2 * Volume Management Device driver
3 * Copyright (c) 2015, Intel Corporation.
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms and conditions of the GNU General Public License,
7 * version 2, as published by the Free Software Foundation.
9 * This program is distributed in the hope it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 #include <linux/device.h>
16 #include <linux/interrupt.h>
17 #include <linux/irq.h>
18 #include <linux/kernel.h>
19 #include <linux/module.h>
20 #include <linux/msi.h>
21 #include <linux/pci.h>
22 #include <linux/rculist.h>
23 #include <linux/rcupdate.h>
25 #include <asm/irqdomain.h>
26 #include <asm/device.h>
28 #include <asm/msidef.h>
35 * Lock for manipulating VMD IRQ lists.
37 static DEFINE_RAW_SPINLOCK(list_lock);
40 * struct vmd_irq - private data to map driver IRQ to the VMD shared vector
41 * @node: list item for parent traversal.
42 * @rcu: RCU callback item for freeing.
43 * @irq: back pointer to parent.
44 * @virq: the virtual IRQ value provided to the requesting driver.
46 * Every MSI/MSI-X IRQ requested for a device in a VMD domain will be mapped to
47 * a VMD IRQ using this structure.
50 struct list_head node;
52 struct vmd_irq_list *irq;
57 * struct vmd_irq_list - list of driver requested IRQs mapping to a VMD vector
58 * @irq_list: the list of irq's the VMD one demuxes to.
59 * @count: number of child IRQs assigned to this vector; used to track
63 struct list_head irq_list;
74 struct vmd_irq_list *irqs;
76 struct pci_sysdata sysdata;
77 struct resource resources[3];
78 struct irq_domain *irq_domain;
81 #ifdef CONFIG_X86_DEV_DMA_OPS
82 struct dma_map_ops dma_ops;
83 struct dma_domain dma_domain;
87 static inline struct vmd_dev *vmd_from_bus(struct pci_bus *bus)
89 return container_of(bus->sysdata, struct vmd_dev, sysdata);
92 static inline unsigned int index_from_irqs(struct vmd_dev *vmd,
93 struct vmd_irq_list *irqs)
95 return irqs - vmd->irqs;
99 * Drivers managing a device in a VMD domain allocate their own IRQs as before,
100 * but the MSI entry for the hardware it's driving will be programmed with a
101 * destination ID for the VMD MSI-X table. The VMD muxes interrupts in its
102 * domain into one of its own, and the VMD driver de-muxes these for the
103 * handlers sharing that VMD IRQ. The vmd irq_domain provides the operations
104 * and irq_chip to set this up.
106 static void vmd_compose_msi_msg(struct irq_data *data, struct msi_msg *msg)
108 struct vmd_irq *vmdirq = data->chip_data;
109 struct vmd_irq_list *irq = vmdirq->irq;
110 struct vmd_dev *vmd = irq_data_get_irq_handler_data(data);
112 msg->address_hi = MSI_ADDR_BASE_HI;
113 msg->address_lo = MSI_ADDR_BASE_LO |
114 MSI_ADDR_DEST_ID(index_from_irqs(vmd, irq));
119 * We rely on MSI_FLAG_USE_DEF_CHIP_OPS to set the IRQ mask/unmask ops.
121 static void vmd_irq_enable(struct irq_data *data)
123 struct vmd_irq *vmdirq = data->chip_data;
126 raw_spin_lock_irqsave(&list_lock, flags);
127 list_add_tail_rcu(&vmdirq->node, &vmdirq->irq->irq_list);
128 raw_spin_unlock_irqrestore(&list_lock, flags);
130 data->chip->irq_unmask(data);
133 static void vmd_irq_disable(struct irq_data *data)
135 struct vmd_irq *vmdirq = data->chip_data;
138 data->chip->irq_mask(data);
140 raw_spin_lock_irqsave(&list_lock, flags);
141 list_del_rcu(&vmdirq->node);
142 INIT_LIST_HEAD_RCU(&vmdirq->node);
143 raw_spin_unlock_irqrestore(&list_lock, flags);
147 * XXX: Stubbed until we develop acceptable way to not create conflicts with
148 * other devices sharing the same vector.
