> | |
66b15db6 | 12 | #include <linux/export.h> |
aa04b4cc PM |
13 | #include <linux/sched.h> |
14 | #include <linux/spinlock.h> | |
aa04b4cc | 15 | #include <linux/init.h> |
fa61a4e3 AK |
16 | #include <linux/memblock.h> |
17 | #include <linux/sizes.h> | |
fc95ca72 | 18 | #include <linux/cma.h> |
90fd09f8 | 19 | #include <linux/bitops.h> |
aa04b4cc PM |
20 | |
21 | #include <asm/cputable.h> | |
22 | #include <asm/kvm_ppc.h> | |
23 | #include <asm/kvm_book3s.h> | |
e928e9cb | 24 | #include <asm/archrandom.h> |
eddb60fb | 25 | #include <asm/xics.h> |
c5742e33 | 26 | #include <asm/xive.h> |
66feed61 PM |
27 | #include <asm/dbell.h> |
28 | #include <asm/cputhreads.h> | |
37f55d30 | 29 | #include <asm/io.h> |
f725758b | 30 | #include <asm/opal.h> |
e2702871 | 31 | #include <asm/smp.h> |
aa04b4cc | 32 | |
fc95ca72 JK |
33 | #define KVM_CMA_CHUNK_ORDER 18 |
34 | ||
fa61a4e3 AK |
35 | /* |
36 | * Hash page table alignment on newer cpus(CPU_FTR_ARCH_206) | |
37 | * should be power of 2. | |
38 | */ | |
39 | #define HPT_ALIGN_PAGES ((1 << 18) >> PAGE_SHIFT) /* 256k */ | |
40 | /* | |
41 | * By default we reserve 5% of memory for hash pagetable allocation. | |
42 | */ | |
43 | static unsigned long kvm_cma_resv_ratio = 5; | |
aa04b4cc | 44 | |
fc95ca72 JK |
45 | static struct cma *kvm_cma; |
46 | ||
fa61a4e3 | 47 | static int __init early_parse_kvm_cma_resv(char *p) |
d2a1b483 | 48 | { |
fa61a4e3 | 49 | pr_debug("%s(%s)\n", __func__, p); |
d2a1b483 | 50 | if (!p) |
fa61a4e3 AK |
51 | return -EINVAL; |
52 | return kstrtoul(p, 0, &kvm_cma_resv_ratio); | |
d2a1b483 | 53 | } |
fa61a4e3 | 54 | early_param("kvm_cma_resv_ratio", early_parse_kvm_cma_resv); |
d2a1b483 | 55 | |
fa61a4e3 | 56 | struct page *kvm_alloc_hpt(unsigned long nr_pages) |
d2a1b483 | 57 | { |
c04fa583 | 58 | VM_BUG_ON(order_base_2(nr_pages) < KVM_CMA_CHUNK_ORDER - PAGE_SHIFT); |
fc95ca72 | 59 | |
c17b98cf | 60 | return cma_alloc(kvm_cma, nr_pages, order_base_2(HPT_ALIGN_PAGES)); |
d2a1b483 AG |
61 | } |
62 | EXPORT_SYMBOL_GPL(kvm_alloc_hpt); | |
63 | ||
fa61a4e3 | 64 | void kvm_release_hpt(struct page *page, unsigned long nr_pages) |
d2a1b483 | 65 | { |
fc95ca72 | 66 | cma_release(kvm_cma, page, nr_pages); |
d2a1b483 AG |
67 | } |
68 | EXPORT_SYMBOL_GPL(kvm_release_hpt); | |
69 | ||
fa61a4e3 AK |
70 | /** |
71 | * kvm_cma_reserve() - reserve area for kvm hash pagetable | |
72 | * | |
73 | * This function reserves memory from early allocator. It should be | |
14ed7409 | 74 | * called by arch specific code once the memblock allocator |
fa61a4e3 AK |
75 | * has been activated and all other subsystems have already allocated/reserved |
76 | * memory. | |
77 | */ | |
78 | void __init kvm_cma_reserve(void) | |
79 | { | |
80 | unsigned long align_size; | |
81 | struct memblock_region *reg; | |
82 | phys_addr_t selected_size = 0; | |
cec26bc3 AK |
83 | |
84 | /* | |
85 | * We need CMA reservation only when we are in HV mode | |
86 | */ | |
87 | if (!cpu_has_feature(CPU_FTR_HVMODE)) | |
88 | return; | |
fa61a4e3 AK |
89 | /* |
