Marc Zyngier | 1a89dd9 | 2013-01-21 19:36:12 -0500 | [diff] [blame] | 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/kvm.h> |
| 20 | #include <linux/kvm_host.h> |
| 21 | #include <linux/interrupt.h> |
| 22 | #include <linux/io.h> |
| 23 | #include <asm/kvm_emulate.h> |
| 24 | |
Marc Zyngier | b47ef92 | 2013-01-21 19:36:14 -0500 | [diff] [blame^] | 25 | /* |
| 26 | * How the whole thing works (courtesy of Christoffer Dall): |
| 27 | * |
| 28 | * - At any time, the dist->irq_pending_on_cpu is the oracle that knows if |
| 29 | * something is pending |
| 30 | * - VGIC pending interrupts are stored on the vgic.irq_state vgic |
| 31 | * bitmap (this bitmap is updated by both user land ioctls and guest |
| 32 | * mmio ops, and other in-kernel peripherals such as the |
| 33 | * arch. timers) and indicate the 'wire' state. |
| 34 | * - Every time the bitmap changes, the irq_pending_on_cpu oracle is |
| 35 | * recalculated |
| 36 | * - To calculate the oracle, we need info for each cpu from |
| 37 | * compute_pending_for_cpu, which considers: |
| 38 | * - PPI: dist->irq_state & dist->irq_enable |
| 39 | * - SPI: dist->irq_state & dist->irq_enable & dist->irq_spi_target |
| 40 | * - irq_spi_target is a 'formatted' version of the GICD_ICFGR |
| 41 | * registers, stored on each vcpu. We only keep one bit of |
| 42 | * information per interrupt, making sure that only one vcpu can |
| 43 | * accept the interrupt. |
| 44 | * - The same is true when injecting an interrupt, except that we only |
| 45 | * consider a single interrupt at a time. The irq_spi_cpu array |
| 46 | * contains the target CPU for each SPI. |
| 47 | * |
| 48 | * The handling of level interrupts adds some extra complexity. We |
| 49 | * need to track when the interrupt has been EOIed, so we can sample |
| 50 | * the 'line' again. This is achieved as such: |
| 51 | * |
| 52 | * - When a level interrupt is moved onto a vcpu, the corresponding |
| 53 | * bit in irq_active is set. As long as this bit is set, the line |
| 54 | * will be ignored for further interrupts. The interrupt is injected |
| 55 | * into the vcpu with the GICH_LR_EOI bit set (generate a |
| 56 | * maintenance interrupt on EOI). |
| 57 | * - When the interrupt is EOIed, the maintenance interrupt fires, |
| 58 | * and clears the corresponding bit in irq_active. This allow the |
| 59 | * interrupt line to be sampled again. |
| 60 | */ |
| 61 | |
Christoffer Dall | 330690c | 2013-01-21 19:36:13 -0500 | [diff] [blame] | 62 | #define VGIC_ADDR_UNDEF (-1) |
| 63 | #define IS_VGIC_ADDR_UNDEF(_x) ((_x) == VGIC_ADDR_UNDEF) |
| 64 | |
Marc Zyngier | 1a89dd9 | 2013-01-21 19:36:12 -0500 | [diff] [blame] | 65 | #define ACCESS_READ_VALUE (1 << 0) |
| 66 | #define ACCESS_READ_RAZ (0 << 0) |
| 67 | #define ACCESS_READ_MASK(x) ((x) & (1 << 0)) |
| 68 | #define ACCESS_WRITE_IGNORED (0 << 1) |
| 69 | #define ACCESS_WRITE_SETBIT (1 << 1) |
| 70 | #define ACCESS_WRITE_CLEARBIT (2 << 1) |
| 71 | #define ACCESS_WRITE_VALUE (3 << 1) |
| 72 | #define ACCESS_WRITE_MASK(x) ((x) & (3 << 1)) |
| 73 | |
Marc Zyngier | b47ef92 | 2013-01-21 19:36:14 -0500 | [diff] [blame^] | 74 | static void vgic_update_state(struct kvm *kvm); |
| 75 | static void vgic_dispatch_sgi(struct kvm_vcpu *vcpu, u32 reg); |
| 76 | |
| 77 | static u32 *vgic_bitmap_get_reg(struct vgic_bitmap *x, |
| 78 | int cpuid, u32 offset) |
| 79 | { |
