aboutsummaryrefslogtreecommitdiffstats
path: root/virt
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@linux-foundation.org>2013-07-03 16:21:40 -0400
committerLinus Torvalds <torvalds@linux-foundation.org>2013-07-03 16:21:40 -0400
commitfe489bf4505ae26d3c6d6a1f1d3064c2a9c5cd85 (patch)
tree46596fd7edf7c4da1dafdb2c62011841e71cf32d /virt
parent3e34131a65127e73fbae68c82748f32c8af7e4a4 (diff)
parenta3ff5fbc94a829680d4aa005cd17add1c1a1fb5b (diff)
Merge tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm
Pull KVM fixes from Paolo Bonzini: "On the x86 side, there are some optimizations and documentation updates. The big ARM/KVM change for 3.11, support for AArch64, will come through Catalin Marinas's tree. s390 and PPC have misc cleanups and bugfixes" * tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm: (87 commits) KVM: PPC: Ignore PIR writes KVM: PPC: Book3S PR: Invalidate SLB entries properly KVM: PPC: Book3S PR: Allow guest to use 1TB segments KVM: PPC: Book3S PR: Don't keep scanning HPTEG after we find a match KVM: PPC: Book3S PR: Fix invalidation of SLB entry 0 on guest entry KVM: PPC: Book3S PR: Fix proto-VSID calculations KVM: PPC: Guard doorbell exception with CONFIG_PPC_DOORBELL KVM: Fix RTC interrupt coalescing tracking kvm: Add a tracepoint write_tsc_offset KVM: MMU: Inform users of mmio generation wraparound KVM: MMU: document fast invalidate all mmio sptes KVM: MMU: document fast invalidate all pages KVM: MMU: document fast page fault KVM: MMU: document mmio page fault KVM: MMU: document write_flooding_count KVM: MMU: document clear_spte_count KVM: MMU: drop kvm_mmu_zap_mmio_sptes KVM: MMU: init kvm generation close to mmio wrap-around value KVM: MMU: add tracepoint for check_mmio_spte KVM: MMU: fast invalidate all mmio sptes ...
Diffstat (limited to 'virt')
-rw-r--r--virt/kvm/arm/arch_timer.c284
-rw-r--r--virt/kvm/arm/vgic.c1499
-rw-r--r--virt/kvm/eventfd.c2
-rw-r--r--virt/kvm/kvm_main.c4
4 files changed, 1788 insertions, 1 deletions
diff --git a/virt/kvm/arm/arch_timer.c b/virt/kvm/arm/arch_timer.c
new file mode 100644
index 000000000000..c2e1ef4604e8
--- /dev/null
+++ b/virt/kvm/arm/arch_timer.c
@@ -0,0 +1,284 @@
1/*
2 * Copyright (C) 2012 ARM Ltd.
3 * Author: Marc Zyngier <marc.zyngier@arm.com>
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License version 2 as
7 * published by the Free Software Foundation.
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, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17 */
18
19#include <linux/cpu.h>
20#include <linux/of_irq.h>
21#include <linux/kvm.h>
22#include <linux/kvm_host.h>
23#include <linux/interrupt.h>
24
25#include <clocksource/arm_arch_timer.h>
26#include <asm/arch_timer.h>
27
28#include <kvm/arm_vgic.h>
29#include <kvm/arm_arch_timer.h>
30
31static struct timecounter *timecounter;
32static struct workqueue_struct *wqueue;
33static unsigned int host_vtimer_irq;
34
35static cycle_t kvm_phys_timer_read(void)
36{
37 return timecounter->cc->read(timecounter->cc);
38}
39
40static bool timer_is_armed(struct arch_timer_cpu *timer)
41{
42 return timer->armed;
43}
44
45/* timer_arm: as in "arm the timer", not as in ARM the company */
46static void timer_arm(struct arch_timer_cpu *timer, u64 ns)
47{
48 timer->armed = true;
49 hrtimer_start(&timer->timer, ktime_add_ns(ktime_get(), ns),
50 HRTIMER_MODE_ABS);
51}
52
53static void timer_disarm(struct arch_timer_cpu *timer)
54{
55 if (timer_is_armed(timer)) {
56 hrtimer_cancel(&timer->timer);
57 cancel_work_sync(&timer->expired);
58 timer->armed = false;
59 }
60}
61
62static void kvm_timer_inject_irq(struct kvm_vcpu *vcpu)
63{
64 struct arch_timer_cpu *timer = &vcpu->arch.timer_cpu;
65
66 timer->cntv_ctl |= ARCH_TIMER_CTRL_IT_MASK;
67 kvm_vgic_inject_irq(vcpu->kvm, vcpu->vcpu_id,
68 timer->irq->irq,
69 timer->irq->level);
70}
71
72static irqreturn_t kvm_arch_timer_handler(int irq, void *dev_id)
73{
74 struct kvm_vcpu *vcpu = *(struct kvm_vcpu **)dev_id;
75
76 /*
77 * We disable the timer in the world switch and let it be
78 * handled by kvm_timer_sync_hwstate(). Getting a timer
79 * interrupt at this point is a sure sign of some major
80 * breakage.
81 */
82 pr_warn("Unexpected interrupt %d on vcpu %p\n", irq, vcpu);
83 return IRQ_HANDLED;
84}
85
86static void kvm_timer_inject_irq_work(struct work_struct *work)
87{
88 struct kvm_vcpu *vcpu;
89
90 vcpu = container_of(work, struct kvm_vcpu, arch.timer_cpu.expired);
91 vcpu->arch.timer_cpu.armed = false;
92 kvm_timer_inject_irq(vcpu);
93}
94
95static enum hrtimer_restart kvm_timer_expire(struct hrtimer *hrt)
96{
97 struct arch_timer_cpu *timer;
98 timer = container_of(hrt, struct arch_timer_cpu, timer);
99 queue_work(wqueue, &timer->expired);
100 return HRTIMER_NORESTART;
101}
102
103/**
104 * kvm_timer_flush_hwstate - prepare to move the virt timer to the cpu
105 * @vcpu: The vcpu pointer
106 *
107 * Disarm any pending soft timers, since the world-switch code will write the
108 * virtual timer state back to the physical CPU.
109 */
110void kvm_timer_flush_hwstate(struct kvm_vcpu *vcpu)
111{
112 struct arch_timer_cpu *timer = &vcpu->arch.timer_cpu;
113
114 /*
115 * We're about to run this vcpu again, so there is no need to
116 * keep the background timer running, as we're about to
117 * populate the CPU timer again.
118 */
119 timer_disarm(timer);
120}
121
122/**
123 * kvm_timer_sync_hwstate - sync timer state from cpu
124 * @vcpu: The vcpu pointer
125 *
126 * Check if the virtual timer was armed and either schedule a corresponding
127 * soft timer or inject directly if already expired.
128 */
129void kvm_timer_sync_hwstate(struct kvm_vcpu *vcpu)
130{
131 struct arch_timer_cpu *timer = &vcpu->arch.timer_cpu;
132 cycle_t cval, now;
133 u64 ns;
134
135 if ((timer->cntv_ctl & ARCH_TIMER_CTRL_IT_MASK) ||
136 !(timer->cntv_ctl & ARCH_TIMER_CTRL_ENABLE))
137 return;
138
139 cval = timer->cntv_cval;
140 now = kvm_phys_timer_read() - vcpu->kvm->arch.timer.cntvoff;
141
142 BUG_ON(timer_is_armed(timer));
143
144 if (cval <= now) {
145 /*
146 * Timer has already expired while we were not
147 * looking. Inject the interrupt and carry on.
148 */
149 kvm_timer_inject_irq(vcpu);
150 return;
151 }
152
153 ns = cyclecounter_cyc2ns(timecounter->cc, cval - now);
154 timer_arm(timer, ns);
155}
156
157void kvm_timer_vcpu_reset(struct kvm_vcpu *vcpu,
158 const struct kvm_irq_level *irq)
159{
160 struct arch_timer_cpu *timer = &vcpu->arch.timer_cpu;
161
162 /*
163 * The vcpu timer irq number cannot be determined in
164 * kvm_timer_vcpu_init() because it is called much before
165 * kvm_vcpu_set_target(). To handle this, we determine
166 * vcpu timer irq number when the vcpu is reset.
