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/interrupt.h>
31 #include <linux/clockchips.h>
32 #include <asm/hyperv.h>
33 #include <asm/mshyperv.h>
34 #include "hyperv_vmbus.h"
36 /* The one and only */
37 struct hv_context hv_context
= {
38 .synic_initialized
= false,
41 #define HV_TIMER_FREQUENCY (10 * 1000 * 1000) /* 100ns period */
42 #define HV_MAX_MAX_DELTA_TICKS 0xffffffff
43 #define HV_MIN_DELTA_TICKS 1
46 * hv_init - Main initialization routine.
48 * This routine must be called before any other routines in here are called
52 if (!hv_is_hypercall_page_setup())
55 hv_context
.cpu_context
= alloc_percpu(struct hv_per_cpu_context
);
56 if (!hv_context
.cpu_context
)
63 * hv_post_message - Post a message using the hypervisor message IPC.
65 * This involves a hypercall.
67 int hv_post_message(union hv_connection_id connection_id
,
68 enum hv_message_type message_type
,
69 void *payload
, size_t payload_size
)
71 struct hv_input_post_message
*aligned_msg
;
72 struct hv_per_cpu_context
*hv_cpu
;
75 if (payload_size
> HV_MESSAGE_PAYLOAD_BYTE_COUNT
)
78 hv_cpu
= get_cpu_ptr(hv_context
.cpu_context
);
79 aligned_msg
= hv_cpu
->post_msg_page
;
80 aligned_msg
->connectionid
= connection_id
;
81 aligned_msg
->reserved
= 0;
82 aligned_msg
->message_type
= message_type
;
83 aligned_msg
->payload_size
= payload_size
;
84 memcpy((void *)aligned_msg
->payload
, payload
, payload_size
);
86 status
= hv_do_hypercall(HVCALL_POST_MESSAGE
, aligned_msg
, NULL
);
88 /* Preemption must remain disabled until after the hypercall
89 * so some other thread can't get scheduled onto this cpu and
90 * corrupt the per-cpu post_msg_page
94 return status
& 0xFFFF;
97 static int hv_ce_set_next_event(unsigned long delta
,
98 struct clock_event_device
*evt
)
102 WARN_ON(!clockevent_state_oneshot(evt
));
104 current_tick
= hyperv_cs
->read(NULL
);
105 current_tick
+= delta
;
106 hv_init_timer(HV_X64_MSR_STIMER0_COUNT
, current_tick
);
110 static int hv_ce_shutdown(struct clock_event_device
*evt
)
112 hv_init_timer(HV_X64_MSR_STIMER0_COUNT
, 0);
113 hv_init_timer_config(HV_X64_MSR_STIMER0_CONFIG
, 0);
118 static int hv_ce_set_oneshot(struct clock_event_device
*evt
)
120 union hv_timer_config timer_cfg
;
122 timer_cfg
.enable
= 1;
123 timer_cfg
.auto_enable
= 1;
124 timer_cfg
.sintx
= VMBUS_MESSAGE_SINT
;
125 hv_init_timer_config(HV_X64_MSR_STIMER0_CONFIG
, timer_cfg
.as_uint64
);
130 static void hv_init_clockevent_device(struct clock_event_device
*dev
, int cpu
)
132 dev
->name
= "Hyper-V clockevent";
133 dev
->features
= CLOCK_EVT_FEAT_ONESHOT
;
134 dev
->cpumask
= cpumask_of(cpu
);
137 * Avoid settint dev->owner = THIS_MODULE deliberately as doing so will
138 * result in clockevents_config_and_register() taking additional
139 * references to the hv_vmbus module making it impossible to unload.
