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 do_IRQ().
 119         *
 120         * This requires to hold vector lock to prevent concurrent updates to
 121         * the affected vector.
 122         */
 123        lock_vector_lock();
 124
 125        /*
 126         * Mark the new target vector on the local CPU if it is currently
 127         * unused. Reuse the VECTOR_RETRIGGERED state which is also used in
 128         * the CPU hotplug path for a similar purpose. This cannot be
 129         * undone here as the current CPU has interrupts disabled and
 130         * cannot handle the interrupt before the whole set_affinity()
 131         * section is done. In the CPU unplug case, the current CPU is
 132         * about to vanish and will not handle any interrupts anymore. The
 133         * vector is cleaned up when the CPU comes online again.
 134         */
 135        if (IS_ERR_OR_NULL(this_cpu_read(vector_irq[cfg->vector])))
 136                this_cpu_write(vector_irq[cfg->vector], VECTOR_RETRIGGERED);
 137
 138        /* Redirect it to the new vector on the local CPU temporarily */
 139        old_cfg.vector = cfg->vector;
 140        irq_msi_update_msg(irqd, &old_cfg);
 141
 142        /* Now transition it to the target CPU */
 143        irq_msi_update_msg(irqd, cfg);
 144
 145        /*
 146         * All interrupts after this point are now targeted at the new
 147         * vector/CPU.
 148         *
 149         * Drop vector lock before testing whether the temporary assignment
 150         * to the local CPU was hit by an interrupt raised in the device,
 151         * because the retrigger function acquires vector lock again.
 152         */
 153        unlock_vector_lock();
 154
 155        /*
 156         * Check whether the transition raced with a device interrupt and
 157         * is pending in the local APICs IRR. It is safe to do this outside
 158         * of vector lock as the irq_desc::lock of this interrupt is still
 159         * held and interrupts are disabled: The check is not accessing the
 160         * underlying vector store. It's just checking the local APIC's
 161         * IRR.
 162         */
 163        if (lapic_vector_set_in_irr(cfg->vector))
 164                irq_data_get_irq_chip(irqd)->irq_retrigger(irqd);
 165
 166        return ret;
 167}
 168
 169/*
 170 * IRQ Chip for MSI PCI/PCI-X/PCI-Express Devices,
 171 * which implement the MSI or MSI-X Capability Structure.
 172 */
 173static struct irq_chip pci_msi_controller = {
 174        .name                   = "PCI-MSI",
 175        .irq_unmask             = pci_msi_unmask_irq,
 176        .irq_mask               = pci_msi_mask_irq,
 177        .irq_ack                = irq_chip_ack_parent,
 178        .irq_retrigger          = irq_chip_retrigger_hierarchy,
 179        .irq_compose_msi_msg    = irq_msi_compose_msg,
 180        .irq_set_affinity       = msi_set_affinity,
 181        .flags                  = IRQCHIP_SKIP_SET_WAKE,
 182};
 183
 184int native_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
 185{
 186        struct irq_domain *domain;
 187        struct irq_alloc_info info;
 188
 189        init_irq_alloc_info(&info, NULL);
 190        info.type = X86_IRQ_ALLOC_TYPE_MSI;
 191        info.msi_dev = dev;
 192
 193        domain = irq_remapping_get_irq_domain(&info);
 194        if (domain == NULL)
 195                domain = msi_default_domain;
 196        if (domain == NULL)
 197                return -ENOSYS;
 198
 199        return msi_domain_alloc_irqs(domain, &dev->dev, nvec);
 200}
 201
 202void native_teardown_msi_irq(unsigned int irq)