| 80 | offset >>= 2; |
| 81 | if (!offset) |
| 82 | return x->percpu[cpuid].reg; |
| 83 | else |
| 84 | return x->shared.reg + offset - 1; |
| 85 | } |
| 86 | |
| 87 | static int vgic_bitmap_get_irq_val(struct vgic_bitmap *x, |
| 88 | int cpuid, int irq) |
| 89 | { |
| 90 | if (irq < VGIC_NR_PRIVATE_IRQS) |
| 91 | return test_bit(irq, x->percpu[cpuid].reg_ul); |
| 92 | |
| 93 | return test_bit(irq - VGIC_NR_PRIVATE_IRQS, x->shared.reg_ul); |
| 94 | } |
| 95 | |
| 96 | static void vgic_bitmap_set_irq_val(struct vgic_bitmap *x, int cpuid, |
| 97 | int irq, int val) |
| 98 | { |
| 99 | unsigned long *reg; |
| 100 | |
| 101 | if (irq < VGIC_NR_PRIVATE_IRQS) { |
| 102 | reg = x->percpu[cpuid].reg_ul; |
| 103 | } else { |
| 104 | reg = x->shared.reg_ul; |
| 105 | irq -= VGIC_NR_PRIVATE_IRQS; |
| 106 | } |
| 107 | |
| 108 | if (val) |
| 109 | set_bit(irq, reg); |
| 110 | else |
| 111 | clear_bit(irq, reg); |
| 112 | } |
| 113 | |
| 114 | static unsigned long *vgic_bitmap_get_cpu_map(struct vgic_bitmap *x, int cpuid) |
| 115 | { |
| 116 | if (unlikely(cpuid >= VGIC_MAX_CPUS)) |
| 117 | return NULL; |
| 118 | return x->percpu[cpuid].reg_ul; |
| 119 | } |
| 120 | |
| 121 | static unsigned long *vgic_bitmap_get_shared_map(struct vgic_bitmap *x) |
| 122 | { |
| 123 | return x->shared.reg_ul; |
| 124 | } |
| 125 | |
| 126 | static u32 *vgic_bytemap_get_reg(struct vgic_bytemap *x, int cpuid, u32 offset) |
| 127 | { |
| 128 | offset >>= 2; |
| 129 | BUG_ON(offset > (VGIC_NR_IRQS / 4)); |
| 130 | if (offset < 4) |
| 131 | return x->percpu[cpuid] + offset; |
| 132 | else |
| 133 | return x->shared + offset - 8; |
| 134 | } |
| 135 | |
| 136 | #define VGIC_CFG_LEVEL 0 |
| 137 | #define VGIC_CFG_EDGE 1 |
| 138 | |
| 139 | static bool vgic_irq_is_edge(struct kvm_vcpu *vcpu, int irq) |
| 140 | { |
| 141 | struct vgic_dist *dist = &vcpu->kvm->arch.vgic; |
| 142 | int irq_val; |
| 143 | |
| 144 | irq_val = vgic_bitmap_get_irq_val(&dist->irq_cfg, vcpu->vcpu_id, irq); |
| 145 | return irq_val == VGIC_CFG_EDGE; |
| 146 | } |
| 147 | |
| 148 | static int vgic_irq_is_enabled(struct kvm_vcpu *vcpu, int irq) |
| 149 | { |
| 150 | struct vgic_dist *dist = &vcpu->kvm->arch.vgic; |
| 151 | |
| 152 | return vgic_bitmap_get_irq_val(&dist->irq_enabled, vcpu->vcpu_id, irq); |
| 153 | } |
| 154 | |
| 155 | static void vgic_dist_irq_set(struct kvm_vcpu *vcpu, int irq) |
| 156 | { |
| 157 | struct vgic_dist *dist = &vcpu->kvm->arch.vgic; |
| 158 | |
| 159 | vgic_bitmap_set_irq_val(&dist->irq_state, vcpu->vcpu_id, irq, 1); |
| 160 | } |
| 161 | |
| 162 | static void vgic_dist_irq_clear(struct kvm_vcpu *vcpu, int irq) |
| 163 | { |
| 164 | struct vgic_dist *dist = &vcpu->kvm->arch.vgic; |
| 165 | |
| 166 | vgic_bitmap_set_irq_val(&dist->irq_state, vcpu->vcpu_id, irq, 0); |
| 167 | } |
| 168 | |
| 169 | static void vgic_cpu_irq_set(struct kvm_vcpu *vcpu, int irq) |
| 170 | { |
| 171 | if (irq < VGIC_NR_PRIVATE_IRQS) |
| 172 | set_bit(irq, vcpu->arch.vgic_cpu.pending_percpu); |
| 173 | else |
| 174 | set_bit(irq - VGIC_NR_PRIVATE_IRQS, |
| 175 | vcpu->arch.vgic_cpu.pending_shared); |
| 176 | } |
| 177 | |
| 178 | static void vgic_cpu_irq_clear(struct kvm_vcpu *vcpu, int irq) |
| 179 | { |
| 180 | if (irq < VGIC_NR_PRIVATE_IRQS) |
| 181 | clear_bit(irq, vcpu->arch.vgic_cpu.pending_percpu); |
| 182 | else |
| 183 | clear_bit(irq - VGIC_NR_PRIVATE_IRQS, |
