2 * Copyright (C) 2012 ARM Ltd.
3 * Author: Marc Zyngier <marc.zyngier@arm.com>
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.
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.
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
19 #include <linux/cpu.h>
20 #include <linux/kvm.h>
21 #include <linux/kvm_host.h>
22 #include <linux/interrupt.h>
25 #include <linux/of_address.h>
26 #include <linux/of_irq.h>
27 #include <linux/rculist.h>
28 #include <linux/uaccess.h>
30 #include <asm/kvm_emulate.h>
31 #include <asm/kvm_arm.h>
32 #include <asm/kvm_mmu.h>
33 #include <trace/events/kvm.h>
35 #include <kvm/iodev.h>
38 * How the whole thing works (courtesy of Christoffer Dall):
40 * - At any time, the dist->irq_pending_on_cpu is the oracle that knows if
41 * something is pending on the CPU interface.
42 * - Interrupts that are pending on the distributor are stored on the
43 * vgic.irq_pending vgic bitmap (this bitmap is updated by both user land
44 * ioctls and guest mmio ops, and other in-kernel peripherals such as the
46 * - Every time the bitmap changes, the irq_pending_on_cpu oracle is
48 * - To calculate the oracle, we need info for each cpu from
49 * compute_pending_for_cpu, which considers:
50 * - PPI: dist->irq_pending & dist->irq_enable
51 * - SPI: dist->irq_pending & dist->irq_enable & dist->irq_spi_target
52 * - irq_spi_target is a 'formatted' version of the GICD_ITARGETSRn
53 * registers, stored on each vcpu. We only keep one bit of
54 * information per interrupt, making sure that only one vcpu can
55 * accept the interrupt.
56 * - If any of the above state changes, we must recalculate the oracle.
57 * - The same is true when injecting an interrupt, except that we only
58 * consider a single interrupt at a time. The irq_spi_cpu array
59 * contains the target CPU for each SPI.
61 * The handling of level interrupts adds some extra complexity. We
62 * need to track when the interrupt has been EOIed, so we can sample
63 * the 'line' again. This is achieved as such:
65 * - When a level interrupt is moved onto a vcpu, the corresponding
66 * bit in irq_queued is set. As long as this bit is set, the line
67 * will be ignored for further interrupts. The interrupt is injected
68 * into the vcpu with the GICH_LR_EOI bit set (generate a
69 * maintenance interrupt on EOI).
70 * - When the interrupt is EOIed, the maintenance interrupt fires,
71 * and clears the corresponding bit in irq_queued. This allows the
72 * interrupt line to be sampled again.
73 * - Note that level-triggered interrupts can also be set to pending from
74 * writes to GICD_ISPENDRn and lowering the external input line does not
75 * cause the interrupt to become inactive in such a situation.
76 * Conversely, writes to GICD_ICPENDRn do not cause the interrupt to become
77 * inactive as long as the external input line is held high.
80 * Initialization rules: there are multiple stages to the vgic
81 * initialization, both for the distributor and the CPU interfaces.
85 * - kvm_vgic_early_init(): initialization of static data that doesn't
86 * depend on any sizing information or emulation type. No allocation
89 * - vgic_init(): allocation and initialization of the generic data
90 * structures that depend on sizing information (number of CPUs,
91 * number of interrupts). Also initializes the vcpu specific data
92 * structures. Can be executed lazily for GICv2.
93 * [to be renamed to kvm_vgic_init??]
97 * - kvm_vgic_cpu_early_init(): initialization of static data that
98 * doesn't depend on any sizing information or emulation type. No
99 * allocation is allowed there.
104 static void vgic_retire_disabled_irqs(struct kvm_vcpu
*vcpu
);
105 static void vgic_retire_lr(int lr_nr
, int irq
, struct kvm_vcpu
*vcpu
);
106 static struct vgic_lr
vgic_get_lr(const struct kvm_vcpu
*vcpu
, int lr
);
107 static void vgic_set_lr(struct kvm_vcpu
*vcpu
, int lr
, struct vgic_lr lr_desc
);
108 static struct irq_phys_map
*vgic_irq_map_search(struct kvm_vcpu
*vcpu
,
111 static const struct vgic_ops
*vgic_ops
;
112 static const struct vgic_params
*vgic
;
114 static void add_sgi_source(struct kvm_vcpu
*vcpu
, int irq
, int source
)
116 vcpu
->kvm
->arch
.vgic
.vm_ops
.add_sgi_source(vcpu
, irq
, source
);
119 static bool queue_sgi(struct kvm_vcpu
*vcpu
, int irq
)
121 return vcpu
->kvm
->arch
.vgic
.vm_ops
.queue_sgi(vcpu
, irq
);
124 int kvm_vgic_map_resources(struct kvm
*kvm
)
126 return kvm
->arch
.vgic
.vm_ops
.map_resources(kvm
, vgic
);
130 * struct vgic_bitmap contains a bitmap made of unsigned longs, but
131 * extracts u32s out of them.
133 * This does not work on 64-bit BE systems, because the bitmap access
134 * will store two consecutive 32-bit words with the higher-addressed
135 * register's bits at the lower index and the lower-addressed register's
136 * bits at the higher index.
138 * Therefore, swizzle the register index when accessing the 32-bit word
139 * registers to access the right register's value.
141 #if defined(CONFIG_CPU_BIG_ENDIAN) && BITS_PER_LONG == 64
142 #define REG_OFFSET_SWIZZLE 1
144 #define REG_OFFSET_SWIZZLE 0
147 static int vgic_init_bitmap(struct vgic_bitmap
*b
, int nr_cpus
, int nr_irqs
)
151 nr_longs
= nr_cpus
+ BITS_TO_LONGS(nr_irqs
- VGIC_NR_PRIVATE_IRQS
);
153 b
->private = kzalloc(sizeof(unsigned long) * nr_longs
, GFP_KERNEL
);
157 b
->shared
= b
->private + nr_cpus
;
162 static void vgic_free_bitmap(struct vgic_bitmap
*b
)
170 * Call this function to convert a u64 value to an unsigned long * bitmask
171 * in a way that works on both 32-bit and 64-bit LE and BE platforms.
173 * Warning: Calling this function may modify *val.
175 static unsigned long *u64_to_bitmask(u64
*val
)
177 #if defined(CONFIG_CPU_BIG_ENDIAN) && BITS_PER_LONG == 32
178 *val
= (*val
>> 32) | (*val
<< 32);
180 return (unsigned long *)val
;
183 u32
*vgic_bitmap_get_reg(struct vgic_bitmap
*x
, int cpuid
, u32 offset
)
187 return (u32
*)(x
->private + cpuid
) + REG_OFFSET_SWIZZLE
;
189 return (u32
*)(x
->shared
) + ((offset
- 1) ^ REG_OFFSET_SWIZZLE
);
192 static int vgic_bitmap_get_irq_val(struct vgic_bitmap
*x
,
195 if (irq
< VGIC_NR_PRIVATE_IRQS
)
196 return test_bit(irq
, x
->private + cpuid
);
198 return test_bit(irq
- VGIC_NR_PRIVATE_IRQS
, x
->shared
);
201 void vgic_bitmap_set_irq_val(struct vgic_bitmap
*x
, int cpuid
,
206 if (irq
< VGIC_NR_PRIVATE_IRQS
) {
207 reg
= x
->private + cpuid
;
210 irq
-= VGIC_NR_PRIVATE_IRQS
;
219 static unsigned long *vgic_bitmap_get_cpu_map(struct vgic_bitmap
*x
, int cpuid
)
221 return x
->private + cpuid
;
224 unsigned long *vgic_bitmap_get_shared_map(struct vgic_bitmap
*x
)
229 static int vgic_init_bytemap(struct vgic_bytemap
*x
, int nr_cpus
, int nr_irqs
)
233 size
= nr_cpus
* VGIC_NR_PRIVATE_IRQS
;
234 size
+= nr_irqs
- VGIC_NR_PRIVATE_IRQS
;
236 x
->private = kzalloc(size
, GFP_KERNEL
);
240 x
->shared
= x
->private + nr_cpus
* VGIC_NR_PRIVATE_IRQS
/ sizeof(u32
);
244 static void vgic_free_bytemap(struct vgic_bytemap
*b
)
251 u32
*vgic_bytemap_get_reg(struct vgic_bytemap
*x
, int cpuid
, u32 offset
)
255 if (offset
< VGIC_NR_PRIVATE_IRQS
) {
257 offset
+= cpuid
* VGIC_NR_PRIVATE_IRQS
;
260 offset
-= VGIC_NR_PRIVATE_IRQS
;
263 return reg
+ (offset
/ sizeof(u32
));
266 #define VGIC_CFG_LEVEL 0
267 #define VGIC_CFG_EDGE 1
269 static bool vgic_irq_is_edge(struct kvm_vcpu
*vcpu
, int irq
)
271 struct vgic_dist
*dist
= &vcpu
->kvm
->arch
.vgic
;
274 irq_val
= vgic_bitmap_get_irq_val(&dist
->irq_cfg
, vcpu
->vcpu_id
, irq
);
275 return irq_val
== VGIC_CFG_EDGE
;
278 static int vgic_irq_is_enabled(struct kvm_vcpu
*vcpu
, int irq
)
280 struct vgic_dist
*dist
= &vcpu
->kvm
->arch
.vgic
;
282 return vgic_bitmap_get_irq_val(&dist
->irq_enabled
, vcpu
->vcpu_id
, irq
);
285 static int vgic_irq_is_queued(struct kvm_vcpu
*vcpu
, int irq
)
287 struct vgic_dist
*dist
= &vcpu
->kvm
->arch
.vgic
;
289 return vgic_bitmap_get_irq_val(&dist
->irq_queued
, vcpu
->vcpu_id
, irq
);
292 static int vgic_irq_is_active(struct kvm_vcpu
*vcpu
, int irq
)
294 struct vgic_dist
*dist
= &vcpu
->kvm
->arch
.vgic
