~hc/RK356X_SDK_RELEASE.git

..	..	@@ -1,27 +1,17 @@
	1	+// SPDX-License-Identifier: GPL-2.0-or-later
1	2	/*
2	3	* KVM paravirt_ops implementation
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 as published by
6		- * the Free Software Foundation; either version 2 of the License, or
7		- * (at your option) any later version.
8		- *
9		- * This program is distributed in the hope that it will be useful,
10		- * but WITHOUT ANY WARRANTY; without even the implied warranty of
11		- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12		- * GNU General Public License for more details.
13		- *
14		- * You should have received a copy of the GNU General Public License
15		- * along with this program; if not, write to the Free Software
16		- * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
17	4	*
18	5	* Copyright (C) 2007, Red Hat, Inc., Ingo Molnar <mingo@redhat.com>
19	6	* Copyright IBM Corporation, 2007
20	7	* Authors: Anthony Liguori <aliguori@us.ibm.com>
21	8	*/
22	9
	10	+#define pr_fmt(fmt) "kvm-guest: " fmt
	11	+
23	12	#include <linux/context_tracking.h>
24	13	#include <linux/init.h>
	14	+#include <linux/irq.h>
25	15	#include <linux/kernel.h>
26	16	#include <linux/kvm_para.h>
27	17	#include <linux/cpu.h>
..	..	@@ -34,9 +24,9 @@
34	24	#include <linux/sched.h>
35	25	#include <linux/slab.h>
36	26	#include <linux/kprobes.h>
37		-#include <linux/debugfs.h>
38	27	#include <linux/nmi.h>
39	28	#include <linux/swait.h>
	29	+#include <linux/syscore_ops.h>
40	30	#include <asm/timer.h>
41	31	#include <asm/cpu.h>
42	32	#include <asm/traps.h>
..	..	@@ -46,6 +36,12 @@
46	36	#include <asm/apicdef.h>
47	37	#include <asm/hypervisor.h>
48	38	#include <asm/tlb.h>
	39	+#include <asm/cpuidle_haltpoll.h>
	40	+#include <asm/ptrace.h>
	41	+#include <asm/reboot.h>
	42	+#include <asm/svm.h>
	43	+
	44	+DEFINE_STATIC_KEY_FALSE(kvm_async_pf_enabled);
49	45
50	46	static int kvmapf = 1;
51	47
..	..	@@ -67,9 +63,10 @@
67	63	early_param("no-steal-acc", parse_no_stealacc);
68	64
69	65	static DEFINE_PER_CPU_DECRYPTED(struct kvm_vcpu_pv_apf_data, apf_reason) __aligned(64);
70		-static DEFINE_PER_CPU_DECRYPTED(struct kvm_steal_time, steal_time) __aligned(64);
	66	+DEFINE_PER_CPU_DECRYPTED(struct kvm_steal_time, steal_time) __aligned(64) __visible;
71	67	static int has_steal_clock = 0;
72	68
	69	+static int has_guest_poll = 0;
73	70	/*
74	71	* No need for any "IO delay" on KVM
75	72	*/
..	..	@@ -85,7 +82,6 @@
85	82	struct swait_queue_head wq;
86	83	u32 token;
87	84	int cpu;
88		- bool halted;
89	85	};
90	86
91	87	static struct kvm_task_sleep_head {
..	..	@@ -108,77 +104,64 @@
108	104	return NULL;
109	105	}
110	106
111		-/*
112		- * @interrupt_kernel: Is this called from a routine which interrupts the kernel
113		- * (other than user space)?
114		- */
115		-void kvm_async_pf_task_wait(u32 token, int interrupt_kernel)
	107	+static bool kvm_async_pf_queue_task(u32 token, struct kvm_task_sleep_node *n)
116	108	{
117	109	u32 key = hash_32(token, KVM_TASK_SLEEP_HASHBITS);
118	110	struct kvm_task_sleep_head *b = &async_pf_sleepers[key];
119		- struct kvm_task_sleep_node n, *e;
120		- DECLARE_SWAITQUEUE(wait);
121		-
122		- rcu_irq_enter();
	111	+ struct kvm_task_sleep_node *e;
123	112
124	113	raw_spin_lock(&b->lock);
125	114	e = _find_apf_task(b, token);
126	115	if (e) {
127	116	/* dummy entry exist -> wake up was delivered ahead of PF */
128	117	hlist_del(&e->link);
129		- kfree(e);
130	118	raw_spin_unlock(&b->lock);
131		-
132		- rcu_irq_exit();
133		- return;
	119	+ kfree(e);
	120	+ return false;
134	121	}
135	122
136		- n.token = token;
137		- n.cpu = smp_processor_id();
138		- n.halted = is_idle_task(current) \|\|
139		- (IS_ENABLED(CONFIG_PREEMPT_COUNT)
140		- ? preempt_count() > 1 \|\| rcu_preempt_depth()
141		- : interrupt_kernel);
142		- init_swait_queue_head(&n.wq);
143		- hlist_add_head(&n.link, &b->list);
	123	+ n->token = token;
	124	+ n->cpu = smp_processor_id();
	125	+ init_swait_queue_head(&n->wq);
	126	+ hlist_add_head(&n->link, &b->list);
144	127	raw_spin_unlock(&b->lock);
	128	+ return true;
	129	+}
	130	+
	131	+/*
	132	+ * kvm_async_pf_task_wait_schedule - Wait for pagefault to be handled
	133	+ * @token: Token to identify the sleep node entry
	134	+ *
	135	+ * Invoked from the async pagefault handling code or from the VM exit page
	136	+ * fault handler. In both cases RCU is watching.
