| 1 | // SPDX-License-Identifier: GPL-2.0 |
| 2 | /* |
| 3 | * Copyright (C) 2022-2024, Advanced Micro Devices, Inc. |
| 4 | */ |
| 5 | |
| 6 | #include <drm/amdxdna_accel.h> |
| 7 | #include <drm/drm_accel.h> |
| 8 | #include <drm/drm_drv.h> |
| 9 | #include <drm/drm_gem.h> |
| 10 | #include <drm/drm_gem_shmem_helper.h> |
| 11 | #include <drm/drm_ioctl.h> |
| 12 | #include <drm/drm_managed.h> |
| 13 | #include <drm/gpu_scheduler.h> |
| 14 | #include <linux/iommu.h> |
| 15 | #include <linux/pci.h> |
| 16 | |
| 17 | #include "amdxdna_ctx.h" |
| 18 | #include "amdxdna_gem.h" |
| 19 | #include "amdxdna_pci_drv.h" |
| 20 | #include "amdxdna_pm.h" |
| 21 | |
| 22 | MODULE_FIRMWARE("amdnpu/1502_00/npu.sbin" ); |
| 23 | MODULE_FIRMWARE("amdnpu/17f0_10/npu.sbin" ); |
| 24 | MODULE_FIRMWARE("amdnpu/17f0_11/npu.sbin" ); |
| 25 | MODULE_FIRMWARE("amdnpu/17f0_20/npu.sbin" ); |
| 26 | |
| 27 | /* |
| 28 | * 0.0: Initial version |
| 29 | * 0.1: Support getting all hardware contexts by DRM_IOCTL_AMDXDNA_GET_ARRAY |
| 30 | * 0.2: Support getting last error hardware error |
| 31 | * 0.3: Support firmware debug buffer |
| 32 | * 0.4: Support getting resource information |
| 33 | * 0.5: Support getting telemetry data |
| 34 | * 0.6: Support preemption |
| 35 | */ |
| 36 | #define AMDXDNA_DRIVER_MAJOR 0 |
| 37 | #define AMDXDNA_DRIVER_MINOR 6 |
| 38 | |
| 39 | /* |
| 40 | * Bind the driver base on (vendor_id, device_id) pair and later use the |
| 41 | * (device_id, rev_id) pair as a key to select the devices. The devices with |
| 42 | * same device_id have very similar interface to host driver. |
| 43 | */ |
| 44 | static const struct pci_device_id pci_ids[] = { |
| 45 | { PCI_DEVICE(PCI_VENDOR_ID_AMD, 0x1502) }, |
| 46 | { PCI_DEVICE(PCI_VENDOR_ID_AMD, 0x17f0) }, |
| 47 | {0} |
| 48 | }; |
| 49 | |
| 50 | MODULE_DEVICE_TABLE(pci, pci_ids); |
| 51 | |
| 52 | static const struct amdxdna_device_id amdxdna_ids[] = { |
| 53 | { 0x1502, 0x0, &dev_npu1_info }, |
| 54 | { 0x17f0, 0x0, &dev_npu2_info }, |
| 55 | { 0x17f0, 0x10, &dev_npu4_info }, |
| 56 | { 0x17f0, 0x11, &dev_npu5_info }, |
| 57 | { 0x17f0, 0x20, &dev_npu6_info }, |
| 58 | {0} |
| 59 | }; |
| 60 | |
| 61 | static int amdxdna_drm_open(struct drm_device *ddev, struct drm_file *filp) |
| 62 | { |
| 63 | struct amdxdna_dev *xdna = to_xdna_dev(ddev); |
| 64 | struct amdxdna_client *client; |
| 65 | int ret; |
| 66 | |
| 67 | client = kzalloc(sizeof(*client), GFP_KERNEL); |
| 68 | if (!client) |
| 69 | return -ENOMEM; |
| 70 | |
| 71 | client->pid = pid_nr(rcu_access_pointer(filp->pid)); |
| 72 | client->xdna = xdna; |
| 73 | |
| 74 | client->sva = iommu_sva_bind_device(dev: xdna->ddev.dev, current->mm); |
