linux/arch/x86/kernel/apic/msi.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * Support of MSI, HPET and DMAR interrupts.
   4 *
   5 * Copyright (C) 1997, 1998, 1999, 2000, 2009 Ingo Molnar, Hajnalka Szabo
   6 *      Moved from arch/x86/kernel/apic/io_apic.c.
   7 * Jiang Liu <jiang.liu@linux.intel.com>
   8 *      Convert to hierarchical irqdomain
   9 */
  10#include <linux/mm.h>
  11#include <linux/interrupt.h>
  12#include <linux/irq.h>
  13#include <linux/pci.h>
  14#include <linux/dmar.h>
  15#include <linux/hpet.h>
  16#include <linux/msi.h>
  17#include <asm/irqdomain.h>
  18#include <asm/msidef.h>
  19#include <asm/hpet.h>
  20#include <asm/hw_irq.h>
  21#include <asm/apic.h>
  22#include <asm/irq_remapping.h>
  23
  24static struct irq_domain *msi_default_domain;
  25
  26static void __irq_msi_compose_msg(struct irq_cfg *cfg, struct msi_msg *msg)
  27{
  28        msg->address_hi = MSI_ADDR_BASE_HI;
  29
  30        if (x2apic_enabled())
  31                msg->address_hi |= MSI_ADDR_EXT_DEST_ID(cfg->dest_apicid);
  32
  33        msg->address_lo =
  34                MSI_ADDR_BASE_LO |
  35                ((apic->irq_dest_mode == 0) ?
  36                        MSI_ADDR_DEST_MODE_PHYSICAL :
  37                        MSI_ADDR_DEST_MODE_LOGICAL) |
  38                MSI_ADDR_REDIRECTION_CPU |
  39                MSI_ADDR_DEST_ID(cfg->dest_apicid);
  40
  41        msg->data =
  42                MSI_DATA_TRIGGER_EDGE |
  43                MSI_DATA_LEVEL_ASSERT |
  44                MSI_DATA_DELIVERY_FIXED |
  45                MSI_DATA_VECTOR(cfg->vector);
  46}
  47
  48static void irq_msi_compose_msg(struct irq_data *data, struct msi_msg *msg)
  49{
  50        __irq_msi_compose_msg(irqd_cfg(data), msg);
  51}
  52
  53static void irq_msi_update_msg(struct irq_data *irqd, struct irq_cfg *cfg)
  54{
  55        struct msi_msg msg[2] = { [1] = { }, };
  56
  57        __irq_msi_compose_msg(cfg, msg);
  58        irq_data_get_irq_chip(irqd)->irq_write_msi_msg(irqd, msg);
  59}
  60
  61static int
  62msi_set_affinity(struct irq_data *irqd, const struct cpumask *mask, bool force)
  63{
  64        struct irq_cfg old_cfg, *cfg = irqd_cfg(irqd);
  65        struct irq_data *parent = irqd->parent_data;
  66        unsigned int cpu;
  67        int ret;
  68
  69        /* Save the current configuration */
  70        cpu = cpumask_first(irq_data_get_effective_affinity_mask(irqd));
  71        old_cfg = *cfg;
  72
  73        /* Allocate a new target vector */
  74        ret = parent->chip->irq_set_affinity(parent, mask, force);
  75        if (ret < 0 || ret == IRQ_SET_MASK_OK_DONE)
  76                return ret;
  77
  78        /*
  79         * For non-maskable and non-remapped MSI interrupts the migration
  80         * to a different destination CPU and a different vector has to be
  81         * done careful to handle the possible stray interrupt which can be
  82         * caused by the non-atomic update of the address/data pair.
  83         *
  84         * Direct update is possible when:
  85         * - The MSI is maskable (remapped MSI does not use this code path)).
  86         *   The quirk bit is not set in this case.
  87         * - The new vector is the same as the old vector
  88         * - The old vector is MANAGED_IRQ_SHUTDOWN_VECTOR (interrupt starts up)
  89         * - The new destination CPU is the same as the old destination CPU
  90         */
  91        if (!irqd_msi_nomask_quirk(irqd) ||
  92            cfg->vector == old_cfg.vector ||
  93            old_cfg.vector == MANAGED_IRQ_SHUTDOWN_VECTOR ||
  94            cfg->dest_apicid == old_cfg.dest_apicid) {
  95                irq_msi_update_msg(irqd, cfg);
  96                return ret;
  97        }
  98
  99        /*
 100         * Paranoia: Validate that the interrupt target is the local
 101         * CPU.