150 static int vmd_irq_set_affinity(struct irq_data *data,
151 const struct cpumask *dest, bool force)
156 static struct irq_chip vmd_msi_controller = {
158 .irq_enable = vmd_irq_enable,
159 .irq_disable = vmd_irq_disable,
160 .irq_compose_msi_msg = vmd_compose_msi_msg,
161 .irq_set_affinity = vmd_irq_set_affinity,
164 static irq_hw_number_t vmd_get_hwirq(struct msi_domain_info *info,
165 msi_alloc_info_t *arg)
171 * XXX: We can be even smarter selecting the best IRQ once we solve the
174 static struct vmd_irq_list *vmd_next_irq(struct vmd_dev *vmd, struct msi_desc *desc)
179 if (!desc->msi_attrib.is_msix || vmd->msix_count == 1)
180 return &vmd->irqs[0];
182 raw_spin_lock_irqsave(&list_lock, flags);
183 for (i = 1; i < vmd->msix_count; i++)
184 if (vmd->irqs[i].count < vmd->irqs[best].count)
186 vmd->irqs[best].count++;
187 raw_spin_unlock_irqrestore(&list_lock, flags);
189 return &vmd->irqs[best];
192 static int vmd_msi_init(struct irq_domain *domain, struct msi_domain_info *info,
193 unsigned int virq, irq_hw_number_t hwirq,
194 msi_alloc_info_t *arg)
196 struct msi_desc *desc = arg->desc;
197 struct vmd_dev *vmd = vmd_from_bus(msi_desc_to_pci_dev(desc)->bus);
198 struct vmd_irq *vmdirq = kzalloc(sizeof(*vmdirq), GFP_KERNEL);
199 unsigned int index, vector;
204 INIT_LIST_HEAD(&vmdirq->node);
205 vmdirq->irq = vmd_next_irq(vmd, desc);
207 index = index_from_irqs(vmd, vmdirq->irq);
208 vector = pci_irq_vector(vmd->dev, index);
210 irq_domain_set_info(domain, virq, vector, info->chip, vmdirq,
211 handle_untracked_irq, vmd, NULL);
215 static void vmd_msi_free(struct irq_domain *domain,
216 struct msi_domain_info *info, unsigned int virq)
218 struct vmd_irq *vmdirq = irq_get_chip_data(virq);
223 /* XXX: Potential optimization to rebalance */
224 raw_spin_lock_irqsave(&list_lock, flags);
225 vmdirq->irq->count--;
226 raw_spin_unlock_irqrestore(&list_lock, flags);
228 kfree_rcu(vmdirq, rcu);
231 static int vmd_msi_prepare(struct irq_domain *domain, struct device *dev,
232 int nvec, msi_alloc_info_t *arg)
234 struct pci_dev *pdev = to_pci_dev(dev);
235 struct vmd_dev *vmd = vmd_from_bus(pdev->bus);
237 if (nvec > vmd->msix_count)
238 return vmd->msix_count;
240 memset(arg, 0, sizeof(*arg));
244 static void vmd_set_desc(msi_alloc_info_t *arg, struct msi_desc *desc)
249 static struct msi_domain_ops vmd_msi_domain_ops = {
250 .get_hwirq = vmd_get_hwirq,
251 .msi_init = vmd_msi_init,
252 .msi_free = vmd_msi_free,
253 .msi_prepare = vmd_msi_prepare,
254 .set_desc = vmd_set_desc,
257 static struct msi_domain_info vmd_msi_domain_info = {
258 .flags = MSI_FLAG_USE_DEF_DOM_OPS | MSI_FLAG_USE_DEF_CHIP_OPS |
260 .ops = &vmd_msi_domain_ops,
261 .chip = &vmd_msi_controller,
264 #ifdef CONFIG_X86_DEV_DMA_OPS
266 * VMD replaces the requester ID with its own. DMA mappings for devices in a
267 * VMD domain need to be mapped for the VMD, not the device requiring