| 184 | vcpu->arch.vgic_cpu.pending_shared); |
| 185 | } |
| 186 | |
Marc Zyngier | 1a89dd9 | 2013-01-21 19:36:12 -0500 | [diff] [blame] | 187 | static u32 mmio_data_read(struct kvm_exit_mmio *mmio, u32 mask) |
| 188 | { |
| 189 | return *((u32 *)mmio->data) & mask; |
| 190 | } |
| 191 | |
| 192 | static void mmio_data_write(struct kvm_exit_mmio *mmio, u32 mask, u32 value) |
| 193 | { |
| 194 | *((u32 *)mmio->data) = value & mask; |
| 195 | } |
| 196 | |
| 197 | /** |
| 198 | * vgic_reg_access - access vgic register |
| 199 | * @mmio: pointer to the data describing the mmio access |
| 200 | * @reg: pointer to the virtual backing of vgic distributor data |
| 201 | * @offset: least significant 2 bits used for word offset |
| 202 | * @mode: ACCESS_ mode (see defines above) |
| 203 | * |
| 204 | * Helper to make vgic register access easier using one of the access |
| 205 | * modes defined for vgic register access |
| 206 | * (read,raz,write-ignored,setbit,clearbit,write) |
| 207 | */ |
| 208 | static void vgic_reg_access(struct kvm_exit_mmio *mmio, u32 *reg, |
| 209 | phys_addr_t offset, int mode) |
| 210 | { |
| 211 | int word_offset = (offset & 3) * 8; |
| 212 | u32 mask = (1UL << (mmio->len * 8)) - 1; |
| 213 | u32 regval; |
| 214 | |
| 215 | /* |
| 216 | * Any alignment fault should have been delivered to the guest |
| 217 | * directly (ARM ARM B3.12.7 "Prioritization of aborts"). |
| 218 | */ |
| 219 | |
| 220 | if (reg) { |
| 221 | regval = *reg; |
| 222 | } else { |
| 223 | BUG_ON(mode != (ACCESS_READ_RAZ | ACCESS_WRITE_IGNORED)); |
| 224 | regval = 0; |
| 225 | } |
| 226 | |
| 227 | if (mmio->is_write) { |
| 228 | u32 data = mmio_data_read(mmio, mask) << word_offset; |
| 229 | switch (ACCESS_WRITE_MASK(mode)) { |
| 230 | case ACCESS_WRITE_IGNORED: |
| 231 | return; |
| 232 | |
| 233 | case ACCESS_WRITE_SETBIT: |
| 234 | regval |= data; |
| 235 | break; |
| 236 | |
| 237 | case ACCESS_WRITE_CLEARBIT: |
| 238 | regval &= ~data; |
| 239 | break; |
| 240 | |
| 241 | case ACCESS_WRITE_VALUE: |
| 242 | regval = (regval & ~(mask << word_offset)) | data; |
| 243 | break; |
| 244 | } |
| 245 | *reg = regval; |
| 246 | } else { |
| 247 | switch (ACCESS_READ_MASK(mode)) { |
| 248 | case ACCESS_READ_RAZ: |
| 249 | regval = 0; |
| 250 | /* fall through */ |
| 251 | |
| 252 | case ACCESS_READ_VALUE: |
| 253 | mmio_data_write(mmio, mask, regval >> word_offset); |
| 254 | } |
| 255 | } |
| 256 | } |
| 257 | |
Marc Zyngier | b47ef92 | 2013-01-21 19:36:14 -0500 | [diff] [blame^] | 258 | static bool handle_mmio_misc(struct kvm_vcpu *vcpu, |
| 259 | struct kvm_exit_mmio *mmio, phys_addr_t offset) |
| 260 | { |
| 261 | u32 reg; |
| 262 | u32 word_offset = offset & 3; |
| 263 | |
| 264 | switch (offset & ~3) { |
| 265 | case 0: /* CTLR */ |
| 266 | reg = vcpu->kvm->arch.vgic.enabled; |
| 267 | vgic_reg_access(mmio, ®, word_offset, |
| 268 | ACCESS_READ_VALUE | ACCESS_WRITE_VALUE); |
| 269 | if (mmio->is_write) { |
| 270 | vcpu->kvm->arch.vgic.enabled = reg & 1; |
| 271 | vgic_update_state(vcpu->kvm); |
| 272 | return true; |
| 273 | } |
| 274 | break; |
| 275 | |
| 276 | case 4: /* TYPER */ |
| 277 | reg = (atomic_read(&vcpu->kvm->online_vcpus) - 1) << 5; |
| 278 | reg |= (VGIC_NR_IRQS >> 5) - 1; |
| 279 | vgic_reg_access(mmio, ®, word_offset, |
| 280 | ACCESS_READ_VALUE | ACCESS_WRITE_IGNORED); |
| 281 | break; |
| 282 | |
| 283 | case 8: /* IIDR */ |