167 */
168 timer->irq = irq;
169}
170
171void kvm_timer_vcpu_init(struct kvm_vcpu *vcpu)
172{
173 struct arch_timer_cpu *timer = &vcpu->arch.timer_cpu;
174
175 INIT_WORK(&timer->expired, kvm_timer_inject_irq_work);
176 hrtimer_init(&timer->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
177 timer->timer.function = kvm_timer_expire;
178}
179
180static void kvm_timer_init_interrupt(void *info)
181{
182 enable_percpu_irq(host_vtimer_irq, 0);
183}
184
185
186static int kvm_timer_cpu_notify(struct notifier_block *self,
187 unsigned long action, void *cpu)
188{
189 switch (action) {
190 case CPU_STARTING:
191 case CPU_STARTING_FROZEN:
192 kvm_timer_init_interrupt(NULL);
193 break;
194 case CPU_DYING:
195 case CPU_DYING_FROZEN:
196 disable_percpu_irq(host_vtimer_irq);
197 break;
198 }
199
200 return NOTIFY_OK;
201}
202
203static struct notifier_block kvm_timer_cpu_nb = {
204 .notifier_call = kvm_timer_cpu_notify,
205};
206
207static const struct of_device_id arch_timer_of_match[] = {
208 { .compatible = "arm,armv7-timer", },
209 { .compatible = "arm,armv8-timer", },
210 {},
211};
212
213int kvm_timer_hyp_init(void)
214{
215 struct device_node *np;
216 unsigned int ppi;
217 int err;
218
219 timecounter = arch_timer_get_timecounter();
220 if (!timecounter)
221 return -ENODEV;
222
223 np = of_find_matching_node(NULL, arch_timer_of_match);
224 if (!np) {
225 kvm_err("kvm_arch_timer: can't find DT node\n");
226 return -ENODEV;
227 }
228
229 ppi = irq_of_parse_and_map(np, 2);
230 if (!ppi) {
231 kvm_err("kvm_arch_timer: no virtual timer interrupt\n");
232 err = -EINVAL;
233 goto out;
234 }
235
236 err = request_percpu_irq(ppi, kvm_arch_timer_handler,
237 "kvm guest timer", kvm_get_running_vcpus());
238 if (err) {
239 kvm_err("kvm_arch_timer: can't request interrupt %d (%d)\n",
240 ppi, err);
241 goto out;
242 }
243
244 host_vtimer_irq = ppi;
245
246 err = register_cpu_notifier(&kvm_timer_cpu_nb);
247 if (err) {
248 kvm_err("Cannot register timer CPU notifier\n");
249 goto out_free;
250 }
251
252 wqueue = create_singlethread_workqueue("kvm_arch_timer");
253 if (!wqueue) {
254 err = -ENOMEM;
255 goto out_free;
256 }
257
258 kvm_info("%s IRQ%d\n", np->name, ppi);
259 on_each_cpu(kvm_timer_init_interrupt, NULL, 1);
260
261 goto out;
262out_free:
263 free_percpu_irq(ppi, kvm_get_running_vcpus());
264out:
265 of_node_put(np);
266 return err;
267}
268
269void kvm_timer_vcpu_terminate(struct kvm_vcpu *vcpu)
270{
271 struct arch_timer_cpu *timer = &vcpu->arch.timer_cpu;
272
273 timer_disarm(timer);
274}
275
276int kvm_timer_init(struct kvm *kvm)
277{
278 if (timecounter && wqueue) {
279 kvm->arch.timer.cntvoff = kvm_phys_timer_read();
280 kvm->arch.timer.enabled = 1;
281 }
282
283 return 0;
284}
diff --git a/virt/kvm/arm/vgic.c b/virt/kvm/arm/vgic.c
new file mode 100644
index 000000000000..17c5ac7d10ed
--- /dev/null
+++ b/virt/kvm/arm/vgic.c
@@ -0,0 +1,1499 @@
1/*
2 * Copyright (C) 2012 ARM Ltd.
3 * Author: Marc Zyngier <marc.zyngier@arm.com>
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License version 2 as
7 * published by the Free Software Foundation.
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, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17 */
18
19#include <linux/cpu.h>
20#include <linux/kvm.h>
21#include <linux/kvm_host.h>
22#include <linux/interrupt.h>
23#include <linux/io.h>
24#include <linux/of.h>
25#include <linux/of_address.h>
26#include <linux/of_irq.h>
27
28#include <linux/irqchip/arm-gic.h>
29
30#include <asm/kvm_emulate.h>
31#include <asm/kvm_arm.h>
32#include <asm/kvm_mmu.h>
33
34/*
35 * How the whole thing works (courtesy of Christoffer Dall):
36 *
37 * - At any time, the dist->irq_pending_on_cpu is the oracle that knows if
38 * something is pending
39 * - VGIC pending interrupts are stored on the vgic.irq_state vgic
40 * bitmap (this bitmap is updated by both user land ioctls and guest
41 * mmio ops, and other in-kernel peripherals such as the
42 * arch. timers) and indicate the 'wire' state.
43 * - Every time the bitmap changes, the irq_pending_on_cpu oracle is
44 * recalculated
45 * - To calculate the oracle, we need info for each cpu from
46 * compute_pending_for_cpu, which considers:
47 * - PPI: dist->irq_state & dist->irq_enable
48 * - SPI: dist->irq_state & dist->irq_enable & dist->irq_spi_target
49 * - irq_spi_target is a 'formatted' version of the GICD_ICFGR
50 * registers, stored on each vcpu. We only keep one bit of
51 * information per interrupt, making sure that only one vcpu can
52 * accept the interrupt.
53 * - The same is true when injecting an interrupt, except that we only
54 * consider a single interrupt at a time. The irq_spi_cpu array
55 * contains the target CPU for each SPI.
56 *
57 * The handling of level interrupts adds some extra complexity. We
58 * need to track when the interrupt has been EOIed, so we can sample
59 * the 'line' again. This is achieved as such:
60 *
61 * - When a level interrupt is moved onto a vcpu, the corresponding
62 * bit in irq_active is set. As long as this bit is set, the line
63 * will be ignored for further interrupts. The interrupt is injected
64 * into the vcpu with the GICH_LR_EOI bit set (generate a
65 * maintenance interrupt on EOI).
66 * - When the interrupt is EOIed, the maintenance interrupt fires,
67 * and clears the corresponding bit in irq_active. This allow the
68 * interrupt line to be sampled again.
69 */
70
71#define VGIC_ADDR_UNDEF (-1)
72#define IS_VGIC_ADDR_UNDEF(_x) ((_x) == VGIC_ADDR_UNDEF)
73
74/* Physical address of vgic virtual cpu interface */
75static phys_addr_t vgic_vcpu_base;
76
77/* Virtual control interface base address */
78static void __iomem *vgic_vctrl_base;
79
80static struct device_node *vgic_node;
81
82#define ACCESS_READ_VALUE (1 << 0)
83#define ACCESS_READ_RAZ (0 << 0)
84#define ACCESS_READ_MASK(x) ((x) & (1 << 0))
85#define ACCESS_WRITE_IGNORED (0 << 1)
86#define ACCESS_WRITE_SETBIT (1 << 1)
87#define ACCESS_WRITE_CLEARBIT (2 << 1)
88#define ACCESS_WRITE_VALUE (3 << 1)
89#define ACCESS_WRITE_MASK(x) ((x) & (3 << 1))
90
91static void vgic_retire_disabled_irqs(struct kvm_vcpu *vcpu);
92static void vgic_update_state(struct kvm *kvm);
93static void vgic_kick_vcpus(struct kvm *kvm);
94static void vgic_dispatch_sgi(struct kvm_vcpu *vcpu, u32 reg);
95static u32 vgic_nr_lr;
96
97static unsigned int vgic_maint_irq;
98
99static u32 *vgic_bitmap_get_reg(struct vgic_bitmap *x,
100 int cpuid, u32 offset)
101{
102 offset >>= 2;
103 if (!offset)
104 return x->percpu[cpuid].reg;
105 else