;
296 return vgic_bitmap_get_irq_val(&dist
->irq_active
, vcpu
->vcpu_id
, irq
);
299 static void vgic_irq_set_queued(struct kvm_vcpu
*vcpu
, int irq
)
301 struct vgic_dist
*dist
= &vcpu
->kvm
->arch
.vgic
;
303 vgic_bitmap_set_irq_val(&dist
->irq_queued
, vcpu
->vcpu_id
, irq
, 1);
306 static void vgic_irq_clear_queued(struct kvm_vcpu
*vcpu
, int irq
)
308 struct vgic_dist
*dist
= &vcpu
->kvm
->arch
.vgic
;
310 vgic_bitmap_set_irq_val(&dist
->irq_queued
, vcpu
->vcpu_id
, irq
, 0);
313 static void vgic_irq_set_active(struct kvm_vcpu
*vcpu
, int irq
)
315 struct vgic_dist
*dist
= &vcpu
->kvm
->arch
.vgic
;
317 vgic_bitmap_set_irq_val(&dist
->irq_active
, vcpu
->vcpu_id
, irq
, 1);
320 static void vgic_irq_clear_active(struct kvm_vcpu
*vcpu
, int irq
)
322 struct vgic_dist
*dist
= &vcpu
->kvm
->arch
.vgic
;
324 vgic_bitmap_set_irq_val(&dist
->irq_active
, vcpu
->vcpu_id
, irq
, 0);
327 static int vgic_dist_irq_get_level(struct kvm_vcpu
*vcpu
, int irq
)
329 struct vgic_dist
*dist
= &vcpu
->kvm
->arch
.vgic
;
331 return vgic_bitmap_get_irq_val(&dist
->irq_level
, vcpu
->vcpu_id
, irq
);
334 static void vgic_dist_irq_set_level(struct kvm_vcpu
*vcpu
, int irq
)
336 struct vgic_dist
*dist
= &vcpu
->kvm
->arch
.vgic
;
338 vgic_bitmap_set_irq_val(&dist
->irq_level
, vcpu
->vcpu_id
, irq
, 1);
341 static void vgic_dist_irq_clear_level(struct kvm_vcpu
*vcpu
, int irq
)
343 struct vgic_dist
*dist
= &vcpu
->kvm
->arch
.vgic
;
345 vgic_bitmap_set_irq_val(&dist
->irq_level
, vcpu
->vcpu_id
, irq
, 0);
348 static int vgic_dist_irq_soft_pend(struct kvm_vcpu
*vcpu
, int irq
)
350 struct vgic_dist
*dist
= &vcpu
->kvm
->arch
.vgic
;
352 return vgic_bitmap_get_irq_val(&dist
->irq_soft_pend
, vcpu
->vcpu_id
, irq
);
355 static void vgic_dist_irq_clear_soft_pend(struct kvm_vcpu
*vcpu
, int irq
)
357 struct vgic_dist
*dist
= &vcpu
->kvm
->arch
.vgic
;
359 vgic_bitmap_set_irq_val(&dist
->irq_soft_pend
, vcpu
->vcpu_id
, irq
, 0);
362 static int vgic_dist_irq_is_pending(struct kvm_vcpu
*vcpu
, int irq
)
364 struct vgic_dist
*dist
= &vcpu
->kvm
->arch
.vgic
;
366 return vgic_bitmap_get_irq_val(&dist
->irq_pending
, vcpu
->vcpu_id
, irq
);
369 void vgic_dist_irq_set_pending(struct kvm_vcpu
*vcpu
, int irq
)
371 struct vgic_dist
*dist
= &vcpu
->kvm
->arch
.vgic
;
373 vgic_bitmap_set_irq_val(&dist
->irq_pending
, vcpu
->vcpu_id
, irq
, 1);
376 void vgic_dist_irq_clear_pending(struct kvm_vcpu
*vcpu
, int irq
)
378 struct vgic_dist
*dist
= &vcpu
->kvm
->arch
.vgic
;
380 vgic_bitmap_set_irq_val(&dist
->irq_pending
, vcpu
->vcpu_id
, irq
, 0);
383 static void vgic_cpu_irq_set(struct kvm_vcpu
*vcpu
, int irq
)
385 if (irq
< VGIC_NR_PRIVATE_IRQS
)
386 set_bit(irq
, vcpu
->arch
.vgic_cpu
.pending_percpu
);
388 set_bit(irq
- VGIC_NR_PRIVATE_IRQS
,
389 vcpu
->arch
.vgic_cpu
.pending_shared
);
392 void vgic_cpu_irq_clear(struct kvm_vcpu
*vcpu
, int irq
)
394 if (irq
< VGIC_NR_PRIVATE_IRQS
)
395 clear_bit(irq
, vcpu
->arch
.vgic_cpu
.pending_percpu
);
397 clear_bit(irq
- VGIC_NR_PRIVATE_IRQS
,
398 vcpu
->arch
.vgic_cpu
.pending_shared
);
401 static bool vgic_can_sample_irq(struct kvm_vcpu
*vcpu
, int irq
)
403 return !vgic_irq_is_queued(vcpu
, irq
);
407 * vgic_reg_access - access vgic register
408 * @mmio: pointer to the data describing the mmio access
409 * @reg: pointer to the virtual backing of vgic distributor data
410 * @offset: least significant 2 bits used for word offset
411 * @mode: ACCESS_ mode (see defines above)
413 * Helper to make vgic register access easier using one of the access
414 * modes defined for vgic register access
415 * (read,raz,write-ignored,setbit,clearbit,write)
417 void vgic_reg_access(struct kvm_exit_mmio
*mmio
, u32
*reg
,
418 phys_addr_t offset
, int mode
)
420 int word_offset
= (offset
& 3) * 8;
421 u32 mask
= (1UL << (mmio
->len
* 8)) - 1;
425 * Any alignment fault should have been delivered to the guest
426 * directly (ARM ARM B3.12.7 "Prioritization of aborts").
432 BUG_ON(mode
!= (ACCESS_READ_RAZ
| ACCESS_WRITE_IGNORED
));
436 if (mmio
->is_write
) {
437 u32 data
= mmio_data_read(mmio
, mask
) << word_offset
;
438 switch (ACCESS_WRITE_MASK(mode
)) {
439 case ACCESS_WRITE_IGNORED
:
442 case ACCESS_WRITE_SETBIT
:
446 case ACCESS_WRITE_CLEARBIT
:
450 case ACCESS_WRITE_VALUE
:
451 regval
= (regval
& ~(mask
<< word_offset
)) | data
;
456 switch (ACCESS_READ_MASK(mode
)) {
457 case ACCESS_READ_RAZ
:
461 case ACCESS_READ_VALUE
:
462 mmio_data_write(mmio
, mask
, regval
>> word_offset
);
467 bool handle_mmio_raz_wi(struct kvm_vcpu
*vcpu
, struct kvm_exit_mmio
*mmio
,
470 vgic_reg_access(mmio
, NULL
, offset
,
471 ACCESS_READ_RAZ
| ACCESS_WRITE_IGNORED
);
475 bool vgic_handle_enable_reg(struct kvm
*kvm
, struct kvm_exit_mmio
*mmio
,
476 phys_addr_t offset
, int vcpu_id
, int access
)
479 int mode
= ACCESS_READ_VALUE
| access
;
480 struct kvm_vcpu
*target_vcpu
= kvm_get_vcpu(kvm
, vcpu_id
);
482 reg
= vgic_bitmap_get_reg(&kvm
->arch
.vgic
.irq_enabled
, vcpu_id
, offset
);
483 vgic_reg_access(mmio
, reg
, offset
, mode
);
484 if (mmio
->is_write
) {
485 if (access
& ACCESS_WRITE_CLEARBIT
) {
486 if (offset
< 4) /* Force SGI enabled */
488 vgic_retire_disabled_irqs(target_vcpu
);
490 vgic_update_state(kvm
);
497 bool vgic_handle_set_pending_reg(struct kvm
*kvm
,
498 struct kvm_exit_mmio
*mmio
,
499 phys_addr_t offset
, int vcpu_id
)
503 int mode
= ACCESS_READ_VALUE
| ACCESS_WRITE_SETBIT
;
504 struct vgic_dist
*dist
= &kvm
->arch
.vgic
;
506 reg
= vgic_bitmap_get_reg(&dist
->irq_cfg
, vcpu_id
, offset
);
507 level_mask
= (~(*reg
));
509 /* Mark both level and edge triggered irqs as pending */
510 reg
= vgic_bitmap_get_reg(&dist
->irq_pending
, vcpu_id
, offset
);
512 vgic_reg_access(mmio
, reg
, offset
, mode
);
514 if (mmio
->is_write
) {
515 /* Set the soft-pending flag only for level-triggered irqs */
516 reg
= vgic_bitmap_get_reg(&dist
->irq_soft_pend
,
518 vgic_reg_access(mmio
, reg
, offset
, mode
);
521 /* Ignore writes to SGIs */
524 *reg
|= orig
& 0xffff;
527 vgic_update_state(kvm
);
534 bool vgic_handle_clear_pending_reg(struct kvm
*kvm
,
535 struct kvm_exit_mmio
*mmio
,
536 phys_addr_t offset
, int vcpu_id
)
540 int mode
= ACCESS_READ_VALUE
| ACCESS_WRITE_CLEARBIT
;
541 struct vgic_dist
*dist
= &kvm
->arch
.vgic
;
543 reg
= vgic_bitmap_get_reg(&dist
->irq_pending
, vcpu_id
, offset
);
545 vgic_reg_access(mmio
, reg
, offset
, mode
);
546 if (mmio
->is_write
) {
547 /* Re-set level triggered level-active interrupts */
548 level_active
= vgic_bitmap_get_reg(&dist
->irq_level
,
550 reg
= vgic_bitmap_get_reg(&dist
->irq_pending
, vcpu_id
, offset
);
551 *reg
|= *level_active
;
553 /* Ignore writes to SGIs */
556 *reg
|= orig
& 0xffff;
559 /* Clear soft-pending flags */
560 reg
= vgic_bitmap_get_reg(&dist
->irq_soft_pend
,
562 vgic_reg_access(mmio
, reg
, offset
, mode
);
564 vgic_update_state(kvm
);
570 bool vgic_handle_set_active_reg(struct kvm
*kvm
,
571 struct kvm_exit_mmio
*mmio
,
572 phys_addr_t offset
, int vcpu_id
)
575 struct vgic_dist
*dist
= &kvm
->arch
.vgic
;
577 reg
= vgic_bitmap_get_reg(&dist
->irq_active
, vcpu_id
, offset
);
578 vgic_reg_access(mmio
, reg
, offset
,
579 ACCESS_READ_VALUE
| ACCESS_WRITE_SETBIT
);
581 if (mmio
->is_write
) {
582 vgic_update_state(kvm
);
589 bool vgic_handle_clear_active_reg(struct kvm
*kvm
,
590 struct kvm_exit_mmio
*mmio
,
591 phys_addr_t offset
, int vcpu_id
)
594 struct vgic_dist
*dist
= &kvm
->arch
.vgic
;
596 reg
= vgic_bitmap_get_reg(&dist
->irq_active
, vcpu_id
, offset
);
597 vgic_reg_access(mmio
, reg
, offset
,
598 ACCESS_READ_VALUE
| ACCESS_WRITE_CLEARBIT
);
600 if (mmio
->is_write
) {
601 vgic_update_state(kvm
);
608 static u32
vgic_cfg_expand(u16 val
)
614 * Turn a 16bit value like abcd...mnop into a 32bit word
615 * a0b0c0d0...m0n0o0p0, which is what the HW cfg register is.
617 for (i
= 0; i
< 16; i
++)
618 res
|= ((val
>> i
) & VGIC_CFG_EDGE
) << (2 * i
+ 1);
623 static u16
vgic_cfg_compress(u32 val
)
629 * Turn a 32bit word a0b0c0d0...m0n0o0p0 into 16bit value like
630 * abcd...mnop which is what we really care about.