| 284 | reg = 0x4B00043B; |
| 285 | vgic_reg_access(mmio, ®, word_offset, |
| 286 | ACCESS_READ_VALUE | ACCESS_WRITE_IGNORED); |
| 287 | break; |
| 288 | } |
| 289 | |
| 290 | return false; |
| 291 | } |
| 292 | |
| 293 | static bool handle_mmio_raz_wi(struct kvm_vcpu *vcpu, |
| 294 | struct kvm_exit_mmio *mmio, phys_addr_t offset) |
| 295 | { |
| 296 | vgic_reg_access(mmio, NULL, offset, |
| 297 | ACCESS_READ_RAZ | ACCESS_WRITE_IGNORED); |
| 298 | return false; |
| 299 | } |
| 300 | |
| 301 | static bool handle_mmio_set_enable_reg(struct kvm_vcpu *vcpu, |
| 302 | struct kvm_exit_mmio *mmio, |
| 303 | phys_addr_t offset) |
| 304 | { |
| 305 | u32 *reg = vgic_bitmap_get_reg(&vcpu->kvm->arch.vgic.irq_enabled, |
| 306 | vcpu->vcpu_id, offset); |
| 307 | vgic_reg_access(mmio, reg, offset, |
| 308 | ACCESS_READ_VALUE | ACCESS_WRITE_SETBIT); |
| 309 | if (mmio->is_write) { |
| 310 | vgic_update_state(vcpu->kvm); |
| 311 | return true; |
| 312 | } |
| 313 | |
| 314 | return false; |
| 315 | } |
| 316 | |
| 317 | static bool handle_mmio_clear_enable_reg(struct kvm_vcpu *vcpu, |
| 318 | struct kvm_exit_mmio *mmio, |
| 319 | phys_addr_t offset) |
| 320 | { |
| 321 | u32 *reg = vgic_bitmap_get_reg(&vcpu->kvm->arch.vgic.irq_enabled, |
| 322 | vcpu->vcpu_id, offset); |
| 323 | vgic_reg_access(mmio, reg, offset, |
| 324 | ACCESS_READ_VALUE | ACCESS_WRITE_CLEARBIT); |
| 325 | if (mmio->is_write) { |
| 326 | if (offset < 4) /* Force SGI enabled */ |
| 327 | *reg |= 0xffff; |
| 328 | vgic_update_state(vcpu->kvm); |
| 329 | return true; |
| 330 | } |
| 331 | |
| 332 | return false; |
| 333 | } |
| 334 | |
| 335 | static bool handle_mmio_set_pending_reg(struct kvm_vcpu *vcpu, |
| 336 | struct kvm_exit_mmio *mmio, |
| 337 | phys_addr_t offset) |
| 338 | { |
| 339 | u32 *reg = vgic_bitmap_get_reg(&vcpu->kvm->arch.vgic.irq_state, |
| 340 | vcpu->vcpu_id, offset); |
| 341 | vgic_reg_access(mmio, reg, offset, |
| 342 | ACCESS_READ_VALUE | ACCESS_WRITE_SETBIT); |
| 343 | if (mmio->is_write) { |
| 344 | vgic_update_state(vcpu->kvm); |
| 345 | return true; |
| 346 | } |
| 347 | |
| 348 | return false; |
| 349 | } |
| 350 | |
| 351 | static bool handle_mmio_clear_pending_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_state, |
| 356 | vcpu->vcpu_id, offset); |
| 357 | vgic_reg_access(mmio, reg, offset, |
| 358 | ACCESS_READ_VALUE | ACCESS_WRITE_CLEARBIT); |
| 359 | if (mmio->is_write) { |
| 360 | vgic_update_state(vcpu->kvm); |
| 361 | return true; |
| 362 | } |
| 363 | |
| 364 | return false; |
| 365 | } |
| 366 | |
| 367 | static bool handle_mmio_priority_reg(struct kvm_vcpu *vcpu, |
| 368 | struct kvm_exit_mmio *mmio, |
| 369 | phys_addr_t offset) |
| 370 | { |
| 371 | u32 *reg = vgic_bytemap_get_reg(&vcpu->kvm->arch.vgic.irq_priority, |
| 372 | vcpu->vcpu_id, offset); |
| 373 | vgic_reg_access(mmio, reg, offset, |
| 374 | ACCESS_READ_VALUE | ACCESS_WRITE_VALUE); |
| 375 | return false; |
| 376 | } |
| 377 | |
| 378 | #define GICD_ITARGETSR_SIZE 32 |
| 379 | #define GICD_CPUTARGETS_BITS 8 |
| 380 | #define GICD_IRQS_PER_ITARGETSR (GICD_ITARGETSR_SIZE / GICD_CPUTARGETS_BITS) |
| 381 | static u32 vgic_get_target_reg(struct kvm *kvm, int irq) |
| 382 | { |
| 383 | struct vgic_dist *dist = &kvm->arch.vgic; |
| 384 | struct kvm_vcpu *vcpu; |
| 385 | int i, c; |
| 386 | unsigned long *bmap; |
| 387 | u32 val = 0; |
| 388 | |
| 389 | irq -= VGIC_NR_PRIVATE_IRQS; |