270 static struct device *to_vmd_dev(struct device *dev)
272 struct pci_dev *pdev = to_pci_dev(dev);
273 struct vmd_dev *vmd = vmd_from_bus(pdev->bus);
275 return &vmd->dev->dev;
278 static struct dma_map_ops *vmd_dma_ops(struct device *dev)
280 return get_dma_ops(to_vmd_dev(dev));
283 static void *vmd_alloc(struct device *dev, size_t size, dma_addr_t *addr,
284 gfp_t flag, unsigned long attrs)
286 return vmd_dma_ops(dev)->alloc(to_vmd_dev(dev), size, addr, flag,
290 static void vmd_free(struct device *dev, size_t size, void *vaddr,
291 dma_addr_t addr, unsigned long attrs)
293 return vmd_dma_ops(dev)->free(to_vmd_dev(dev), size, vaddr, addr,
297 static int vmd_mmap(struct device *dev, struct vm_area_struct *vma,
298 void *cpu_addr, dma_addr_t addr, size_t size,
301 return vmd_dma_ops(dev)->mmap(to_vmd_dev(dev), vma, cpu_addr, addr,
305 static int vmd_get_sgtable(struct device *dev, struct sg_table *sgt,
306 void *cpu_addr, dma_addr_t addr, size_t size,
309 return vmd_dma_ops(dev)->get_sgtable(to_vmd_dev(dev), sgt, cpu_addr,
313 static dma_addr_t vmd_map_page(struct device *dev, struct page *page,
314 unsigned long offset, size_t size,
315 enum dma_data_direction dir,
318 return vmd_dma_ops(dev)->map_page(to_vmd_dev(dev), page, offset, size,
322 static void vmd_unmap_page(struct device *dev, dma_addr_t addr, size_t size,
323 enum dma_data_direction dir, unsigned long attrs)
325 vmd_dma_ops(dev)->unmap_page(to_vmd_dev(dev), addr, size, dir, attrs);
328 static int vmd_map_sg(struct device *dev, struct scatterlist *sg, int nents,
329 enum dma_data_direction dir, unsigned long attrs)
331 return vmd_dma_ops(dev)->map_sg(to_vmd_dev(dev), sg, nents, dir, attrs);
334 static void vmd_unmap_sg(struct device *dev, struct scatterlist *sg, int nents,
335 enum dma_data_direction dir, unsigned long attrs)
337 vmd_dma_ops(dev)->unmap_sg(to_vmd_dev(dev), sg, nents, dir, attrs);
340 static void vmd_sync_single_for_cpu(struct device *dev, dma_addr_t addr,
341 size_t size, enum dma_data_direction dir)
343 vmd_dma_ops(dev)->sync_single_for_cpu(to_vmd_dev(dev), addr, size, dir);
346 static void vmd_sync_single_for_device(struct device *dev, dma_addr_t addr,
347 size_t size, enum dma_data_direction dir)
349 vmd_dma_ops(dev)->sync_single_for_device(to_vmd_dev(dev), addr, size,
353 static void vmd_sync_sg_for_cpu(struct device *dev, struct scatterlist *sg,
354 int nents, enum dma_data_direction dir)
356 vmd_dma_ops(dev)->sync_sg_for_cpu(to_vmd_dev(dev), sg, nents, dir);
359 static void vmd_sync_sg_for_device(struct device *dev, struct scatterlist *sg,
360 int nents, enum dma_data_direction dir)
362 vmd_dma_ops(dev)->sync_sg_for_device(to_vmd_dev(dev), sg, nents, dir);
365 static int vmd_mapping_error(struct device *dev, dma_addr_t addr)
367 return vmd_dma_ops(dev)->mapping_error(to_vmd_dev(dev), addr);