	137	+ */
	138	+void kvm_async_pf_task_wait_schedule(u32 token)
	139	+{
	140	+ struct kvm_task_sleep_node n;
	141	+ DECLARE_SWAITQUEUE(wait);
	142	+
	143	+ lockdep_assert_irqs_disabled();
	144	+
	145	+ if (!kvm_async_pf_queue_task(token, &n))
	146	+ return;
145	147
146	148	for (;;) {
147		- if (!n.halted)
148		- prepare_to_swait_exclusive(&n.wq, &wait, TASK_UNINTERRUPTIBLE);
	149	+ prepare_to_swait_exclusive(&n.wq, &wait, TASK_UNINTERRUPTIBLE);
149	150	if (hlist_unhashed(&n.link))
150	151	break;
151	152
152		- rcu_irq_exit();
153		-
154		- if (!n.halted) {
155		- local_irq_enable();
156		- schedule();
157		- local_irq_disable();
158		- } else {
159		- /*
160		- * We cannot reschedule. So halt.
161		- */
162		- native_safe_halt();
163		- local_irq_disable();
164		- }
165		-
166		- rcu_irq_enter();
	153	+ local_irq_enable();
	154	+ schedule();
	155	+ local_irq_disable();
167	156	}
168		- if (!n.halted)
169		- finish_swait(&n.wq, &wait);
170		-
171		- rcu_irq_exit();
172		- return;
	157	+ finish_swait(&n.wq, &wait);
173	158	}
174		-EXPORT_SYMBOL_GPL(kvm_async_pf_task_wait);
	159	+EXPORT_SYMBOL_GPL(kvm_async_pf_task_wait_schedule);
175	160
176	161	static void apf_task_wake_one(struct kvm_task_sleep_node *n)
177	162	{
178	163	hlist_del_init(&n->link);
179		- if (n->halted)
180		- smp_send_reschedule(n->cpu);
181		- else if (swq_has_sleeper(&n->wq))
	164	+ if (swq_has_sleeper(&n->wq))
182	165	swake_up_one(&n->wq);
183	166	}
184	167
..	..	@@ -187,12 +170,13 @@
187	170	int i;
188	171
189	172	for (i = 0; i < KVM_TASK_SLEEP_HASHSIZE; i++) {
190		- struct hlist_node p, next;
191	173	struct kvm_task_sleep_head *b = &async_pf_sleepers[i];
	174	+ struct kvm_task_sleep_node *n;
	175	+ struct hlist_node p, next;
	176	+
192	177	raw_spin_lock(&b->lock);
193	178	hlist_for_each_safe(p, next, &b->list) {
194		- struct kvm_task_sleep_node *n =
195		- hlist_entry(p, typeof(*n), link);
	179	+ n = hlist_entry(p, typeof(*n), link);
196	180	if (n->cpu == smp_processor_id())
197	181	apf_task_wake_one(n);
198	182	}
..	..	@@ -204,7 +188,7 @@
204	188	{
205	189	u32 key = hash_32(token, KVM_TASK_SLEEP_HASHBITS);
206	190	struct kvm_task_sleep_head *b = &async_pf_sleepers[key];
207		- struct kvm_task_sleep_node *n;
	191	+ struct kvm_task_sleep_node n, dummy = NULL;
208	192
209	193	if (token == ~0) {
210	194	apf_task_wake_all();
..	..	@@ -216,74 +200,115 @@
216	200	n = _find_apf_task(b, token);
217	201	if (!n) {
218	202	/*
219		- * async PF was not yet handled.
220		- * Add dummy entry for the token.
	203	+ * Async #PF not yet handled, add a dummy entry for the token.
	204	+ * Allocating the token must be down outside of the raw lock
	205	+ * as the allocator is preemptible on PREEMPT_RT kernels.
221	206	*/
222		- n = kzalloc(sizeof(*n), GFP_ATOMIC);
223		- if (!n) {
224		- /*
225		- * Allocation failed! Busy wait while other cpu
226		- * handles async PF.
227		- */
	207	+ if (!dummy) {
228	208	raw_spin_unlock(&b->lock);
229		- cpu_relax();
	209	+ dummy = kzalloc(sizeof(*dummy), GFP_ATOMIC);
	210	+
	211	+ /*
	212	+ * Continue looping on allocation failure, eventually
	213	+ * the async #PF will be handled and allocating a new
	214	+ * node will be unnecessary.
	215	+ */
	216	+ if (!dummy)
	217	+ cpu_relax();
	218	+
	219	+ /*
	220	+ * Recheck for async #PF completion before enqueueing
	221	+ * the dummy token to avoid duplicate list entries.