106 return x->shared.reg + offset - 1;
107}
108
109static int vgic_bitmap_get_irq_val(struct vgic_bitmap *x,
110 int cpuid, int irq)
111{
112 if (irq < VGIC_NR_PRIVATE_IRQS)
113 return test_bit(irq, x->percpu[cpuid].reg_ul);
114
115 return test_bit(irq - VGIC_NR_PRIVATE_IRQS, x->shared.reg_ul);
116}
117
118static void vgic_bitmap_set_irq_val(struct vgic_bitmap *x, int cpuid,
119 int irq, int val)
120{
121 unsigned long *reg;
122
123 if (irq < VGIC_NR_PRIVATE_IRQS) {
124 reg = x->percpu[cpuid].reg_ul;
125 } else {
126 reg = x->shared.reg_ul;
127 irq -= VGIC_NR_PRIVATE_IRQS;
128 }
129
130 if (val)
131 set_bit(irq, reg);
132 else
133 clear_bit(irq, reg);
134}
135
136static unsigned long *vgic_bitmap_get_cpu_map(struct vgic_bitmap *x, int cpuid)
137{
138 if (unlikely(cpuid >= VGIC_MAX_CPUS))
139 return NULL;
140 return x->percpu[cpuid].reg_ul;
141}
142
143static unsigned long *vgic_bitmap_get_shared_map(struct vgic_bitmap *x)
144{
145 return x->shared.reg_ul;
146}
147
148static u32 *vgic_bytemap_get_reg(struct vgic_bytemap *x, int cpuid, u32 offset)
149{
150 offset >>= 2;
151 BUG_ON(offset > (VGIC_NR_IRQS / 4));
152 if (offset < 4)
153 return x->percpu[cpuid] + offset;
154 else
155 return x->shared + offset - 8;
156}
157
158#define VGIC_CFG_LEVEL 0
159#define VGIC_CFG_EDGE 1
160
161static bool vgic_irq_is_edge(struct kvm_vcpu *vcpu, int irq)
162{
163 struct vgic_dist *dist = &vcpu->kvm->arch.vgic;
164 int irq_val;
165
166 irq_val = vgic_bitmap_get_irq_val(&dist->irq_cfg, vcpu->vcpu_id, irq);
167 return irq_val == VGIC_CFG_EDGE;
168}
169
170static int vgic_irq_is_enabled(struct kvm_vcpu *vcpu, int irq)
171{
172 struct vgic_dist *dist = &vcpu->kvm->arch.vgic;
173
174 return vgic_bitmap_get_irq_val(&dist->irq_enabled, vcpu->vcpu_id, irq);
175}
176
177static int vgic_irq_is_active(struct kvm_vcpu *vcpu, int irq)
178{
179 struct vgic_dist *dist = &vcpu->kvm->arch.vgic;
180
181 return vgic_bitmap_get_irq_val(&dist->irq_active, vcpu->vcpu_id, irq);
182}
183
184static void vgic_irq_set_active(struct kvm_vcpu *vcpu, int irq)
185{
186 struct vgic_dist *dist = &vcpu->kvm->arch.vgic;
187
188 vgic_bitmap_set_irq_val(&dist->irq_active, vcpu->vcpu_id, irq, 1);
189}
190
191static void vgic_irq_clear_active(struct kvm_vcpu *vcpu, int irq)
192{
193 struct vgic_dist *dist = &vcpu->kvm->arch.vgic;
194
195 vgic_bitmap_set_irq_val(&dist->irq_active, vcpu->vcpu_id, irq, 0);
196}
197
198static int vgic_dist_irq_is_pending(struct kvm_vcpu *vcpu, int irq)
199{
200 struct vgic_dist *dist = &vcpu->kvm->arch.vgic;
201
202 return vgic_bitmap_get_irq_val(&dist->irq_state, vcpu->vcpu_id, irq);
203}
204
205static void vgic_dist_irq_set(struct kvm_vcpu *vcpu, int irq)
206{
207 struct vgic_dist *dist = &vcpu->kvm->arch.vgic;
208
209 vgic_bitmap_set_irq_val(&dist->irq_state, vcpu->vcpu_id, irq, 1);
210}
211
212static void vgic_dist_irq_clear(struct kvm_vcpu *vcpu, int irq)
213{
214 struct vgic_dist *dist = &vcpu->kvm->arch.vgic;
215
216 vgic_bitmap_set_irq_val(&dist->irq_state, vcpu->vcpu_id, irq, 0);
217}
218
219static void vgic_cpu_irq_set(struct kvm_vcpu *vcpu, int irq)
220{
221 if (irq < VGIC_NR_PRIVATE_IRQS)
222 set_bit(irq, vcpu->arch.vgic_cpu.pending_percpu);
223 else
224 set_bit(irq - VGIC_NR_PRIVATE_IRQS,
225 vcpu->arch.vgic_cpu.pending_shared);
226}
227
228static void vgic_cpu_irq_clear(struct kvm_vcpu *vcpu, int irq)
229{
230 if (irq < VGIC_NR_PRIVATE_IRQS)
231 clear_bit(irq, vcpu->arch.vgic_cpu.pending_percpu);
232 else
233 clear_bit(irq - VGIC_NR_PRIVATE_IRQS,
234 vcpu->arch.vgic_cpu.pending_shared);
235}
236
237static u32 mmio_data_read(struct kvm_exit_mmio *mmio, u32 mask)
238{
239 return *((u32 *)mmio->data) & mask;
240}
241
242static void mmio_data_write(struct kvm_exit_mmio *mmio, u32 mask, u32 value)
243{
244 *((u32 *)mmio->data) = value & mask;
245}
246
247/**
248 * vgic_reg_access - access vgic register
249 * @mmio: pointer to the data describing the mmio access
250 * @reg: pointer to the virtual backing of vgic distributor data
251 * @offset: least significant 2 bits used for word offset
252 * @mode: ACCESS_ mode (see defines above)
253 *
254 * Helper to make vgic register access easier using one of the access
255 * modes defined for vgic register access
256 * (read,raz,write-ignored,setbit,clearbit,write)
257 */
258static void vgic_reg_access(struct kvm_exit_mmio *mmio, u32 *reg,
259 phys_addr_t offset, int mode)
260{
261 int word_offset = (offset & 3) * 8;
262 u32 mask = (1UL << (mmio->len * 8)) - 1;
263 u32 regval;
264
265 /*
266 * Any alignment fault should have been delivered to the guest
267 * directly (ARM ARM B3.12.7 "Prioritization of aborts").
268 */
269
270 if (reg) {
271 regval = *reg;
272 } else {
273 BUG_ON(mode != (ACCESS_READ_RAZ | ACCESS_WRITE_IGNORED));
274 regval = 0;
275 }
276
277 if (mmio->is_write) {
278 u32 data = mmio_data_read(mmio, mask) << word_offset;
279 switch (ACCESS_WRITE_MASK(mode)) {
280 case ACCESS_WRITE_IGNORED:
281 return;
282
283 case ACCESS_WRITE_SETBIT:
284 regval |= data;
285 break;
286
287 case ACCESS_WRITE_CLEARBIT:
288 regval &= ~data;
289 break;
290
291 case ACCESS_WRITE_VALUE:
292 regval = (regval & ~(mask << word_offset)) | data;
293 break;
294 }
295 *reg = regval;
296 } else {
297 switch (ACCESS_READ_MASK(mode)) {
298 case ACCESS_READ_RAZ:
299 regval = 0;
300 /* fall through */
301
302 case ACCESS_READ_VALUE:
303 mmio_data_write(mmio, mask, regval >> word_offset);
304 }
305 }
306}
307
308static bool handle_mmio_misc(struct kvm_vcpu *vcpu,
309 struct kvm_exit_mmio *mmio, phys_addr_t offset)
310{
311 u32 reg;
312 u32 word_offset = offset & 3;
313
314 switch (offset & ~3) {
315 case 0: /* CTLR */
316 reg = vcpu->kvm->arch.vgic.enabled;
317 vgic_reg_access(mmio, &reg, word_offset,
318 ACCESS_READ_VALUE | ACCESS_WRITE_VALUE);
319 if (mmio->is_write) {
320 vcpu->kvm->arch.vgic.enabled = reg & 1;
321 vgic_update_state(vcpu->kvm);
322 return true;
323 }
324 break;
325
326 case 4: /* TYPER */
327 reg = (atomic_read(&vcpu->kvm->online_vcpus) - 1) << 5;
328 reg |= (VGIC_NR_IRQS >> 5) - 1;
329 vgic_reg_access(mmio, &reg, word_offset,
330 ACCESS_READ_VALUE | ACCESS_WRITE_IGNORED);
331 break;
332
333 case 8: /* IIDR */
334 reg = 0x4B00043B;
335 vgic_reg_access(mmio, &reg, word_offset,
336 ACCESS_READ_VALUE | ACCESS_WRITE_IGNORED);
337 break;
338 }
339
340 return false;
341}
342
343static bool handle_mmio_raz_wi(struct kvm_vcpu *vcpu,
344 struct kvm_exit_mmio *mmio, phys_addr_t offset)
345{
346 vgic_reg_access(mmio, NULL, offset,
347 ACCESS_READ_RAZ | ACCESS_WRITE_IGNORED);
348 return false;
349}
350