142 dev
->set_state_shutdown
= hv_ce_shutdown
;
143 dev
->set_state_oneshot
= hv_ce_set_oneshot
;
144 dev
->set_next_event
= hv_ce_set_next_event
;
148 int hv_synic_alloc(void)
152 hv_context
.hv_numa_map
= kzalloc(sizeof(struct cpumask
) * nr_node_ids
,
154 if (hv_context
.hv_numa_map
== NULL
) {
155 pr_err("Unable to allocate NUMA map\n");
159 for_each_present_cpu(cpu
) {
160 struct hv_per_cpu_context
*hv_cpu
161 = per_cpu_ptr(hv_context
.cpu_context
, cpu
);
163 memset(hv_cpu
, 0, sizeof(*hv_cpu
));
164 tasklet_init(&hv_cpu
->msg_dpc
,
165 vmbus_on_msg_dpc
, (unsigned long) hv_cpu
);
167 hv_cpu
->clk_evt
= kzalloc(sizeof(struct clock_event_device
),
169 if (hv_cpu
->clk_evt
== NULL
) {
170 pr_err("Unable to allocate clock event device\n");
173 hv_init_clockevent_device(hv_cpu
->clk_evt
, cpu
);
175 hv_cpu
->synic_message_page
=
176 (void *)get_zeroed_page(GFP_ATOMIC
);
177 if (hv_cpu
->synic_message_page
== NULL
) {
178 pr_err("Unable to allocate SYNIC message page\n");
182 hv_cpu
->synic_event_page
= (void *)get_zeroed_page(GFP_ATOMIC
);
183 if (hv_cpu
->synic_event_page
== NULL
) {
184 pr_err("Unable to allocate SYNIC event page\n");
188 hv_cpu
->post_msg_page
= (void *)get_zeroed_page(GFP_ATOMIC
);
189 if (hv_cpu
->post_msg_page
== NULL
) {
190 pr_err("Unable to allocate post msg page\n");
194 INIT_LIST_HEAD(&hv_cpu
->chan_list
);
203 void hv_synic_free(void)
207 for_each_present_cpu(cpu
) {
208 struct hv_per_cpu_context
*hv_cpu
209 = per_cpu_ptr(hv_context
.cpu_context
, cpu
);
211 if (hv_cpu
->synic_event_page
)
212 free_page((unsigned long)hv_cpu
->synic_event_page
);
213 if (hv_cpu
->synic_message_page
)
214 free_page((unsigned long)hv_cpu
->synic_message_page
);
215 if (hv_cpu
->post_msg_page
)
216 free_page((unsigned long)hv_cpu
->post_msg_page
);
219 kfree(hv_context
.hv_numa_map
);
223 * hv_synic_init - Initialize the Synthethic Interrupt Controller.
225 * If it is already initialized by another entity (ie x2v shim), we need to
226 * retrieve the initialized message and event pages. Otherwise, we create and
227 * initialize the message and event pages.
229 int hv_synic_init(unsigned int cpu
)
231 struct hv_per_cpu_context
*hv_cpu
232 = per_cpu_ptr(hv_context
.cpu_context
, cpu
);
233 union hv_synic_simp simp
;
234 union hv_synic_siefp siefp
;
235 union hv_synic_sint shared_sint
;
236 union hv_synic_scontrol sctrl
;
238 /* Setup the Synic's message page */
239 hv_get_simp(simp
.as_uint64
);
240 simp
.simp_enabled
= 1;
241 simp
.base_simp_gpa
= virt_to_phys(hv_cpu
->synic_message_page
)
244 hv_set_simp(simp
.as_uint64
);
246 /* Setup the Synic's event page */
247 hv_get_siefp(siefp
.as_uint64
);
248 siefp
.siefp_enabled
= 1;
249 siefp
.base_siefp_gpa
= virt_to_phys(hv_cpu
->synic_event_page
)
252 hv_set_siefp(siefp
.as_uint64
);
254 /* Setup the shared SINT. */
255 hv_get_synint_state(HV_X64_MSR_SINT0
+ VMBUS_MESSAGE_SINT
,
256 shared_sint
.as_uint64
);
258 shared_sint
.as_uint64
= 0;
259 shared_sint
.vector
= HYPERVISOR_CALLBACK_VECTOR
;
260 shared_sint
.masked
= false;
261 if (ms_hyperv
.hints
& HV_X64_DEPRECATING_AEOI_RECOMMENDED
)
262 shared_sint
.auto_eoi
= false;
264 shared_sint
.auto_eoi
= true;
266 hv_set_synint_state(HV_X64_MSR_SINT0
+ VMBUS_MESSAGE_SINT
,
267 shared_sint
.as_uint64
);
269 /* Enable the global synic bit */
270 hv_get_synic_state(sctrl
.as_uint64
);
273 hv_set_synic_state(sctrl
.as_uint64
);
275 hv_context
.synic_initialized
= true;
278 * Register the per-cpu clockevent source.