| 75 | if (IS_ERR(ptr: client->sva)) { |
| 76 | ret = PTR_ERR(ptr: client->sva); |
| 77 | XDNA_ERR(xdna, "SVA bind device failed, ret %d" , ret); |
| 78 | goto failed; |
| 79 | } |
| 80 | client->pasid = iommu_sva_get_pasid(handle: client->sva); |
| 81 | if (client->pasid == IOMMU_PASID_INVALID) { |
| 82 | XDNA_ERR(xdna, "SVA get pasid failed" ); |
| 83 | ret = -ENODEV; |
| 84 | goto unbind_sva; |
| 85 | } |
| 86 | init_srcu_struct(&client->hwctx_srcu); |
| 87 | xa_init_flags(xa: &client->hwctx_xa, XA_FLAGS_ALLOC); |
| 88 | mutex_init(&client->mm_lock); |
| 89 | |
| 90 | mutex_lock(&xdna->dev_lock); |
| 91 | list_add_tail(new: &client->node, head: &xdna->client_list); |
| 92 | mutex_unlock(lock: &xdna->dev_lock); |
| 93 | |
| 94 | filp->driver_priv = client; |
| 95 | client->filp = filp; |
| 96 | |
| 97 | XDNA_DBG(xdna, "pid %d opened" , client->pid); |
| 98 | return 0; |
| 99 | |
| 100 | unbind_sva: |
| 101 | iommu_sva_unbind_device(handle: client->sva); |
| 102 | failed: |
| 103 | kfree(objp: client); |
| 104 | |
| 105 | return ret; |
| 106 | } |
| 107 | |
| 108 | static void amdxdna_drm_close(struct drm_device *ddev, struct drm_file *filp) |
| 109 | { |
| 110 | struct amdxdna_client *client = filp->driver_priv; |
| 111 | struct amdxdna_dev *xdna = to_xdna_dev(ddev); |
| 112 | |
| 113 | XDNA_DBG(xdna, "closing pid %d" , client->pid); |
| 114 | |
| 115 | xa_destroy(&client->hwctx_xa); |
| 116 | cleanup_srcu_struct(ssp: &client->hwctx_srcu); |
| 117 | mutex_destroy(lock: &client->mm_lock); |
| 118 | if (client->dev_heap) |
| 119 | drm_gem_object_put(to_gobj(client->dev_heap)); |
| 120 | |
| 121 | iommu_sva_unbind_device(handle: client->sva); |
| 122 | |
| 123 | XDNA_DBG(xdna, "pid %d closed" , client->pid); |
| 124 | kfree(objp: client); |
| 125 | } |
| 126 | |
| 127 | static int amdxdna_flush(struct file *f, fl_owner_t id) |
| 128 | { |
| 129 | struct drm_file *filp = f->private_data; |
| 130 | struct amdxdna_client *client = filp->driver_priv; |
| 131 | struct amdxdna_dev *xdna = client->xdna; |
| 132 | int idx; |
| 133 | |
| 134 | XDNA_DBG(xdna, "PID %d flushing..." , client->pid); |
| 135 | if (!drm_dev_enter(dev: &xdna->ddev, idx: &idx)) |
| 136 | return 0; |
| 137 | |
| 138 | mutex_lock(&xdna->dev_lock); |
| 139 | list_del_init(entry: &client->node); |
| 140 | amdxdna_hwctx_remove_all(client); |
| 141 | mutex_unlock(lock: &xdna->dev_lock); |
| 142 | |
| 143 | drm_dev_exit(idx); |
| 144 | return 0; |
| 145 | } |
| 146 | |
| 147 | static int amdxdna_drm_get_info_ioctl(struct drm_device *dev, void *data, struct drm_file *filp) |
| 148 | { |
| 149 | struct amdxdna_client *client = filp->driver_priv; |