 203{
 204        irq_domain_free_irqs(irq, 1);
 205}
 206
 207static irq_hw_number_t pci_msi_get_hwirq(struct msi_domain_info *info,
 208                                         msi_alloc_info_t *arg)
 209{
 210        return arg->msi_hwirq;
 211}
 212
 213int pci_msi_prepare(struct irq_domain *domain, struct device *dev, int nvec,
 214                    msi_alloc_info_t *arg)
 215{
 216        struct pci_dev *pdev = to_pci_dev(dev);
 217        struct msi_desc *desc = first_pci_msi_entry(pdev);
 218
 219        init_irq_alloc_info(arg, NULL);
 220        arg->msi_dev = pdev;
 221        if (desc->msi_attrib.is_msix) {
 222                arg->type = X86_IRQ_ALLOC_TYPE_MSIX;
 223        } else {
 224                arg->type = X86_IRQ_ALLOC_TYPE_MSI;
 225                arg->flags |= X86_IRQ_ALLOC_CONTIGUOUS_VECTORS;
 226        }
 227
 228        return 0;
 229}
 230EXPORT_SYMBOL_GPL(pci_msi_prepare);
 231
 232void pci_msi_set_desc(msi_alloc_info_t *arg, struct msi_desc *desc)
 233{
 234        arg->msi_hwirq = pci_msi_domain_calc_hwirq(arg->msi_dev, desc);
 235}
 236EXPORT_SYMBOL_GPL(pci_msi_set_desc);
 237
 238static struct msi_domain_ops pci_msi_domain_ops = {
 239        .get_hwirq      = pci_msi_get_hwirq,
 240        .msi_prepare    = pci_msi_prepare,
 241        .set_desc       = pci_msi_set_desc,
 242};
 243
 244static struct msi_domain_info pci_msi_domain_info = {
 245        .flags          = MSI_FLAG_USE_DEF_DOM_OPS | MSI_FLAG_USE_DEF_CHIP_OPS |
 246                          MSI_FLAG_PCI_MSIX,
 247        .ops            = &pci_msi_domain_ops,
 248        .chip           = &pci_msi_controller,
 249        .handler        = handle_edge_irq,
 250        .handler_name   = "edge",
 251};
 252
 253void __init arch_init_msi_domain(struct irq_domain *parent)
 254{
 255        struct fwnode_handle *fn;
 256
 257        if (disable_apic)
 258                return;
 259
 260        fn = irq_domain_alloc_named_fwnode("PCI-MSI");
 261        if (fn) {
 262                msi_default_domain =
 263                        pci_msi_create_irq_domain(fn, &pci_msi_domain_info,
 264                                                  parent);
 265                irq_domain_free_fwnode(fn);
 266        }
 267        if (!msi_default_domain)
 268                pr_warn("failed to initialize irqdomain for MSI/MSI-x.\n");
 269        else
 270                msi_default_domain->flags |= IRQ_DOMAIN_MSI_NOMASK_QUIRK;
 271}
 272
 273#ifdef CONFIG_IRQ_REMAP
 274static struct irq_chip pci_msi_ir_controller = {
 275        .name                   = "IR-PCI-MSI",
 276        .irq_unmask             = pci_msi_unmask_irq,
 277        .irq_mask               = pci_msi_mask_irq,
 278        .irq_ack                = irq_chip_ack_parent,
 279        .irq_retrigger          = irq_chip_retrigger_hierarchy,
 280        .irq_set_vcpu_affinity  = irq_chip_set_vcpu_affinity_parent,
 281        .flags                  = IRQCHIP_SKIP_SET_WAKE,
 282};
 283
 284static struct msi_domain_info pci_msi_ir_domain_info = {
 285        .flags          = MSI_FLAG_USE_DEF_DOM_OPS | MSI_FLAG_USE_DEF_CHIP_OPS |
 286                          MSI_FLAG_MULTI_PCI_MSI | MSI_FLAG_PCI_MSIX,
 287        .ops            = &pci_msi_domain_ops,
 288        .chip           = &pci_msi_ir_controller,
 289        .handler        = handle_edge_irq,
 290        .handler_name   = "edge",
 291};
 292
 293struct irq_domain *arch_create_remap_msi_irq_domain(struct irq_domain *parent,
 294                                                    const char *name, int id)
 295{
 296        struct fwnode_handle *fn;
