2 * Copyright (c) 2009, Microsoft Corporation.
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms and conditions of the GNU General Public License,
6 * version 2, as published by the Free Software Foundation.
8 * This program is distributed in the hope it will be useful, but WITHOUT
9 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * You should have received a copy of the GNU General Public License along with
14 * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
15 * Place - Suite 330, Boston, MA 02111-1307 USA.
18 * Haiyang Zhang <haiyangz@microsoft.com>
19 * Hank Janssen <hjanssen@microsoft.com>
22 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
24 #include <linux/kernel.h>
26 #include <linux/slab.h>
27 #include <linux/vmalloc.h>
28 #include <linux/hyperv.h>
29 #include <linux/version.h>
30 #include <linux/random.h>
31 #include <linux/clockchips.h>
32 #include <asm/mshyperv.h>
33 #include "hyperv_vmbus.h"
35 /* The one and only */
36 struct hv_context hv_context
= {
37 .synic_initialized
= false,
41 * If false, we're using the old mechanism for stimer0 interrupts
42 * where it sends a VMbus message when it expires. The old
43 * mechanism is used when running on older versions of Hyper-V
44 * that don't support Direct Mode. While Hyper-V provides
45 * four stimer's per CPU, Linux uses only stimer0.
47 static bool direct_mode_enabled
;
48 static int stimer0_irq
;
49 static int stimer0_vector
;
51 #define HV_TIMER_FREQUENCY (10 * 1000 * 1000) /* 100ns period */
52 #define HV_MAX_MAX_DELTA_TICKS 0xffffffff
53 #define HV_MIN_DELTA_TICKS 1
56 * hv_init - Main initialization routine.
58 * This routine must be called before any other routines in here are called
62 hv_context
.cpu_context
= alloc_percpu(struct hv_per_cpu_context
);
63 if (!hv_context
.cpu_context
)
66 direct_mode_enabled
= ms_hyperv
.misc_features
&
67 HV_X64_STIMER_DIRECT_MODE_AVAILABLE
;
72 * hv_post_message - Post a message using the hypervisor message IPC.
74 * This involves a hypercall.
76 int hv_post_message(union hv_connection_id connection_id
,
77 enum hv_message_type message_type
,
78 void *payload
, size_t payload_size
)
80 struct hv_input_post_message
*aligned_msg
;
81 struct hv_per_cpu_context
*hv_cpu
;
84 if (payload_size
> HV_MESSAGE_PAYLOAD_BYTE_COUNT
)
87 hv_cpu
= get_cpu_ptr(hv_context
.cpu_context
);
88 aligned_msg
= hv_cpu
->post_msg_page
;
89 aligned_msg
->connectionid
= connection_id
;
90 aligned_msg
->reserved
= 0;
91 aligned_msg
->message_type
= message_type
;
92 aligned_msg
->payload_size
= payload_size
;
93 memcpy((void *)aligned_msg
->payload
, payload
, payload_size
);
95 status
= hv_do_hypercall(HVCALL_POST_MESSAGE
, aligned_msg
, NULL
);
97 /* Preemption must remain disabled until after the hypercall
98 * so some other thread can't get scheduled onto this cpu and
99 * corrupt the per-cpu post_msg_page
103 return status
& 0xFFFF;
107 * ISR for when stimer0 is operating in Direct Mode. Direct Mode
108 * does not use VMbus or any VMbus messages, so process here and not
109 * in the VMbus driver code.
112 static void hv_stimer0_isr(void)
114 struct hv_per_cpu_context
*hv_cpu
;
116 hv_cpu
= this_cpu_ptr(hv_context
.cpu_context
);
117 hv_cpu
->clk_evt
->event_handler(hv_cpu
->clk_evt
);
118 add_interrupt_randomness(stimer0_vector
, 0);
121 static int hv_ce_set_next_event(unsigned long delta
,
122 struct clock_event_device
*evt
)
126 WARN_ON(!clockevent_state_oneshot(evt
));
128 current_tick
= hyperv_cs
->read(NULL
);
129 current_tick
+= delta
;
130 hv_init_timer(HV_X64_MSR_STIMER0_COUNT
, current_tick
);
134 static int hv_ce_shutdown(struct clock_event_device
*evt
)
136 hv_init_timer(HV_X64_MSR_STIMER0_COUNT
, 0);
137 hv_init_timer_config(HV_X64_MSR_STIMER0_CONFIG
, 0);
138 if (direct_mode_enabled
)
139 hv_disable_stimer0_percpu_irq(stimer0_irq
);
144 static int hv_ce_set_oneshot(struct clock_event_device
*evt
)
146 union hv_timer_config timer_cfg
;
148 timer_cfg
.as_uint64
= 0;
149 timer_cfg
.enable
= 1;
150 timer_cfg
.auto_enable
= 1;
151 if (direct_mode_enabled
) {
153 * When it expires, the timer will directly interrupt
154 * on the specified hardware vector/IRQ.