632 for (i
= 0; i
< 16; i
++)
633 res
|= ((val
>> (i
* 2 + 1)) & VGIC_CFG_EDGE
) << i
;
639 * The distributor uses 2 bits per IRQ for the CFG register, but the
640 * LSB is always 0. As such, we only keep the upper bit, and use the
641 * two above functions to compress/expand the bits
643 bool vgic_handle_cfg_reg(u32
*reg
, struct kvm_exit_mmio
*mmio
,
653 val
= vgic_cfg_expand(val
);
654 vgic_reg_access(mmio
, &val
, offset
,
655 ACCESS_READ_VALUE
| ACCESS_WRITE_VALUE
);
656 if (mmio
->is_write
) {
658 *reg
= ~0U; /* Force PPIs/SGIs to 1 */
662 val
= vgic_cfg_compress(val
);
667 *reg
&= 0xffff << 16;
676 * vgic_unqueue_irqs - move pending/active IRQs from LRs to the distributor
677 * @vgic_cpu: Pointer to the vgic_cpu struct holding the LRs
679 * Move any IRQs that have already been assigned to LRs back to the
680 * emulated distributor state so that the complete emulated state can be read
681 * from the main emulation structures without investigating the LRs.
683 void vgic_unqueue_irqs(struct kvm_vcpu
*vcpu
)
685 struct vgic_cpu
*vgic_cpu
= &vcpu
->arch
.vgic_cpu
;
688 for_each_set_bit(i
, vgic_cpu
->lr_used
, vgic_cpu
->nr_lr
) {
689 struct vgic_lr lr
= vgic_get_lr(vcpu
, i
);
692 * There are three options for the state bits:
696 * 11: pending and active
698 BUG_ON(!(lr
.state
& LR_STATE_MASK
));
700 /* Reestablish SGI source for pending and active IRQs */
701 if (lr
.irq
< VGIC_NR_SGIS
)
702 add_sgi_source(vcpu
, lr
.irq
, lr
.source
);
705 * If the LR holds an active (10) or a pending and active (11)
706 * interrupt then move the active state to the
707 * distributor tracking bit.
709 if (lr
.state
& LR_STATE_ACTIVE
) {
710 vgic_irq_set_active(vcpu
, lr
.irq
);
711 lr
.state
&= ~LR_STATE_ACTIVE
;
715 * Reestablish the pending state on the distributor and the
716 * CPU interface. It may have already been pending, but that
717 * is fine, then we are only setting a few bits that were
720 if (lr
.state
& LR_STATE_PENDING
) {
721 vgic_dist_irq_set_pending(vcpu
, lr
.irq
);
722 lr
.state
&= ~LR_STATE_PENDING
;
725 vgic_set_lr(vcpu
, i
, lr
);
728 * Mark the LR as free for other use.
730 BUG_ON(lr
.state
& LR_STATE_MASK
);
731 vgic_retire_lr(i
, lr
.irq
, vcpu
);
732 vgic_irq_clear_queued(vcpu
, lr
.irq
);
734 /* Finally update the VGIC state. */
735 vgic_update_state(vcpu
->kvm
);
740 struct vgic_io_range
*vgic_find_range(const struct vgic_io_range
*ranges
,
741 int len
, gpa_t offset
)
743 while (ranges
->len
) {
744 if (offset
>= ranges
->base
&&
745 (offset
+ len
) <= (ranges
->base
+ ranges
->len
))
753 static bool vgic_validate_access(const struct vgic_dist
*dist
,
754 const struct vgic_io_range
*range
,
755 unsigned long offset
)
759 if (!range
->bits_per_irq
)
760 return true; /* Not an irq-based access */
762 irq
= offset
* 8 / range
->bits_per_irq
;
763 if (irq
>= dist
->nr_irqs
)
770 * Call the respective handler function for the given range.
771 * We split up any 64 bit accesses into two consecutive 32 bit
772 * handler calls and merge the result afterwards.
773 * We do this in a little endian fashion regardless of the host's
774 * or guest's endianness, because the GIC is always LE and the rest of
775 * the code (vgic_reg_access) also puts it in a LE fashion already.
776 * At this point we have already identified the handle function, so
777 * range points to that one entry and offset is relative to this.
779 static bool call_range_handler(struct kvm_vcpu
*vcpu
,
780 struct kvm_exit_mmio
*mmio
,
781 unsigned long offset
,
782 const struct vgic_io_range
*range
)
784 struct kvm_exit_mmio mmio32
;
787 if (likely(mmio
->len
<= 4))
788 return range
->handle_mmio(vcpu
, mmio
, offset
);
791 * Any access bigger than 4 bytes (that we currently handle in KVM)
792 * is actually 8 bytes long, caused by a 64-bit access
796 mmio32
.is_write
= mmio
->is_write
;
797 mmio32
.private = mmio
->private;
799 mmio32
.phys_addr
= mmio
->phys_addr
+ 4;
800 mmio32
.data
= &((u32
*)mmio
->data
)[1];
801 ret
= range
->handle_mmio(vcpu
, &mmio32
, offset
+ 4);
803 mmio32
.phys_addr
= mmio
->phys_addr
;
804 mmio32
.data
= &((u32
*)mmio
->data
)[0];
805 ret
|= range
->handle_mmio(vcpu
, &mmio32
, offset
);
811 * vgic_handle_mmio_access - handle an in-kernel MMIO access
812 * This is called by the read/write KVM IO device wrappers below.
813 * @vcpu: pointer to the vcpu performing the access
814 * @this: pointer to the KVM IO device in charge
815 * @addr: guest physical address of the access
816 * @len: size of the access
817 * @val: pointer to the data region
818 * @is_write: read or write access
820 * returns true if the MMIO access could be performed
822 static int vgic_handle_mmio_access(struct kvm_vcpu
*vcpu
,
823 struct kvm_io_device
*this, gpa_t addr
,
824 int len
, void *val
, bool is_write
)
826 struct vgic_dist
*dist
= &vcpu
->kvm
->arch
.vgic
;
827 struct vgic_io_device
*iodev
= container_of(this,
828 struct vgic_io_device
, dev
);
829 struct kvm_run
*run
= vcpu
->run
;
830 const struct vgic_io_range
*range
;
831 struct kvm_exit_mmio mmio
;
835 offset
= addr
- iodev
->addr
;
836 range
= vgic_find_range(iodev
->reg_ranges
, len
, offset
);
837 if (unlikely(!range
|| !range
->handle_mmio
)) {
838 pr_warn("Unhandled access %d %08llx %d\n", is_write
, addr
, len
);
842 mmio
.phys_addr
= addr
;
844 mmio
.is_write
= is_write
;
846 mmio
.private = iodev
->redist_vcpu
;
848 spin_lock(&dist
->lock
);
849 offset
-= range
->base
;
850 if (vgic_validate_access(dist
, range
, offset
)) {
851 updated_state
= call_range_handler(vcpu
, &mmio
, offset
, range
);
855 updated_state
= false;
857 spin_unlock(&dist
->lock
);
858 run
->mmio
.is_write
= is_write
;
860 run
->mmio
.phys_addr
= addr
;
861 memcpy(run
->mmio
.data
, val
, len
);
863 kvm_handle_mmio_return(vcpu
, run
);
866 vgic_kick_vcpus(vcpu
->kvm
);
871 static int vgic_handle_mmio_read(struct kvm_vcpu
*vcpu
,
872 struct kvm_io_device
*this,
873 gpa_t addr
, int len
, void *val
)
875 return vgic_handle_mmio_access(vcpu
, this, addr
, len
, val
, false);
878 static int vgic_handle_mmio_write(struct kvm_vcpu
*vcpu
,
879 struct kvm_io_device
*this,
880 gpa_t addr
, int len
, const void *val
)
882 return vgic_handle_mmio_access(vcpu
, this, addr
, len
, (void *)val
,
886 struct kvm_io_device_ops vgic_io_ops
= {
887 .read
= vgic_handle_mmio_read
,
888 .write
= vgic_handle_mmio_write
,
892 * vgic_register_kvm_io_dev - register VGIC register frame on the KVM I/O bus
893 * @kvm: The VM structure pointer
894 * @base: The (guest) base address for the register frame
895 * @len: Length of the register frame window
896 * @ranges: Describing the handler functions for each register
897 * @redist_vcpu_id: The VCPU ID to pass on to the handlers on call
898 * @iodev: Points to memory to be passed on to the handler
900 * @iodev stores the parameters of this function to be usable by the handler
901 * respectively the dispatcher function (since the KVM I/O bus framework lacks
902 * an opaque parameter). Initialization is done in this function, but the
903 * reference should be valid and unique for the whole VGIC lifetime.