90 | * We cannot use memblock_phys_mem_size() here, because | |
91 | * memblock_analyze() has not been called yet. | |
92 | */ | |
93 | for_each_memblock(memory, reg) | |
94 | selected_size += memblock_region_memory_end_pfn(reg) - | |
95 | memblock_region_memory_base_pfn(reg); | |
96 | ||
97 | selected_size = (selected_size * kvm_cma_resv_ratio / 100) << PAGE_SHIFT; | |
98 | if (selected_size) { | |
99 | pr_debug("%s: reserving %ld MiB for global area\n", __func__, | |
100 | (unsigned long)selected_size / SZ_1M); | |
c17b98cf | 101 | align_size = HPT_ALIGN_PAGES << PAGE_SHIFT; |
c1f733aa JK |
102 | cma_declare_contiguous(0, selected_size, 0, align_size, |
103 | KVM_CMA_CHUNK_ORDER - PAGE_SHIFT, false, &kvm_cma); | |
fa61a4e3 AK |
104 | } |
105 | } | |
441c19c8 | 106 | |
90fd09f8 SB |
107 | /* |
108 | * Real-mode H_CONFER implementation. | |
109 | * We check if we are the only vcpu out of this virtual core | |
110 | * still running in the guest and not ceded. If so, we pop up | |
111 | * to the virtual-mode implementation; if not, just return to | |
112 | * the guest. | |
113 | */ | |
114 | long int kvmppc_rm_h_confer(struct kvm_vcpu *vcpu, int target, | |
115 | unsigned int yield_count) | |
116 | { | |
ec257165 PM |
117 | struct kvmppc_vcore *vc = local_paca->kvm_hstate.kvm_vcore; |
118 | int ptid = local_paca->kvm_hstate.ptid; | |
90fd09f8 SB |
119 | int threads_running; |
120 | int threads_ceded; | |
121 | int threads_conferring; | |
122 | u64 stop = get_tb() + 10 * tb_ticks_per_usec; | |
123 | int rv = H_SUCCESS; /* => don't yield */ | |
124 | ||
ec257165 | 125 | set_bit(ptid, &vc->conferring_threads); |
7d6c40da PM |
126 | while ((get_tb() < stop) && !VCORE_IS_EXITING(vc)) { |
127 | threads_running = VCORE_ENTRY_MAP(vc); | |
128 | threads_ceded = vc->napping_threads; | |
129 | threads_conferring = vc->conferring_threads; | |
130 | if ((threads_ceded | threads_conferring) == threads_running) { | |
90fd09f8 SB |
131 | rv = H_TOO_HARD; /* => do yield */ |
132 | break; | |
133 | } | |
134 | } | |
ec257165 | 135 | clear_bit(ptid, &vc->conferring_threads); |
90fd09f8 SB |
136 | return rv; |
137 | } | |
138 | ||
441c19c8 ME |
139 | /* |
140 | * When running HV mode KVM we need to block certain operations while KVM VMs | |
141 | * exist in the system. We use a counter of VMs to track this. | |
142 | * | |
143 | * One of the operations we need to block is onlining of secondaries, so we | |
144 | * protect hv_vm_count with get/put_online_cpus(). | |
145 | */ | |
146 | static atomic_t hv_vm_count; | |
147 | ||
148 | void kvm_hv_vm_activated(void) | |
149 | { | |
150 | get_online_cpus(); | |
151 | atomic_inc(&hv_vm_count); | |
152 | put_online_cpus(); | |
153 | } | |
154 | EXPORT_SYMBOL_GPL(kvm_hv_vm_activated); | |
155 | ||
156 | void kvm_hv_vm_deactivated(void) | |
157 | { | |
158 | get_online_cpus(); | |
159 | atomic_dec(&hv_vm_count); | |
160 | put_online_cpus(); | |
161 | } | |
162 | EXPORT_SYMBOL_GPL(kvm_hv_vm_deactivated); | |
163 | ||
164 | bool kvm_hv_mode_active(void) | |