280 if (ms_hyperv
.features
& HV_X64_MSR_SYNTIMER_AVAILABLE
)
281 clockevents_config_and_register(hv_cpu
->clk_evt
,
284 HV_MAX_MAX_DELTA_TICKS
);
289 * hv_synic_clockevents_cleanup - Cleanup clockevent devices
291 void hv_synic_clockevents_cleanup(void)
295 if (!(ms_hyperv
.features
& HV_X64_MSR_SYNTIMER_AVAILABLE
))
298 for_each_present_cpu(cpu
) {
299 struct hv_per_cpu_context
*hv_cpu
300 = per_cpu_ptr(hv_context
.cpu_context
, cpu
);
302 clockevents_unbind_device(hv_cpu
->clk_evt
, cpu
);
307 * hv_synic_cleanup - Cleanup routine for hv_synic_init().
309 int hv_synic_cleanup(unsigned int cpu
)
311 union hv_synic_sint shared_sint
;
312 union hv_synic_simp simp
;
313 union hv_synic_siefp siefp
;
314 union hv_synic_scontrol sctrl
;
315 struct vmbus_channel
*channel
, *sc
;
316 bool channel_found
= false;
319 if (!hv_context
.synic_initialized
)
323 * Search for channels which are bound to the CPU we're about to
324 * cleanup. In case we find one and vmbus is still connected we need to
325 * fail, this will effectively prevent CPU offlining. There is no way
326 * we can re-bind channels to different CPUs for now.
328 mutex_lock(&vmbus_connection
.channel_mutex
);
329 list_for_each_entry(channel
, &vmbus_connection
.chn_list
, listentry
) {
330 if (channel
->target_cpu
== cpu
) {
331 channel_found
= true;
334 spin_lock_irqsave(&channel
->lock
, flags
);
335 list_for_each_entry(sc
, &channel
->sc_list
, sc_list
) {
336 if (sc
->target_cpu
== cpu
) {
337 channel_found
= true;
341 spin_unlock_irqrestore(&channel
->lock
, flags
);
345 mutex_unlock(&vmbus_connection
.channel_mutex
);
347 if (channel_found
&& vmbus_connection
.conn_state
== CONNECTED
)
350 /* Turn off clockevent device */
351 if (ms_hyperv
.features
& HV_X64_MSR_SYNTIMER_AVAILABLE
) {
352 struct hv_per_cpu_context
*hv_cpu
353 = this_cpu_ptr(hv_context
.cpu_context
);
355 clockevents_unbind_device(hv_cpu
->clk_evt
, cpu
);
356 hv_ce_shutdown(hv_cpu
->clk_evt
);
360 hv_get_synint_state(HV_X64_MSR_SINT0
+ VMBUS_MESSAGE_SINT
,
361 shared_sint
.as_uint64
);
363 shared_sint
.masked
= 1;
365 /* Need to correctly cleanup in the case of SMP!!! */
366 /* Disable the interrupt */
367 hv_set_synint_state(HV_X64_MSR_SINT0
+ VMBUS_MESSAGE_SINT
,
368 shared_sint
.as_uint64
);
370 hv_get_simp(simp
.as_uint64
);
371 simp
.simp_enabled
= 0;
372 simp
.base_simp_gpa
= 0;
374 hv_set_simp(simp
.as_uint64
);
376 hv_get_siefp(siefp
.as_uint64
);
377 siefp
.siefp_enabled
= 0;
378 siefp
.base_siefp_gpa
= 0;
380 hv_set_siefp(siefp
.as_uint64
);
382 /* Disable the global synic bit */
383 hv_get_synic_state(sctrl
.as_uint64
);
385 hv_set_synic_state(sctrl
.as_uint64
);