904 * If the register frame is not mapped for a specific VCPU, pass -1 to
907 int vgic_register_kvm_io_dev(struct kvm
*kvm
, gpa_t base
, int len
,
908 const struct vgic_io_range
*ranges
,
910 struct vgic_io_device
*iodev
)
912 struct kvm_vcpu
*vcpu
= NULL
;
915 if (redist_vcpu_id
>= 0)
916 vcpu
= kvm_get_vcpu(kvm
, redist_vcpu_id
);
920 iodev
->reg_ranges
= ranges
;
921 iodev
->redist_vcpu
= vcpu
;
923 kvm_iodevice_init(&iodev
->dev
, &vgic_io_ops
);
925 mutex_lock(&kvm
->slots_lock
);
927 ret
= kvm_io_bus_register_dev(kvm
, KVM_MMIO_BUS
, base
, len
,
929 mutex_unlock(&kvm
->slots_lock
);
931 /* Mark the iodev as invalid if registration fails. */
933 iodev
->dev
.ops
= NULL
;
938 static int vgic_nr_shared_irqs(struct vgic_dist
*dist
)
940 return dist
->nr_irqs
- VGIC_NR_PRIVATE_IRQS
;
943 static int compute_active_for_cpu(struct kvm_vcpu
*vcpu
)
945 struct vgic_dist
*dist
= &vcpu
->kvm
->arch
.vgic
;
946 unsigned long *active
, *enabled
, *act_percpu
, *act_shared
;
947 unsigned long active_private
, active_shared
;
948 int nr_shared
= vgic_nr_shared_irqs(dist
);
951 vcpu_id
= vcpu
->vcpu_id
;
952 act_percpu
= vcpu
->arch
.vgic_cpu
.active_percpu
;
953 act_shared
= vcpu
->arch
.vgic_cpu
.active_shared
;
955 active
= vgic_bitmap_get_cpu_map(&dist
->irq_active
, vcpu_id
);
956 enabled
= vgic_bitmap_get_cpu_map(&dist
->irq_enabled
, vcpu_id
);
957 bitmap_and(act_percpu
, active
, enabled
, VGIC_NR_PRIVATE_IRQS
);
959 active
= vgic_bitmap_get_shared_map(&dist
->irq_active
);
960 enabled
= vgic_bitmap_get_shared_map(&dist
->irq_enabled
);
961 bitmap_and(act_shared
, active
, enabled
, nr_shared
);
962 bitmap_and(act_shared
, act_shared
,
963 vgic_bitmap_get_shared_map(&dist
->irq_spi_target
[vcpu_id
]),
966 active_private
= find_first_bit(act_percpu
, VGIC_NR_PRIVATE_IRQS
);
967 active_shared
= find_first_bit(act_shared
, nr_shared
);
969 return (active_private
< VGIC_NR_PRIVATE_IRQS
||
970 active_shared
< nr_shared
);
973 static int compute_pending_for_cpu(struct kvm_vcpu
*vcpu
)
975 struct vgic_dist
*dist
= &vcpu
->kvm
->arch
.vgic
;
976 unsigned long *pending
, *enabled
, *pend_percpu
, *pend_shared
;
977 unsigned long pending_private
, pending_shared
;
978 int nr_shared
= vgic_nr_shared_irqs(dist
);
981 vcpu_id
= vcpu
->vcpu_id
;
982 pend_percpu
= vcpu
->arch
.vgic_cpu
.pending_percpu
;
983 pend_shared
= vcpu
->arch
.vgic_cpu
.pending_shared
;
985 pending
= vgic_bitmap_get_cpu_map(&dist
->irq_pending
, vcpu_id
);
986 enabled
= vgic_bitmap_get_cpu_map(&dist
->irq_enabled
, vcpu_id
);
987 bitmap_and(pend_percpu
, pending
, enabled
, VGIC_NR_PRIVATE_IRQS
);
989 pending
= vgic_bitmap_get_shared_map(&dist
->irq_pending
);
990 enabled
= vgic_bitmap_get_shared_map(&dist
->irq_enabled
);
991 bitmap_and(pend_shared
, pending
, enabled
, nr_shared
);
992 bitmap_and(pend_shared
, pend_shared
,
993 vgic_bitmap_get_shared_map(&dist
->irq_spi_target
[vcpu_id
]),
996 pending_private
= find_first_bit(pend_percpu
, VGIC_NR_PRIVATE_IRQS
);
997 pending_shared
= find_first_bit(pend_shared
, nr_shared
);
998 return (pending_private
< VGIC_NR_PRIVATE_IRQS
||
999 pending_shared
< vgic_nr_shared_irqs(dist
));
1003 * Update the interrupt state and determine which CPUs have pending
1004 * or active interrupts. Must be called with distributor lock held.
1006 void vgic_update_state(struct kvm
*kvm
)
1008 struct vgic_dist
*dist
= &kvm
->arch
.vgic
;
1009 struct kvm_vcpu
*vcpu
;
1012 if (!dist
->enabled
) {
1013 set_bit(0, dist
->irq_pending_on_cpu
);
1017 kvm_for_each_vcpu(c
, vcpu
, kvm
) {
1018 if (compute_pending_for_cpu(vcpu
))
1019 set_bit(c
, dist
->irq_pending_on_cpu
);
1021 if (compute_active_for_cpu(vcpu
))
1022 set_bit(c
, dist
->irq_active_on_cpu
);
1024 clear_bit(c
, dist
->irq_active_on_cpu
);
1028 static struct vgic_lr
vgic_get_lr(const struct kvm_vcpu
*vcpu
, int lr
)
1030 return vgic_ops
->get_lr(vcpu
, lr
);
1033 static void vgic_set_lr(struct kvm_vcpu
*vcpu
, int lr
,
1036 vgic_ops
->set_lr(vcpu
, lr
, vlr
);
1039 static void vgic_sync_lr_elrsr(struct kvm_vcpu
*vcpu
, int lr
,
1042 vgic_ops
->sync_lr_elrsr(vcpu
, lr
, vlr
);
1045 static inline u64
vgic_get_elrsr(struct kvm_vcpu
*vcpu
)
1047 return vgic_ops
->get_elrsr(vcpu
);
1050 static inline u64
vgic_get_eisr(struct kvm_vcpu
*vcpu
)
1052 return vgic_ops
->get_eisr(vcpu
);
1055 static inline void vgic_clear_eisr(struct kvm_vcpu
*vcpu
)
1057 vgic_ops
->clear_eisr(vcpu
);
1060 static inline u32
vgic_get_interrupt_status(struct kvm_vcpu
*vcpu
)
1062 return vgic_ops
->get_interrupt_status(vcpu
);
1065 static inline void vgic_enable_underflow(struct kvm_vcpu
*vcpu
)
1067 vgic_ops
->enable_underflow(vcpu
);
1070 static inline void vgic_disable_underflow(struct kvm_vcpu
*vcpu
)
1072 vgic_ops
->disable_underflow(vcpu
);
1075 void vgic_get_vmcr(struct kvm_vcpu
*vcpu
, struct vgic_vmcr
*vmcr
)
1077 vgic_ops
->get_vmcr(vcpu
, vmcr
);
1080 void vgic_set_vmcr(struct kvm_vcpu
*vcpu
, struct vgic_vmcr
*vmcr
)
1082 vgic_ops
->set_vmcr(vcpu
, vmcr
);
1085 static inline void vgic_enable(struct kvm_vcpu
*vcpu
)
1087 vgic_ops
->enable(vcpu
);
1090 static void vgic_retire_lr(int lr_nr
, int irq
, struct kvm_vcpu
*vcpu
)
1092 struct vgic_cpu
*vgic_cpu
= &vcpu
->arch
.vgic_cpu
;
1093 struct vgic_lr vlr
= vgic_get_lr(vcpu
, lr_nr
);
1096 * We must transfer the pending state back to the distributor before
1097 * retiring the LR, otherwise we may loose edge-triggered interrupts.
1099 if (vlr
.state
& LR_STATE_PENDING
) {
1100 vgic_dist_irq_set_pending(vcpu
, irq
);
1105 vgic_set_lr(vcpu
, lr_nr
, vlr
);
1106 clear_bit(lr_nr
, vgic_cpu
->lr_used
);
1107 vgic_cpu
->vgic_irq_lr_map
[irq
] = LR_EMPTY
;
1108 vgic_sync_lr_elrsr(vcpu
, lr_nr
, vlr
);
1112 * An interrupt may have been disabled after being made pending on the
1113 * CPU interface (the classic case is a timer running while we're
1114 * rebooting the guest - the interrupt would kick as soon as the CPU
1115 * interface gets enabled, with deadly consequences).
1117 * The solution is to examine already active LRs, and check the
1118 * interrupt is still enabled. If not, just retire it.
1120 static void vgic_retire_disabled_irqs(struct kvm_vcpu
*vcpu
)
1122 struct vgic_cpu
*vgic_cpu
= &vcpu
->arch
.vgic_cpu
;
1125 for_each_set_bit(lr
, vgic_cpu
->lr_used
, vgic
->nr_lr
) {
1126 struct vgic_lr vlr
= vgic_get_lr(vcpu
, lr
);
1128 if (!vgic_irq_is_enabled(vcpu
, vlr
.irq
)) {
1129 vgic_retire_lr(lr
, vlr
.irq
, vcpu
);
1130 if (vgic_irq_is_queued(vcpu
, vlr
.irq
))
1131 vgic_irq_clear_queued(vcpu
, vlr
.irq
);
1136 static void vgic_queue_irq_to_lr(struct kvm_vcpu
*vcpu
, int irq
,
1137 int lr_nr
, struct vgic_lr vlr
)
1139 if (vgic_irq_is_active(vcpu
, irq
)) {
1140 vlr
.state
|= LR_STATE_ACTIVE
;
1141 kvm_debug("Set active, clear distributor: 0x%x\n", vlr
.state
);
1142 vgic_irq_clear_active(vcpu
, irq
);
1143 vgic_update_state(vcpu
->kvm
);
1145 WARN_ON(!vgic_dist_irq_is_pending(vcpu
, irq
));
1146 vlr
.state
|= LR_STATE_PENDING
;
1147 kvm_debug("Set pending: 0x%x\n", vlr
.state
);
1150 if (!vgic_irq_is_edge(vcpu
, irq
))
1151 vlr
.state
|= LR_EOI_INT
;
1153 if (vlr
.irq
>= VGIC_NR_SGIS
) {
1154 struct irq_phys_map
*map
;
1155 map
= vgic_irq_map_search(vcpu
, irq
);
1158 vlr
.hwirq
= map
->phys_irq
;
1160 vlr
.state
&= ~LR_EOI_INT
;
1163 * Make sure we're not going to sample this
1164 * again, as a HW-backed interrupt cannot be
1165 * in the PENDING_ACTIVE stage.
1167 vgic_irq_set_queued(vcpu
, irq
);
1171 vgic_set_lr(vcpu
, lr_nr
, vlr
);
1172 vgic_sync_lr_elrsr(vcpu
, lr_nr
, vlr
);
1176 * Queue an interrupt to a CPU virtual interface. Return true on success,
1177 * or false if it wasn't possible to queue it.
1178 * sgi_source must be zero for any non-SGI interrupts.