	222	+ */
230	223	goto again;
231	224	}
232		- n->token = token;
233		- n->cpu = smp_processor_id();
234		- init_swait_queue_head(&n->wq);
235		- hlist_add_head(&n->link, &b->list);
236		- } else
	225	+ dummy->token = token;
	226	+ dummy->cpu = smp_processor_id();
	227	+ init_swait_queue_head(&dummy->wq);
	228	+ hlist_add_head(&dummy->link, &b->list);
	229	+ dummy = NULL;
	230	+ } else {
237	231	apf_task_wake_one(n);
	232	+ }
238	233	raw_spin_unlock(&b->lock);
239		- return;
	234	+
	235	+ /* A dummy token might be allocated and ultimately not used. */
	236	+ if (dummy)
	237	+ kfree(dummy);
240	238	}
241	239	EXPORT_SYMBOL_GPL(kvm_async_pf_task_wake);
242	240
243		-u32 kvm_read_and_reset_pf_reason(void)
	241	+noinstr u32 kvm_read_and_reset_apf_flags(void)
244	242	{
245		- u32 reason = 0;
	243	+ u32 flags = 0;
246	244
247	245	if (__this_cpu_read(apf_reason.enabled)) {
248		- reason = __this_cpu_read(apf_reason.reason);
249		- __this_cpu_write(apf_reason.reason, 0);
	246	+ flags = __this_cpu_read(apf_reason.flags);
	247	+ __this_cpu_write(apf_reason.flags, 0);
250	248	}
251	249
252		- return reason;
	250	+ return flags;
253	251	}
254		-EXPORT_SYMBOL_GPL(kvm_read_and_reset_pf_reason);
255		-NOKPROBE_SYMBOL(kvm_read_and_reset_pf_reason);
	252	+EXPORT_SYMBOL_GPL(kvm_read_and_reset_apf_flags);
256	253
257		-dotraplinkage void
258		-do_async_page_fault(struct pt_regs *regs, unsigned long error_code)
	254	+noinstr bool __kvm_handle_async_pf(struct pt_regs *regs, u32 token)
259	255	{
260		- enum ctx_state prev_state;
	256	+ u32 flags = kvm_read_and_reset_apf_flags();
	257	+ irqentry_state_t state;
261	258
262		- switch (kvm_read_and_reset_pf_reason()) {
263		- default:
264		- do_page_fault(regs, error_code);
265		- break;
266		- case KVM_PV_REASON_PAGE_NOT_PRESENT:
267		- /* page is swapped out by the host. */
268		- prev_state = exception_enter();
269		- kvm_async_pf_task_wait((u32)read_cr2(), !user_mode(regs));
270		- exception_exit(prev_state);
271		- break;
272		- case KVM_PV_REASON_PAGE_READY:
273		- rcu_irq_enter();
274		- kvm_async_pf_task_wake((u32)read_cr2());
275		- rcu_irq_exit();
276		- break;
	259	+ if (!flags)
	260	+ return false;
	261	+
	262	+ state = irqentry_enter(regs);
	263	+ instrumentation_begin();
	264	+
	265	+ /*
	266	+ * If the host managed to inject an async #PF into an interrupt
	267	+ * disabled region, then die hard as this is not going to end well
	268	+ * and the host side is seriously broken.
	269	+ */
	270	+ if (unlikely(!(regs->flags & X86_EFLAGS_IF)))
	271	+ panic("Host injected async #PF in interrupt disabled region\n");
	272	+
	273	+ if (flags & KVM_PV_REASON_PAGE_NOT_PRESENT) {
	274	+ if (unlikely(!(user_mode(regs))))
	275	+ panic("Host injected async #PF in kernel mode\n");
	276	+ /* Page is swapped out by the host. */
	277	+ kvm_async_pf_task_wait_schedule(token);
	278	+ } else {
	279	+ WARN_ONCE(1, "Unexpected async PF flags: %x\n", flags);
277	280	}
	281	+
	282	+ instrumentation_end();
	283	+ irqentry_exit(regs, state);
	284	+ return true;
278	285	}
279		-NOKPROBE_SYMBOL(do_async_page_fault);
	286	+
	287	+DEFINE_IDTENTRY_SYSVEC(sysvec_kvm_asyncpf_interrupt)
	288	+{
	289	+ struct pt_regs *old_regs = set_irq_regs(regs);
	290	+ u32 token;
	291	+
	292	+ ack_APIC_irq();
	293	+
	294	+ inc_irq_stat(irq_hv_callback_count);
	295	+
	296	+ if (__this_cpu_read(apf_reason.enabled)) {
	297	+ token = __this_cpu_read(apf_reason.token);
	298	+ kvm_async_pf_task_wake(token);
	299	+ __this_cpu_write(apf_reason.token, 0);
	300	+ wrmsrl(MSR_KVM_ASYNC_PF_ACK, 1);
	301	+ }
	302	+
	303	+ set_irq_regs(old_regs);
	304	+}
280	305
281	306	static void __init paravirt_ops_setup(void)
282	307	{
283	308	pv_info.name = "KVM";
284	309
285	310	if (kvm_para_has_feature(KVM_FEATURE_NOP_IO_DELAY))
286		- pv_cpu_ops.io_delay = kvm_io_delay;
	311	+ pv_ops.cpu.io_delay = kvm_io_delay;
287	312
288	313	#ifdef CONFIG_X86_IO_APIC
289	314	no_timer_check = 1;
..	..	@@ -299,8 +324,8 @@
299	324	return;
300	325
