Dave Hansen | c221c0b | 2019-02-25 10:57:40 -0800 | [diff] [blame] | 1 | // SPDX-License-Identifier: GPL-2.0 |
| 2 | /* Copyright(c) 2016-2019 Intel Corporation. All rights reserved. */ |
| 3 | #include <linux/memremap.h> |
| 4 | #include <linux/pagemap.h> |
| 5 | #include <linux/memory.h> |
| 6 | #include <linux/module.h> |
| 7 | #include <linux/device.h> |
| 8 | #include <linux/pfn_t.h> |
| 9 | #include <linux/slab.h> |
| 10 | #include <linux/dax.h> |
| 11 | #include <linux/fs.h> |
| 12 | #include <linux/mm.h> |
| 13 | #include <linux/mman.h> |
| 14 | #include "dax-private.h" |
| 15 | #include "bus.h" |
| 16 | |
| 17 | int dev_dax_kmem_probe(struct device *dev) |
| 18 | { |
| 19 | struct dev_dax *dev_dax = to_dev_dax(dev); |
| 20 | struct resource *res = &dev_dax->region->res; |
| 21 | resource_size_t kmem_start; |
| 22 | resource_size_t kmem_size; |
| 23 | resource_size_t kmem_end; |
| 24 | struct resource *new_res; |
| 25 | int numa_node; |
| 26 | int rc; |
| 27 | |
| 28 | /* |
| 29 | * Ensure good NUMA information for the persistent memory. |
| 30 | * Without this check, there is a risk that slow memory |
| 31 | * could be mixed in a node with faster memory, causing |
| 32 | * unavoidable performance issues. |
| 33 | */ |
| 34 | numa_node = dev_dax->target_node; |
| 35 | if (numa_node < 0) { |
| 36 | dev_warn(dev, "rejecting DAX region %pR with invalid node: %d\n", |
| 37 | res, numa_node); |
| 38 | return -EINVAL; |
| 39 | } |
| 40 | |
| 41 | /* Hotplug starting at the beginning of the next block: */ |
| 42 | kmem_start = ALIGN(res->start, memory_block_size_bytes()); |
| 43 | |
| 44 | kmem_size = resource_size(res); |
| 45 | /* Adjust the size down to compensate for moving up kmem_start: */ |
| 46 | kmem_size -= kmem_start - res->start; |
| 47 | /* Align the size down to cover only complete blocks: */ |
| 48 | kmem_size &= ~(memory_block_size_bytes() - 1); |
| 49 | kmem_end = kmem_start + kmem_size; |
| 50 | |
| 51 | /* Region is permanently reserved. Hot-remove not yet implemented. */ |
| 52 | new_res = request_mem_region(kmem_start, kmem_size, dev_name(dev)); |
| 53 | if (!new_res) { |
| 54 | dev_warn(dev, "could not reserve region [%pa-%pa]\n", |
| 55 | &kmem_start, &kmem_end); |
| 56 | return -EBUSY; |
| 57 | } |
| 58 | |
| 59 | /* |
| 60 | * Set flags appropriate for System RAM. Leave ..._BUSY clear |
| 61 | * so that add_memory() can add a child resource. Do not |
| 62 | * inherit flags from the parent since it may set new flags |
| 63 | * unknown to us that will break add_memory() below. |
| 64 | */ |
| 65 | new_res->flags = IORESOURCE_SYSTEM_RAM; |
| 66 | new_res->name = dev_name(dev); |
| 67 | |
| 68 | rc = add_memory(numa_node, new_res->start, resource_size(new_res)); |
| 69 | if (rc) |
| 70 | return rc; |
| 71 | |
| 72 | return 0; |
| 73 | } |
| 74 | |
| 75 | static int dev_dax_kmem_remove(struct device *dev) |
| 76 | { |
| 77 | /* |
| 78 | * Purposely leak the request_mem_region() for the device-dax |
| 79 | * range and return '0' to ->remove() attempts. The removal of |
| 80 | * the device from the driver always succeeds, but the region |
| 81 | * is permanently pinned as reserved by the unreleased |
| 82 | * request_mem_region(). |
| 83 | */ |
| 84 | return 0; |
| 85 | } |
| 86 | |
| 87 | static struct dax_device_driver device_dax_kmem_driver = { |
| 88 | .drv = { |
| 89 | .probe = dev_dax_kmem_probe, |
| 90 | .remove = dev_dax_kmem_remove, |
| 91 | }, |
| 92 | }; |
| 93 | |
| 94 | static int __init dax_kmem_init(void) |
| 95 | { |
| 96 | return dax_driver_register(&device_dax_kmem_driver); |
| 97 | } |
| 98 | |
| 99 | static void __exit dax_kmem_exit(void) |
| 100 | { |
| 101 | dax_driver_unregister(&device_dax_kmem_driver); |
| 102 | } |
| 103 | |
| 104 | MODULE_AUTHOR("Intel Corporation"); |
| 105 | MODULE_LICENSE("GPL v2"); |
| 106 | module_init(dax_kmem_init); |
| 107 | module_exit(dax_kmem_exit); |
| 108 | MODULE_ALIAS_DAX_DEVICE(0); |