1180 bool vgic_queue_irq(struct kvm_vcpu
*vcpu
, u8 sgi_source_id
, int irq
)
1182 struct vgic_cpu
*vgic_cpu
= &vcpu
->arch
.vgic_cpu
;
1183 struct vgic_dist
*dist
= &vcpu
->kvm
->arch
.vgic
;
1187 /* Sanitize the input... */
1188 BUG_ON(sgi_source_id
& ~7);
1189 BUG_ON(sgi_source_id
&& irq
>= VGIC_NR_SGIS
);
1190 BUG_ON(irq
>= dist
->nr_irqs
);
1192 kvm_debug("Queue IRQ%d\n", irq
);
1194 lr
= vgic_cpu
->vgic_irq_lr_map
[irq
];
1196 /* Do we have an active interrupt for the same CPUID? */
1197 if (lr
!= LR_EMPTY
) {
1198 vlr
= vgic_get_lr(vcpu
, lr
);
1199 if (vlr
.source
== sgi_source_id
) {
1200 kvm_debug("LR%d piggyback for IRQ%d\n", lr
, vlr
.irq
);
1201 BUG_ON(!test_bit(lr
, vgic_cpu
->lr_used
));
1202 vgic_queue_irq_to_lr(vcpu
, irq
, lr
, vlr
);
1207 /* Try to use another LR for this interrupt */
1208 lr
= find_first_zero_bit((unsigned long *)vgic_cpu
->lr_used
,
1210 if (lr
>= vgic
->nr_lr
)
1213 kvm_debug("LR%d allocated for IRQ%d %x\n", lr
, irq
, sgi_source_id
);
1214 vgic_cpu
->vgic_irq_lr_map
[irq
] = lr
;
1215 set_bit(lr
, vgic_cpu
->lr_used
);
1218 vlr
.source
= sgi_source_id
;
1220 vgic_queue_irq_to_lr(vcpu
, irq
, lr
, vlr
);
1225 static bool vgic_queue_hwirq(struct kvm_vcpu
*vcpu
, int irq
)
1227 if (!vgic_can_sample_irq(vcpu
, irq
))
1228 return true; /* level interrupt, already queued */
1230 if (vgic_queue_irq(vcpu
, 0, irq
)) {
1231 if (vgic_irq_is_edge(vcpu
, irq
)) {
1232 vgic_dist_irq_clear_pending(vcpu
, irq
);
1233 vgic_cpu_irq_clear(vcpu
, irq
);
1235 vgic_irq_set_queued(vcpu
, irq
);
1245 * Fill the list registers with pending interrupts before running the
1248 static void __kvm_vgic_flush_hwstate(struct kvm_vcpu
*vcpu
)
1250 struct vgic_cpu
*vgic_cpu
= &vcpu
->arch
.vgic_cpu
;
1251 struct vgic_dist
*dist
= &vcpu
->kvm
->arch
.vgic
;
1252 unsigned long *pa_percpu
, *pa_shared
;
1255 int nr_shared
= vgic_nr_shared_irqs(dist
);
1257 vcpu_id
= vcpu
->vcpu_id
;
1259 pa_percpu
= vcpu
->arch
.vgic_cpu
.pend_act_percpu
;
1260 pa_shared
= vcpu
->arch
.vgic_cpu
.pend_act_shared
;
1262 bitmap_or(pa_percpu
, vgic_cpu
->pending_percpu
, vgic_cpu
->active_percpu
,
1263 VGIC_NR_PRIVATE_IRQS
);
1264 bitmap_or(pa_shared
, vgic_cpu
->pending_shared
, vgic_cpu
->active_shared
,
1267 * We may not have any pending interrupt, or the interrupts
1268 * may have been serviced from another vcpu. In all cases,
1271 if (!kvm_vgic_vcpu_pending_irq(vcpu
) && !kvm_vgic_vcpu_active_irq(vcpu
))
1275 for_each_set_bit(i
, pa_percpu
, VGIC_NR_SGIS
) {
1276 if (!queue_sgi(vcpu
, i
))
1281 for_each_set_bit_from(i
, pa_percpu
, VGIC_NR_PRIVATE_IRQS
) {
1282 if (!vgic_queue_hwirq(vcpu
, i
))
1287 for_each_set_bit(i
, pa_shared
, nr_shared
) {
1288 if (!vgic_queue_hwirq(vcpu
, i
+ VGIC_NR_PRIVATE_IRQS
))
1297 vgic_enable_underflow(vcpu
);
1299 vgic_disable_underflow(vcpu
);
1301 * We're about to run this VCPU, and we've consumed
1302 * everything the distributor had in store for
1303 * us. Claim we don't have anything pending. We'll
1304 * adjust that if needed while exiting.
1306 clear_bit(vcpu_id
, dist
->irq_pending_on_cpu
);
1310 static bool vgic_process_maintenance(struct kvm_vcpu
*vcpu
)
1312 u32 status
= vgic_get_interrupt_status(vcpu
);
1313 struct vgic_dist
*dist
= &vcpu
->kvm
->arch
.vgic
;
1314 bool level_pending
= false;
1315 struct kvm
*kvm
= vcpu
->kvm
;
1317 kvm_debug("STATUS = %08x\n", status
);
1319 if (status
& INT_STATUS_EOI
) {
1321 * Some level interrupts have been EOIed. Clear their
1324 u64 eisr
= vgic_get_eisr(vcpu
);
1325 unsigned long *eisr_ptr
= u64_to_bitmask(&eisr
);
1328 for_each_set_bit(lr
, eisr_ptr
, vgic
->nr_lr
) {
1329 struct vgic_lr vlr
= vgic_get_lr(vcpu
, lr
);
1330 WARN_ON(vgic_irq_is_edge(vcpu
, vlr
.irq
));
1332 spin_lock(&dist
->lock
);
1333 vgic_irq_clear_queued(vcpu
, vlr
.irq
);
1334 WARN_ON(vlr
.state
& LR_STATE_MASK
);
1336 vgic_set_lr(vcpu
, lr
, vlr
);
1339 * If the IRQ was EOIed it was also ACKed and we we
1340 * therefore assume we can clear the soft pending
1341 * state (should it had been set) for this interrupt.
1343 * Note: if the IRQ soft pending state was set after
1344 * the IRQ was acked, it actually shouldn't be
1345 * cleared, but we have no way of knowing that unless
1346 * we start trapping ACKs when the soft-pending state
1349 vgic_dist_irq_clear_soft_pend(vcpu
, vlr
.irq
);
1352 * kvm_notify_acked_irq calls kvm_set_irq()
1353 * to reset the IRQ level. Need to release the
1354 * lock for kvm_set_irq to grab it.
1356 spin_unlock(&dist
->lock
);
1358 kvm_notify_acked_irq(kvm
, 0,
1359 vlr
.irq
- VGIC_NR_PRIVATE_IRQS
);
1360 spin_lock(&dist
->lock
);
1362 /* Any additional pending interrupt? */
1363 if (vgic_dist_irq_get_level(vcpu
, vlr
.irq
)) {
1364 vgic_cpu_irq_set(vcpu
, vlr
.irq
);
1365 level_pending
= true;
1367 vgic_dist_irq_clear_pending(vcpu
, vlr
.irq
);
1368 vgic_cpu_irq_clear(vcpu
, vlr
.irq
);
1371 spin_unlock(&dist
->lock
);
1374 * Despite being EOIed, the LR may not have
1375 * been marked as empty.
1377 vgic_sync_lr_elrsr(vcpu
, lr
, vlr
);
1381 if (status
& INT_STATUS_UNDERFLOW
)
1382 vgic_disable_underflow(vcpu
);
1385 * In the next iterations of the vcpu loop, if we sync the vgic state
1386 * after flushing it, but before entering the guest (this happens for
1387 * pending signals and vmid rollovers), then make sure we don't pick
1388 * up any old maintenance interrupts here.
1390 vgic_clear_eisr(vcpu
);
1392 return level_pending
;
1396 * Save the physical active state, and reset it to inactive.
1398 * Return 1 if HW interrupt went from active to inactive, and 0 otherwise.
1400 static int vgic_sync_hwirq(struct kvm_vcpu
*vcpu
, struct vgic_lr vlr
)
1402 struct irq_phys_map
*map
;
1405 if (!(vlr
.state
& LR_HW
))
1408 map
= vgic_irq_map_search(vcpu
, vlr
.irq
);
1409 BUG_ON(!map
|| !map
->active
);
1411 ret
= irq_get_irqchip_state(map
->irq
,
1412 IRQCHIP_STATE_ACTIVE
,
1423 /* Sync back the VGIC state after a guest run */
1424 static void __kvm_vgic_sync_hwstate(struct kvm_vcpu
*vcpu
)
1426 struct vgic_cpu
*vgic_cpu
= &vcpu
->arch
.vgic_cpu
;
1427 struct vgic_dist
*dist
= &vcpu
->kvm
->arch
.vgic
;
1429 unsigned long *elrsr_ptr
;
1433 level_pending
= vgic_process_maintenance(vcpu
);
1434 elrsr
= vgic_get_elrsr(vcpu
);
1435 elrsr_ptr
= u64_to_bitmask(&elrsr
);
1437 /* Deal with HW interrupts, and clear mappings for empty LRs */
1438 for (lr
= 0; lr
< vgic
->nr_lr
; lr
++) {
1441 if (!test_bit(lr
, vgic_cpu
->lr_used
))
1444 vlr
= vgic_get_lr(vcpu
, lr
);
1445 if (vgic_sync_hwirq(vcpu
, vlr
)) {
1447 * So this is a HW interrupt that the guest
1448 * EOI-ed. Clean the LR state and allow the
1449 * interrupt to be sampled again.