301	326	wrmsrl(MSR_KVM_STEAL_TIME, (slow_virt_to_phys(st) \| KVM_MSR_ENABLED));
302		- pr_info("kvm-stealtime: cpu %d, msr %llx\n",
303		- cpu, (unsigned long long) slow_virt_to_phys(st));
	327	+ pr_info("stealtime: cpu %d, msr %llx\n", cpu,
	328	+ (unsigned long long) slow_virt_to_phys(st));
304	329	}
305	330
306	331	static DEFINE_PER_CPU_DECRYPTED(unsigned long, kvm_apic_eoi) = KVM_PV_EOI_DISABLED;
..	..	@@ -321,28 +346,27 @@
321	346
322	347	static void kvm_guest_cpu_init(void)
323	348	{
324		- if (!kvm_para_available())
325		- return;
326		-
327		- if (kvm_para_has_feature(KVM_FEATURE_ASYNC_PF) && kvmapf) {
	349	+ if (kvm_para_has_feature(KVM_FEATURE_ASYNC_PF_INT) && kvmapf) {
328	350	u64 pa = slow_virt_to_phys(this_cpu_ptr(&apf_reason));
329	351
330		-#ifdef CONFIG_PREEMPT
331		- pa \|= KVM_ASYNC_PF_SEND_ALWAYS;
332		-#endif
333		- pa \|= KVM_ASYNC_PF_ENABLED;
	352	+ WARN_ON_ONCE(!static_branch_likely(&kvm_async_pf_enabled));
	353	+
	354	+ pa = slow_virt_to_phys(this_cpu_ptr(&apf_reason));
	355	+ pa \|= KVM_ASYNC_PF_ENABLED \| KVM_ASYNC_PF_DELIVERY_AS_INT;
334	356
335	357	if (kvm_para_has_feature(KVM_FEATURE_ASYNC_PF_VMEXIT))
336	358	pa \|= KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
337	359
	360	+ wrmsrl(MSR_KVM_ASYNC_PF_INT, HYPERVISOR_CALLBACK_VECTOR);
	361	+
338	362	wrmsrl(MSR_KVM_ASYNC_PF_EN, pa);
339	363	__this_cpu_write(apf_reason.enabled, 1);
340		- printk(KERN_INFO"KVM setup async PF for cpu %d\n",
341		- smp_processor_id());
	364	+ pr_info("KVM setup async PF for cpu %d\n", smp_processor_id());
342	365	}
343	366
344	367	if (kvm_para_has_feature(KVM_FEATURE_PV_EOI)) {
345	368	unsigned long pa;
	369	+
346	370	/* Size alignment is implied but just to make it explicit. */
347	371	BUILD_BUG_ON(__alignof__(kvm_apic_eoi) < 4);
348	372	__this_cpu_write(kvm_apic_eoi, 0);
..	..	@@ -363,8 +387,15 @@
363	387	wrmsrl(MSR_KVM_ASYNC_PF_EN, 0);
364	388	__this_cpu_write(apf_reason.enabled, 0);
365	389
366		- printk(KERN_INFO"Unregister pv shared memory for cpu %d\n",
367		- smp_processor_id());
	390	+ pr_info("Unregister pv shared memory for cpu %d\n", smp_processor_id());
	391	+}
	392	+
	393	+static void kvm_disable_steal_time(void)
	394	+{
	395	+ if (!has_steal_clock)
	396	+ return;
	397	+
	398	+ wrmsr(MSR_KVM_STEAL_TIME, 0, 0);
368	399	}
369	400
370	401	static void kvm_pv_guest_cpu_reboot(void *unused)
..	..	@@ -409,14 +440,6 @@
409	440	return steal;
410	441	}
411	442
412		-void kvm_disable_steal_time(void)
413		-{
414		- if (!has_steal_clock)
415		- return;
416		-
417		- wrmsr(MSR_KVM_STEAL_TIME, 0, 0);
418		-}
419		-
420	443	static inline void __set_percpu_decrypted(void *ptr, unsigned long size)
421	444	{
422	445	early_set_memory_decrypted((unsigned long) ptr, size);
..	..	@@ -444,7 +467,50 @@
444	467	}
445	468	}
446	469
	470	+static bool pv_tlb_flush_supported(void)
	471	+{
	472	+ return (kvm_para_has_feature(KVM_FEATURE_PV_TLB_FLUSH) &&
	473	+ !kvm_para_has_hint(KVM_HINTS_REALTIME) &&
	474	+ kvm_para_has_feature(KVM_FEATURE_STEAL_TIME));
	475	+}
	476	+
	477	+static DEFINE_PER_CPU(cpumask_var_t, __pv_cpu_mask);
	478	+
	479	+static void kvm_guest_cpu_offline(bool shutdown)
	480	+{
	481	+ kvm_disable_steal_time();
	482	+ if (kvm_para_has_feature(KVM_FEATURE_PV_EOI))
	483	+ wrmsrl(MSR_KVM_PV_EOI_EN, 0);
	484	+ kvm_pv_disable_apf();
	485	+ if (!shutdown)
	486	+ apf_task_wake_all();
	487	+ kvmclock_disable();
	488	+}
	489	+
	490	+static int kvm_cpu_online(unsigned int cpu)
	491	+{
	492	+ unsigned long flags;
	493	+
	494	+ local_irq_save(flags);
	495	+ kvm_guest_cpu_init();
	496	+ local_irq_restore(flags);
	497	+ return 0;
	498	+}
	499	+
447	500	#ifdef CONFIG_SMP
	501	+
	502	+static bool pv_ipi_supported(void)
	503	+{
	504	+ return kvm_para_has_feature(KVM_FEATURE_PV_SEND_IPI);
	505	+}
	506	+
	507	+static bool pv_sched_yield_supported(void)
	508	+{
	509	+ return (kvm_para_has_feature(KVM_FEATURE_PV_SCHED_YIELD) &&
	510	+ !kvm_para_has_hint(KVM_HINTS_REALTIME) &&
	511	+ kvm_para_has_feature(KVM_FEATURE_STEAL_TIME));
	512	+}
	513	+