| 390 | |
| 391 | kvm_for_each_vcpu(c, vcpu, kvm) { |
| 392 | bmap = vgic_bitmap_get_shared_map(&dist->irq_spi_target[c]); |
| 393 | for (i = 0; i < GICD_IRQS_PER_ITARGETSR; i++) |
| 394 | if (test_bit(irq + i, bmap)) |
| 395 | val |= 1 << (c + i * 8); |
| 396 | } |
| 397 | |
| 398 | return val; |
| 399 | } |
| 400 | |
| 401 | static void vgic_set_target_reg(struct kvm *kvm, u32 val, int irq) |
| 402 | { |
| 403 | struct vgic_dist *dist = &kvm->arch.vgic; |
| 404 | struct kvm_vcpu *vcpu; |
| 405 | int i, c; |
| 406 | unsigned long *bmap; |
| 407 | u32 target; |
| 408 | |
| 409 | irq -= VGIC_NR_PRIVATE_IRQS; |
| 410 | |
| 411 | /* |
| 412 | * Pick the LSB in each byte. This ensures we target exactly |
| 413 | * one vcpu per IRQ. If the byte is null, assume we target |
| 414 | * CPU0. |
| 415 | */ |
| 416 | for (i = 0; i < GICD_IRQS_PER_ITARGETSR; i++) { |
| 417 | int shift = i * GICD_CPUTARGETS_BITS; |
| 418 | target = ffs((val >> shift) & 0xffU); |
| 419 | target = target ? (target - 1) : 0; |
| 420 | dist->irq_spi_cpu[irq + i] = target; |
| 421 | kvm_for_each_vcpu(c, vcpu, kvm) { |
| 422 | bmap = vgic_bitmap_get_shared_map(&dist->irq_spi_target[c]); |
| 423 | if (c == target) |
| 424 | set_bit(irq + i, bmap); |
| 425 | else |
| 426 | clear_bit(irq + i, bmap); |
| 427 | } |
| 428 | } |
| 429 | } |
| 430 | |
| 431 | static bool handle_mmio_target_reg(struct kvm_vcpu *vcpu, |
| 432 | struct kvm_exit_mmio *mmio, |
| 433 | phys_addr_t offset) |
| 434 | { |
| 435 | u32 reg; |
| 436 | |
| 437 | /* We treat the banked interrupts targets as read-only */ |
| 438 | if (offset < 32) { |
| 439 | u32 roreg = 1 << vcpu->vcpu_id; |
| 440 | roreg |= roreg << 8; |
| 441 | roreg |= roreg << 16; |
| 442 | |
| 443 | vgic_reg_access(mmio, &roreg, offset, |
| 444 | ACCESS_READ_VALUE | ACCESS_WRITE_IGNORED); |
| 445 | return false; |
| 446 | } |
| 447 | |
| 448 | reg = vgic_get_target_reg(vcpu->kvm, offset & ~3U); |
| 449 | vgic_reg_access(mmio, ®, offset, |
| 450 | ACCESS_READ_VALUE | ACCESS_WRITE_VALUE); |
| 451 | if (mmio->is_write) { |
| 452 | vgic_set_target_reg(vcpu->kvm, reg, offset & ~3U); |
| 453 | vgic_update_state(vcpu->kvm); |
| 454 | return true; |
| 455 | } |
| 456 | |
| 457 | return false; |
| 458 | } |
| 459 | |
| 460 | static u32 vgic_cfg_expand(u16 val) |
| 461 | { |
| 462 | u32 res = 0; |
| 463 | int i; |
| 464 | |
| 465 | /* |
| 466 | * Turn a 16bit value like abcd...mnop into a 32bit word |
| 467 | * a0b0c0d0...m0n0o0p0, which is what the HW cfg register is. |
| 468 | */ |
| 469 | for (i = 0; i < 16; i++) |
| 470 | res |= ((val >> i) & VGIC_CFG_EDGE) << (2 * i + 1); |
| 471 | |
| 472 | return res; |
| 473 | } |
| 474 | |
| 475 | static u16 vgic_cfg_compress(u32 val) |
| 476 | { |
| 477 | u16 res = 0; |
| 478 | int i; |
| 479 | |
| 480 | /* |
| 481 | * Turn a 32bit word a0b0c0d0...m0n0o0p0 into 16bit value like |
| 482 | * abcd...mnop which is what we really care about. |
| 483 | */ |
| 484 | for (i = 0; i < 16; i++) |
| 485 | res |= ((val >> (i * 2 + 1)) & VGIC_CFG_EDGE) << i; |
| 486 | |
| 487 | return res; |
| 488 | } |
| 489 | |
| 490 | /* |
| 491 | * The distributor uses 2 bits per IRQ for the CFG register, but the |
| 492 | * LSB is always 0. As such, we only keep the upper bit, and use the |
| 493 | * two above functions to compress/expand the bits |
| 494 | */ |
| 495 | static bool handle_mmio_cfg_reg(struct kvm_vcpu *vcpu, |