1453 vgic_set_lr(vcpu
, lr
, vlr
);
1454 vgic_irq_clear_queued(vcpu
, vlr
.irq
);
1455 set_bit(lr
, elrsr_ptr
);
1458 if (!test_bit(lr
, elrsr_ptr
))
1461 clear_bit(lr
, vgic_cpu
->lr_used
);
1463 BUG_ON(vlr
.irq
>= dist
->nr_irqs
);
1464 vgic_cpu
->vgic_irq_lr_map
[vlr
.irq
] = LR_EMPTY
;
1467 /* Check if we still have something up our sleeve... */
1468 pending
= find_first_zero_bit(elrsr_ptr
, vgic
->nr_lr
);
1469 if (level_pending
|| pending
< vgic
->nr_lr
)
1470 set_bit(vcpu
->vcpu_id
, dist
->irq_pending_on_cpu
);
1473 void kvm_vgic_flush_hwstate(struct kvm_vcpu
*vcpu
)
1475 struct vgic_dist
*dist
= &vcpu
->kvm
->arch
.vgic
;
1477 if (!irqchip_in_kernel(vcpu
->kvm
))
1480 spin_lock(&dist
->lock
);
1481 __kvm_vgic_flush_hwstate(vcpu
);
1482 spin_unlock(&dist
->lock
);
1485 void kvm_vgic_sync_hwstate(struct kvm_vcpu
*vcpu
)
1487 if (!irqchip_in_kernel(vcpu
->kvm
))
1490 __kvm_vgic_sync_hwstate(vcpu
);
1493 int kvm_vgic_vcpu_pending_irq(struct kvm_vcpu
*vcpu
)
1495 struct vgic_dist
*dist
= &vcpu
->kvm
->arch
.vgic
;
1497 if (!irqchip_in_kernel(vcpu
->kvm
))
1500 return test_bit(vcpu
->vcpu_id
, dist
->irq_pending_on_cpu
);
1503 int kvm_vgic_vcpu_active_irq(struct kvm_vcpu
*vcpu
)
1505 struct vgic_dist
*dist
= &vcpu
->kvm
->arch
.vgic
;
1507 if (!irqchip_in_kernel(vcpu
->kvm
))
1510 return test_bit(vcpu
->vcpu_id
, dist
->irq_active_on_cpu
);
1514 void vgic_kick_vcpus(struct kvm
*kvm
)
1516 struct kvm_vcpu
*vcpu
;
1520 * We've injected an interrupt, time to find out who deserves
1523 kvm_for_each_vcpu(c
, vcpu
, kvm
) {
1524 if (kvm_vgic_vcpu_pending_irq(vcpu
))
1525 kvm_vcpu_kick(vcpu
);
1529 static int vgic_validate_injection(struct kvm_vcpu
*vcpu
, int irq
, int level
)
1531 int edge_triggered
= vgic_irq_is_edge(vcpu
, irq
);
1534 * Only inject an interrupt if:
1535 * - edge triggered and we have a rising edge
1536 * - level triggered and we change level
1538 if (edge_triggered
) {
1539 int state
= vgic_dist_irq_is_pending(vcpu
, irq
);
1540 return level
> state
;
1542 int state
= vgic_dist_irq_get_level(vcpu
, irq
);
1543 return level
!= state
;
1547 static int vgic_update_irq_pending(struct kvm
*kvm
, int cpuid
,
1548 struct irq_phys_map
*map
,
1549 unsigned int irq_num
, bool level
)
1551 struct vgic_dist
*dist
= &kvm
->arch
.vgic
;
1552 struct kvm_vcpu
*vcpu
;
1553 int edge_triggered
, level_triggered
;
1555 bool ret
= true, can_inject
= true;
1557 if (irq_num
>= min(kvm
->arch
.vgic
.nr_irqs
, 1020))
1560 spin_lock(&dist
->lock
);
1562 vcpu
= kvm_get_vcpu(kvm
, cpuid
);
1563 edge_triggered
= vgic_irq_is_edge(vcpu
, irq_num
);
1564 level_triggered
= !edge_triggered
;
1566 if (!vgic_validate_injection(vcpu
, irq_num
, level
)) {
1571 if (irq_num
>= VGIC_NR_PRIVATE_IRQS
) {
1572 cpuid
= dist
->irq_spi_cpu
[irq_num
- VGIC_NR_PRIVATE_IRQS
];
1573 if (cpuid
== VCPU_NOT_ALLOCATED
) {
1574 /* Pretend we use CPU0, and prevent injection */
1578 vcpu
= kvm_get_vcpu(kvm
, cpuid
);
1581 kvm_debug("Inject IRQ%d level %d CPU%d\n", irq_num
, level
, cpuid
);
1584 if (level_triggered
)
1585 vgic_dist_irq_set_level(vcpu
, irq_num
);
1586 vgic_dist_irq_set_pending(vcpu
, irq_num
);
1588 if (level_triggered
) {
1589 vgic_dist_irq_clear_level(vcpu
, irq_num
);
1590 if (!vgic_dist_irq_soft_pend(vcpu
, irq_num
)) {
1591 vgic_dist_irq_clear_pending(vcpu
, irq_num
);
1592 vgic_cpu_irq_clear(vcpu
, irq_num
);
1593 if (!compute_pending_for_cpu(vcpu
))
1594 clear_bit(cpuid
, dist
->irq_pending_on_cpu
);
1602 enabled
= vgic_irq_is_enabled(vcpu
, irq_num
);
1604 if (!enabled
|| !can_inject
) {
1609 if (!vgic_can_sample_irq(vcpu
, irq_num
)) {
1611 * Level interrupt in progress, will be picked up
1619 vgic_cpu_irq_set(vcpu
, irq_num
);
1620 set_bit(cpuid
, dist
->irq_pending_on_cpu
);
1624 spin_unlock(&dist
->lock
);
1627 /* kick the specified vcpu */
1628 kvm_vcpu_kick(kvm_get_vcpu(kvm
, cpuid
));
1634 static int vgic_lazy_init(struct kvm
*kvm
)
1638 if (unlikely(!vgic_initialized(kvm
))) {
1640 * We only provide the automatic initialization of the VGIC
1641 * for the legacy case of a GICv2. Any other type must
1642 * be explicitly initialized once setup with the respective
1645 if (kvm
->arch
.vgic
.vgic_model
!= KVM_DEV_TYPE_ARM_VGIC_V2
)
1648 mutex_lock(&kvm
->lock
);
1649 ret
= vgic_init(kvm
);
1650 mutex_unlock(&kvm
->lock
);
1657 * kvm_vgic_inject_irq - Inject an IRQ from a device to the vgic
1658 * @kvm: The VM structure pointer
1659 * @cpuid: The CPU for PPIs
1660 * @irq_num: The IRQ number that is assigned to the device. This IRQ
1661 * must not be mapped to a HW interrupt.
1662 * @level: Edge-triggered: true: to trigger the interrupt
1663 * false: to ignore the call
1664 * Level-sensitive true: raise the input signal
1665 * false: lower the input signal
1667 * The GIC is not concerned with devices being active-LOW or active-HIGH for
1668 * level-sensitive interrupts. You can think of the level parameter as 1
1669 * being HIGH and 0 being LOW and all devices being active-HIGH.
1671 int kvm_vgic_inject_irq(struct kvm
*kvm
, int cpuid
, unsigned int irq_num
,
1674 struct irq_phys_map
*map
;
1677 ret
= vgic_lazy_init(kvm
);
1681 map
= vgic_irq_map_search(kvm_get_vcpu(kvm
, cpuid
), irq_num
);
1685 return vgic_update_irq_pending(kvm
, cpuid
, NULL
, irq_num
, level
);
1689 * kvm_vgic_inject_mapped_irq - Inject a physically mapped IRQ to the vgic
1690 * @kvm: The VM structure pointer
1691 * @cpuid: The CPU for PPIs
1692 * @map: Pointer to a irq_phys_map structure describing the mapping
1693 * @level: Edge-triggered: true: to trigger the interrupt
1694 * false: to ignore the call
1695 * Level-sensitive true: raise the input signal
1696 * false: lower the input signal
1698 * The GIC is not concerned with devices being active-LOW or active-HIGH for
1699 * level-sensitive interrupts. You can think of the level parameter as 1
1700 * being HIGH and 0 being LOW and all devices being active-HIGH.
1702 int kvm_vgic_inject_mapped_irq(struct kvm
*kvm
, int cpuid
,
1703 struct irq_phys_map
*map
, bool level
)
1707 ret
= vgic_lazy_init(kvm
);
1711 return vgic_update_irq_pending(kvm
, cpuid
, map
, map
->virt_irq
, level
);
1714 static irqreturn_t
vgic_maintenance_handler(int irq
, void *data
)
1717 * We cannot rely on the vgic maintenance interrupt to be
1718 * delivered synchronously. This means we can only use it to
1719 * exit the VM, and we perform the handling of EOIed
1720 * interrupts on the exit path (see vgic_process_maintenance).
1725 static struct list_head
*vgic_get_irq_phys_map_list(struct kvm_vcpu
*vcpu
,
1728 if (virt_irq
< VGIC_NR_PRIVATE_IRQS
)
1729 return &vcpu
->arch
.vgic_cpu
.irq_phys_map_list
;
1731 return &vcpu
->kvm
->arch
.vgic
.irq_phys_map_list
;
1735 * kvm_vgic_map_phys_irq - map a virtual IRQ to a physical IRQ
1736 * @vcpu: The VCPU pointer
1737 * @virt_irq: The virtual irq number
1738 * @irq: The Linux IRQ number
1740 * Establish a mapping between a guest visible irq (@virt_irq) and a
1741 * Linux irq (@irq). On injection, @virt_irq will be associated with
1742 * the physical interrupt represented by @irq. This mapping can be
1743 * established multiple times as long as the parameters are the same.
1745 * Returns a valid pointer on success, and an error pointer otherwise
1747 struct irq_phys_map
*kvm_vgic_map_phys_irq(struct kvm_vcpu
*vcpu
,
1748 int virt_irq
, int irq
)
1750 struct vgic_dist
*dist
= &vcpu
->kvm
->arch
.vgic
;
1751 struct list_head
*root
= vgic_get_irq_phys_map_list(vcpu
, virt_irq
);
1752 struct irq_phys_map
*map
;
1753 struct irq_phys_map_entry
*entry
;
1754 struct irq_desc
*desc
;
1755 struct irq_data
*data
;
1758 desc
= irq_to_desc(irq
);
1760 kvm_err("%s: no interrupt descriptor\n", __func__
);
1761 return ERR_PTR(-EINVAL
);
1764 data
= irq_desc_get_irq_data(desc
);
1765 while (data
->parent_data
)
1766 data
= data
->parent_data
;
1768 phys_irq
= data
->hwirq
;
1770 /* Create a new mapping */
1771 entry
= kzalloc(sizeof(*entry
), GFP_KERNEL
);
1773 return ERR_PTR(-ENOMEM
);
1775 spin_lock(&dist
->irq_phys_map_lock
);
1777 /* Try to match an existing mapping */
1778 map
= vgic_irq_map_search(vcpu
, virt_irq
);
1780 /* Make sure this mapping matches */
1781 if (map
->phys_irq
!= phys_irq
||
1783 map
= ERR_PTR(-EINVAL
);
1785 /* Found an existing, valid mapping */
1790 map
->virt_irq
= virt_irq
;
1791 map
->phys_irq
= phys_irq
;
1794 list_add_tail_rcu(&entry
->entry
, root
);
1797 spin_unlock(&dist
->irq_phys_map_lock
);
1798 /* If we've found a hit in the existing list, free the useless
1800 if (IS_ERR(map
) || map
!= &entry
->map
)
1805 static struct irq_phys_map
*vgic_irq_map_search(struct kvm_vcpu
*vcpu
,
1808 struct list_head
*root
= vgic_get_irq_phys_map_list(vcpu
, virt_irq
);
1809 struct irq_phys_map_entry
*entry
;
1810 struct irq_phys_map
*map
;
1814 list_for_each_entry_rcu(entry
, root
, entry
) {
1816 if (map
->virt_irq
== virt_irq
) {
1827 static void vgic_free_phys_irq_map_rcu(struct rcu_head
*rcu
)
1829 struct irq_phys_map_entry
*entry
;
1831 entry
= container_of(rcu
, struct irq_phys_map_entry
, rcu
);
1836 * kvm_vgic_get_phys_irq_active - Return the active state of a mapped IRQ
1838 * Return the logical active state of a mapped interrupt. This doesn't
1839 * necessarily reflects the current HW state.
1841 bool kvm_vgic_get_phys_irq_active(struct irq_phys_map
*map
)
1848 * kvm_vgic_set_phys_irq_active - Set the active state of a mapped IRQ
1850 * Set the logical active state of a mapped interrupt. This doesn't
1851 * immediately affects the HW state.
1853 void kvm_vgic_set_phys_irq_active(struct irq_phys_map
*map
, bool active
)
1856 map
->active
= active
;
1860 * kvm_vgic_unmap_phys_irq - Remove a virtual to physical IRQ mapping
1861 * @vcpu: The VCPU pointer
1862 * @map: The pointer to a mapping obtained through kvm_vgic_map_phys_irq
1864 * Remove an existing mapping between virtual and physical interrupts.