448	514	#define KVM_IPI_CLUSTER_SIZE (2 * BITS_PER_LONG)
449	515
450	516	static void __send_ipi_mask(const struct cpumask *mask, int vector)
..	..	@@ -480,12 +546,13 @@
480	546	} else if (apic_id < min && max - apic_id < KVM_IPI_CLUSTER_SIZE) {
481	547	ipi_bitmap <<= min - apic_id;
482	548	min = apic_id;
483		- } else if (apic_id < min + KVM_IPI_CLUSTER_SIZE) {
	549	+ } else if (apic_id > min && apic_id < min + KVM_IPI_CLUSTER_SIZE) {
484	550	max = apic_id < max ? max : apic_id;
485	551	} else {
486	552	ret = kvm_hypercall4(KVM_HC_SEND_IPI, (unsigned long)ipi_bitmap,
487	553	(unsigned long)(ipi_bitmap >> BITS_PER_LONG), min, icr);
488		- WARN_ONCE(ret < 0, "KVM: failed to send PV IPI: %ld", ret);
	554	+ WARN_ONCE(ret < 0, "kvm-guest: failed to send PV IPI: %ld",
	555	+ ret);
489	556	min = max = apic_id;
490	557	ipi_bitmap = 0;
491	558	}
..	..	@@ -495,7 +562,8 @@
495	562	if (ipi_bitmap) {
496	563	ret = kvm_hypercall4(KVM_HC_SEND_IPI, (unsigned long)ipi_bitmap,
497	564	(unsigned long)(ipi_bitmap >> BITS_PER_LONG), min, icr);
498		- WARN_ONCE(ret < 0, "KVM: failed to send PV IPI: %ld", ret);
	565	+ WARN_ONCE(ret < 0, "kvm-guest: failed to send PV IPI: %ld",
	566	+ ret);
499	567	}
500	568
501	569	local_irq_restore(flags);
..	..	@@ -509,23 +577,13 @@
509	577	static void kvm_send_ipi_mask_allbutself(const struct cpumask *mask, int vector)
510	578	{
511	579	unsigned int this_cpu = smp_processor_id();
512		- struct cpumask new_mask;
	580	+ struct cpumask *new_mask = this_cpu_cpumask_var_ptr(__pv_cpu_mask);
513	581	const struct cpumask *local_mask;
514	582
515		- cpumask_copy(&new_mask, mask);
516		- cpumask_clear_cpu(this_cpu, &new_mask);
517		- local_mask = &new_mask;
	583	+ cpumask_copy(new_mask, mask);
	584	+ cpumask_clear_cpu(this_cpu, new_mask);
	585	+ local_mask = new_mask;
518	586	__send_ipi_mask(local_mask, vector);
519		-}
520		-
521		-static void kvm_send_ipi_allbutself(int vector)
522		-{
523		- kvm_send_ipi_mask_allbutself(cpu_online_mask, vector);
524		-}
525		-
526		-static void kvm_send_ipi_all(int vector)
527		-{
528		- __send_ipi_mask(cpu_online_mask, vector);
529	587	}
530	588
531	589	/*
..	..	@@ -535,16 +593,22 @@
535	593	{
536	594	apic->send_IPI_mask = kvm_send_ipi_mask;
537	595	apic->send_IPI_mask_allbutself = kvm_send_ipi_mask_allbutself;
538		- apic->send_IPI_allbutself = kvm_send_ipi_allbutself;
539		- apic->send_IPI_all = kvm_send_ipi_all;
540		- pr_info("KVM setup pv IPIs\n");
	596	+ pr_info("setup PV IPIs\n");
541	597	}
542	598
543		-static void __init kvm_smp_prepare_cpus(unsigned int max_cpus)
	599	+static void kvm_smp_send_call_func_ipi(const struct cpumask *mask)
544	600	{
545		- native_smp_prepare_cpus(max_cpus);
546		- if (kvm_para_has_hint(KVM_HINTS_REALTIME))
547		- static_branch_disable(&virt_spin_lock_key);
	601	+ int cpu;
	602	+
	603	+ native_send_call_func_ipi(mask);
	604	+
	605	+ /* Make sure other vCPUs get a chance to run if they need to. */
	606	+ for_each_cpu(cpu, mask) {
	607	+ if (vcpu_is_preempted(cpu)) {
	608	+ kvm_hypercall1(KVM_HC_SCHED_YIELD, per_cpu(x86_cpu_to_apicid, cpu));
	609	+ break;
	610	+ }
	611	+ }
548	612	}
549	613
550	614	static void __init kvm_smp_prepare_boot_cpu(void)
..	..	@@ -560,38 +624,60 @@
560	624	kvm_spinlock_init();
561	625	}
562	626
563		-static void kvm_guest_cpu_offline(void)
564		-{
565		- kvm_disable_steal_time();
566		- if (kvm_para_has_feature(KVM_FEATURE_PV_EOI))
567		- wrmsrl(MSR_KVM_PV_EOI_EN, 0);
568		- kvm_pv_disable_apf();
569		- apf_task_wake_all();
570		-}
571		-
572		-static int kvm_cpu_online(unsigned int cpu)
573		-{
574		- local_irq_disable();
575		- kvm_guest_cpu_init();
576		- local_irq_enable();
577		- return 0;
578		-}
579		-
580	627	static int kvm_cpu_down_prepare(unsigned int cpu)
581	628	{
582		- local_irq_disable();
583		- kvm_guest_cpu_offline();
584		- local_irq_enable();
	629	+ unsigned long flags;
	630	+
	631	+ local_irq_save(flags);
	632	+ kvm_guest_cpu_offline(false);
	633	+ local_irq_restore(flags);
585	634	return 0;
586	635	}
	636	+
587	637	#endif