351static bool handle_mmio_set_enable_reg(struct kvm_vcpu *vcpu,
352 struct kvm_exit_mmio *mmio,
353 phys_addr_t offset)
354{
355 u32 *reg = vgic_bitmap_get_reg(&vcpu->kvm->arch.vgic.irq_enabled,
356 vcpu->vcpu_id, offset);
357 vgic_reg_access(mmio, reg, offset,
358 ACCESS_READ_VALUE | ACCESS_WRITE_SETBIT);
359 if (mmio->is_write) {
360 vgic_update_state(vcpu->kvm);
361 return true;
362 }
363
364 return false;
365}
366
367static bool handle_mmio_clear_enable_reg(struct kvm_vcpu *vcpu,
368 struct kvm_exit_mmio *mmio,
369 phys_addr_t offset)
370{
371 u32 *reg = vgic_bitmap_get_reg(&vcpu->kvm->arch.vgic.irq_enabled,
372 vcpu->vcpu_id, offset);
373 vgic_reg_access(mmio, reg, offset,
374 ACCESS_READ_VALUE | ACCESS_WRITE_CLEARBIT);
375 if (mmio->is_write) {
376 if (offset < 4) /* Force SGI enabled */
377 *reg |= 0xffff;
378 vgic_retire_disabled_irqs(vcpu);
379 vgic_update_state(vcpu->kvm);
380 return true;
381 }
382
383 return false;
384}
385
386static bool handle_mmio_set_pending_reg(struct kvm_vcpu *vcpu,
387 struct kvm_exit_mmio *mmio,
388 phys_addr_t offset)
389{
390 u32 *reg = vgic_bitmap_get_reg(&vcpu->kvm->arch.vgic.irq_state,
391 vcpu->vcpu_id, offset);
392 vgic_reg_access(mmio, reg, offset,
393 ACCESS_READ_VALUE | ACCESS_WRITE_SETBIT);
394 if (mmio->is_write) {
395 vgic_update_state(vcpu->kvm);
396 return true;
397 }
398
399 return false;
400}
401
402static bool handle_mmio_clear_pending_reg(struct kvm_vcpu *vcpu,
403 struct kvm_exit_mmio *mmio,
404 phys_addr_t offset)
405{
406 u32 *reg = vgic_bitmap_get_reg(&vcpu->kvm->arch.vgic.irq_state,
407 vcpu->vcpu_id, offset);
408 vgic_reg_access(mmio, reg, offset,
409 ACCESS_READ_VALUE | ACCESS_WRITE_CLEARBIT);
410 if (mmio->is_write) {
411 vgic_update_state(vcpu->kvm);
412 return true;
413 }
414
415 return false;
416}
417
418static bool handle_mmio_priority_reg(struct kvm_vcpu *vcpu,
419 struct kvm_exit_mmio *mmio,
420 phys_addr_t offset)
421{
422 u32 *reg = vgic_bytemap_get_reg(&vcpu->kvm->arch.vgic.irq_priority,
423 vcpu->vcpu_id, offset);
424 vgic_reg_access(mmio, reg, offset,
425 ACCESS_READ_VALUE | ACCESS_WRITE_VALUE);
426 return false;
427}
428
429#define GICD_ITARGETSR_SIZE 32
430#define GICD_CPUTARGETS_BITS 8
431#define GICD_IRQS_PER_ITARGETSR (GICD_ITARGETSR_SIZE / GICD_CPUTARGETS_BITS)
432static u32 vgic_get_target_reg(struct kvm *kvm, int irq)
433{
434 struct vgic_dist *dist = &kvm->arch.vgic;
435 struct kvm_vcpu *vcpu;
436 int i, c;
437 unsigned long *bmap;
438 u32 val = 0;
439
440 irq -= VGIC_NR_PRIVATE_IRQS;
441
442 kvm_for_each_vcpu(c, vcpu, kvm) {
443 bmap = vgic_bitmap_get_shared_map(&dist->irq_spi_target[c]);
444 for (i = 0; i < GICD_IRQS_PER_ITARGETSR; i++)
445 if (test_bit(irq + i, bmap))
446 val |= 1 << (c + i * 8);
447 }
448
449 return val;
450}
451
452static void vgic_set_target_reg(struct kvm *kvm, u32 val, int irq)
453{
454 struct vgic_dist *dist = &kvm->arch.vgic;
455 struct kvm_vcpu *vcpu;
456 int i, c;
457 unsigned long *bmap;
458 u32 target;
459
460 irq -= VGIC_NR_PRIVATE_IRQS;
461
462 /*
463 * Pick the LSB in each byte. This ensures we target exactly
464 * one vcpu per IRQ. If the byte is null, assume we target
465 * CPU0.
466 */
467 for (i = 0; i < GICD_IRQS_PER_ITARGETSR; i++) {
468 int shift = i * GICD_CPUTARGETS_BITS;
469 target = ffs((val >> shift) & 0xffU);
470 target = target ? (target - 1) : 0;
471 dist->irq_spi_cpu[irq + i] = target;
472 kvm_for_each_vcpu(c, vcpu, kvm) {
473 bmap = vgic_bitmap_get_shared_map(&dist->irq_spi_target[c]);
474 if (c == target)
475 set_bit(irq + i, bmap);
476 else
477 clear_bit(irq + i, bmap);
478 }
479 }
480}
481
482static bool handle_mmio_target_reg(struct kvm_vcpu *vcpu,
483 struct kvm_exit_mmio *mmio,
484 phys_addr_t offset)
485{
486 u32 reg;
487
488 /* We treat the banked interrupts targets as read-only */
489 if (offset < 32) {
490 u32 roreg = 1 << vcpu->vcpu_id;
491 roreg |= roreg << 8;
492 roreg |= roreg << 16;
493
494 vgic_reg_access(mmio, &roreg, offset,
495 ACCESS_READ_VALUE | ACCESS_WRITE_IGNORED);
496 return false;
497 }
498
499 reg = vgic_get_target_reg(vcpu->kvm, offset & ~3U);
500 vgic_reg_access(mmio, &reg, offset,
501 ACCESS_READ_VALUE | ACCESS_WRITE_VALUE);
502 if (mmio->is_write) {
503 vgic_set_target_reg(vcpu->kvm, reg, offset & ~3U);
504 vgic_update_state(vcpu->kvm);
505 return true;
506 }
507
508 return false;
509}
510
511static u32 vgic_cfg_expand(u16 val)
512{
513 u32 res = 0;
514 int i;
515
516 /*
517 * Turn a 16bit value like abcd...mnop into a 32bit word
518 * a0b0c0d0...m0n0o0p0, which is what the HW cfg register is.
519 */
520 for (i = 0; i < 16; i++)
521 res |= ((val >> i) & VGIC_CFG_EDGE) << (2 * i + 1);
522
523 return res;
524}
525
526static u16 vgic_cfg_compress(u32 val)
527{
528 u16 res = 0;
529 int i;
530
531 /*
532 * Turn a 32bit word a0b0c0d0...m0n0o0p0 into 16bit value like
533 * abcd...mnop which is what we really care about.
534 */
535 for (i = 0; i < 16; i++)
536 res |= ((val >> (i * 2 + 1)) & VGIC_CFG_EDGE) << i;
537
538 return res;
539}
540
541/*
542 * The distributor uses 2 bits per IRQ for the CFG register, but the
543 * LSB is always 0. As such, we only keep the upper bit, and use the
544 * two above functions to compress/expand the bits
545 */
546static bool handle_mmio_cfg_reg(struct kvm_vcpu *vcpu,
547 struct kvm_exit_mmio *mmio, phys_addr_t offset)
548{
549 u32 val;
550 u32 *reg = vgic_bitmap_get_reg(&vcpu->kvm->arch.vgic.irq_cfg,
551 vcpu->vcpu_id, offset >> 1);
552 if (offset & 2)
553 val = *reg >> 16;
554 else
555 val = *reg & 0xffff;
556
557 val = vgic_cfg_expand(val);
558 vgic_reg_access(mmio, &val, offset,
559 ACCESS_READ_VALUE | ACCESS_WRITE_VALUE);
560 if (mmio->is_write) {
561 if (offset < 4) {
562 *reg = ~0U; /* Force PPIs/SGIs to 1 */
563 return false;
564 }
565
566 val = vgic_cfg_compress(val);
567 if (offset & 2) {
568 *reg &= 0xffff;
569 *reg |= val << 16;
570 } else {
571 *reg &= 0xffff << 16;
572 *reg |= val;
573 }
574 }
575
576 return false;
577}
578
579static bool handle_mmio_sgi_reg(struct kvm_vcpu *vcpu,
580 struct kvm_exit_mmio *mmio, phys_addr_t offset)
581{
582 u32 reg;
583 vgic_reg_access(mmio, &reg, offset,
584 ACCESS_READ_RAZ | ACCESS_WRITE_VALUE);
585 if (mmio->is_write) {
586 vgic_dispatch_sgi(vcpu, reg);
587 vgic_update_state(vcpu->kvm);
588 return true;
589 }
590
591 return false;
592}
593
594/*
595 * I would have liked to use the kvm_bus_io_*() API instead, but it
596 * cannot cope with banked registers (only the VM pointer is passed
597 * around, and we need the vcpu). One of these days, someone please
598 * fix it!