156 timer_cfg
.direct_mode
= 1;
157 timer_cfg
.apic_vector
= stimer0_vector
;
158 hv_enable_stimer0_percpu_irq(stimer0_irq
);
161 * When it expires, the timer will generate a VMbus message,
162 * to be handled by the normal VMbus interrupt handler.
164 timer_cfg
.direct_mode
= 0;
165 timer_cfg
.sintx
= VMBUS_MESSAGE_SINT
;
167 hv_init_timer_config(HV_X64_MSR_STIMER0_CONFIG
, timer_cfg
.as_uint64
);
171 static void hv_init_clockevent_device(struct clock_event_device
*dev
, int cpu
)
173 dev
->name
= "Hyper-V clockevent";
174 dev
->features
= CLOCK_EVT_FEAT_ONESHOT
;
175 dev
->cpumask
= cpumask_of(cpu
);
178 * Avoid settint dev->owner = THIS_MODULE deliberately as doing so will
179 * result in clockevents_config_and_register() taking additional
180 * references to the hv_vmbus module making it impossible to unload.
183 dev
->set_state_shutdown
= hv_ce_shutdown
;
184 dev
->set_state_oneshot
= hv_ce_set_oneshot
;
185 dev
->set_next_event
= hv_ce_set_next_event
;
189 int hv_synic_alloc(void)
193 hv_context
.hv_numa_map
= kcalloc(nr_node_ids
, sizeof(struct cpumask
),
195 if (hv_context
.hv_numa_map
== NULL
) {
196 pr_err("Unable to allocate NUMA map\n");
200 for_each_present_cpu(cpu
) {
201 struct hv_per_cpu_context
*hv_cpu
202 = per_cpu_ptr(hv_context
.cpu_context
, cpu
);
204 memset(hv_cpu
, 0, sizeof(*hv_cpu
));
205 tasklet_init(&hv_cpu
->msg_dpc
,
206 vmbus_on_msg_dpc
, (unsigned long) hv_cpu
);
208 hv_cpu
->clk_evt
= kzalloc(sizeof(struct clock_event_device
),
210 if (hv_cpu
->clk_evt
== NULL
) {
211 pr_err("Unable to allocate clock event device\n");
214 hv_init_clockevent_device(hv_cpu
->clk_evt
, cpu
);
216 hv_cpu
->synic_message_page
=
217 (void *)get_zeroed_page(GFP_ATOMIC
);
218 if (hv_cpu
->synic_message_page
== NULL
) {
219 pr_err("Unable to allocate SYNIC message page\n");
223 hv_cpu
->synic_event_page
= (void *)get_zeroed_page(GFP_ATOMIC
);
224 if (hv_cpu
->synic_event_page
== NULL
) {
225 pr_err("Unable to allocate SYNIC event page\n");
229 hv_cpu
->post_msg_page
= (void *)get_zeroed_page(GFP_ATOMIC
);
230 if (hv_cpu
->post_msg_page
== NULL
) {
231 pr_err("Unable to allocate post msg page\n");
235 INIT_LIST_HEAD(&hv_cpu
->chan_list
);
238 if (direct_mode_enabled
&&
239 hv_setup_stimer0_irq(&stimer0_irq
, &stimer0_vector
,
246 * Any memory allocations that succeeded will be freed when
247 * the caller cleans up by calling hv_synic_free()
253 void hv_synic_free(void)
257 for_each_present_cpu(cpu
) {
258 struct hv_per_cpu_context
*hv_cpu
259 = per_cpu_ptr(hv_context
.cpu_context
, cpu
);
261 kfree(hv_cpu
->clk_evt
);
262 free_page((unsigned long)hv_cpu
->synic_event_page
);
263 free_page((unsigned long)hv_cpu
->synic_message_page
);
264 free_page((unsigned long)hv_cpu
->post_msg_page
);
267 kfree(hv_context
.hv_numa_map
);
271 * hv_synic_init - Initialize the Synthetic Interrupt Controller.
273 * If it is already initialized by another entity (ie x2v shim), we need to
274 * retrieve the initialized message and event pages. Otherwise, we create and
275 * initialize the message and event pages.