588	638
589		-static void __init kvm_apf_trap_init(void)
	639	+static int kvm_suspend(void)
590	640	{
591		- update_intr_gate(X86_TRAP_PF, async_page_fault);
	641	+ u64 val = 0;
	642	+
	643	+ kvm_guest_cpu_offline(false);
	644	+
	645	+#ifdef CONFIG_ARCH_CPUIDLE_HALTPOLL
	646	+ if (kvm_para_has_feature(KVM_FEATURE_POLL_CONTROL))
	647	+ rdmsrl(MSR_KVM_POLL_CONTROL, val);
	648	+ has_guest_poll = !(val & 1);
	649	+#endif
	650	+ return 0;
592	651	}
593	652
594		-static DEFINE_PER_CPU(cpumask_var_t, __pv_tlb_mask);
	653	+static void kvm_resume(void)
	654	+{
	655	+ kvm_cpu_online(raw_smp_processor_id());
	656	+
	657	+#ifdef CONFIG_ARCH_CPUIDLE_HALTPOLL
	658	+ if (kvm_para_has_feature(KVM_FEATURE_POLL_CONTROL) && has_guest_poll)
	659	+ wrmsrl(MSR_KVM_POLL_CONTROL, 0);
	660	+#endif
	661	+}
	662	+
	663	+static struct syscore_ops kvm_syscore_ops = {
	664	+ .suspend = kvm_suspend,
	665	+ .resume = kvm_resume,
	666	+};
	667	+
	668	+/*
	669	+ * After a PV feature is registered, the host will keep writing to the
	670	+ * registered memory location. If the guest happens to shutdown, this memory
	671	+ * won't be valid. In cases like kexec, in which you install a new kernel, this
	672	+ * means a random memory location will be kept being written.
	673	+ */
	674	+#ifdef CONFIG_KEXEC_CORE
	675	+static void kvm_crash_shutdown(struct pt_regs *regs)
	676	+{
	677	+ kvm_guest_cpu_offline(true);
	678	+ native_machine_crash_shutdown(regs);
	679	+}
	680	+#endif
595	681
596	682	static void kvm_flush_tlb_others(const struct cpumask *cpumask,
597	683	const struct flush_tlb_info *info)
..	..	@@ -599,7 +685,7 @@
599	685	u8 state;
600	686	int cpu;
601	687	struct kvm_steal_time *src;
602		- struct cpumask *flushmask = this_cpu_cpumask_var_ptr(__pv_tlb_mask);
	688	+ struct cpumask *flushmask = this_cpu_cpumask_var_ptr(__pv_cpu_mask);
603	689
604	690	cpumask_copy(flushmask, cpumask);
605	691	/*
..	..	@@ -623,41 +709,49 @@
623	709	{
624	710	int i;
625	711
626		- if (!kvm_para_available())
627		- return;
628		-
629	712	paravirt_ops_setup();
630	713	register_reboot_notifier(&kvm_pv_reboot_nb);
631	714	for (i = 0; i < KVM_TASK_SLEEP_HASHSIZE; i++)
632	715	raw_spin_lock_init(&async_pf_sleepers[i].lock);
633		- if (kvm_para_has_feature(KVM_FEATURE_ASYNC_PF))
634		- x86_init.irqs.trap_init = kvm_apf_trap_init;
635	716
636	717	if (kvm_para_has_feature(KVM_FEATURE_STEAL_TIME)) {
637	718	has_steal_clock = 1;
638		- pv_time_ops.steal_clock = kvm_steal_clock;
	719	+ pv_ops.time.steal_clock = kvm_steal_clock;
639	720	}
640	721
641		- if (kvm_para_has_feature(KVM_FEATURE_PV_TLB_FLUSH) &&
642		- !kvm_para_has_hint(KVM_HINTS_REALTIME) &&
643		- kvm_para_has_feature(KVM_FEATURE_STEAL_TIME)) {
644		- pv_mmu_ops.flush_tlb_others = kvm_flush_tlb_others;
645		- pv_mmu_ops.tlb_remove_table = tlb_remove_table;
	722	+ if (pv_tlb_flush_supported()) {
	723	+ pv_ops.mmu.flush_tlb_others = kvm_flush_tlb_others;
	724	+ pv_ops.mmu.tlb_remove_table = tlb_remove_table;
	725	+ pr_info("KVM setup pv remote TLB flush\n");
646	726	}
647	727
648	728	if (kvm_para_has_feature(KVM_FEATURE_PV_EOI))
649	729	apic_set_eoi_write(kvm_guest_apic_eoi_write);
650	730
	731	+ if (kvm_para_has_feature(KVM_FEATURE_ASYNC_PF_INT) && kvmapf) {
	732	+ static_branch_enable(&kvm_async_pf_enabled);
	733	+ alloc_intr_gate(HYPERVISOR_CALLBACK_VECTOR, asm_sysvec_kvm_asyncpf_interrupt);
	734	+ }
	735	+
651	736	#ifdef CONFIG_SMP
652		- smp_ops.smp_prepare_cpus = kvm_smp_prepare_cpus;
653	737	smp_ops.smp_prepare_boot_cpu = kvm_smp_prepare_boot_cpu;
	738	+ if (pv_sched_yield_supported()) {
	739	+ smp_ops.send_call_func_ipi = kvm_smp_send_call_func_ipi;
	740	+ pr_info("setup PV sched yield\n");
	741	+ }
654	742	if (cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "x86/kvm:online",
655	743	kvm_cpu_online, kvm_cpu_down_prepare) < 0)
656		- pr_err("kvm_guest: Failed to install cpu hotplug callbacks\n");
	744	+ pr_err("failed to install cpu hotplug callbacks\n");
657	745	#else
658	746	sev_map_percpu_data();