 102         */
 103        if (WARN_ON_ONCE(cpu != smp_processor_id())) {
 104                irq_msi_update_msg(irqd, cfg);
 105                return ret;
 106        }
 107
 108        /*
 109         * Redirect the interrupt to the new vector on the current CPU
 110         * first. This might cause a spurious interrupt on this vector if
 111         * the device raises an interrupt right between this update and the
 112         * update to the final destination CPU.
 113         *
 114         * If the vector is in use then the installed device handler will
 115         * denote it as spurious which is no harm as this is a rare event
 116         * and interrupt handlers have to cope with spurious interrupts
 117         * anyway. If the vector is unused, then it is marked so it won't
 118         * trigger the 'No irq handler for vector' warning in
 119         * common_interrupt().
 120         *
 121         * This requires to hold vector lock to prevent concurrent updates to
 122         * the affected vector.
 123         */
 124        lock_vector_lock();
 125
 126        /*
 127         * Mark the new target vector on the local CPU if it is currently
 128         * unused. Reuse the VECTOR_RETRIGGERED state which is also used in
 129         * the CPU hotplug path for a similar purpose. This cannot be
 130         * undone here as the current CPU has interrupts disabled and
 131         * cannot handle the interrupt before the whole set_affinity()
 132         * section is done. In the CPU unplug case, the current CPU is
 133         * about to vanish and will not handle any interrupts anymore. The
 134         * vector is cleaned up when the CPU comes online again.
 135         */
 136        if (IS_ERR_OR_NULL(this_cpu_read(vector_irq[cfg->vector])))
 137                this_cpu_write(vector_irq[cfg->vector], VECTOR_RETRIGGERED);
 138
 139        /* Redirect it to the new vector on the local CPU temporarily */
 140        old_cfg.vector = cfg->vector;
 141        irq_msi_update_msg(irqd, &old_cfg);
 142
 143        /* Now transition it to the target CPU */
 144        irq_msi_update_msg(irqd, cfg);
 145
 146        /*
 147         * All interrupts after this point are now targeted at the new
 148         * vector/CPU.
 149         *
 150         * Drop vector lock before testing whether the temporary assignment
 151         * to the local CPU was hit by an interrupt raised in the device,
 152         * because the retrigger function acquires vector lock again.
 153         */
 154        unlock_vector_lock();
 155
 156        /*
 157         * Check whether the transition raced with a device interrupt and
 158         * is pending in the local APICs IRR. It is safe to do this outside
 159         * of vector lock as the irq_desc::lock of this interrupt is still
 160         * held and interrupts are disabled: The check is not accessing the
 161         * underlying vector store. It's just checking the local APIC's
 162         * IRR.
 163         */
 164        if (lapic_vector_set_in_irr(cfg->vector))
 165                irq_data_get_irq_chip(irqd)->irq_retrigger(irqd);
 166
 167        return ret;
 168}
 169
 170/*
 171 * IRQ Chip for MSI PCI/PCI-X/PCI-Express Devices,
 172 * which implement the MSI or MSI-X Capability Structure.