165 | { | |
166 | return atomic_read(&hv_vm_count) != 0; | |
167 | } | |
ae2113a4 PM |
168 | |
169 | extern int hcall_real_table[], hcall_real_table_end[]; | |
170 | ||
171 | int kvmppc_hcall_impl_hv_realmode(unsigned long cmd) | |
172 | { | |
173 | cmd /= 4; | |
174 | if (cmd < hcall_real_table_end - hcall_real_table && | |
175 | hcall_real_table[cmd]) | |
176 | return 1; | |
177 | ||
178 | return 0; | |
179 | } | |
180 | EXPORT_SYMBOL_GPL(kvmppc_hcall_impl_hv_realmode); | |
e928e9cb ME |
181 | |
182 | int kvmppc_hwrng_present(void) | |
183 | { | |
184 | return powernv_hwrng_present(); | |
185 | } | |
186 | EXPORT_SYMBOL_GPL(kvmppc_hwrng_present); | |
187 | ||
188 | long kvmppc_h_random(struct kvm_vcpu *vcpu) | |
189 | { | |
791d5db2 PM |
190 | int r; |
191 | ||
192 | /* Only need to do the expensive mfmsr() on radix */ | |
193 | if (kvm_is_radix(vcpu->kvm) && (mfmsr() & MSR_IR)) | |
194 | r = powernv_get_random_long(&vcpu->arch.gpr[4]); | |
195 | else | |
196 | r = powernv_get_random_real_mode(&vcpu->arch.gpr[4]); | |
197 | if (r) | |
e928e9cb ME |
198 | return H_SUCCESS; |
199 | ||
200 | return H_HARDWARE; | |
201 | } | |
eddb60fb PM |
202 | |
203 | static inline void rm_writeb(unsigned long paddr, u8 val) | |
204 | { | |
205 | __asm__ __volatile__("stbcix %0,0,%1" | |
206 | : : "r" (val), "r" (paddr) : "memory"); | |
207 | } | |
208 | ||
209 | /* | |
66feed61 | 210 | * Send an interrupt or message to another CPU. |
eddb60fb PM |
211 | * The caller needs to include any barrier needed to order writes |
212 | * to memory vs. the IPI/message. | |
213 | */ | |
214 | void kvmhv_rm_send_ipi(int cpu) | |
215 | { | |
216 | unsigned long xics_phys; | |
1704a81c | 217 | unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER); |
eddb60fb | 218 | |
1704a81c PM |
219 | /* On POWER9 we can use msgsnd for any destination cpu. */ |
220 | if (cpu_has_feature(CPU_FTR_ARCH_300)) { | |
221 | msg |= get_hard_smp_processor_id(cpu); | |
222 | __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg)); | |
223 | return; | |
224 | } | |
225 | /* On POWER8 for IPIs to threads in the same core, use msgsnd. */ | |
66feed61 PM |
226 | if (cpu_has_feature(CPU_FTR_ARCH_207S) && |
227 | cpu_first_thread_sibling(cpu) == | |
228 | cpu_first_thread_sibling(raw_smp_processor_id())) { | |
66feed61 PM |
229 | msg |= cpu_thread_in_core(cpu); |
230 | __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg)); | |
231 | return; | |
232 | } | |
233 | ||
c5742e33 BH |
234 | /* We should never reach this */ |
235 | if (WARN_ON_ONCE(xive_enabled())) | |
236 | return; | |
237 | ||
66feed61 | 238 | /* Else poke the target with an IPI */ |
eddb60fb | 239 | xics_phys = paca[cpu].kvm_hstate.xics_phys; |
f0cb5203 | 240 | if (xics_phys) |
f725758b PM |
241 | rm_writeb(xics_phys + XICS_MFRR, IPI_PRIORITY); |
242 | else | |
f0cb5203 | 243 | opal_int_set_mfrr(get_hard_smp_processor_id(cpu), IPI_PRIORITY); |
eddb60fb PM |
244 | } |
245 | ||
246 | /* | |
247 | * The following functions are called from the assembly code | |