| 150 | struct amdxdna_dev *xdna = to_xdna_dev(dev); |
| 151 | struct amdxdna_drm_get_info *args = data; |
| 152 | int ret; |
| 153 | |
| 154 | if (!xdna->dev_info->ops->get_aie_info) |
| 155 | return -EOPNOTSUPP; |
| 156 | |
| 157 | XDNA_DBG(xdna, "Request parameter %u" , args->param); |
| 158 | mutex_lock(&xdna->dev_lock); |
| 159 | ret = xdna->dev_info->ops->get_aie_info(client, args); |
| 160 | mutex_unlock(lock: &xdna->dev_lock); |
| 161 | return ret; |
| 162 | } |
| 163 | |
| 164 | static int amdxdna_drm_get_array_ioctl(struct drm_device *dev, void *data, |
| 165 | struct drm_file *filp) |
| 166 | { |
| 167 | struct amdxdna_client *client = filp->driver_priv; |
| 168 | struct amdxdna_dev *xdna = to_xdna_dev(dev); |
| 169 | struct amdxdna_drm_get_array *args = data; |
| 170 | |
| 171 | if (!xdna->dev_info->ops->get_array) |
| 172 | return -EOPNOTSUPP; |
| 173 | |
| 174 | if (args->pad || !args->num_element || !args->element_size) |
| 175 | return -EINVAL; |
| 176 | |
| 177 | guard(mutex)(T: &xdna->dev_lock); |
| 178 | return xdna->dev_info->ops->get_array(client, args); |
| 179 | } |
| 180 | |
| 181 | static int amdxdna_drm_set_state_ioctl(struct drm_device *dev, void *data, struct drm_file *filp) |
| 182 | { |
| 183 | struct amdxdna_client *client = filp->driver_priv; |
| 184 | struct amdxdna_dev *xdna = to_xdna_dev(dev); |
| 185 | struct amdxdna_drm_set_state *args = data; |
| 186 | int ret; |
| 187 | |
| 188 | if (!xdna->dev_info->ops->set_aie_state) |
| 189 | return -EOPNOTSUPP; |
| 190 | |
| 191 | XDNA_DBG(xdna, "Request parameter %u" , args->param); |
| 192 | mutex_lock(&xdna->dev_lock); |
| 193 | ret = xdna->dev_info->ops->set_aie_state(client, args); |
| 194 | mutex_unlock(lock: &xdna->dev_lock); |
| 195 | |
| 196 | return ret; |
| 197 | } |
| 198 | |
| 199 | static const struct drm_ioctl_desc amdxdna_drm_ioctls[] = { |
| 200 | /* Context */ |
| 201 | DRM_IOCTL_DEF_DRV(AMDXDNA_CREATE_HWCTX, amdxdna_drm_create_hwctx_ioctl, 0), |
| 202 | DRM_IOCTL_DEF_DRV(AMDXDNA_DESTROY_HWCTX, amdxdna_drm_destroy_hwctx_ioctl, 0), |
| 203 | DRM_IOCTL_DEF_DRV(AMDXDNA_CONFIG_HWCTX, amdxdna_drm_config_hwctx_ioctl, 0), |
| 204 | /* BO */ |
| 205 | DRM_IOCTL_DEF_DRV(AMDXDNA_CREATE_BO, amdxdna_drm_create_bo_ioctl, 0), |
| 206 | DRM_IOCTL_DEF_DRV(AMDXDNA_GET_BO_INFO, amdxdna_drm_get_bo_info_ioctl, 0), |
| 207 | DRM_IOCTL_DEF_DRV(AMDXDNA_SYNC_BO, amdxdna_drm_sync_bo_ioctl, 0), |
| 208 | /* Execution */ |
| 209 | DRM_IOCTL_DEF_DRV(AMDXDNA_EXEC_CMD, amdxdna_drm_submit_cmd_ioctl, 0), |
| 210 | /* AIE hardware */ |
| 211 | DRM_IOCTL_DEF_DRV(AMDXDNA_GET_INFO, amdxdna_drm_get_info_ioctl, 0), |
| 212 | DRM_IOCTL_DEF_DRV(AMDXDNA_GET_ARRAY, amdxdna_drm_get_array_ioctl, 0), |