 173 */
 174static struct irq_chip pci_msi_controller = {
 175        .name                   = "PCI-MSI",
 176        .irq_unmask             = pci_msi_unmask_irq,
 177        .irq_mask               = pci_msi_mask_irq,
 178        .irq_ack                = irq_chip_ack_parent,
 179        .irq_retrigger          = irq_chip_retrigger_hierarchy,
 180        .irq_compose_msi_msg    = irq_msi_compose_msg,
 181        .irq_set_affinity       = msi_set_affinity,
 182        .flags                  = IRQCHIP_SKIP_SET_WAKE,
 183};
 184
 185int native_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
 186{
 187        struct irq_domain *domain;
 188        struct irq_alloc_info info;
 189
 190        init_irq_alloc_info(&info, NULL);
 191        info.type = X86_IRQ_ALLOC_TYPE_MSI;
 192        info.msi_dev = dev;
 193
 194        domain = irq_remapping_get_irq_domain(&info);
 195        if (domain == NULL)
 196                domain = msi_default_domain;
 197        if (domain == NULL)
 198                return -ENOSYS;
 199
 200        return msi_domain_alloc_irqs(domain, &dev->dev, nvec);
 201}
 202
 203void native_teardown_msi_irq(unsigned int irq)
 204{
 205        irq_domain_free_irqs(irq, 1);
 206}
 207
 208static irq_hw_number_t pci_msi_get_hwirq(struct msi_domain_info *info,
 209                                         msi_alloc_info_t *arg)
 210{
 211        return arg->msi_hwirq;
 212}
 213
 214int pci_msi_prepare(struct irq_domain *domain, struct device *dev, int nvec,
 215                    msi_alloc_info_t *arg)
 216{
 217        struct pci_dev *pdev = to_pci_dev(dev);
 218        struct msi_desc *desc = first_pci_msi_entry(pdev);
 219
 220        init_irq_alloc_info(arg, NULL);
 221        arg->msi_dev = pdev;
 222        if (desc->msi_attrib.is_msix) {
 223                arg->type = X86_IRQ_ALLOC_TYPE_MSIX;
 224        } else {
 225                arg->type = X86_IRQ_ALLOC_TYPE_MSI;
 226                arg->flags |= X86_IRQ_ALLOC_CONTIGUOUS_VECTORS;
 227        }
 228
 229        return 0;
 230}
 231EXPORT_SYMBOL_GPL(pci_msi_prepare);
 232
 233void pci_msi_set_desc(msi_alloc_info_t *arg, struct msi_desc *desc)
 234{
 235        arg->msi_hwirq = pci_msi_domain_calc_hwirq(arg->msi_dev, desc);
 236}
 237EXPORT_SYMBOL_GPL(pci_msi_set_desc);
 238
 239static struct msi_domain_ops pci_msi_domain_ops = {
 240        .get_hwirq      = pci_msi_get_hwirq,
 241        .msi_prepare    = pci_msi_prepare,
 242        .set_desc       = pci_msi_set_desc,
 243};
 244
 245static struct msi_domain_info pci_msi_domain_info = {
 246        .flags          = MSI_FLAG_USE_DEF_DOM_OPS | MSI_FLAG_USE_DEF_CHIP_OPS |
 247                          MSI_FLAG_PCI_MSIX,
 248        .ops            = &pci_msi_domain_ops,
 249        .chip           = &pci_msi_controller,
 250        .handler        = handle_edge_irq,
 251        .handler_name   = "edge",
 252};
 253
 254void __init arch_init_msi_domain(struct irq_domain *parent)
 255{
 256        struct fwnode_handle *fn;
 257
 258        if (disable_apic)
 259                return;
 260
 261        fn = irq_domain_alloc_named_fwnode("PCI-MSI");
 262        if (fn) {
 263                msi_default_domain =
 264                        pci_msi_create_irq_domain(fn, &pci_msi_domain_info,
 265                                                  parent);
 266        }
 267        if (!msi_default_domain) {
 268                irq_domain_free_fwnode(fn);
 269                pr_warn("failed to initialize irqdomain for MSI/MSI-x.\n");
 270        } else {
 271                msi_default_domain->flags |= IRQ_DOMAIN_MSI_NOMASK_QUIRK;
 272        }
 273}
 274
 275#ifdef CONFIG_IRQ_REMAP
 276static struct irq_chip pci_msi_ir_controller = {
 277        .name                   = "IR-PCI-MSI",
 278        .irq_unmask             = pci_msi_unmask_irq,
 279        .irq_mask               = pci_msi_mask_irq,
 280        .irq_ack                = irq_chip_ack_parent,
 281        .irq_retrigger          = irq_chip_retrigger_hierarchy,
 282        .irq_set_vcpu_affinity  = irq_chip_set_vcpu_affinity_parent,