1866 int kvm_vgic_unmap_phys_irq(struct kvm_vcpu
*vcpu
, struct irq_phys_map
*map
)
1868 struct vgic_dist
*dist
= &vcpu
->kvm
->arch
.vgic
;
1869 struct irq_phys_map_entry
*entry
;
1870 struct list_head
*root
;
1875 root
= vgic_get_irq_phys_map_list(vcpu
, map
->virt_irq
);
1877 spin_lock(&dist
->irq_phys_map_lock
);
1879 list_for_each_entry(entry
, root
, entry
) {
1880 if (&entry
->map
== map
) {
1881 list_del_rcu(&entry
->entry
);
1882 call_rcu(&entry
->rcu
, vgic_free_phys_irq_map_rcu
);
1887 spin_unlock(&dist
->irq_phys_map_lock
);
1892 static void vgic_destroy_irq_phys_map(struct kvm
*kvm
, struct list_head
*root
)
1894 struct vgic_dist
*dist
= &kvm
->arch
.vgic
;
1895 struct irq_phys_map_entry
*entry
;
1897 spin_lock(&dist
->irq_phys_map_lock
);
1899 list_for_each_entry(entry
, root
, entry
) {
1900 list_del_rcu(&entry
->entry
);
1901 call_rcu(&entry
->rcu
, vgic_free_phys_irq_map_rcu
);
1904 spin_unlock(&dist
->irq_phys_map_lock
);
1907 void kvm_vgic_vcpu_destroy(struct kvm_vcpu
*vcpu
)
1909 struct vgic_cpu
*vgic_cpu
= &vcpu
->arch
.vgic_cpu
;
1911 kfree(vgic_cpu
->pending_shared
);
1912 kfree(vgic_cpu
->active_shared
);
1913 kfree(vgic_cpu
->pend_act_shared
);
1914 kfree(vgic_cpu
->vgic_irq_lr_map
);
1915 vgic_destroy_irq_phys_map(vcpu
->kvm
, &vgic_cpu
->irq_phys_map_list
);
1916 vgic_cpu
->pending_shared
= NULL
;
1917 vgic_cpu
->active_shared
= NULL
;
1918 vgic_cpu
->pend_act_shared
= NULL
;
1919 vgic_cpu
->vgic_irq_lr_map
= NULL
;
1922 static int vgic_vcpu_init_maps(struct kvm_vcpu
*vcpu
, int nr_irqs
)
1924 struct vgic_cpu
*vgic_cpu
= &vcpu
->arch
.vgic_cpu
;
1926 int sz
= (nr_irqs
- VGIC_NR_PRIVATE_IRQS
) / 8;
1927 vgic_cpu
->pending_shared
= kzalloc(sz
, GFP_KERNEL
);
1928 vgic_cpu
->active_shared
= kzalloc(sz
, GFP_KERNEL
);
1929 vgic_cpu
->pend_act_shared
= kzalloc(sz
, GFP_KERNEL
);
1930 vgic_cpu
->vgic_irq_lr_map
= kmalloc(nr_irqs
, GFP_KERNEL
);
1932 if (!vgic_cpu
->pending_shared
1933 || !vgic_cpu
->active_shared
1934 || !vgic_cpu
->pend_act_shared
1935 || !vgic_cpu
->vgic_irq_lr_map
) {
1936 kvm_vgic_vcpu_destroy(vcpu
);
1940 memset(vgic_cpu
->vgic_irq_lr_map
, LR_EMPTY
, nr_irqs
);
1943 * Store the number of LRs per vcpu, so we don't have to go
1944 * all the way to the distributor structure to find out. Only
1945 * assembly code should use this one.
1947 vgic_cpu
->nr_lr
= vgic
->nr_lr
;
1953 * kvm_vgic_vcpu_early_init - Earliest possible per-vcpu vgic init stage
1955 * No memory allocation should be performed here, only static init.
1957 void kvm_vgic_vcpu_early_init(struct kvm_vcpu
*vcpu
)
1959 struct vgic_cpu
*vgic_cpu
= &vcpu
->arch
.vgic_cpu
;
1960 INIT_LIST_HEAD(&vgic_cpu
->irq_phys_map_list
);
1964 * kvm_vgic_get_max_vcpus - Get the maximum number of VCPUs allowed by HW
1966 * The host's GIC naturally limits the maximum amount of VCPUs a guest
1969 int kvm_vgic_get_max_vcpus(void)
1971 return vgic
->max_gic_vcpus
;
1974 void kvm_vgic_destroy(struct kvm
*kvm
)
1976 struct vgic_dist
*dist
= &kvm
->arch
.vgic
;
1977 struct kvm_vcpu
*vcpu
;
1980 kvm_for_each_vcpu(i
, vcpu
, kvm
)
1981 kvm_vgic_vcpu_destroy(vcpu
);
1983 vgic_free_bitmap(&dist
->irq_enabled
);
1984 vgic_free_bitmap(&dist
->irq_level
);
1985 vgic_free_bitmap(&dist
->irq_pending
);
1986 vgic_free_bitmap(&dist
->irq_soft_pend
);
1987 vgic_free_bitmap(&dist
->irq_queued
);
1988 vgic_free_bitmap(&dist
->irq_cfg
);
1989 vgic_free_bytemap(&dist
->irq_priority
);
1990 if (dist
->irq_spi_target
) {
1991 for (i
= 0; i
< dist
->nr_cpus
; i
++)
1992 vgic_free_bitmap(&dist
->irq_spi_target
[i
]);
1994 kfree(dist
->irq_sgi_sources
);
1995 kfree(dist
->irq_spi_cpu
);
1996 kfree(dist
->irq_spi_mpidr
);
1997 kfree(dist
->irq_spi_target
);
1998 kfree(dist
->irq_pending_on_cpu
);
1999 kfree(dist
->irq_active_on_cpu
);
2000 vgic_destroy_irq_phys_map(kvm
, &dist
->irq_phys_map_list
);
2001 dist
->irq_sgi_sources
= NULL
;
2002 dist
->irq_spi_cpu
= NULL
;
2003 dist
->irq_spi_target
= NULL
;
2004 dist
->irq_pending_on_cpu
= NULL
;
2005 dist
->irq_active_on_cpu
= NULL
;
2010 * Allocate and initialize the various data structures. Must be called
2011 * with kvm->lock held!
2013 int vgic_init(struct kvm
*kvm
)
2015 struct vgic_dist
*dist
= &kvm
->arch
.vgic
;
2016 struct kvm_vcpu
*vcpu
;
2017 int nr_cpus
, nr_irqs
;
2018 int ret
, i
, vcpu_id
;
2020 if (vgic_initialized(kvm
))
2023 nr_cpus
= dist
->nr_cpus
= atomic_read(&kvm
->online_vcpus
);
2024 if (!nr_cpus
) /* No vcpus? Can't be good... */
2028 * If nobody configured the number of interrupts, use the
2032 dist
->nr_irqs
= VGIC_NR_IRQS_LEGACY
;
2034 nr_irqs
= dist
->nr_irqs
;
2036 ret
= vgic_init_bitmap(&dist
->irq_enabled
, nr_cpus
, nr_irqs
);
2037 ret
|= vgic_init_bitmap(&dist
->irq_level
, nr_cpus
, nr_irqs
);
2038 ret
|= vgic_init_bitmap(&dist
->irq_pending
, nr_cpus
, nr_irqs
);
2039 ret
|= vgic_init_bitmap(&dist
->irq_soft_pend
, nr_cpus
, nr_irqs
);
2040 ret
|= vgic_init_bitmap(&dist
->irq_queued
, nr_cpus
, nr_irqs
);
2041 ret
|= vgic_init_bitmap(&dist
->irq_active
, nr_cpus
, nr_irqs
);
2042 ret
|= vgic_init_bitmap(&dist
->irq_cfg
, nr_cpus
, nr_irqs
);
2043 ret
|= vgic_init_bytemap(&dist
->irq_priority
, nr_cpus
, nr_irqs
);
2048 dist
->irq_sgi_sources
= kzalloc(nr_cpus
* VGIC_NR_SGIS
, GFP_KERNEL
);
2049 dist
->irq_spi_cpu
= kzalloc(nr_irqs
- VGIC_NR_PRIVATE_IRQS
, GFP_KERNEL
);
2050 dist
->irq_spi_target
= kzalloc(sizeof(*dist
->irq_spi_target
) * nr_cpus
,
2052 dist
->irq_pending_on_cpu
= kzalloc(BITS_TO_LONGS(nr_cpus
) * sizeof(long),
2054 dist
->irq_active_on_cpu
= kzalloc(BITS_TO_LONGS(nr_cpus
) * sizeof(long),
2056 if (!dist
->irq_sgi_sources
||
2057 !dist
->irq_spi_cpu
||
2058 !dist
->irq_spi_target
||
2059 !dist
->irq_pending_on_cpu
||
2060 !dist
->irq_active_on_cpu
) {
2065 for (i
= 0; i
< nr_cpus
; i
++)
2066 ret
|= vgic_init_bitmap(&dist
->irq_spi_target
[i
],
2072 ret
= kvm
->arch
.vgic
.vm_ops
.init_model(kvm
);
2076 kvm_for_each_vcpu(vcpu_id
, vcpu
, kvm
) {
2077 ret
= vgic_vcpu_init_maps(vcpu
, nr_irqs
);
2079 kvm_err("VGIC: Failed to allocate vcpu memory\n");
2083 for (i
= 0; i
< dist
->nr_irqs
; i
++) {
2084 if (i
< VGIC_NR_PPIS
)
2085 vgic_bitmap_set_irq_val(&dist
->irq_enabled
,
2086 vcpu
->vcpu_id
, i
, 1);
2087 if (i
< VGIC_NR_PRIVATE_IRQS
)
2088 vgic_bitmap_set_irq_val(&dist
->irq_cfg
,
2098 kvm_vgic_destroy(kvm
);
2103 static int init_vgic_model(struct kvm
*kvm
, int type
)
2106 case KVM_DEV_TYPE_ARM_VGIC_V2
:
2107 vgic_v2_init_emulation(kvm
);
2109 #ifdef CONFIG_ARM_GIC_V3
2110 case KVM_DEV_TYPE_ARM_VGIC_V3
:
2111 vgic_v3_init_emulation(kvm
);
2118 if (atomic_read(&kvm
->online_vcpus
) > kvm
->arch
.max_vcpus
)
2125 * kvm_vgic_early_init - Earliest possible vgic initialization stage
2127 * No memory allocation should be performed here, only static init.
2129 void kvm_vgic_early_init(struct kvm
*kvm
)
2131 spin_lock_init(&kvm
->arch
.vgic
.lock
);
2132 spin_lock_init(&kvm
->arch
.vgic
.irq_phys_map_lock
);
2133 INIT_LIST_HEAD(&kvm
->arch
.vgic
.irq_phys_map_list
);
2136 int kvm_vgic_create(struct kvm
*kvm
, u32 type
)
2138 int i
, vcpu_lock_idx
= -1, ret
;
2139 struct kvm_vcpu
*vcpu
;
2141 mutex_lock(&kvm
->lock
);
2143 if (irqchip_in_kernel(kvm
)) {
2149 * This function is also called by the KVM_CREATE_IRQCHIP handler,
2150 * which had no chance yet to check the availability of the GICv2
2151 * emulation. So check this here again. KVM_CREATE_DEVICE does
2152 * the proper checks already.
2154 if (type
== KVM_DEV_TYPE_ARM_VGIC_V2
&& !vgic
->can_emulate_gicv2
) {
2160 * Any time a vcpu is run, vcpu_load is called which tries to grab the
2161 * vcpu->mutex. By grabbing the vcpu->mutex of all VCPUs we ensure
2162 * that no other VCPUs are run while we create the vgic.