| 213 | DRM_IOCTL_DEF_DRV(AMDXDNA_SET_STATE, amdxdna_drm_set_state_ioctl, DRM_ROOT_ONLY), |
| 214 | }; |
| 215 | |
| 216 | static const struct file_operations amdxdna_fops = { |
| 217 | .owner = THIS_MODULE, |
| 218 | .open = accel_open, |
| 219 | .release = drm_release, |
| 220 | .flush = amdxdna_flush, |
| 221 | .unlocked_ioctl = drm_ioctl, |
| 222 | .compat_ioctl = drm_compat_ioctl, |
| 223 | .poll = drm_poll, |
| 224 | .read = drm_read, |
| 225 | .llseek = noop_llseek, |
| 226 | .mmap = drm_gem_mmap, |
| 227 | .fop_flags = FOP_UNSIGNED_OFFSET, |
| 228 | }; |
| 229 | |
| 230 | const struct drm_driver amdxdna_drm_drv = { |
| 231 | .driver_features = DRIVER_GEM | DRIVER_COMPUTE_ACCEL | |
| 232 | DRIVER_SYNCOBJ | DRIVER_SYNCOBJ_TIMELINE, |
| 233 | .fops = &amdxdna_fops, |
| 234 | .name = "amdxdna_accel_driver" , |
| 235 | .desc = "AMD XDNA DRM implementation" , |
| 236 | .major = AMDXDNA_DRIVER_MAJOR, |
| 237 | .minor = AMDXDNA_DRIVER_MINOR, |
| 238 | .open = amdxdna_drm_open, |
| 239 | .postclose = amdxdna_drm_close, |
| 240 | .ioctls = amdxdna_drm_ioctls, |
| 241 | .num_ioctls = ARRAY_SIZE(amdxdna_drm_ioctls), |
| 242 | |
| 243 | .gem_create_object = amdxdna_gem_create_object_cb, |
| 244 | .gem_prime_import = amdxdna_gem_prime_import, |
| 245 | }; |
| 246 | |
| 247 | static const struct amdxdna_dev_info * |
| 248 | amdxdna_get_dev_info(struct pci_dev *pdev) |
| 249 | { |
| 250 | int i; |
| 251 | |
| 252 | for (i = 0; i < ARRAY_SIZE(amdxdna_ids); i++) { |
| 253 | if (pdev->device == amdxdna_ids[i].device && |
| 254 | pdev->revision == amdxdna_ids[i].revision) |
| 255 | return amdxdna_ids[i].dev_info; |
| 256 | } |
| 257 | return NULL; |
| 258 | } |
| 259 | |
| 260 | static int amdxdna_probe(struct pci_dev *pdev, const struct pci_device_id *id) |
| 261 | { |
| 262 | struct device *dev = &pdev->dev; |
| 263 | struct amdxdna_dev *xdna; |
| 264 | int ret; |
| 265 | |
| 266 | xdna = devm_drm_dev_alloc(dev, &amdxdna_drm_drv, typeof(*xdna), ddev); |
| 267 | if (IS_ERR(ptr: xdna)) |
| 268 | return PTR_ERR(ptr: xdna); |
| 269 | |
| 270 | xdna->dev_info = amdxdna_get_dev_info(pdev); |
| 271 | if (!xdna->dev_info) |
| 272 | return -ENODEV; |
| 273 | |
| 274 | drmm_mutex_init(&xdna->ddev, &xdna->dev_lock); |
| 275 | init_rwsem(&xdna->notifier_lock); |
| 276 | INIT_LIST_HEAD(list: &xdna->client_list); |
| 277 | pci_set_drvdata(pdev, data: xdna); |
| 278 | |
| 279 | if (IS_ENABLED(CONFIG_LOCKDEP)) { |
| 280 | fs_reclaim_acquire(GFP_KERNEL); |
| 281 | might_lock(&xdna->notifier_lock); |
| 282 | fs_reclaim_release(GFP_KERNEL); |
| 283 | } |
| 284 | |
| 285 | xdna->notifier_wq = alloc_ordered_workqueue("notifier_wq" , 0); |