| 496 | struct kvm_exit_mmio *mmio, phys_addr_t offset) |
| 497 | { |
| 498 | u32 val; |
| 499 | u32 *reg = vgic_bitmap_get_reg(&vcpu->kvm->arch.vgic.irq_cfg, |
| 500 | vcpu->vcpu_id, offset >> 1); |
| 501 | if (offset & 2) |
| 502 | val = *reg >> 16; |
| 503 | else |
| 504 | val = *reg & 0xffff; |
| 505 | |
| 506 | val = vgic_cfg_expand(val); |
| 507 | vgic_reg_access(mmio, &val, offset, |
| 508 | ACCESS_READ_VALUE | ACCESS_WRITE_VALUE); |
| 509 | if (mmio->is_write) { |
| 510 | if (offset < 4) { |
| 511 | *reg = ~0U; /* Force PPIs/SGIs to 1 */ |
| 512 | return false; |
| 513 | } |
| 514 | |
| 515 | val = vgic_cfg_compress(val); |
| 516 | if (offset & 2) { |
| 517 | *reg &= 0xffff; |
| 518 | *reg |= val << 16; |
| 519 | } else { |
| 520 | *reg &= 0xffff << 16; |
| 521 | *reg |= val; |
| 522 | } |
| 523 | } |
| 524 | |
| 525 | return false; |
| 526 | } |
| 527 | |
| 528 | static bool handle_mmio_sgi_reg(struct kvm_vcpu *vcpu, |
| 529 | struct kvm_exit_mmio *mmio, phys_addr_t offset) |
| 530 | { |
| 531 | u32 reg; |
| 532 | vgic_reg_access(mmio, ®, offset, |
| 533 | ACCESS_READ_RAZ | ACCESS_WRITE_VALUE); |
| 534 | if (mmio->is_write) { |
| 535 | vgic_dispatch_sgi(vcpu, reg); |
| 536 | vgic_update_state(vcpu->kvm); |
| 537 | return true; |
| 538 | } |
| 539 | |
| 540 | return false; |
| 541 | } |
| 542 | |
Marc Zyngier | 1a89dd9 | 2013-01-21 19:36:12 -0500 | [diff] [blame] | 543 | /* |
| 544 | * I would have liked to use the kvm_bus_io_*() API instead, but it |
| 545 | * cannot cope with banked registers (only the VM pointer is passed |
| 546 | * around, and we need the vcpu). One of these days, someone please |
| 547 | * fix it! |
| 548 | */ |
| 549 | struct mmio_range { |
| 550 | phys_addr_t base; |
| 551 | unsigned long len; |
| 552 | bool (*handle_mmio)(struct kvm_vcpu *vcpu, struct kvm_exit_mmio *mmio, |
| 553 | phys_addr_t offset); |
| 554 | }; |
| 555 | |
| 556 | static const struct mmio_range vgic_ranges[] = { |
Marc Zyngier | b47ef92 | 2013-01-21 19:36:14 -0500 | [diff] [blame^] | 557 | { |
| 558 | .base = GIC_DIST_CTRL, |
| 559 | .len = 12, |
| 560 | .handle_mmio = handle_mmio_misc, |
| 561 | }, |
| 562 | { |
| 563 | .base = GIC_DIST_IGROUP, |
| 564 | .len = VGIC_NR_IRQS / 8, |
| 565 | .handle_mmio = handle_mmio_raz_wi, |
| 566 | }, |
| 567 | { |
| 568 | .base = GIC_DIST_ENABLE_SET, |
| 569 | .len = VGIC_NR_IRQS / 8, |
| 570 | .handle_mmio = handle_mmio_set_enable_reg, |
| 571 | }, |
| 572 | { |
| 573 | .base = GIC_DIST_ENABLE_CLEAR, |
| 574 | .len = VGIC_NR_IRQS / 8, |
| 575 | .handle_mmio = handle_mmio_clear_enable_reg, |
| 576 | }, |
| 577 | { |
| 578 | .base = GIC_DIST_PENDING_SET, |
| 579 | .len = VGIC_NR_IRQS / 8, |
| 580 | .handle_mmio = handle_mmio_set_pending_reg, |
| 581 | }, |
| 582 | { |
| 583 | .base = GIC_DIST_PENDING_CLEAR, |
| 584 | .len = VGIC_NR_IRQS / 8, |
| 585 | .handle_mmio = handle_mmio_clear_pending_reg, |
| 586 | }, |
| 587 | { |
| 588 | .base = GIC_DIST_ACTIVE_SET, |
| 589 | .len = VGIC_NR_IRQS / 8, |
| 590 | .handle_mmio = handle_mmio_raz_wi, |
| 591 | }, |
| 592 | { |
| 593 | .base = GIC_DIST_ACTIVE_CLEAR, |
| 594 | .len = VGIC_NR_IRQS / 8, |
| 595 | .handle_mmio = handle_mmio_raz_wi, |
| 596 | }, |
| 597 | { |
| 598 | .base = GIC_DIST_PRI, |
| 599 | .len = VGIC_NR_IRQS, |
| 600 | .handle_mmio = handle_mmio_priority_reg, |
| 601 | }, |
| 602 | { |