370 static int vmd_dma_supported(struct device *dev, u64 mask)
372 return vmd_dma_ops(dev)->dma_supported(to_vmd_dev(dev), mask);
375 #ifdef ARCH_HAS_DMA_GET_REQUIRED_MASK
376 static u64 vmd_get_required_mask(struct device *dev)
378 return vmd_dma_ops(dev)->get_required_mask(to_vmd_dev(dev));
382 static void vmd_teardown_dma_ops(struct vmd_dev *vmd)
384 struct dma_domain *domain = &vmd->dma_domain;
386 if (get_dma_ops(&vmd->dev->dev))
387 del_dma_domain(domain);
390 #define ASSIGN_VMD_DMA_OPS(source, dest, fn) \
393 dest->fn = vmd_##fn; \
396 static void vmd_setup_dma_ops(struct vmd_dev *vmd)
398 const struct dma_map_ops *source = get_dma_ops(&vmd->dev->dev);
399 struct dma_map_ops *dest = &vmd->dma_ops;
400 struct dma_domain *domain = &vmd->dma_domain;
402 domain->domain_nr = vmd->sysdata.domain;
403 domain->dma_ops = dest;
407 ASSIGN_VMD_DMA_OPS(source, dest, alloc);
408 ASSIGN_VMD_DMA_OPS(source, dest, free);
409 ASSIGN_VMD_DMA_OPS(source, dest, mmap);
410 ASSIGN_VMD_DMA_OPS(source, dest, get_sgtable);
411 ASSIGN_VMD_DMA_OPS(source, dest, map_page);
412 ASSIGN_VMD_DMA_OPS(source, dest, unmap_page);
413 ASSIGN_VMD_DMA_OPS(source, dest, map_sg);
414 ASSIGN_VMD_DMA_OPS(source, dest, unmap_sg);
415 ASSIGN_VMD_DMA_OPS(source, dest, sync_single_for_cpu);
416 ASSIGN_VMD_DMA_OPS(source, dest, sync_single_for_device);
417 ASSIGN_VMD_DMA_OPS(source, dest, sync_sg_for_cpu);
418 ASSIGN_VMD_DMA_OPS(source, dest, sync_sg_for_device);
419 ASSIGN_VMD_DMA_OPS(source, dest, mapping_error);
420 ASSIGN_VMD_DMA_OPS(source, dest, dma_supported);
421 #ifdef ARCH_HAS_DMA_GET_REQUIRED_MASK
422 ASSIGN_VMD_DMA_OPS(source, dest, get_required_mask);
424 add_dma_domain(domain);
426 #undef ASSIGN_VMD_DMA_OPS
428 static void vmd_teardown_dma_ops(struct vmd_dev *vmd) {}
429 static void vmd_setup_dma_ops(struct vmd_dev *vmd) {}
432 static char __iomem *vmd_cfg_addr(struct vmd_dev *vmd, struct pci_bus *bus,
433 unsigned int devfn, int reg, int len)
435 char __iomem *addr = vmd->cfgbar +
436 (bus->number << 20) + (devfn << 12) + reg;
438 if ((addr - vmd->cfgbar) + len >=
439 resource_size(&vmd->dev->resource[VMD_CFGBAR]))
446 * CPU may deadlock if config space is not serialized on some versions of this
447 * hardware, so all config space access is done under a spinlock.
449 static int vmd_pci_read(struct pci_bus *bus, unsigned int devfn, int reg,
452 struct vmd_dev *vmd = vmd_from_bus(bus);
453 char __iomem *addr = vmd_cfg_addr(vmd, bus, devfn, reg, len);
460 spin_lock_irqsave(&vmd->cfg_lock, flags);
463 *value = readb(addr);
466 *value = readw(addr);
469 *value = readl(addr);
475 spin_unlock_irqrestore(&vmd->cfg_lock, flags);
480 * VMD h/w converts non-posted config writes to posted memory writes. The
481 * read-back in this function forces the completion so it returns only after
482 * the config space was written, as expected.