277 int hv_synic_init(unsigned int cpu
)
279 struct hv_per_cpu_context
*hv_cpu
280 = per_cpu_ptr(hv_context
.cpu_context
, cpu
);
281 union hv_synic_simp simp
;
282 union hv_synic_siefp siefp
;
283 union hv_synic_sint shared_sint
;
284 union hv_synic_scontrol sctrl
;
286 /* Setup the Synic's message page */
287 hv_get_simp(simp
.as_uint64
);
288 simp
.simp_enabled
= 1;
289 simp
.base_simp_gpa
= virt_to_phys(hv_cpu
->synic_message_page
)
292 hv_set_simp(simp
.as_uint64
);
294 /* Setup the Synic's event page */
295 hv_get_siefp(siefp
.as_uint64
);
296 siefp
.siefp_enabled
= 1;
297 siefp
.base_siefp_gpa
= virt_to_phys(hv_cpu
->synic_event_page
)
300 hv_set_siefp(siefp
.as_uint64
);
302 /* Setup the shared SINT. */
303 hv_get_synint_state(HV_X64_MSR_SINT0
+ VMBUS_MESSAGE_SINT
,
304 shared_sint
.as_uint64
);
306 shared_sint
.vector
= HYPERVISOR_CALLBACK_VECTOR
;
307 shared_sint
.masked
= false;
308 if (ms_hyperv
.hints
& HV_X64_DEPRECATING_AEOI_RECOMMENDED
)
309 shared_sint
.auto_eoi
= false;
311 shared_sint
.auto_eoi
= true;
313 hv_set_synint_state(HV_X64_MSR_SINT0
+ VMBUS_MESSAGE_SINT
,
314 shared_sint
.as_uint64
);
316 /* Enable the global synic bit */
317 hv_get_synic_state(sctrl
.as_uint64
);
320 hv_set_synic_state(sctrl
.as_uint64
);
322 hv_context
.synic_initialized
= true;
325 * Register the per-cpu clockevent source.
327 if (ms_hyperv
.features
& HV_X64_MSR_SYNTIMER_AVAILABLE
)
328 clockevents_config_and_register(hv_cpu
->clk_evt
,
331 HV_MAX_MAX_DELTA_TICKS
);
336 * hv_synic_clockevents_cleanup - Cleanup clockevent devices
338 void hv_synic_clockevents_cleanup(void)
342 if (!(ms_hyperv
.features
& HV_X64_MSR_SYNTIMER_AVAILABLE
))
345 if (direct_mode_enabled
)
346 hv_remove_stimer0_irq(stimer0_irq
);
348 for_each_present_cpu(cpu
) {
349 struct hv_per_cpu_context
*hv_cpu
350 = per_cpu_ptr(hv_context
.cpu_context
, cpu
);
352 clockevents_unbind_device(hv_cpu
->clk_evt
, cpu
);
357 * hv_synic_cleanup - Cleanup routine for hv_synic_init().
359 int hv_synic_cleanup(unsigned int cpu
)
361 union hv_synic_sint shared_sint
;
362 union hv_synic_simp simp
;
363 union hv_synic_siefp siefp
;
364 union hv_synic_scontrol sctrl
;
365 struct vmbus_channel
*channel
, *sc
;
366 bool channel_found
= false;
369 if (!hv_context
.synic_initialized
)
373 * Search for channels which are bound to the CPU we're about to
374 * cleanup. In case we find one and vmbus is still connected we need to
375 * fail, this will effectively prevent CPU offlining. There is no way
376 * we can re-bind channels to different CPUs for now.
378 mutex_lock(&vmbus_connection
.channel_mutex
);
379 list_for_each_entry(channel
, &vmbus_connection
.chn_list
, listentry
) {
380 if (channel
->target_cpu
== cpu
) {
381 channel_found
= true;
384 spin_lock_irqsave(&channel
->lock
, flags
);
385 list_for_each_entry(sc
, &channel
->sc_list
, sc_list
) {
386 if (sc
->target_cpu
== cpu
) {
387 channel_found
= true;
391 spin_unlock_irqrestore(&channel
->lock
, flags
);
395 mutex_unlock(&vmbus_connection
.channel_mutex
);
397 if (channel_found
&& vmbus_connection
.conn_state
== CONNECTED
)
400 /* Turn off clockevent device */
401 if (ms_hyperv
.features
& HV_X64_MSR_SYNTIMER_AVAILABLE
) {
402 struct hv_per_cpu_context
*hv_cpu
403 = this_cpu_ptr(hv_context
.cpu_context
);
405 clockevents_unbind_device(hv_cpu
->clk_evt
, cpu
);
406 hv_ce_shutdown(hv_cpu
->clk_evt
);
410 hv_get_synint_state(HV_X64_MSR_SINT0
+ VMBUS_MESSAGE_SINT
,
411 shared_sint
.as_uint64
);
413 shared_sint
.masked
= 1;
415 /* Need to correctly cleanup in the case of SMP!!! */
416 /* Disable the interrupt */
417 hv_set_synint_state(HV_X64_MSR_SINT0
+ VMBUS_MESSAGE_SINT
,
418 shared_sint
.as_uint64
);
420 hv_get_simp(simp
.as_uint64
);
421 simp
.simp_enabled
= 0;
422 simp
.base_simp_gpa
= 0;
424 hv_set_simp(simp
.as_uint64
);
426 hv_get_siefp(siefp
.as_uint64
);
427 siefp
.siefp_enabled
= 0;
428 siefp
.base_siefp_gpa
= 0;
430 hv_set_siefp(siefp
.as_uint64
);
432 /* Disable the global synic bit */
433 hv_get_synic_state(sctrl
.as_uint64
);
435 hv_set_synic_state(sctrl
.as_uint64
);