599 */
600struct mmio_range {
601 phys_addr_t base;
602 unsigned long len;
603 bool (*handle_mmio)(struct kvm_vcpu *vcpu, struct kvm_exit_mmio *mmio,
604 phys_addr_t offset);
605};
606
607static const struct mmio_range vgic_ranges[] = {
608 {
609 .base = GIC_DIST_CTRL,
610 .len = 12,
611 .handle_mmio = handle_mmio_misc,
612 },
613 {
614 .base = GIC_DIST_IGROUP,
615 .len = VGIC_NR_IRQS / 8,
616 .handle_mmio = handle_mmio_raz_wi,
617 },
618 {
619 .base = GIC_DIST_ENABLE_SET,
620 .len = VGIC_NR_IRQS / 8,
621 .handle_mmio = handle_mmio_set_enable_reg,
622 },
623 {
624 .base = GIC_DIST_ENABLE_CLEAR,
625 .len = VGIC_NR_IRQS / 8,
626 .handle_mmio = handle_mmio_clear_enable_reg,
627 },
628 {
629 .base = GIC_DIST_PENDING_SET,
630 .len = VGIC_NR_IRQS / 8,
631 .handle_mmio = handle_mmio_set_pending_reg,
632 },
633 {
634 .base = GIC_DIST_PENDING_CLEAR,
635 .len = VGIC_NR_IRQS / 8,
636 .handle_mmio = handle_mmio_clear_pending_reg,
637 },
638 {
639 .base = GIC_DIST_ACTIVE_SET,
640 .len = VGIC_NR_IRQS / 8,
641 .handle_mmio = handle_mmio_raz_wi,
642 },
643 {
644 .base = GIC_DIST_ACTIVE_CLEAR,
645 .len = VGIC_NR_IRQS / 8,
646 .handle_mmio = handle_mmio_raz_wi,
647 },
648 {
649 .base = GIC_DIST_PRI,
650 .len = VGIC_NR_IRQS,
651 .handle_mmio = handle_mmio_priority_reg,
652 },
653 {
654 .base = GIC_DIST_TARGET,
655 .len = VGIC_NR_IRQS,
656 .handle_mmio = handle_mmio_target_reg,
657 },
658 {
659 .base = GIC_DIST_CONFIG,
660 .len = VGIC_NR_IRQS / 4,
661 .handle_mmio = handle_mmio_cfg_reg,
662 },
663 {
664 .base = GIC_DIST_SOFTINT,
665 .len = 4,
666 .handle_mmio = handle_mmio_sgi_reg,
667 },
668 {}
669};
670
671static const
672struct mmio_range *find_matching_range(const struct mmio_range *ranges,
673 struct kvm_exit_mmio *mmio,
674 phys_addr_t base)
675{
676 const struct mmio_range *r = ranges;
677 phys_addr_t addr = mmio->phys_addr - base;
678
679 while (r->len) {
680 if (addr >= r->base &&
681 (addr + mmio->len) <= (r->base + r->len))
682 return r;
683 r++;
684 }
685
686 return NULL;
687}
688
689/**
690 * vgic_handle_mmio - handle an in-kernel MMIO access
691 * @vcpu: pointer to the vcpu performing the access
692 * @run: pointer to the kvm_run structure
693 * @mmio: pointer to the data describing the access
694 *
695 * returns true if the MMIO access has been performed in kernel space,
696 * and false if it needs to be emulated in user space.
697 */
698bool vgic_handle_mmio(struct kvm_vcpu *vcpu, struct kvm_run *run,
699 struct kvm_exit_mmio *mmio)
700{
701 const struct mmio_range *range;
702 struct vgic_dist *dist = &vcpu->kvm->arch.vgic;
703 unsigned long base = dist->vgic_dist_base;
704 bool updated_state;
705 unsigned long offset;
706
707 if (!irqchip_in_kernel(vcpu->kvm) ||
708 mmio->phys_addr < base ||
709 (mmio->phys_addr + mmio->len) > (base + KVM_VGIC_V2_DIST_SIZE))
710 return false;
711
712 /* We don't support ldrd / strd or ldm / stm to the emulated vgic */
713 if (mmio->len > 4) {
714 kvm_inject_dabt(vcpu, mmio->phys_addr);
715 return true;
716 }
717
718 range = find_matching_range(vgic_ranges, mmio, base);
719 if (unlikely(!range || !range->handle_mmio)) {
720 pr_warn("Unhandled access %d %08llx %d\n",
721 mmio->is_write, mmio->phys_addr, mmio->len);
722 return false;
723 }
724
725 spin_lock(&vcpu->kvm->arch.vgic.lock);
726 offset = mmio->phys_addr - range->base - base;
727 updated_state = range->handle_mmio(vcpu, mmio, offset);
728 spin_unlock(&vcpu->kvm->arch.vgic.lock);
729 kvm_prepare_mmio(run, mmio);
730 kvm_handle_mmio_return(vcpu, run);
731
732 if (updated_state)
733 vgic_kick_vcpus(vcpu->kvm);
734
735 return true;
736}
737
738static void vgic_dispatch_sgi(struct kvm_vcpu *vcpu, u32 reg)
739{
740 struct kvm *kvm = vcpu->kvm;
741 struct vgic_dist *dist = &kvm->arch.vgic;
742 int nrcpus = atomic_read(&kvm->online_vcpus);
743 u8 target_cpus;
744 int sgi, mode, c, vcpu_id;
745
746 vcpu_id = vcpu->vcpu_id;
747
748 sgi = reg & 0xf;
749 target_cpus = (reg >> 16) & 0xff;
750 mode = (reg >> 24) & 3;
751
752 switch (mode) {
753 case 0:
754 if (!target_cpus)
755 return;
756
757 case 1:
758 target_cpus = ((1 << nrcpus) - 1) & ~(1 << vcpu_id) & 0xff;
759 break;
760
761 case 2:
762 target_cpus = 1 << vcpu_id;
763 break;
764 }
765
766 kvm_for_each_vcpu(c, vcpu, kvm) {
767 if (target_cpus & 1) {
768 /* Flag the SGI as pending */
769 vgic_dist_irq_set(vcpu, sgi);
770 dist->irq_sgi_sources[c][sgi] |= 1 << vcpu_id;
771 kvm_debug("SGI%d from CPU%d to CPU%d\n", sgi, vcpu_id, c);
772 }
773
774 target_cpus >>= 1;
775 }
776}
777
778static int compute_pending_for_cpu(struct kvm_vcpu *vcpu)
779{
780 struct vgic_dist *dist = &vcpu->kvm->arch.vgic;
781 unsigned long *pending, *enabled, *pend_percpu, *pend_shared;
782 unsigned long pending_private, pending_shared;
783 int vcpu_id;
784
785 vcpu_id = vcpu->vcpu_id;
786 pend_percpu = vcpu->arch.vgic_cpu.pending_percpu;
787 pend_shared = vcpu->arch.vgic_cpu.pending_shared;
788
789 pending = vgic_bitmap_get_cpu_map(&dist->irq_state, vcpu_id);
790 enabled = vgic_bitmap_get_cpu_map(&dist->irq_enabled, vcpu_id);
791 bitmap_and(pend_percpu, pending, enabled, VGIC_NR_PRIVATE_IRQS);
792
793 pending = vgic_bitmap_get_shared_map(&dist->irq_state);
794 enabled = vgic_bitmap_get_shared_map(&dist->irq_enabled);
795 bitmap_and(pend_shared, pending, enabled, VGIC_NR_SHARED_IRQS);
796 bitmap_and(pend_shared, pend_shared,
797 vgic_bitmap_get_shared_map(&dist->irq_spi_target[vcpu_id]),
798 VGIC_NR_SHARED_IRQS);
799
800 pending_private = find_first_bit(pend_percpu, VGIC_NR_PRIVATE_IRQS);
801 pending_shared = find_first_bit(pend_shared, VGIC_NR_SHARED_IRQS);
802 return (pending_private < VGIC_NR_PRIVATE_IRQS ||
803 pending_shared < VGIC_NR_SHARED_IRQS);
804}
805
806/*
807 * Update the interrupt state and determine which CPUs have pending
808 * interrupts. Must be called with distributor lock held.
809 */
810static void vgic_update_state(struct kvm *kvm)
811{
812 struct vgic_dist *dist = &kvm->arch.vgic;
813 struct kvm_vcpu *vcpu;
814 int c;
815
816 if (!dist->enabled) {
817 set_bit(0, &dist->irq_pending_on_cpu);
818 return;
819 }
820
821 kvm_for_each_vcpu(c, vcpu, kvm) {
822 if (compute_pending_for_cpu(vcpu)) {
823 pr_debug("CPU%d has pending interrupts\n", c);
824 set_bit(c, &dist->irq_pending_on_cpu);
825 }
826 }
827}
828
829#define LR_CPUID(lr) \
830 (((lr) & GICH_LR_PHYSID_CPUID) >> GICH_LR_PHYSID_CPUID_SHIFT)
831#define MK_LR_PEND(src, irq) \
832 (GICH_LR_PENDING_BIT | ((src) << GICH_LR_PHYSID_CPUID_SHIFT) | (irq))
833
834/*
835 * An interrupt may have been disabled after being made pending on the
836 * CPU interface (the classic case is a timer running while we're
837 * rebooting the guest - the interrupt would kick as soon as the CPU
838 * interface gets enabled, with deadly consequences).
839 *
840 * The solution is to examine already active LRs, and check the
841 * interrupt is still enabled. If not, just retire it.
842 */
843static void vgic_retire_disabled_irqs(struct kvm_vcpu *vcpu)
844{
845 struct vgic_cpu *vgic_cpu = &vcpu->arch.vgic_cpu;
846 int lr;
847
848 for_each_set_bit(lr, vgic_cpu->lr_used, vgic_cpu->nr_lr) {
849 int irq = vgic_cpu->vgic_lr[lr] & GICH_LR_VIRTUALID;
850
851 if (!vgic_irq_is_enabled(vcpu, irq)) {
852 vgic_cpu->vgic_irq_lr_map[irq] = LR_EMPTY;
853 clear_bit(lr, vgic_cpu->lr_used);
854 vgic_cpu->vgic_lr[lr] &= ~GICH_LR_STATE;
855 if (vgic_irq_is_active(vcpu, irq))
856 vgic_irq_clear_active(vcpu, irq);
857 }
858 }
859}
860
861/*
862 * Queue an interrupt to a CPU virtual interface. Return true on success,
863 * or false if it wasn't possible to queue it.