484 static int vmd_pci_write(struct pci_bus *bus, unsigned int devfn, int reg,
487 struct vmd_dev *vmd = vmd_from_bus(bus);
488 char __iomem *addr = vmd_cfg_addr(vmd, bus, devfn, reg, len);
495 spin_lock_irqsave(&vmd->cfg_lock, flags);
513 spin_unlock_irqrestore(&vmd->cfg_lock, flags);
517 static struct pci_ops vmd_ops = {
518 .read = vmd_pci_read,
519 .write = vmd_pci_write,
522 static void vmd_attach_resources(struct vmd_dev *vmd)
524 vmd->dev->resource[VMD_MEMBAR1].child = &vmd->resources[1];
525 vmd->dev->resource[VMD_MEMBAR2].child = &vmd->resources[2];
528 static void vmd_detach_resources(struct vmd_dev *vmd)
530 vmd->dev->resource[VMD_MEMBAR1].child = NULL;
531 vmd->dev->resource[VMD_MEMBAR2].child = NULL;
535 * VMD domains start at 0x1000 to not clash with ACPI _SEG domains.
537 static int vmd_find_free_domain(void)
540 struct pci_bus *bus = NULL;
542 while ((bus = pci_find_next_bus(bus)) != NULL)
543 domain = max_t(int, domain, pci_domain_nr(bus));
547 static int vmd_enable_domain(struct vmd_dev *vmd)
549 struct pci_sysdata *sd = &vmd->sysdata;
550 struct resource *res;
553 LIST_HEAD(resources);
555 res = &vmd->dev->resource[VMD_CFGBAR];
556 vmd->resources[0] = (struct resource) {
557 .name = "VMD CFGBAR",
559 .end = (resource_size(res) >> 20) - 1,
560 .flags = IORESOURCE_BUS | IORESOURCE_PCI_FIXED,
564 * If the window is below 4GB, clear IORESOURCE_MEM_64 so we can
565 * put 32-bit resources in the window.
567 * There's no hardware reason why a 64-bit window *couldn't*
568 * contain a 32-bit resource, but pbus_size_mem() computes the
569 * bridge window size assuming a 64-bit window will contain no
570 * 32-bit resources. __pci_assign_resource() enforces that
571 * artificial restriction to make sure everything will fit.
573 * The only way we could use a 64-bit non-prefechable MEMBAR is
574 * if its address is <4GB so that we can convert it to a 32-bit
575 * resource. To be visible to the host OS, all VMD endpoints must
576 * be initially configured by platform BIOS, which includes setting
577 * up these resources. We can assume the device is configured
578 * according to the platform needs.
580 res = &vmd->dev->resource[VMD_MEMBAR1];
581 upper_bits = upper_32_bits(res->end);
582 flags = res->flags & ~IORESOURCE_SIZEALIGN;
584 flags &= ~IORESOURCE_MEM_64;
585 vmd->resources[1] = (struct resource) {
586 .name = "VMD MEMBAR1",
593 res = &vmd->dev->resource[VMD_MEMBAR2];
594 upper_bits = upper_32_bits(res->end);
595 flags = res->flags & ~IORESOURCE_SIZEALIGN;
597 flags &= ~IORESOURCE_MEM_64;
598 vmd->resources[2] = (struct resource) {
599 .name = "VMD MEMBAR2",
600 .start = res->start + 0x2000,
606 sd->domain = vmd_find_free_domain();
610 sd->node = pcibus_to_node(vmd->dev->bus);
612 vmd->irq_domain = pci_msi_create_irq_domain(NULL, &vmd_msi_domain_info,
614 if (!vmd->irq_domain)
617 pci_add_resource(&resources, &vmd->resources[0]);