 283        .flags                  = IRQCHIP_SKIP_SET_WAKE,
 284};
 285
 286static struct msi_domain_info pci_msi_ir_domain_info = {
 287        .flags          = MSI_FLAG_USE_DEF_DOM_OPS | MSI_FLAG_USE_DEF_CHIP_OPS |
 288                          MSI_FLAG_MULTI_PCI_MSI | MSI_FLAG_PCI_MSIX,
 289        .ops            = &pci_msi_domain_ops,
 290        .chip           = &pci_msi_ir_controller,
 291        .handler        = handle_edge_irq,
 292        .handler_name   = "edge",
 293};
 294
 295struct irq_domain *arch_create_remap_msi_irq_domain(struct irq_domain *parent,
 296                                                    const char *name, int id)
 297{
 298        struct fwnode_handle *fn;
 299        struct irq_domain *d;
 300
 301        fn = irq_domain_alloc_named_id_fwnode(name, id);
 302        if (!fn)
 303                return NULL;
 304        d = pci_msi_create_irq_domain(fn, &pci_msi_ir_domain_info, parent);
 305        if (!d)
 306                irq_domain_free_fwnode(fn);
 307        return d;
 308}
 309#endif
 310
 311#ifdef CONFIG_DMAR_TABLE
 312static void dmar_msi_write_msg(struct irq_data *data, struct msi_msg *msg)
 313{
 314        dmar_msi_write(data->irq, msg);
 315}
 316
 317static struct irq_chip dmar_msi_controller = {
 318        .name                   = "DMAR-MSI",
 319        .irq_unmask             = dmar_msi_unmask,
 320        .irq_mask               = dmar_msi_mask,
 321        .irq_ack                = irq_chip_ack_parent,
 322        .irq_set_affinity       = msi_domain_set_affinity,
 323        .irq_retrigger          = irq_chip_retrigger_hierarchy,
 324        .irq_compose_msi_msg    = irq_msi_compose_msg,
 325        .irq_write_msi_msg      = dmar_msi_write_msg,
 326        .flags                  = IRQCHIP_SKIP_SET_WAKE,
 327};
 328
 329static irq_hw_number_t dmar_msi_get_hwirq(struct msi_domain_info *info,
 330                                          msi_alloc_info_t *arg)
 331{
 332        return arg->dmar_id;
 333}
 334
 335static int dmar_msi_init(struct irq_domain *domain,
 336                         struct msi_domain_info *info, unsigned int virq,
 337                         irq_hw_number_t hwirq, msi_alloc_info_t *arg)
 338{
 339        irq_domain_set_info(domain, virq, arg->dmar_id, info->chip, NULL,
 340                            handle_edge_irq, arg->dmar_data, "edge");
 341
 342        return 0;
 343}
 344
 345static struct msi_domain_ops dmar_msi_domain_ops = {
 346        .get_hwirq      = dmar_msi_get_hwirq,
 347        .msi_init       = dmar_msi_init,
 348};
 349
 350static struct msi_domain_info dmar_msi_domain_info = {
 351        .ops            = &dmar_msi_domain_ops,
 352        .chip           = &dmar_msi_controller,
 353};
 354
 355static struct irq_domain *dmar_get_irq_domain(void)
 356{
 357        static struct irq_domain *dmar_domain;
 358        static DEFINE_MUTEX(dmar_lock);
 359        struct fwnode_handle *fn;
 360
 361        mutex_lock(&dmar_lock);
 362        if (dmar_domain)
 363                goto out;
 364
 365        fn = irq_domain_alloc_named_fwnode("DMAR-MSI");
 366        if (fn) {
 367                dmar_domain = msi_create_irq_domain(fn, &dmar_msi_domain_info,
 368                                                    x86_vector_domain);
 369                if (!dmar_domain)
 370                        irq_domain_free_fwnode(fn);
 371        }
 372out:
 373        mutex_unlock(&dmar_lock);
 374        return dmar_domain;
 375}
 376
 377int dmar_alloc_hwirq(int id, int node, void *arg)
 378{
 379        struct irq_domain *domain = dmar_get_irq_domain();
 380        struct irq_alloc_info info;
 381
 382        if (!domain)
 383                return -1;
 384
 385        init_irq_alloc_info(&info, NULL);
 386        info.type = X86_IRQ_ALLOC_TYPE_DMAR;
 387        info.dmar_id = id;
 388        info.dmar_data = arg;
 389
 390        return irq_domain_alloc_irqs(domain, 1, node, &info);
 391}
 392
 393void dmar_free_hwirq(int irq)
 394{
 395        irq_domain_free_irqs(irq, 1);
 396}
 397#endif
 398