| 603 | .base = GIC_DIST_TARGET, |
| 604 | .len = VGIC_NR_IRQS, |
| 605 | .handle_mmio = handle_mmio_target_reg, |
| 606 | }, |
| 607 | { |
| 608 | .base = GIC_DIST_CONFIG, |
| 609 | .len = VGIC_NR_IRQS / 4, |
| 610 | .handle_mmio = handle_mmio_cfg_reg, |
| 611 | }, |
| 612 | { |
| 613 | .base = GIC_DIST_SOFTINT, |
| 614 | .len = 4, |
| 615 | .handle_mmio = handle_mmio_sgi_reg, |
| 616 | }, |
Marc Zyngier | 1a89dd9 | 2013-01-21 19:36:12 -0500 | [diff] [blame] | 617 | {} |
| 618 | }; |
| 619 | |
| 620 | static const |
| 621 | struct mmio_range *find_matching_range(const struct mmio_range *ranges, |
| 622 | struct kvm_exit_mmio *mmio, |
| 623 | phys_addr_t base) |
| 624 | { |
| 625 | const struct mmio_range *r = ranges; |
| 626 | phys_addr_t addr = mmio->phys_addr - base; |
| 627 | |
| 628 | while (r->len) { |
| 629 | if (addr >= r->base && |
| 630 | (addr + mmio->len) <= (r->base + r->len)) |
| 631 | return r; |
| 632 | r++; |
| 633 | } |
| 634 | |
| 635 | return NULL; |
| 636 | } |
| 637 | |
| 638 | /** |
| 639 | * vgic_handle_mmio - handle an in-kernel MMIO access |
| 640 | * @vcpu: pointer to the vcpu performing the access |
| 641 | * @run: pointer to the kvm_run structure |
| 642 | * @mmio: pointer to the data describing the access |
| 643 | * |
| 644 | * returns true if the MMIO access has been performed in kernel space, |
| 645 | * and false if it needs to be emulated in user space. |
| 646 | */ |
| 647 | bool vgic_handle_mmio(struct kvm_vcpu *vcpu, struct kvm_run *run, |
| 648 | struct kvm_exit_mmio *mmio) |
| 649 | { |
Marc Zyngier | b47ef92 | 2013-01-21 19:36:14 -0500 | [diff] [blame^] | 650 | const struct mmio_range *range; |
| 651 | struct vgic_dist *dist = &vcpu->kvm->arch.vgic; |
| 652 | unsigned long base = dist->vgic_dist_base; |
| 653 | bool updated_state; |
| 654 | unsigned long offset; |
| 655 | |
| 656 | if (!irqchip_in_kernel(vcpu->kvm) || |
| 657 | mmio->phys_addr < base || |
| 658 | (mmio->phys_addr + mmio->len) > (base + KVM_VGIC_V2_DIST_SIZE)) |
| 659 | return false; |
| 660 | |
| 661 | /* We don't support ldrd / strd or ldm / stm to the emulated vgic */ |
| 662 | if (mmio->len > 4) { |
| 663 | kvm_inject_dabt(vcpu, mmio->phys_addr); |
| 664 | return true; |
| 665 | } |
| 666 | |
| 667 | range = find_matching_range(vgic_ranges, mmio, base); |
| 668 | if (unlikely(!range || !range->handle_mmio)) { |
| 669 | pr_warn("Unhandled access %d %08llx %d\n", |
| 670 | mmio->is_write, mmio->phys_addr, mmio->len); |
| 671 | return false; |
| 672 | } |
| 673 | |
| 674 | spin_lock(&vcpu->kvm->arch.vgic.lock); |
| 675 | offset = mmio->phys_addr - range->base - base; |
| 676 | updated_state = range->handle_mmio(vcpu, mmio, offset); |
| 677 | spin_unlock(&vcpu->kvm->arch.vgic.lock); |
| 678 | kvm_prepare_mmio(run, mmio); |
| 679 | kvm_handle_mmio_return(vcpu, run); |
| 680 | |
| 681 | return true; |
| 682 | } |
| 683 | |
| 684 | static void vgic_dispatch_sgi(struct kvm_vcpu *vcpu, u32 reg) |
| 685 | { |
| 686 | struct kvm *kvm = vcpu->kvm; |
| 687 | struct vgic_dist *dist = &kvm->arch.vgic; |
| 688 | int nrcpus = atomic_read(&kvm->online_vcpus); |
| 689 | u8 target_cpus; |
| 690 | int sgi, mode, c, vcpu_id; |
| 691 | |
| 692 | vcpu_id = vcpu->vcpu_id; |
| 693 | |
| 694 | sgi = reg & 0xf; |
| 695 | target_cpus = (reg >> 16) & 0xff; |
| 696 | mode = (reg >> 24) & 3; |
| 697 | |
| 698 | switch (mode) { |
| 699 | case 0: |