2165 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
2166 if (!mutex_trylock(&vcpu
->mutex
))
2171 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
2172 if (vcpu
->arch
.has_run_once
)
2177 ret
= init_vgic_model(kvm
, type
);
2181 kvm
->arch
.vgic
.in_kernel
= true;
2182 kvm
->arch
.vgic
.vgic_model
= type
;
2183 kvm
->arch
.vgic
.vctrl_base
= vgic
->vctrl_base
;
2184 kvm
->arch
.vgic
.vgic_dist_base
= VGIC_ADDR_UNDEF
;
2185 kvm
->arch
.vgic
.vgic_cpu_base
= VGIC_ADDR_UNDEF
;
2186 kvm
->arch
.vgic
.vgic_redist_base
= VGIC_ADDR_UNDEF
;
2189 for (; vcpu_lock_idx
>= 0; vcpu_lock_idx
--) {
2190 vcpu
= kvm_get_vcpu(kvm
, vcpu_lock_idx
);
2191 mutex_unlock(&vcpu
->mutex
);
2195 mutex_unlock(&kvm
->lock
);
2199 static int vgic_ioaddr_overlap(struct kvm
*kvm
)
2201 phys_addr_t dist
= kvm
->arch
.vgic
.vgic_dist_base
;
2202 phys_addr_t cpu
= kvm
->arch
.vgic
.vgic_cpu_base
;
2204 if (IS_VGIC_ADDR_UNDEF(dist
) || IS_VGIC_ADDR_UNDEF(cpu
))
2206 if ((dist
<= cpu
&& dist
+ KVM_VGIC_V2_DIST_SIZE
> cpu
) ||
2207 (cpu
<= dist
&& cpu
+ KVM_VGIC_V2_CPU_SIZE
> dist
))
2212 static int vgic_ioaddr_assign(struct kvm
*kvm
, phys_addr_t
*ioaddr
,
2213 phys_addr_t addr
, phys_addr_t size
)
2217 if (addr
& ~KVM_PHYS_MASK
)
2220 if (addr
& (SZ_4K
- 1))
2223 if (!IS_VGIC_ADDR_UNDEF(*ioaddr
))
2225 if (addr
+ size
< addr
)
2229 ret
= vgic_ioaddr_overlap(kvm
);
2231 *ioaddr
= VGIC_ADDR_UNDEF
;
2237 * kvm_vgic_addr - set or get vgic VM base addresses
2238 * @kvm: pointer to the vm struct
2239 * @type: the VGIC addr type, one of KVM_VGIC_V[23]_ADDR_TYPE_XXX
2240 * @addr: pointer to address value
2241 * @write: if true set the address in the VM address space, if false read the
2244 * Set or get the vgic base addresses for the distributor and the virtual CPU
2245 * interface in the VM physical address space. These addresses are properties
2246 * of the emulated core/SoC and therefore user space initially knows this
2249 int kvm_vgic_addr(struct kvm
*kvm
, unsigned long type
, u64
*addr
, bool write
)
2252 struct vgic_dist
*vgic
= &kvm
->arch
.vgic
;
2254 phys_addr_t
*addr_ptr
, block_size
;
2255 phys_addr_t alignment
;
2257 mutex_lock(&kvm
->lock
);
2259 case KVM_VGIC_V2_ADDR_TYPE_DIST
:
2260 type_needed
= KVM_DEV_TYPE_ARM_VGIC_V2
;
2261 addr_ptr
= &vgic
->vgic_dist_base
;
2262 block_size
= KVM_VGIC_V2_DIST_SIZE
;
2265 case KVM_VGIC_V2_ADDR_TYPE_CPU
:
2266 type_needed
= KVM_DEV_TYPE_ARM_VGIC_V2
;
2267 addr_ptr
= &vgic
->vgic_cpu_base
;
2268 block_size
= KVM_VGIC_V2_CPU_SIZE
;
2271 #ifdef CONFIG_ARM_GIC_V3
2272 case KVM_VGIC_V3_ADDR_TYPE_DIST
:
2273 type_needed
= KVM_DEV_TYPE_ARM_VGIC_V3
;
2274 addr_ptr
= &vgic
->vgic_dist_base
;
2275 block_size
= KVM_VGIC_V3_DIST_SIZE
;
2278 case KVM_VGIC_V3_ADDR_TYPE_REDIST
:
2279 type_needed
= KVM_DEV_TYPE_ARM_VGIC_V3
;
2280 addr_ptr
= &vgic
->vgic_redist_base
;
2281 block_size
= KVM_VGIC_V3_REDIST_SIZE
;
2290 if (vgic
->vgic_model
!= type_needed
) {
2296 if (!IS_ALIGNED(*addr
, alignment
))
2299 r
= vgic_ioaddr_assign(kvm
, addr_ptr
, *addr
,
2306 mutex_unlock(&kvm
->lock
);
2310 int vgic_set_common_attr(struct kvm_device
*dev
, struct kvm_device_attr
*attr
)
2314 switch (attr
->group
) {
2315 case KVM_DEV_ARM_VGIC_GRP_ADDR
: {
2316 u64 __user
*uaddr
= (u64 __user
*)(long)attr
->addr
;
2318 unsigned long type
= (unsigned long)attr
->attr
;
2320 if (copy_from_user(&addr
, uaddr
, sizeof(addr
)))
2323 r
= kvm_vgic_addr(dev
->kvm
, type
, &addr
, true);
2324 return (r
== -ENODEV
) ? -ENXIO
: r
;
2326 case KVM_DEV_ARM_VGIC_GRP_NR_IRQS
: {
2327 u32 __user
*uaddr
= (u32 __user
*)(long)attr
->addr
;
2331 if (get_user(val
, uaddr
))
2336 * - at least 32 SPIs on top of the 16 SGIs and 16 PPIs
2337 * - at most 1024 interrupts
2338 * - a multiple of 32 interrupts
2340 if (val
< (VGIC_NR_PRIVATE_IRQS
+ 32) ||
2341 val
> VGIC_MAX_IRQS
||
2345 mutex_lock(&dev
->kvm
->lock
);
2347 if (vgic_ready(dev
->kvm
) || dev
->kvm
->arch
.vgic
.nr_irqs
)
2350 dev
->kvm
->arch
.vgic
.nr_irqs
= val
;
2352 mutex_unlock(&dev
->kvm
->lock
);
2356 case KVM_DEV_ARM_VGIC_GRP_CTRL
: {
2357 switch (attr
->attr
) {
2358 case KVM_DEV_ARM_VGIC_CTRL_INIT
:
2359 r
= vgic_init(dev
->kvm
);
2369 int vgic_get_common_attr(struct kvm_device
*dev
, struct kvm_device_attr
*attr
)
2373 switch (attr
->group
) {
2374 case KVM_DEV_ARM_VGIC_GRP_ADDR
: {
2375 u64 __user
*uaddr
= (u64 __user
*)(long)attr
->addr
;
2377 unsigned long type
= (unsigned long)attr
->attr
;
2379 r
= kvm_vgic_addr(dev
->kvm
, type
, &addr
, false);
2381 return (r
== -ENODEV
) ? -ENXIO
: r
;
2383 if (copy_to_user(uaddr
, &addr
, sizeof(addr
)))
2387 case KVM_DEV_ARM_VGIC_GRP_NR_IRQS
: {
2388 u32 __user
*uaddr
= (u32 __user
*)(long)attr
->addr
;
2390 r
= put_user(dev
->kvm
->arch
.vgic
.nr_irqs
, uaddr
);
2399 int vgic_has_attr_regs(const struct vgic_io_range
*ranges
, phys_addr_t offset
)
2401 if (vgic_find_range(ranges
, 4, offset
))
2407 static void vgic_init_maintenance_interrupt(void *info
)
2409 enable_percpu_irq(vgic
->maint_irq
, 0);
2412 static int vgic_cpu_notify(struct notifier_block
*self
,
2413 unsigned long action
, void *cpu
)
2417 case CPU_STARTING_FROZEN
:
2418 vgic_init_maintenance_interrupt(NULL
);
2421 case CPU_DYING_FROZEN
:
2422 disable_percpu_irq(vgic
->maint_irq
);
2429 static struct notifier_block vgic_cpu_nb
= {
2430 .notifier_call
= vgic_cpu_notify
,
2433 static const struct of_device_id vgic_ids
[] = {
2434 { .compatible
= "arm,cortex-a15-gic", .data
= vgic_v2_probe
, },
2435 { .compatible
= "arm,cortex-a7-gic", .data
= vgic_v2_probe
, },
2436 { .compatible
= "arm,gic-400", .data
= vgic_v2_probe
, },
2437 { .compatible
= "arm,gic-v3", .data
= vgic_v3_probe
, },
2441 int kvm_vgic_hyp_init(void)
2443 const struct of_device_id
*matched_id
;
2444 const int (*vgic_probe
)(struct device_node
*,const struct vgic_ops
**,
2445 const struct vgic_params
**);
2446 struct device_node
*vgic_node
;
2449 vgic_node
= of_find_matching_node_and_match(NULL
,
2450 vgic_ids
, &matched_id
);
2452 kvm_err("error: no compatible GIC node found\n");
2456 vgic_probe
= matched_id
->data
;
2457 ret
= vgic_probe(vgic_node
, &vgic_ops
, &vgic
);
2461 ret
= request_percpu_irq(vgic
->maint_irq
, vgic_maintenance_handler
,
2462 "vgic", kvm_get_running_vcpus());
2464 kvm_err("Cannot register interrupt %d\n", vgic
->maint_irq
);
2468 ret
= __register_cpu_notifier(&vgic_cpu_nb
);
2470 kvm_err("Cannot register vgic CPU notifier\n");
2474 on_each_cpu(vgic_init_maintenance_interrupt
, NULL
, 1);
2479 free_percpu_irq(vgic
->maint_irq
, kvm_get_running_vcpus());
2483 int kvm_irq_map_gsi(struct kvm
*kvm
,
2484 struct kvm_kernel_irq_routing_entry
*entries
,
2490 int kvm_irq_map_chip_pin(struct kvm
*kvm
, unsigned irqchip
, unsigned pin
)
2495 int kvm_set_irq(struct kvm
*kvm
, int irq_source_id
,
2496 u32 irq
, int level
, bool line_status
)
2498 unsigned int spi
= irq
+ VGIC_NR_PRIVATE_IRQS
;
2500 trace_kvm_set_irq(irq
, level
, irq_source_id
);
2502 BUG_ON(!vgic_initialized(kvm
));
2504 return kvm_vgic_inject_irq(kvm
, 0, spi
, level
);
2507 /* MSI not implemented yet */
2508 int kvm_set_msi(struct kvm_kernel_irq_routing_entry
*e
,
2509 struct kvm
*kvm
, int irq_source_id
,
2510 int level
, bool line_status
)