864 */
865static bool vgic_queue_irq(struct kvm_vcpu *vcpu, u8 sgi_source_id, int irq)
866{
867 struct vgic_cpu *vgic_cpu = &vcpu->arch.vgic_cpu;
868 int lr;
869
870 /* Sanitize the input... */
871 BUG_ON(sgi_source_id & ~7);
872 BUG_ON(sgi_source_id && irq >= VGIC_NR_SGIS);
873 BUG_ON(irq >= VGIC_NR_IRQS);
874
875 kvm_debug("Queue IRQ%d\n", irq);
876
877 lr = vgic_cpu->vgic_irq_lr_map[irq];
878
879 /* Do we have an active interrupt for the same CPUID? */
880 if (lr != LR_EMPTY &&
881 (LR_CPUID(vgic_cpu->vgic_lr[lr]) == sgi_source_id)) {
882 kvm_debug("LR%d piggyback for IRQ%d %x\n",
883 lr, irq, vgic_cpu->vgic_lr[lr]);
884 BUG_ON(!test_bit(lr, vgic_cpu->lr_used));
885 vgic_cpu->vgic_lr[lr] |= GICH_LR_PENDING_BIT;
886 return true;
887 }
888
889 /* Try to use another LR for this interrupt */
890 lr = find_first_zero_bit((unsigned long *)vgic_cpu->lr_used,
891 vgic_cpu->nr_lr);
892 if (lr >= vgic_cpu->nr_lr)
893 return false;
894
895 kvm_debug("LR%d allocated for IRQ%d %x\n", lr, irq, sgi_source_id);
896 vgic_cpu->vgic_lr[lr] = MK_LR_PEND(sgi_source_id, irq);
897 vgic_cpu->vgic_irq_lr_map[irq] = lr;
898 set_bit(lr, vgic_cpu->lr_used);
899
900 if (!vgic_irq_is_edge(vcpu, irq))
901 vgic_cpu->vgic_lr[lr] |= GICH_LR_EOI;
902
903 return true;
904}
905
906static bool vgic_queue_sgi(struct kvm_vcpu *vcpu, int irq)
907{
908 struct vgic_dist *dist = &vcpu->kvm->arch.vgic;
909 unsigned long sources;
910 int vcpu_id = vcpu->vcpu_id;
911 int c;
912
913 sources = dist->irq_sgi_sources[vcpu_id][irq];
914
915 for_each_set_bit(c, &sources, VGIC_MAX_CPUS) {
916 if (vgic_queue_irq(vcpu, c, irq))
917 clear_bit(c, &sources);
918 }
919
920 dist->irq_sgi_sources[vcpu_id][irq] = sources;
921
922 /*
923 * If the sources bitmap has been cleared it means that we
924 * could queue all the SGIs onto link registers (see the
925 * clear_bit above), and therefore we are done with them in
926 * our emulated gic and can get rid of them.
927 */
928 if (!sources) {
929 vgic_dist_irq_clear(vcpu, irq);
930 vgic_cpu_irq_clear(vcpu, irq);
931 return true;
932 }
933
934 return false;
935}
936
937static bool vgic_queue_hwirq(struct kvm_vcpu *vcpu, int irq)
938{
939 if (vgic_irq_is_active(vcpu, irq))
940 return true; /* level interrupt, already queued */
941
942 if (vgic_queue_irq(vcpu, 0, irq)) {
943 if (vgic_irq_is_edge(vcpu, irq)) {
944 vgic_dist_irq_clear(vcpu, irq);
945 vgic_cpu_irq_clear(vcpu, irq);
946 } else {
947 vgic_irq_set_active(vcpu, irq);
948 }
949
950 return true;
951 }
952
953 return false;
954}
955
956/*
957 * Fill the list registers with pending interrupts before running the
958 * guest.
959 */
960static void __kvm_vgic_flush_hwstate(struct kvm_vcpu *vcpu)
961{
962 struct vgic_cpu *vgic_cpu = &vcpu->arch.vgic_cpu;
963 struct vgic_dist *dist = &vcpu->kvm->arch.vgic;
964 int i, vcpu_id;
965 int overflow = 0;
966
967 vcpu_id = vcpu->vcpu_id;
968
969 /*
970 * We may not have any pending interrupt, or the interrupts
971 * may have been serviced from another vcpu. In all cases,
972 * move along.
973 */
974 if (!kvm_vgic_vcpu_pending_irq(vcpu)) {
975 pr_debug("CPU%d has no pending interrupt\n", vcpu_id);
976 goto epilog;
977 }
978
979 /* SGIs */
980 for_each_set_bit(i, vgic_cpu->pending_percpu, VGIC_NR_SGIS) {
981 if (!vgic_queue_sgi(vcpu, i))
982 overflow = 1;
983 }
984
985 /* PPIs */
986 for_each_set_bit_from(i, vgic_cpu->pending_percpu, VGIC_NR_PRIVATE_IRQS) {
987 if (!vgic_queue_hwirq(vcpu, i))
988 overflow = 1;
989 }
990
991 /* SPIs */
992 for_each_set_bit(i, vgic_cpu->pending_shared, VGIC_NR_SHARED_IRQS) {
993 if (!vgic_queue_hwirq(vcpu, i + VGIC_NR_PRIVATE_IRQS))
994 overflow = 1;
995 }
996
997epilog:
998 if (overflow) {
999 vgic_cpu->vgic_hcr |= GICH_HCR_UIE;
1000 } else {
1001 vgic_cpu->vgic_hcr &= ~GICH_HCR_UIE;
1002 /*
1003 * We're about to run this VCPU, and we've consumed
1004 * everything the distributor had in store for
1005 * us. Claim we don't have anything pending. We'll
1006 * adjust that if needed while exiting.
1007 */
1008 clear_bit(vcpu_id, &dist->irq_pending_on_cpu);
1009 }
1010}
1011
1012static bool vgic_process_maintenance(struct kvm_vcpu *vcpu)
1013{
1014 struct vgic_cpu *vgic_cpu = &vcpu->arch.vgic_cpu;
1015 bool level_pending = false;
1016
1017 kvm_debug("MISR = %08x\n", vgic_cpu->vgic_misr);
1018
1019 if (vgic_cpu->vgic_misr & GICH_MISR_EOI) {
1020 /*
1021 * Some level interrupts have been EOIed. Clear their
1022 * active bit.
1023 */
1024 int lr, irq;
1025
1026 for_each_set_bit(lr, (unsigned long *)vgic_cpu->vgic_eisr,
1027 vgic_cpu->nr_lr) {
1028 irq = vgic_cpu->vgic_lr[lr] & GICH_LR_VIRTUALID;
1029
1030 vgic_irq_clear_active(vcpu, irq);
1031 vgic_cpu->vgic_lr[lr] &= ~GICH_LR_EOI;
1032
1033 /* Any additional pending interrupt? */
1034 if (vgic_dist_irq_is_pending(vcpu, irq)) {
1035 vgic_cpu_irq_set(vcpu, irq);
1036 level_pending = true;
1037 } else {
1038 vgic_cpu_irq_clear(vcpu, irq);
1039 }
1040
1041 /*
1042 * Despite being EOIed, the LR may not have
1043 * been marked as empty.
1044 */
1045 set_bit(lr, (unsigned long *)vgic_cpu->vgic_elrsr);
1046 vgic_cpu->vgic_lr[lr] &= ~GICH_LR_ACTIVE_BIT;
1047 }
1048 }
1049
1050 if (vgic_cpu->vgic_misr & GICH_MISR_U)
1051 vgic_cpu->vgic_hcr &= ~GICH_HCR_UIE;
1052
1053 return level_pending;
1054}
1055
1056/*
1057 * Sync back the VGIC state after a guest run. The distributor lock is
1058 * needed so we don't get preempted in the middle of the state processing.