| 700 | if (!target_cpus) |
| 701 | return; |
| 702 | |
| 703 | case 1: |
| 704 | target_cpus = ((1 << nrcpus) - 1) & ~(1 << vcpu_id) & 0xff; |
| 705 | break; |
| 706 | |
| 707 | case 2: |
| 708 | target_cpus = 1 << vcpu_id; |
| 709 | break; |
| 710 | } |
| 711 | |
| 712 | kvm_for_each_vcpu(c, vcpu, kvm) { |
| 713 | if (target_cpus & 1) { |
| 714 | /* Flag the SGI as pending */ |
| 715 | vgic_dist_irq_set(vcpu, sgi); |
| 716 | dist->irq_sgi_sources[c][sgi] |= 1 << vcpu_id; |
| 717 | kvm_debug("SGI%d from CPU%d to CPU%d\n", sgi, vcpu_id, c); |
| 718 | } |
| 719 | |
| 720 | target_cpus >>= 1; |
| 721 | } |
| 722 | } |
| 723 | |
| 724 | static int compute_pending_for_cpu(struct kvm_vcpu *vcpu) |
| 725 | { |
| 726 | return 0; |
| 727 | } |
| 728 | |
| 729 | /* |
| 730 | * Update the interrupt state and determine which CPUs have pending |
| 731 | * interrupts. Must be called with distributor lock held. |
| 732 | */ |
| 733 | static void vgic_update_state(struct kvm *kvm) |
| 734 | { |
| 735 | struct vgic_dist *dist = &kvm->arch.vgic; |
| 736 | struct kvm_vcpu *vcpu; |
| 737 | int c; |
| 738 | |
| 739 | if (!dist->enabled) { |
| 740 | set_bit(0, &dist->irq_pending_on_cpu); |
| 741 | return; |
| 742 | } |
| 743 | |
| 744 | kvm_for_each_vcpu(c, vcpu, kvm) { |
| 745 | if (compute_pending_for_cpu(vcpu)) { |
| 746 | pr_debug("CPU%d has pending interrupts\n", c); |
| 747 | set_bit(c, &dist->irq_pending_on_cpu); |
| 748 | } |
| 749 | } |
Marc Zyngier | 1a89dd9 | 2013-01-21 19:36:12 -0500 | [diff] [blame] | 750 | } |
Christoffer Dall | 330690c | 2013-01-21 19:36:13 -0500 | [diff] [blame] | 751 | |
| 752 | static bool vgic_ioaddr_overlap(struct kvm *kvm) |
| 753 | { |
| 754 | phys_addr_t dist = kvm->arch.vgic.vgic_dist_base; |
| 755 | phys_addr_t cpu = kvm->arch.vgic.vgic_cpu_base; |
| 756 | |
| 757 | if (IS_VGIC_ADDR_UNDEF(dist) || IS_VGIC_ADDR_UNDEF(cpu)) |
| 758 | return 0; |
| 759 | if ((dist <= cpu && dist + KVM_VGIC_V2_DIST_SIZE > cpu) || |
| 760 | (cpu <= dist && cpu + KVM_VGIC_V2_CPU_SIZE > dist)) |
| 761 | return -EBUSY; |
| 762 | return 0; |
| 763 | } |
| 764 | |
| 765 | static int vgic_ioaddr_assign(struct kvm *kvm, phys_addr_t *ioaddr, |
| 766 | phys_addr_t addr, phys_addr_t size) |
| 767 | { |
| 768 | int ret; |
| 769 | |
| 770 | if (!IS_VGIC_ADDR_UNDEF(*ioaddr)) |
| 771 | return -EEXIST; |
| 772 | if (addr + size < addr) |
| 773 | return -EINVAL; |
| 774 | |
| 775 | ret = vgic_ioaddr_overlap(kvm); |
| 776 | if (ret) |
| 777 | return ret; |
| 778 | *ioaddr = addr; |
| 779 | return ret; |
| 780 | } |
| 781 | |
| 782 | int kvm_vgic_set_addr(struct kvm *kvm, unsigned long type, u64 addr) |
| 783 | { |
| 784 | int r = 0; |
| 785 | struct vgic_dist *vgic = &kvm->arch.vgic; |
| 786 | |
| 787 | if (addr & ~KVM_PHYS_MASK) |
| 788 | return -E2BIG; |
| 789 | |
| 790 | if (addr & ~PAGE_MASK) |
| 791 | return -EINVAL; |
| 792 | |
| 793 | mutex_lock(&kvm->lock); |
| 794 | switch (type) { |
| 795 | case KVM_VGIC_V2_ADDR_TYPE_DIST: |
| 796 | r = vgic_ioaddr_assign(kvm, &vgic->vgic_dist_base, |
| 797 | addr, KVM_VGIC_V2_DIST_SIZE); |
| 798 | break; |
| 799 | case KVM_VGIC_V2_ADDR_TYPE_CPU: |
| 800 | r = vgic_ioaddr_assign(kvm, &vgic->vgic_cpu_base, |
| 801 | addr, KVM_VGIC_V2_CPU_SIZE); |
| 802 | break; |
| 803 | default: |
| 804 | r = -ENODEV; |
| 805 | } |
| 806 | |
| 807 | mutex_unlock(&kvm->lock); |
| 808 | return r; |
| 809 | } |