659	747	kvm_guest_cpu_init();
660	748	#endif
	749	+
	750	+#ifdef CONFIG_KEXEC_CORE
	751	+ machine_ops.crash_shutdown = kvm_crash_shutdown;
	752	+#endif
	753	+
	754	+ register_syscore_ops(&kvm_syscore_ops);
661	755
662	756	/*
663	757	* Hard lockup detection is enabled by default. Disable it, as guests
..	..	@@ -703,6 +797,7 @@
703	797	{
704	798	return cpuid_edx(kvm_cpuid_base() \| KVM_CPUID_FEATURES);
705	799	}
	800	+EXPORT_SYMBOL_GPL(kvm_arch_para_hints);
706	801
707	802	static uint32_t __init kvm_detect(void)
708	803	{
..	..	@@ -712,7 +807,7 @@
712	807	static void __init kvm_apic_init(void)
713	808	{
714	809	#if defined(CONFIG_SMP)
715		- if (kvm_para_has_feature(KVM_FEATURE_PV_SEND_IPI))
	810	+ if (pv_ipi_supported())
716	811	kvm_setup_pv_ipi();
717	812	#endif
718	813	}
..	..	@@ -723,13 +818,34 @@
723	818	x86_platform.apic_post_init = kvm_apic_init;
724	819	}
725	820
	821	+#if defined(CONFIG_AMD_MEM_ENCRYPT)
	822	+static void kvm_sev_es_hcall_prepare(struct ghcb ghcb, struct pt_regs regs)
	823	+{
	824	+ /* RAX and CPL are already in the GHCB */
	825	+ ghcb_set_rbx(ghcb, regs->bx);
	826	+ ghcb_set_rcx(ghcb, regs->cx);
	827	+ ghcb_set_rdx(ghcb, regs->dx);
	828	+ ghcb_set_rsi(ghcb, regs->si);
	829	+}
	830	+
	831	+static bool kvm_sev_es_hcall_finish(struct ghcb ghcb, struct pt_regs regs)
	832	+{
	833	+ /* No checking of the return state needed */
	834	+ return true;
	835	+}
	836	+#endif
	837	+
726	838	const __initconst struct hypervisor_x86 x86_hyper_kvm = {
727		- .name = "KVM",
728		- .detect = kvm_detect,
729		- .type = X86_HYPER_KVM,
730		- .init.guest_late_init = kvm_guest_init,
731		- .init.x2apic_available = kvm_para_available,
732		- .init.init_platform = kvm_init_platform,
	839	+ .name = "KVM",
	840	+ .detect = kvm_detect,
	841	+ .type = X86_HYPER_KVM,
	842	+ .init.guest_late_init = kvm_guest_init,
	843	+ .init.x2apic_available = kvm_para_available,
	844	+ .init.init_platform = kvm_init_platform,
	845	+#if defined(CONFIG_AMD_MEM_ENCRYPT)
	846	+ .runtime.sev_es_hcall_prepare = kvm_sev_es_hcall_prepare,
	847	+ .runtime.sev_es_hcall_finish = kvm_sev_es_hcall_finish,
	848	+#endif
733	849	};
734	850
735	851	static __init int activate_jump_labels(void)
..	..	@@ -744,23 +860,31 @@
744	860	}
745	861	arch_initcall(activate_jump_labels);
746	862
747		-static __init int kvm_setup_pv_tlb_flush(void)
	863	+static __init int kvm_alloc_cpumask(void)
748	864	{
749	865	int cpu;
	866	+ bool alloc = false;
750	867
751		- if (kvm_para_has_feature(KVM_FEATURE_PV_TLB_FLUSH) &&
752		- !kvm_para_has_hint(KVM_HINTS_REALTIME) &&
753		- kvm_para_has_feature(KVM_FEATURE_STEAL_TIME)) {
	868	+ if (!kvm_para_available() \|\| nopv)
	869	+ return 0;
	870	+
	871	+ if (pv_tlb_flush_supported())
	872	+ alloc = true;
	873	+
	874	+#if defined(CONFIG_SMP)
	875	+ if (pv_ipi_supported())
	876	+ alloc = true;
	877	+#endif
	878	+
	879	+ if (alloc)
754	880	for_each_possible_cpu(cpu) {
755		- zalloc_cpumask_var_node(per_cpu_ptr(&__pv_tlb_mask, cpu),
	881	+ zalloc_cpumask_var_node(per_cpu_ptr(&__pv_cpu_mask, cpu),
756	882	GFP_KERNEL, cpu_to_node(cpu));
757	883	}
758		- pr_info("KVM setup pv remote TLB flush\n");
759		- }
760	884
761	885	return 0;
762	886	}
763		-arch_initcall(kvm_setup_pv_tlb_flush);
	887	+arch_initcall(kvm_alloc_cpumask);
764	888
765	889	#ifdef CONFIG_PARAVIRT_SPINLOCKS
766	890
..	..	@@ -829,7 +953,7 @@
829	953	"movq __per_cpu_offset(,%rdi,8), %rax;"
830	954	"cmpb $0, " __stringify(KVM_STEAL_TIME_preempted) "+steal_time(%rax);"
831	955	"setne %al;"
832		-"ret;"
	956	+ASM_RET
833	957	".size __raw_callee_save___kvm_vcpu_is_preempted, .-__raw_callee_save___kvm_vcpu_is_preempted;"
834	958	".popsection");
835	959
..	..	@@ -840,29 +964,91 @@
840	964	*/
841	965	void __init kvm_spinlock_init(void)
842	966	{
843		- if (!kvm_para_available())
	967	+ /*
	968	+ * In case host doesn't support KVM_FEATURE_PV_UNHALT there is still an
	969	+ * advantage of keeping virt_spin_lock_key enabled: virt_spin_lock() is
	970	+ * preferred over native qspinlock when vCPU is preempted.