 297        struct irq_domain *d;
 298
 299        fn = irq_domain_alloc_named_id_fwnode(name, id);
 300        if (!fn)
 301                return NULL;
 302        d = pci_msi_create_irq_domain(fn, &pci_msi_ir_domain_info, parent);
 303        irq_domain_free_fwnode(fn);
 304        return d;
 305}
 306#endif
 307
 308#ifdef CONFIG_DMAR_TABLE
 309static void dmar_msi_write_msg(struct irq_data *data, struct msi_msg *msg)
 310{
 311        dmar_msi_write(data->irq, msg);
 312}
 313
 314static struct irq_chip dmar_msi_controller = {
 315        .name                   = "DMAR-MSI",
 316        .irq_unmask             = dmar_msi_unmask,
 317        .irq_mask               = dmar_msi_mask,
 318        .irq_ack                = irq_chip_ack_parent,
 319        .irq_set_affinity       = msi_domain_set_affinity,
 320        .irq_retrigger          = irq_chip_retrigger_hierarchy,
 321        .irq_compose_msi_msg    = irq_msi_compose_msg,
 322        .irq_write_msi_msg      = dmar_msi_write_msg,
 323        .flags                  = IRQCHIP_SKIP_SET_WAKE,
 324};
 325
 326static irq_hw_number_t dmar_msi_get_hwirq(struct msi_domain_info *info,
 327                                          msi_alloc_info_t *arg)
 328{
 329        return arg->dmar_id;
 330}
 331
 332static int dmar_msi_init(struct irq_domain *domain,
 333                         struct msi_domain_info *info, unsigned int virq,
 334                         irq_hw_number_t hwirq, msi_alloc_info_t *arg)
 335{
 336        irq_domain_set_info(domain, virq, arg->dmar_id, info->chip, NULL,
 337                            handle_edge_irq, arg->dmar_data, "edge");
 338
 339        return 0;
 340}
 341
 342static struct msi_domain_ops dmar_msi_domain_ops = {
 343        .get_hwirq      = dmar_msi_get_hwirq,
 344        .msi_init       = dmar_msi_init,
 345};
 346
 347static struct msi_domain_info dmar_msi_domain_info = {
 348        .ops            = &dmar_msi_domain_ops,
 349        .chip           = &dmar_msi_controller,
 350};
 351
 352static struct irq_domain *dmar_get_irq_domain(void)
 353{
 354        static struct irq_domain *dmar_domain;
 355        static DEFINE_MUTEX(dmar_lock);
 356        struct fwnode_handle *fn;
 357
 358        mutex_lock(&dmar_lock);
 359        if (dmar_domain)
 360                goto out;
 361
 362        fn = irq_domain_alloc_named_fwnode("DMAR-MSI");
 363        if (fn) {
 364                dmar_domain = msi_create_irq_domain(fn, &dmar_msi_domain_info,
 365                                                    x86_vector_domain);
 366                irq_domain_free_fwnode(fn);
 367        }
 368out:
 369        mutex_unlock(&dmar_lock);
 370        return dmar_domain;
 371}
 372
 373int dmar_alloc_hwirq(int id, int node, void *arg)
 374{
 375        struct irq_domain *domain = dmar_get_irq_domain();
 376        struct irq_alloc_info info;
 377
 378        if (!domain)
 379                return -1;
 380
 381        init_irq_alloc_info(&info, NULL);
 382        info.type = X86_IRQ_ALLOC_TYPE_DMAR;
 383        info.dmar_id = id;
 384        info.dmar_data = arg;
 385
 386        return irq_domain_alloc_irqs(domain, 1, node, &info);
 387}
 388
 389void dmar_free_hwirq(int irq)
 390{
 391        irq_domain_free_irqs(irq, 1);
 392}
 393#endif
 394
 395/*
 396 * MSI message composition
 397 */
 398#ifdef CONFIG_HPET_TIMER
 399static inline int hpet_dev_id(struct irq_domain *domain)
 400{
 401        struct msi_domain_info *info = msi_get_domain_info(domain);
 402
 403        return (int)(long)info->data;
 404}
 405
 406static void hpet_msi_write_msg(struct irq_data *data, struct msi_msg *msg)
 407{