248 | * in book3s_hv_rmhandlers.S. | |
249 | */ | |
250 | static void kvmhv_interrupt_vcore(struct kvmppc_vcore *vc, int active) | |
251 | { | |
252 | int cpu = vc->pcpu; | |
253 | ||
254 | /* Order setting of exit map vs. msgsnd/IPI */ | |
255 | smp_mb(); | |
256 | for (; active; active >>= 1, ++cpu) | |
257 | if (active & 1) | |
258 | kvmhv_rm_send_ipi(cpu); | |
259 | } | |
260 | ||
261 | void kvmhv_commence_exit(int trap) | |
262 | { | |
263 | struct kvmppc_vcore *vc = local_paca->kvm_hstate.kvm_vcore; | |
264 | int ptid = local_paca->kvm_hstate.ptid; | |
b4deba5c PM |
265 | struct kvm_split_mode *sip = local_paca->kvm_hstate.kvm_split_mode; |
266 | int me, ee, i; | |
eddb60fb PM |
267 | |
268 | /* Set our bit in the threads-exiting-guest map in the 0xff00 | |
269 | bits of vcore->entry_exit_map */ | |
270 | me = 0x100 << ptid; | |
271 | do { | |
272 | ee = vc->entry_exit_map; | |
273 | } while (cmpxchg(&vc->entry_exit_map, ee, ee | me) != ee); | |
274 | ||
275 | /* Are we the first here? */ | |
276 | if ((ee >> 8) != 0) | |
277 | return; | |
278 | ||
279 | /* | |
280 | * Trigger the other threads in this vcore to exit the guest. | |
281 | * If this is a hypervisor decrementer interrupt then they | |
282 | * will be already on their way out of the guest. | |
283 | */ | |
284 | if (trap != BOOK3S_INTERRUPT_HV_DECREMENTER) | |
285 | kvmhv_interrupt_vcore(vc, ee & ~(1 << ptid)); | |
b4deba5c PM |
286 | |
287 | /* | |
288 | * If we are doing dynamic micro-threading, interrupt the other | |
289 | * subcores to pull them out of their guests too. | |
290 | */ | |
291 | if (!sip) | |
292 | return; | |
293 | ||
294 | for (i = 0; i < MAX_SUBCORES; ++i) { | |
295 | vc = sip->master_vcs[i]; | |
296 | if (!vc) | |
297 | break; | |
298 | do { | |
299 | ee = vc->entry_exit_map; | |
300 | /* Already asked to exit? */ | |
301 | if ((ee >> 8) != 0) | |
302 | break; | |
303 | } while (cmpxchg(&vc->entry_exit_map, ee, | |
304 | ee | VCORE_EXIT_REQ) != ee); | |
305 | if ((ee >> 8) == 0) | |
306 | kvmhv_interrupt_vcore(vc, ee); | |
307 | } | |
eddb60fb | 308 | } |
79b6c247 SW |
309 | |
310 | struct kvmppc_host_rm_ops *kvmppc_host_rm_ops_hv; | |
311 | EXPORT_SYMBOL_GPL(kvmppc_host_rm_ops_hv); | |
37f55d30 | 312 | |
e3c13e56 SW |
313 | #ifdef CONFIG_KVM_XICS |
314 | static struct kvmppc_irq_map *get_irqmap(struct kvmppc_passthru_irqmap *pimap, | |
315 | u32 xisr) | |
316 | { | |
317 | int i; | |
318 | ||
319 | /* | |
320 | * We access the mapped array here without a lock. That | |
321 | * is safe because we never reduce the number of entries | |
322 | * in the array and we never change the v_hwirq field of | |
323 | * an entry once it is set. | |
324 | * | |
325 | * We have also carefully ordered the stores in the writer | |
326 | * and the loads here in the reader, so that if we find a matching | |
327 | * hwirq here, the associated GSI and irq_desc fields are valid. | |
328 | */ | |
329 | for (i = 0; i < pimap->n_mapped; i++) { | |
330 | if (xisr == pimap->mapped[i].r_hwirq) { | |
331 | /* | |