	971	+ */
	972	+ if (!kvm_para_has_feature(KVM_FEATURE_PV_UNHALT)) {
	973	+ pr_info("PV spinlocks disabled, no host support\n");
844	974	return;
845		- /* Does host kernel support KVM_FEATURE_PV_UNHALT? */
846		- if (!kvm_para_has_feature(KVM_FEATURE_PV_UNHALT))
847		- return;
	975	+ }
848	976
849		- if (kvm_para_has_hint(KVM_HINTS_REALTIME))
850		- return;
	977	+ /*
	978	+ * Disable PV spinlocks and use native qspinlock when dedicated pCPUs
	979	+ * are available.
	980	+ */
	981	+ if (kvm_para_has_hint(KVM_HINTS_REALTIME)) {
	982	+ pr_info("PV spinlocks disabled with KVM_HINTS_REALTIME hints\n");
	983	+ goto out;
	984	+ }
851	985
852		- /* Don't use the pvqspinlock code if there is only 1 vCPU. */
853		- if (num_possible_cpus() == 1)
854		- return;
	986	+ if (num_possible_cpus() == 1) {
	987	+ pr_info("PV spinlocks disabled, single CPU\n");
	988	+ goto out;
	989	+ }
	990	+
	991	+ if (nopvspin) {
	992	+ pr_info("PV spinlocks disabled, forced by \"nopvspin\" parameter\n");
	993	+ goto out;
	994	+ }
	995	+
	996	+ pr_info("PV spinlocks enabled\n");
855	997
856	998	__pv_init_lock_hash();
857		- pv_lock_ops.queued_spin_lock_slowpath = __pv_queued_spin_lock_slowpath;
858		- pv_lock_ops.queued_spin_unlock = PV_CALLEE_SAVE(__pv_queued_spin_unlock);
859		- pv_lock_ops.wait = kvm_wait;
860		- pv_lock_ops.kick = kvm_kick_cpu;
	999	+ pv_ops.lock.queued_spin_lock_slowpath = __pv_queued_spin_lock_slowpath;
	1000	+ pv_ops.lock.queued_spin_unlock =
	1001	+ PV_CALLEE_SAVE(__pv_queued_spin_unlock);
	1002	+ pv_ops.lock.wait = kvm_wait;
	1003	+ pv_ops.lock.kick = kvm_kick_cpu;
861	1004
862	1005	if (kvm_para_has_feature(KVM_FEATURE_STEAL_TIME)) {
863		- pv_lock_ops.vcpu_is_preempted =
	1006	+ pv_ops.lock.vcpu_is_preempted =
864	1007	PV_CALLEE_SAVE(__kvm_vcpu_is_preempted);
865	1008	}
	1009	+ /*
	1010	+ * When PV spinlock is enabled which is preferred over
	1011	+ * virt_spin_lock(), virt_spin_lock_key's value is meaningless.
	1012	+ * Just disable it anyway.
	1013	+ */
	1014	+out:
	1015	+ static_branch_disable(&virt_spin_lock_key);
866	1016	}
867	1017
868	1018	#endif /* CONFIG_PARAVIRT_SPINLOCKS */
	1019	+
	1020	+#ifdef CONFIG_ARCH_CPUIDLE_HALTPOLL
	1021	+
	1022	+static void kvm_disable_host_haltpoll(void *i)
	1023	+{
	1024	+ wrmsrl(MSR_KVM_POLL_CONTROL, 0);
	1025	+}
	1026	+
	1027	+static void kvm_enable_host_haltpoll(void *i)
	1028	+{
	1029	+ wrmsrl(MSR_KVM_POLL_CONTROL, 1);
	1030	+}
	1031	+
	1032	+void arch_haltpoll_enable(unsigned int cpu)
	1033	+{
	1034	+ if (!kvm_para_has_feature(KVM_FEATURE_POLL_CONTROL)) {
	1035	+ pr_err_once("host does not support poll control\n");
	1036	+ pr_err_once("host upgrade recommended\n");
	1037	+ return;
	1038	+ }
	1039	+
	1040	+ /* Enable guest halt poll disables host halt poll */
	1041	+ smp_call_function_single(cpu, kvm_disable_host_haltpoll, NULL, 1);
	1042	+}
	1043	+EXPORT_SYMBOL_GPL(arch_haltpoll_enable);
	1044	+
	1045	+void arch_haltpoll_disable(unsigned int cpu)
	1046	+{
	1047	+ if (!kvm_para_has_feature(KVM_FEATURE_POLL_CONTROL))
	1048	+ return;
	1049	+
	1050	+ /* Disable guest halt poll enables host halt poll */
	1051	+ smp_call_function_single(cpu, kvm_enable_host_haltpoll, NULL, 1);
	1052	+}
	1053	+EXPORT_SYMBOL_GPL(arch_haltpoll_disable);
	1054	+#endif