 399/*
 400 * MSI message composition
 401 */
 402#ifdef CONFIG_HPET_TIMER
 403static inline int hpet_dev_id(struct irq_domain *domain)
 404{
 405        struct msi_domain_info *info = msi_get_domain_info(domain);
 406
 407        return (int)(long)info->data;
 408}
 409
 410static void hpet_msi_write_msg(struct irq_data *data, struct msi_msg *msg)
 411{
 412        hpet_msi_write(irq_data_get_irq_handler_data(data), msg);
 413}
 414
 415static struct irq_chip hpet_msi_controller __ro_after_init = {
 416        .name = "HPET-MSI",
 417        .irq_unmask = hpet_msi_unmask,
 418        .irq_mask = hpet_msi_mask,
 419        .irq_ack = irq_chip_ack_parent,
 420        .irq_set_affinity = msi_domain_set_affinity,
 421        .irq_retrigger = irq_chip_retrigger_hierarchy,
 422        .irq_compose_msi_msg = irq_msi_compose_msg,
 423        .irq_write_msi_msg = hpet_msi_write_msg,
 424        .flags = IRQCHIP_SKIP_SET_WAKE,
 425};
 426
 427static irq_hw_number_t hpet_msi_get_hwirq(struct msi_domain_info *info,
 428                                          msi_alloc_info_t *arg)
 429{
 430        return arg->hpet_index;
 431}
 432
 433static int hpet_msi_init(struct irq_domain *domain,
 434                         struct msi_domain_info *info, unsigned int virq,
 435                         irq_hw_number_t hwirq, msi_alloc_info_t *arg)
 436{
 437        irq_set_status_flags(virq, IRQ_MOVE_PCNTXT);
 438        irq_domain_set_info(domain, virq, arg->hpet_index, info->chip, NULL,
 439                            handle_edge_irq, arg->hpet_data, "edge");
 440
 441        return 0;
 442}
 443
 444static void hpet_msi_free(struct irq_domain *domain,
 445                          struct msi_domain_info *info, unsigned int virq)
 446{
 447        irq_clear_status_flags(virq, IRQ_MOVE_PCNTXT);
 448}
 449
 450static struct msi_domain_ops hpet_msi_domain_ops = {
 451        .get_hwirq      = hpet_msi_get_hwirq,
 452        .msi_init       = hpet_msi_init,
 453        .msi_free       = hpet_msi_free,
 454};
 455
 456static struct msi_domain_info hpet_msi_domain_info = {
 457        .ops            = &hpet_msi_domain_ops,
 458        .chip           = &hpet_msi_controller,
 459};
 460
 461struct irq_domain *hpet_create_irq_domain(int hpet_id)
 462{
 463        struct msi_domain_info *domain_info;
 464        struct irq_domain *parent, *d;
 465        struct irq_alloc_info info;
 466        struct fwnode_handle *fn;
 467
 468        if (x86_vector_domain == NULL)
 469                return NULL;
 470
 471        domain_info = kzalloc(sizeof(*domain_info), GFP_KERNEL);
 472        if (!domain_info)
 473                return NULL;
 474
 475        *domain_info = hpet_msi_domain_info;
 476        domain_info->data = (void *)(long)hpet_id;
 477
 478        init_irq_alloc_info(&info, NULL);
 479        info.type = X86_IRQ_ALLOC_TYPE_HPET;
 480        info.hpet_id = hpet_id;
 481        parent = irq_remapping_get_ir_irq_domain(&info);
 482        if (parent == NULL)
 483                parent = x86_vector_domain;
 484        else
 485                hpet_msi_controller.name = "IR-HPET-MSI";
 486
 487        fn = irq_domain_alloc_named_id_fwnode(hpet_msi_controller.name,
 488                                              hpet_id);
 489        if (!fn) {
 490                kfree(domain_info);
 491                return NULL;
 492        }
 493
 494        d = msi_create_irq_domain(fn, domain_info, parent);
 495        if (!d) {
 496                irq_domain_free_fwnode(fn);
 497                kfree(domain_info);
 498        }
 499        return d;
 500}
 501
 502int hpet_assign_irq(struct irq_domain *domain, struct hpet_channel *hc,
 503                    int dev_num)
 504{
 505        struct irq_alloc_info info;
 506
 507        init_irq_alloc_info(&info, NULL);
 508        info.type = X86_IRQ_ALLOC_TYPE_HPET;
 509        info.hpet_data = hc;
 510        info.hpet_id = hpet_dev_id(domain);
 511        info.hpet_index = dev_num;
 512
 513        return irq_domain_alloc_irqs(domain, 1, NUMA_NO_NODE, &info);
 514}
 515#endif
 516