| 286 | if (!xdna->notifier_wq) |
| 287 | return -ENOMEM; |
| 288 | |
| 289 | mutex_lock(&xdna->dev_lock); |
| 290 | ret = xdna->dev_info->ops->init(xdna); |
| 291 | mutex_unlock(lock: &xdna->dev_lock); |
| 292 | if (ret) { |
| 293 | XDNA_ERR(xdna, "Hardware init failed, ret %d" , ret); |
| 294 | goto destroy_notifier_wq; |
| 295 | } |
| 296 | |
| 297 | ret = amdxdna_sysfs_init(xdna); |
| 298 | if (ret) { |
| 299 | XDNA_ERR(xdna, "Create amdxdna attrs failed: %d" , ret); |
| 300 | goto failed_dev_fini; |
| 301 | } |
| 302 | |
| 303 | ret = drm_dev_register(dev: &xdna->ddev, flags: 0); |
| 304 | if (ret) { |
| 305 | XDNA_ERR(xdna, "DRM register failed, ret %d" , ret); |
| 306 | goto failed_sysfs_fini; |
| 307 | } |
| 308 | |
| 309 | return 0; |
| 310 | |
| 311 | failed_sysfs_fini: |
| 312 | amdxdna_sysfs_fini(xdna); |
| 313 | failed_dev_fini: |
| 314 | mutex_lock(&xdna->dev_lock); |
| 315 | xdna->dev_info->ops->fini(xdna); |
| 316 | mutex_unlock(lock: &xdna->dev_lock); |
| 317 | destroy_notifier_wq: |
| 318 | destroy_workqueue(wq: xdna->notifier_wq); |
| 319 | return ret; |
| 320 | } |
| 321 | |
| 322 | static void amdxdna_remove(struct pci_dev *pdev) |
| 323 | { |
| 324 | struct amdxdna_dev *xdna = pci_get_drvdata(pdev); |
| 325 | struct amdxdna_client *client; |
| 326 | |
| 327 | destroy_workqueue(wq: xdna->notifier_wq); |
| 328 | |
| 329 | drm_dev_unplug(dev: &xdna->ddev); |
| 330 | amdxdna_sysfs_fini(xdna); |
| 331 | |
| 332 | mutex_lock(&xdna->dev_lock); |
| 333 | client = list_first_entry_or_null(&xdna->client_list, |
| 334 | struct amdxdna_client, node); |
| 335 | while (client) { |
| 336 | list_del_init(entry: &client->node); |
| 337 | amdxdna_hwctx_remove_all(client); |
| 338 | |
| 339 | client = list_first_entry_or_null(&xdna->client_list, |
| 340 | struct amdxdna_client, node); |
| 341 | } |
| 342 | |
| 343 | xdna->dev_info->ops->fini(xdna); |
| 344 | mutex_unlock(lock: &xdna->dev_lock); |
| 345 | } |
| 346 | |
| 347 | static const struct dev_pm_ops amdxdna_pm_ops = { |
| 348 | SYSTEM_SLEEP_PM_OPS(amdxdna_pm_suspend, amdxdna_pm_resume) |
| 349 | RUNTIME_PM_OPS(amdxdna_pm_suspend, amdxdna_pm_resume, NULL) |
| 350 | }; |
| 351 | |
| 352 | static struct pci_driver amdxdna_pci_driver = { |
| 353 | .name = KBUILD_MODNAME, |
| 354 | .id_table = pci_ids, |
| 355 | .probe = amdxdna_probe, |
| 356 | .remove = amdxdna_remove, |
| 357 | .driver.pm = &amdxdna_pm_ops, |
| 358 | }; |
| 359 | |
| 360 | module_pci_driver(amdxdna_pci_driver); |
| 361 | |
| 362 | MODULE_LICENSE("GPL" ); |
| 363 | MODULE_AUTHOR("XRT Team <runtimeca39d@amd.com>" ); |
| 364 | MODULE_DESCRIPTION("amdxdna driver" ); |
| 365 | |