332 | * Order subsequent reads in the caller to serialize | |
333 | * with the writer. | |
334 | */ | |
335 | smp_rmb(); | |
336 | return &pimap->mapped[i]; | |
337 | } | |
338 | } | |
339 | return NULL; | |
340 | } | |
341 | ||
342 | /* | |
343 | * If we have an interrupt that's not an IPI, check if we have a | |
344 | * passthrough adapter and if so, check if this external interrupt | |
345 | * is for the adapter. | |
346 | * We will attempt to deliver the IRQ directly to the target VCPU's | |
347 | * ICP, the virtual ICP (based on affinity - the xive value in ICS). | |
348 | * | |
349 | * If the delivery fails or if this is not for a passthrough adapter, | |
350 | * return to the host to handle this interrupt. We earlier | |
351 | * saved a copy of the XIRR in the PACA, it will be picked up by | |
352 | * the host ICP driver. | |
353 | */ | |
f725758b | 354 | static int kvmppc_check_passthru(u32 xisr, __be32 xirr, bool *again) |
e3c13e56 SW |
355 | { |
356 | struct kvmppc_passthru_irqmap *pimap; | |
357 | struct kvmppc_irq_map *irq_map; | |
358 | struct kvm_vcpu *vcpu; | |
359 | ||
360 | vcpu = local_paca->kvm_hstate.kvm_vcpu; | |
361 | if (!vcpu) | |
362 | return 1; | |
363 | pimap = kvmppc_get_passthru_irqmap(vcpu->kvm); | |
364 | if (!pimap) | |
365 | return 1; | |
366 | irq_map = get_irqmap(pimap, xisr); | |
367 | if (!irq_map) | |
368 | return 1; | |
369 | ||
370 | /* We're handling this interrupt, generic code doesn't need to */ | |
371 | local_paca->kvm_hstate.saved_xirr = 0; | |
372 | ||
f725758b | 373 | return kvmppc_deliver_irq_passthru(vcpu, xirr, irq_map, pimap, again); |
e3c13e56 SW |
374 | } |
375 | ||
376 | #else | |
e2702871 | 377 | static inline int kvmppc_check_passthru(u32 xisr, __be32 xirr, bool *again) |
e3c13e56 SW |
378 | { |
379 | return 1; | |
380 | } | |
381 | #endif | |
382 | ||
37f55d30 SW |
383 | /* |
384 | * Determine what sort of external interrupt is pending (if any). | |
385 | * Returns: | |
386 | * 0 if no interrupt is pending | |
387 | * 1 if an interrupt is pending that needs to be handled by the host | |
f7af5209 | 388 | * 2 Passthrough that needs completion in the host |
37f55d30 | 389 | * -1 if there was a guest wakeup IPI (which has now been cleared) |
e3c13e56 | 390 | * -2 if there is PCI passthrough external interrupt that was handled |
37f55d30 | 391 | */ |
f725758b | 392 | static long kvmppc_read_one_intr(bool *again); |
37f55d30 SW |
393 | |
394 | long kvmppc_read_intr(void) | |
f725758b PM |
395 | { |
396 | long ret = 0; | |
397 | long rc; | |
398 | bool again; | |
399 | ||
c5742e33 BH |
400 | if (xive_enabled()) |
401 | return 1; | |
402 | ||
f725758b PM |
403 | do { |
404 | again = false; | |
405 | rc = kvmppc_read_one_intr(&again); | |
406 | if (rc && (ret == 0 || rc > ret)) | |
407 | ret = rc; | |
408 | } while (again); | |
409 | return ret; | |
410 | } | |
411 | ||
412 | static long kvmppc_read_one_intr(bool *again) | |
37f55d30 SW |
413 | { |
414 | unsigned long xics_phys; | |
415 | u32 h_xirr; | |
416 | __be32 xirr; | |