1059 */
1060static void __kvm_vgic_sync_hwstate(struct kvm_vcpu *vcpu)
1061{
1062 struct vgic_cpu *vgic_cpu = &vcpu->arch.vgic_cpu;
1063 struct vgic_dist *dist = &vcpu->kvm->arch.vgic;
1064 int lr, pending;
1065 bool level_pending;
1066
1067 level_pending = vgic_process_maintenance(vcpu);
1068
1069 /* Clear mappings for empty LRs */
1070 for_each_set_bit(lr, (unsigned long *)vgic_cpu->vgic_elrsr,
1071 vgic_cpu->nr_lr) {
1072 int irq;
1073
1074 if (!test_and_clear_bit(lr, vgic_cpu->lr_used))
1075 continue;
1076
1077 irq = vgic_cpu->vgic_lr[lr] & GICH_LR_VIRTUALID;
1078
1079 BUG_ON(irq >= VGIC_NR_IRQS);
1080 vgic_cpu->vgic_irq_lr_map[irq] = LR_EMPTY;
1081 }
1082
1083 /* Check if we still have something up our sleeve... */
1084 pending = find_first_zero_bit((unsigned long *)vgic_cpu->vgic_elrsr,
1085 vgic_cpu->nr_lr);
1086 if (level_pending || pending < vgic_cpu->nr_lr)
1087 set_bit(vcpu->vcpu_id, &dist->irq_pending_on_cpu);
1088}
1089
1090void kvm_vgic_flush_hwstate(struct kvm_vcpu *vcpu)
1091{
1092 struct vgic_dist *dist = &vcpu->kvm->arch.vgic;
1093
1094 if (!irqchip_in_kernel(vcpu->kvm))
1095 return;
1096
1097 spin_lock(&dist->lock);
1098 __kvm_vgic_flush_hwstate(vcpu);
1099 spin_unlock(&dist->lock);
1100}
1101
1102void kvm_vgic_sync_hwstate(struct kvm_vcpu *vcpu)
1103{
1104 struct vgic_dist *dist = &vcpu->kvm->arch.vgic;
1105
1106 if (!irqchip_in_kernel(vcpu->kvm))
1107 return;
1108
1109 spin_lock(&dist->lock);
1110 __kvm_vgic_sync_hwstate(vcpu);
1111 spin_unlock(&dist->lock);
1112}
1113
1114int kvm_vgic_vcpu_pending_irq(struct kvm_vcpu *vcpu)
1115{
1116 struct vgic_dist *dist = &vcpu->kvm->arch.vgic;
1117
1118 if (!irqchip_in_kernel(vcpu->kvm))
1119 return 0;
1120
1121 return test_bit(vcpu->vcpu_id, &dist->irq_pending_on_cpu);
1122}
1123
1124static void vgic_kick_vcpus(struct kvm *kvm)
1125{
1126 struct kvm_vcpu *vcpu;
1127 int c;
1128
1129 /*
1130 * We've injected an interrupt, time to find out who deserves
1131 * a good kick...
1132 */
1133 kvm_for_each_vcpu(c, vcpu, kvm) {
1134 if (kvm_vgic_vcpu_pending_irq(vcpu))
1135 kvm_vcpu_kick(vcpu);
1136 }
1137}
1138
1139static int vgic_validate_injection(struct kvm_vcpu *vcpu, int irq, int level)
1140{
1141 int is_edge = vgic_irq_is_edge(vcpu, irq);
1142 int state = vgic_dist_irq_is_pending(vcpu, irq);
1143
1144 /*
1145 * Only inject an interrupt if:
1146 * - edge triggered and we have a rising edge
1147 * - level triggered and we change level
1148 */
1149 if (is_edge)
1150 return level > state;
1151 else
1152 return level != state;
1153}
1154
1155static bool vgic_update_irq_state(struct kvm *kvm, int cpuid,
1156 unsigned int irq_num, bool level)
1157{
1158 struct vgic_dist *dist = &kvm->arch.vgic;
1159 struct kvm_vcpu *vcpu;
1160 int is_edge, is_level;
1161 int enabled;
1162 bool ret = true;
1163
1164 spin_lock(&dist->lock);
1165
1166 vcpu = kvm_get_vcpu(kvm, cpuid);
1167 is_edge = vgic_irq_is_edge(vcpu, irq_num);
1168 is_level = !is_edge;
1169
1170 if (!vgic_validate_injection(vcpu, irq_num, level)) {
1171 ret = false;
1172 goto out;
1173 }
1174
1175 if (irq_num >= VGIC_NR_PRIVATE_IRQS) {
1176 cpuid = dist->irq_spi_cpu[irq_num - VGIC_NR_PRIVATE_IRQS];
1177 vcpu = kvm_get_vcpu(kvm, cpuid);
1178 }
1179
1180 kvm_debug("Inject IRQ%d level %d CPU%d\n", irq_num, level, cpuid);
1181
1182 if (level)
1183 vgic_dist_irq_set(vcpu, irq_num);
1184 else
1185 vgic_dist_irq_clear(vcpu, irq_num);
1186
1187 enabled = vgic_irq_is_enabled(vcpu, irq_num);
1188
1189 if (!enabled) {
1190 ret = false;
1191 goto out;
1192 }
1193
1194 if (is_level && vgic_irq_is_active(vcpu, irq_num)) {
1195 /*
1196 * Level interrupt in progress, will be picked up
1197 * when EOId.
1198 */
1199 ret = false;
1200 goto out;
1201 }
1202
1203 if (level) {
1204 vgic_cpu_irq_set(vcpu, irq_num);
1205 set_bit(cpuid, &dist->irq_pending_on_cpu);
1206 }
1207
1208out:
1209 spin_unlock(&dist->lock);
1210
1211 return ret;
1212}
1213
1214/**
1215 * kvm_vgic_inject_irq - Inject an IRQ from a device to the vgic
1216 * @kvm: The VM structure pointer
1217 * @cpuid: The CPU for PPIs
1218 * @irq_num: The IRQ number that is assigned to the device
1219 * @level: Edge-triggered: true: to trigger the interrupt
1220 * false: to ignore the call
1221 * Level-sensitive true: activates an interrupt
1222 * false: deactivates an interrupt
1223 *
1224 * The GIC is not concerned with devices being active-LOW or active-HIGH for
1225 * level-sensitive interrupts. You can think of the level parameter as 1
1226 * being HIGH and 0 being LOW and all devices being active-HIGH.
1227 */
1228int kvm_vgic_inject_irq(struct kvm *kvm, int cpuid, unsigned int irq_num,
1229 bool level)
1230{
1231 if (vgic_update_irq_state(kvm, cpuid, irq_num, level))
1232 vgic_kick_vcpus(kvm);
1233
1234 return 0;
1235}
1236
1237static irqreturn_t vgic_maintenance_handler(int irq, void *data)
1238{
1239 /*
1240 * We cannot rely on the vgic maintenance interrupt to be
1241 * delivered synchronously. This means we can only use it to
1242 * exit the VM, and we perform the handling of EOIed
1243 * interrupts on the exit path (see vgic_process_maintenance).
1244 */
1245 return IRQ_HANDLED;
1246}
1247
1248int kvm_vgic_vcpu_init(struct kvm_vcpu *vcpu)
1249{
1250 struct vgic_cpu *vgic_cpu = &vcpu->arch.vgic_cpu;
1251 struct vgic_dist *dist = &vcpu->kvm->arch.vgic;
1252 int i;
1253
1254 if (!irqchip_in_kernel(vcpu->kvm))
1255 return 0;
1256
1257 if (vcpu->vcpu_id >= VGIC_MAX_CPUS)
1258 return -EBUSY;
1259
1260 for (i = 0; i < VGIC_NR_IRQS; i++) {
1261 if (i < VGIC_NR_PPIS)
1262 vgic_bitmap_set_irq_val(&dist->irq_enabled,
1263 vcpu->vcpu_id, i, 1);
1264 if (i < VGIC_NR_PRIVATE_IRQS)
1265 vgic_bitmap_set_irq_val(&dist->irq_cfg,
1266 vcpu->vcpu_id, i, VGIC_CFG_EDGE);
1267
1268 vgic_cpu->vgic_irq_lr_map[i] = LR_EMPTY;
1269 }
1270
1271 /*
1272 * By forcing VMCR to zero, the GIC will restore the binary
1273 * points to their reset values. Anything else resets to zero
1274 * anyway.