618 pci_add_resource(&resources, &vmd->resources[1]);
619 pci_add_resource(&resources, &vmd->resources[2]);
620 vmd->bus = pci_create_root_bus(&vmd->dev->dev, 0, &vmd_ops, sd,
623 pci_free_resource_list(&resources);
624 irq_domain_remove(vmd->irq_domain);
628 vmd_attach_resources(vmd);
629 vmd_setup_dma_ops(vmd);
630 dev_set_msi_domain(&vmd->bus->dev, vmd->irq_domain);
631 pci_rescan_bus(vmd->bus);
633 WARN(sysfs_create_link(&vmd->dev->dev.kobj, &vmd->bus->dev.kobj,
634 "domain"), "Can't create symlink to domain\n");
638 static irqreturn_t vmd_irq(int irq, void *data)
640 struct vmd_irq_list *irqs = data;
641 struct vmd_irq *vmdirq;
644 list_for_each_entry_rcu(vmdirq, &irqs->irq_list, node)
645 generic_handle_irq(vmdirq->virq);
651 static int vmd_probe(struct pci_dev *dev, const struct pci_device_id *id)
656 if (resource_size(&dev->resource[VMD_CFGBAR]) < (1 << 20))
659 vmd = devm_kzalloc(&dev->dev, sizeof(*vmd), GFP_KERNEL);
664 err = pcim_enable_device(dev);
668 vmd->cfgbar = pcim_iomap(dev, VMD_CFGBAR, 0);
673 if (dma_set_mask_and_coherent(&dev->dev, DMA_BIT_MASK(64)) &&
674 dma_set_mask_and_coherent(&dev->dev, DMA_BIT_MASK(32)))
677 vmd->msix_count = pci_msix_vec_count(dev);
678 if (vmd->msix_count < 0)
681 vmd->msix_count = pci_alloc_irq_vectors(dev, 1, vmd->msix_count,
682 PCI_IRQ_MSIX | PCI_IRQ_AFFINITY);
683 if (vmd->msix_count < 0)
684 return vmd->msix_count;
686 vmd->irqs = devm_kcalloc(&dev->dev, vmd->msix_count, sizeof(*vmd->irqs),
691 for (i = 0; i < vmd->msix_count; i++) {
692 INIT_LIST_HEAD(&vmd->irqs[i].irq_list);
693 err = devm_request_irq(&dev->dev, pci_irq_vector(dev, i),
694 vmd_irq, 0, "vmd", &vmd->irqs[i]);
699 spin_lock_init(&vmd->cfg_lock);
700 pci_set_drvdata(dev, vmd);
701 err = vmd_enable_domain(vmd);
705 dev_info(&vmd->dev->dev, "Bound to PCI domain %04x\n",
706 vmd->sysdata.domain);
710 static void vmd_remove(struct pci_dev *dev)
712 struct vmd_dev *vmd = pci_get_drvdata(dev);
714 vmd_detach_resources(vmd);
715 pci_set_drvdata(dev, NULL);
716 sysfs_remove_link(&vmd->dev->dev.kobj, "domain");
717 pci_stop_root_bus(vmd->bus);
718 pci_remove_root_bus(vmd->bus);
719 vmd_teardown_dma_ops(vmd);
720 irq_domain_remove(vmd->irq_domain);
724 static int vmd_suspend(struct device *dev)
726 struct pci_dev *pdev = to_pci_dev(dev);
728 pci_save_state(pdev);
732 static int vmd_resume(struct device *dev)
734 struct pci_dev *pdev = to_pci_dev(dev);
736 pci_restore_state(pdev);
740 static SIMPLE_DEV_PM_OPS(vmd_dev_pm_ops, vmd_suspend, vmd_resume);
742 static const struct pci_device_id vmd_ids[] = {
743 {PCI_DEVICE(PCI_VENDOR_ID_INTEL, 0x201d),},
746 MODULE_DEVICE_TABLE(pci, vmd_ids);
748 static struct pci_driver vmd_drv = {
752 .remove = vmd_remove,
754 .pm = &vmd_dev_pm_ops,
757 module_pci_driver(vmd_drv);
759 MODULE_AUTHOR("Intel Corporation");
760 MODULE_LICENSE("GPL v2");
761 MODULE_VERSION("0.6");