 408        hpet_msi_write(irq_data_get_irq_handler_data(data), msg);
 409}
 410
 411static struct irq_chip hpet_msi_controller __ro_after_init = {
 412        .name = "HPET-MSI",
 413        .irq_unmask = hpet_msi_unmask,
 414        .irq_mask = hpet_msi_mask,
 415        .irq_ack = irq_chip_ack_parent,
 416        .irq_set_affinity = msi_domain_set_affinity,
 417        .irq_retrigger = irq_chip_retrigger_hierarchy,
 418        .irq_compose_msi_msg = irq_msi_compose_msg,
 419        .irq_write_msi_msg = hpet_msi_write_msg,
 420        .flags = IRQCHIP_SKIP_SET_WAKE,
 421};
 422
 423static irq_hw_number_t hpet_msi_get_hwirq(struct msi_domain_info *info,
 424                                          msi_alloc_info_t *arg)
 425{
 426        return arg->hpet_index;
 427}
 428
 429static int hpet_msi_init(struct irq_domain *domain,
 430                         struct msi_domain_info *info, unsigned int virq,
 431                         irq_hw_number_t hwirq, msi_alloc_info_t *arg)
 432{
 433        irq_set_status_flags(virq, IRQ_MOVE_PCNTXT);
 434        irq_domain_set_info(domain, virq, arg->hpet_index, info->chip, NULL,
 435                            handle_edge_irq, arg->hpet_data, "edge");
 436
 437        return 0;
 438}
 439
 440static void hpet_msi_free(struct irq_domain *domain,
 441                          struct msi_domain_info *info, unsigned int virq)
 442{
 443        irq_clear_status_flags(virq, IRQ_MOVE_PCNTXT);
 444}
 445
 446static struct msi_domain_ops hpet_msi_domain_ops = {
 447        .get_hwirq      = hpet_msi_get_hwirq,
 448        .msi_init       = hpet_msi_init,
 449        .msi_free       = hpet_msi_free,
 450};
 451
 452static struct msi_domain_info hpet_msi_domain_info = {
 453        .ops            = &hpet_msi_domain_ops,
 454        .chip           = &hpet_msi_controller,
 455};
 456
 457struct irq_domain *hpet_create_irq_domain(int hpet_id)
 458{
 459        struct msi_domain_info *domain_info;
 460        struct irq_domain *parent, *d;
 461        struct irq_alloc_info info;
 462        struct fwnode_handle *fn;
 463
 464        if (x86_vector_domain == NULL)
 465                return NULL;
 466
 467        domain_info = kzalloc(sizeof(*domain_info), GFP_KERNEL);
 468        if (!domain_info)
 469                return NULL;
 470
 471        *domain_info = hpet_msi_domain_info;
 472        domain_info->data = (void *)(long)hpet_id;
 473
 474        init_irq_alloc_info(&info, NULL);
 475        info.type = X86_IRQ_ALLOC_TYPE_HPET;
 476        info.hpet_id = hpet_id;
 477        parent = irq_remapping_get_ir_irq_domain(&info);
 478        if (parent == NULL)
 479                parent = x86_vector_domain;
 480        else
 481                hpet_msi_controller.name = "IR-HPET-MSI";
 482
 483        fn = irq_domain_alloc_named_id_fwnode(hpet_msi_controller.name,
 484                                              hpet_id);
 485        if (!fn) {
 486                kfree(domain_info);
 487                return NULL;
 488        }
 489
 490        d = msi_create_irq_domain(fn, domain_info, parent);
 491        irq_domain_free_fwnode(fn);
 492        return d;
 493}
 494
 495int hpet_assign_irq(struct irq_domain *domain, struct hpet_channel *hc,
 496                    int dev_num)
 497{
 498        struct irq_alloc_info info;
 499
 500        init_irq_alloc_info(&info, NULL);
 501        info.type = X86_IRQ_ALLOC_TYPE_HPET;
 502        info.hpet_data = hc;
 503        info.hpet_id = hpet_dev_id(domain);
 504        info.hpet_index = dev_num;
 505
 506        return irq_domain_alloc_irqs(domain, 1, NUMA_NO_NODE, &info);
 507}
 508#endif
 509