1275 */
1276 vgic_cpu->vgic_vmcr = 0;
1277
1278 vgic_cpu->nr_lr = vgic_nr_lr;
1279 vgic_cpu->vgic_hcr = GICH_HCR_EN; /* Get the show on the road... */
1280
1281 return 0;
1282}
1283
1284static void vgic_init_maintenance_interrupt(void *info)
1285{
1286 enable_percpu_irq(vgic_maint_irq, 0);
1287}
1288
1289static int vgic_cpu_notify(struct notifier_block *self,
1290 unsigned long action, void *cpu)
1291{
1292 switch (action) {
1293 case CPU_STARTING:
1294 case CPU_STARTING_FROZEN:
1295 vgic_init_maintenance_interrupt(NULL);
1296 break;
1297 case CPU_DYING:
1298 case CPU_DYING_FROZEN:
1299 disable_percpu_irq(vgic_maint_irq);
1300 break;
1301 }
1302
1303 return NOTIFY_OK;
1304}
1305
1306static struct notifier_block vgic_cpu_nb = {
1307 .notifier_call = vgic_cpu_notify,
1308};
1309
1310int kvm_vgic_hyp_init(void)
1311{
1312 int ret;
1313 struct resource vctrl_res;
1314 struct resource vcpu_res;
1315
1316 vgic_node = of_find_compatible_node(NULL, NULL, "arm,cortex-a15-gic");
1317 if (!vgic_node) {
1318 kvm_err("error: no compatible vgic node in DT\n");
1319 return -ENODEV;
1320 }
1321
1322 vgic_maint_irq = irq_of_parse_and_map(vgic_node, 0);
1323 if (!vgic_maint_irq) {
1324 kvm_err("error getting vgic maintenance irq from DT\n");
1325 ret = -ENXIO;
1326 goto out;
1327 }
1328
1329 ret = request_percpu_irq(vgic_maint_irq, vgic_maintenance_handler,
1330 "vgic", kvm_get_running_vcpus());
1331 if (ret) {
1332 kvm_err("Cannot register interrupt %d\n", vgic_maint_irq);
1333 goto out;
1334 }
1335
1336 ret = register_cpu_notifier(&vgic_cpu_nb);
1337 if (ret) {
1338 kvm_err("Cannot register vgic CPU notifier\n");
1339 goto out_free_irq;
1340 }
1341
1342 ret = of_address_to_resource(vgic_node, 2, &vctrl_res);
1343 if (ret) {
1344 kvm_err("Cannot obtain VCTRL resource\n");
1345 goto out_free_irq;
1346 }
1347
1348 vgic_vctrl_base = of_iomap(vgic_node, 2);
1349 if (!vgic_vctrl_base) {
1350 kvm_err("Cannot ioremap VCTRL\n");
1351 ret = -ENOMEM;
1352 goto out_free_irq;
1353 }
1354
1355 vgic_nr_lr = readl_relaxed(vgic_vctrl_base + GICH_VTR);
1356 vgic_nr_lr = (vgic_nr_lr & 0x3f) + 1;
1357
1358 ret = create_hyp_io_mappings(vgic_vctrl_base,
1359 vgic_vctrl_base + resource_size(&vctrl_res),
1360 vctrl_res.start);
1361 if (ret) {
1362 kvm_err("Cannot map VCTRL into hyp\n");
1363 goto out_unmap;
1364 }
1365
1366 kvm_info("%s@%llx IRQ%d\n", vgic_node->name,
1367 vctrl_res.start, vgic_maint_irq);
1368 on_each_cpu(vgic_init_maintenance_interrupt, NULL, 1);
1369
1370 if (of_address_to_resource(vgic_node, 3, &vcpu_res)) {
1371 kvm_err("Cannot obtain VCPU resource\n");
1372 ret = -ENXIO;
1373 goto out_unmap;
1374 }
1375 vgic_vcpu_base = vcpu_res.start;
1376
1377 goto out;
1378
1379out_unmap:
1380 iounmap(vgic_vctrl_base);
1381out_free_irq:
1382 free_percpu_irq(vgic_maint_irq, kvm_get_running_vcpus());
1383out:
1384 of_node_put(vgic_node);
1385 return ret;
1386}
1387
1388int kvm_vgic_init(struct kvm *kvm)
1389{
1390 int ret = 0, i;
1391
1392 mutex_lock(&kvm->lock);
1393
1394 if (vgic_initialized(kvm))
1395 goto out;
1396
1397 if (IS_VGIC_ADDR_UNDEF(kvm->arch.vgic.vgic_dist_base) ||
1398 IS_VGIC_ADDR_UNDEF(kvm->arch.vgic.vgic_cpu_base)) {
1399 kvm_err("Need to set vgic cpu and dist addresses first\n");
1400 ret = -ENXIO;
1401 goto out;
1402 }
1403
1404 ret = kvm_phys_addr_ioremap(kvm, kvm->arch.vgic.vgic_cpu_base,
1405 vgic_vcpu_base, KVM_VGIC_V2_CPU_SIZE);
1406 if (ret) {
1407 kvm_err("Unable to remap VGIC CPU to VCPU\n");
1408 goto out;
1409 }
1410
1411 for (i = VGIC_NR_PRIVATE_IRQS; i < VGIC_NR_IRQS; i += 4)
1412 vgic_set_target_reg(kvm, 0, i);
1413
1414 kvm_timer_init(kvm);
1415 kvm->arch.vgic.ready = true;
1416out:
1417 mutex_unlock(&kvm->lock);
1418 return ret;
1419}
1420
1421int kvm_vgic_create(struct kvm *kvm)
1422{
1423 int ret = 0;
1424
1425 mutex_lock(&kvm->lock);
1426
1427 if (atomic_read(&kvm->online_vcpus) || kvm->arch.vgic.vctrl_base) {
1428 ret = -EEXIST;
1429 goto out;
1430 }
1431
1432 spin_lock_init(&kvm->arch.vgic.lock);
1433 kvm->arch.vgic.vctrl_base = vgic_vctrl_base;
1434 kvm->arch.vgic.vgic_dist_base = VGIC_ADDR_UNDEF;
1435 kvm->arch.vgic.vgic_cpu_base = VGIC_ADDR_UNDEF;
1436
1437out:
1438 mutex_unlock(&kvm->lock);
1439 return ret;
1440}
1441
1442static bool vgic_ioaddr_overlap(struct kvm *kvm)
1443{
1444 phys_addr_t dist = kvm->arch.vgic.vgic_dist_base;
1445 phys_addr_t cpu = kvm->arch.vgic.vgic_cpu_base;
1446
1447 if (IS_VGIC_ADDR_UNDEF(dist) || IS_VGIC_ADDR_UNDEF(cpu))
1448 return 0;
1449 if ((dist <= cpu && dist + KVM_VGIC_V2_DIST_SIZE > cpu) ||
1450 (cpu <= dist && cpu + KVM_VGIC_V2_CPU_SIZE > dist))
1451 return -EBUSY;
1452 return 0;
1453}
1454
1455static int vgic_ioaddr_assign(struct kvm *kvm, phys_addr_t *ioaddr,
1456 phys_addr_t addr, phys_addr_t size)
1457{
1458 int ret;
1459
1460 if (!IS_VGIC_ADDR_UNDEF(*ioaddr))
1461 return -EEXIST;
1462 if (addr + size < addr)
1463 return -EINVAL;
1464
1465 ret = vgic_ioaddr_overlap(kvm);
1466 if (ret)
1467 return ret;
1468 *ioaddr = addr;
1469 return ret;
1470}
1471
1472int kvm_vgic_set_addr(struct kvm *kvm, unsigned long type, u64 addr)
1473{
1474 int r = 0;
1475 struct vgic_dist *vgic = &kvm->arch.vgic;
1476
1477 if (addr & ~KVM_PHYS_MASK)
1478 return -E2BIG;
1479
1480 if (addr & (SZ_4K - 1))
1481 return -EINVAL;
1482
1483 mutex_lock(&kvm->lock);
1484 switch (type) {
1485 case KVM_VGIC_V2_ADDR_TYPE_DIST:
1486 r = vgic_ioaddr_assign(kvm, &vgic->vgic_dist_base,
1487 addr, KVM_VGIC_V2_DIST_SIZE);
1488 break;
1489 case KVM_VGIC_V2_ADDR_TYPE_CPU:
1490 r = vgic_ioaddr_assign(kvm, &vgic->vgic_cpu_base,
1491 addr, KVM_VGIC_V2_CPU_SIZE);
1492 break;
1493 default:
1494 r = -ENODEV;
1495 }
1496
1497 mutex_unlock(&kvm->lock);
1498 return r;
1499}
diff --git a/virt/kvm/eventfd.c b/virt/kvm/eventfd.c
index 64ee720b75c7..1550637d1b10 100644
--- a/virt/kvm/eventfd.c
+++ b/virt/kvm/eventfd.c
@@ -753,6 +753,7 @@ kvm_assign_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args)
753 if (ret < 0) 753 if (ret < 0)
754 goto unlock_fail; 754 goto unlock_fail;
755 755
756 kvm->buses[bus_idx]->ioeventfd_count++;
756 list_add_tail(&p->list, &kvm->ioeventfds); 757 list_add_tail(&p->list, &kvm->ioeventfds);
757 758
758 mutex_unlock(&kvm->slots_lock); 759 mutex_unlock(&kvm->slots_lock);
@@ -798,6 +799,7 @@ kvm_deassign_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args)
798 continue; 799 continue;
799 800
800 kvm_io_bus_unregister_dev(kvm, bus_idx, &p->dev); 801 kvm_io_bus_unregister_dev(kvm, bus_idx, &p->dev);
802 kvm->buses[bus_idx]->ioeventfd_count--;
801 ioeventfd_release(p); 803 ioeventfd_release(p);
802 ret = 0; 804 ret = 0;
803 break; 805 break;
diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
index 302681c4aa44..1580dd4ace4e 100644
--- a/virt/kvm/kvm_main.c
+++ b/virt/kvm/kvm_main.c
@@ -2926,7 +2926,8 @@ int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2926 struct kvm_io_bus *new_bus, *bus; 2926 struct kvm_io_bus *new_bus, *bus;
2927 2927
2928 bus = kvm->buses[bus_idx]; 2928 bus = kvm->buses[bus_idx];
2929 if (bus->dev_count > NR_IOBUS_DEVS - 1) 2929 /* exclude ioeventfd which is limited by maximum fd */
2930 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
2930 return -ENOSPC; 2931 return -ENOSPC;
2931 2932
2932 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) * 2933 new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
@@ -3181,6 +3182,7 @@ int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3181 3182
3182out_undebugfs: 3183out_undebugfs:
3183 unregister_syscore_ops(&kvm_syscore_ops); 3184 unregister_syscore_ops(&kvm_syscore_ops);
3185 misc_deregister(&kvm_dev);
3184out_unreg: 3186out_unreg:
3185 kvm_async_pf_deinit(); 3187 kvm_async_pf_deinit();
3186out_free: 3188out_free:
generated by cgit v1.2.2 (git 2.25.0) at 2024-11-10 23:11:23 -0500