417 | u32 xisr; | |
418 | u8 host_ipi; | |
f725758b | 419 | int64_t rc; |
37f55d30 SW |
420 | |
421 | /* see if a host IPI is pending */ | |
422 | host_ipi = local_paca->kvm_hstate.host_ipi; | |
423 | if (host_ipi) | |
424 | return 1; | |
425 | ||
426 | /* Now read the interrupt from the ICP */ | |
427 | xics_phys = local_paca->kvm_hstate.xics_phys; | |
ddf05501 | 428 | rc = 0; |
f0cb5203 | 429 | if (!xics_phys) |
ddf05501 | 430 | rc = opal_int_get_xirr(&xirr, false); |
ddf05501 | 431 | else |
f725758b | 432 | xirr = _lwzcix(xics_phys + XICS_XIRR); |
ddf05501 PM |
433 | if (rc < 0) |
434 | return 1; | |
37f55d30 SW |
435 | |
436 | /* | |
437 | * Save XIRR for later. Since we get control in reverse endian | |
438 | * on LE systems, save it byte reversed and fetch it back in | |
439 | * host endian. Note that xirr is the value read from the | |
440 | * XIRR register, while h_xirr is the host endian version. | |
441 | */ | |
37f55d30 SW |
442 | h_xirr = be32_to_cpu(xirr); |
443 | local_paca->kvm_hstate.saved_xirr = h_xirr; | |
444 | xisr = h_xirr & 0xffffff; | |
445 | /* | |
446 | * Ensure that the store/load complete to guarantee all side | |
447 | * effects of loading from XIRR has completed | |
448 | */ | |
449 | smp_mb(); | |
450 | ||
451 | /* if nothing pending in the ICP */ | |
452 | if (!xisr) | |
453 | return 0; | |
454 | ||
455 | /* We found something in the ICP... | |
456 | * | |
457 | * If it is an IPI, clear the MFRR and EOI it. | |
458 | */ | |
459 | if (xisr == XICS_IPI) { | |
ddf05501 | 460 | rc = 0; |
f0cb5203 | 461 | if (xics_phys) { |
f725758b PM |
462 | _stbcix(xics_phys + XICS_MFRR, 0xff); |
463 | _stwcix(xics_phys + XICS_XIRR, xirr); | |
464 | } else { | |
f0cb5203 BH |
465 | opal_int_set_mfrr(hard_smp_processor_id(), 0xff); |
466 | rc = opal_int_eoi(h_xirr); | |
f725758b | 467 | } |
ddf05501 PM |
468 | /* If rc > 0, there is another interrupt pending */ |
469 | *again = rc > 0; | |
f725758b | 470 | |
37f55d30 SW |
471 | /* |
472 | * Need to ensure side effects of above stores | |
473 | * complete before proceeding. | |
474 | */ | |
475 | smp_mb(); | |
476 | ||
477 | /* | |
478 | * We need to re-check host IPI now in case it got set in the | |
479 | * meantime. If it's clear, we bounce the interrupt to the | |
480 | * guest | |
481 | */ | |
482 | host_ipi = local_paca->kvm_hstate.host_ipi; | |
483 | if (unlikely(host_ipi != 0)) { | |
484 | /* We raced with the host, | |
485 | * we need to resend that IPI, bummer | |
486 | */ | |
f0cb5203 | 487 | if (xics_phys) |
f725758b PM |
488 | _stbcix(xics_phys + XICS_MFRR, IPI_PRIORITY); |
489 | else | |
f0cb5203 BH |
490 | opal_int_set_mfrr(hard_smp_processor_id(), |
491 | IPI_PRIORITY); | |
37f55d30 SW |
492 | /* Let side effects complete */ |
493 | smp_mb(); | |
494 | return 1; | |
495 | } | |
496 | ||
497 | /* OK, it's an IPI for us */ | |
498 | local_paca->kvm_hstate.saved_xirr = 0; | |
499 | return -1; | |
500 | } | |
501 | ||
f725758b | 502 | return kvmppc_check_passthru(xisr, xirr, again); |
37f55d30 | 503 | } |