These changes are the raw update to linux-4.4.6-rt14. Kernel sources
[kvmfornfv.git] / kernel / drivers / gpu / drm / amd / amdgpu / amdgpu_kms.c
diff --git a/kernel/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c b/kernel/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
new file mode 100644 (file)
index 0000000..e23843f
--- /dev/null
@@ -0,0 +1,752 @@
+/*
+ * Copyright 2008 Advanced Micro Devices, Inc.
+ * Copyright 2008 Red Hat Inc.
+ * Copyright 2009 Jerome Glisse.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Dave Airlie
+ *          Alex Deucher
+ *          Jerome Glisse
+ */
+#include <drm/drmP.h>
+#include "amdgpu.h"
+#include <drm/amdgpu_drm.h>
+#include "amdgpu_uvd.h"
+#include "amdgpu_vce.h"
+
+#include <linux/vga_switcheroo.h>
+#include <linux/slab.h>
+#include <linux/pm_runtime.h>
+#include "amdgpu_amdkfd.h"
+
+#if defined(CONFIG_VGA_SWITCHEROO)
+bool amdgpu_has_atpx(void);
+#else
+static inline bool amdgpu_has_atpx(void) { return false; }
+#endif
+
+/**
+ * amdgpu_driver_unload_kms - Main unload function for KMS.
+ *
+ * @dev: drm dev pointer
+ *
+ * This is the main unload function for KMS (all asics).
+ * Returns 0 on success.
+ */
+int amdgpu_driver_unload_kms(struct drm_device *dev)
+{
+       struct amdgpu_device *adev = dev->dev_private;
+
+       if (adev == NULL)
+               return 0;
+
+       if (adev->rmmio == NULL)
+               goto done_free;
+
+       pm_runtime_get_sync(dev->dev);
+
+       amdgpu_amdkfd_device_fini(adev);
+
+       amdgpu_acpi_fini(adev);
+
+       amdgpu_device_fini(adev);
+
+done_free:
+       kfree(adev);
+       dev->dev_private = NULL;
+       return 0;
+}
+
+/**
+ * amdgpu_driver_load_kms - Main load function for KMS.
+ *
+ * @dev: drm dev pointer
+ * @flags: device flags
+ *
+ * This is the main load function for KMS (all asics).
+ * Returns 0 on success, error on failure.
+ */
+int amdgpu_driver_load_kms(struct drm_device *dev, unsigned long flags)
+{
+       struct amdgpu_device *adev;
+       int r, acpi_status;
+
+       adev = kzalloc(sizeof(struct amdgpu_device), GFP_KERNEL);
+       if (adev == NULL) {
+               return -ENOMEM;
+       }
+       dev->dev_private = (void *)adev;
+
+       if ((amdgpu_runtime_pm != 0) &&
+           amdgpu_has_atpx() &&
+           ((flags & AMD_IS_APU) == 0))
+               flags |= AMD_IS_PX;
+
+       /* amdgpu_device_init should report only fatal error
+        * like memory allocation failure or iomapping failure,
+        * or memory manager initialization failure, it must
+        * properly initialize the GPU MC controller and permit
+        * VRAM allocation
+        */
+       r = amdgpu_device_init(adev, dev, dev->pdev, flags);
+       if (r) {
+               dev_err(&dev->pdev->dev, "Fatal error during GPU init\n");
+               goto out;
+       }
+
+       /* Call ACPI methods: require modeset init
+        * but failure is not fatal
+        */
+       if (!r) {
+               acpi_status = amdgpu_acpi_init(adev);
+               if (acpi_status)
+               dev_dbg(&dev->pdev->dev,
+                               "Error during ACPI methods call\n");
+       }
+
+       amdgpu_amdkfd_load_interface(adev);
+       amdgpu_amdkfd_device_probe(adev);
+       amdgpu_amdkfd_device_init(adev);
+
+       if (amdgpu_device_is_px(dev)) {
+               pm_runtime_use_autosuspend(dev->dev);
+               pm_runtime_set_autosuspend_delay(dev->dev, 5000);
+               pm_runtime_set_active(dev->dev);
+               pm_runtime_allow(dev->dev);
+               pm_runtime_mark_last_busy(dev->dev);
+               pm_runtime_put_autosuspend(dev->dev);
+       }
+
+out:
+       if (r)
+               amdgpu_driver_unload_kms(dev);
+
+
+       return r;
+}
+
+/*
+ * Userspace get information ioctl
+ */
+/**
+ * amdgpu_info_ioctl - answer a device specific request.
+ *
+ * @adev: amdgpu device pointer
+ * @data: request object
+ * @filp: drm filp
+ *
+ * This function is used to pass device specific parameters to the userspace
+ * drivers.  Examples include: pci device id, pipeline parms, tiling params,
+ * etc. (all asics).
+ * Returns 0 on success, -EINVAL on failure.
+ */
+static int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
+{
+       struct amdgpu_device *adev = dev->dev_private;
+       struct drm_amdgpu_info *info = data;
+       struct amdgpu_mode_info *minfo = &adev->mode_info;
+       void __user *out = (void __user *)(long)info->return_pointer;
+       uint32_t size = info->return_size;
+       struct drm_crtc *crtc;
+       uint32_t ui32 = 0;
+       uint64_t ui64 = 0;
+       int i, found;
+
+       if (!info->return_size || !info->return_pointer)
+               return -EINVAL;
+
+       switch (info->query) {
+       case AMDGPU_INFO_ACCEL_WORKING:
+               ui32 = adev->accel_working;
+               return copy_to_user(out, &ui32, min(size, 4u)) ? -EFAULT : 0;
+       case AMDGPU_INFO_CRTC_FROM_ID:
+               for (i = 0, found = 0; i < adev->mode_info.num_crtc; i++) {
+                       crtc = (struct drm_crtc *)minfo->crtcs[i];
+                       if (crtc && crtc->base.id == info->mode_crtc.id) {
+                               struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc);
+                               ui32 = amdgpu_crtc->crtc_id;
+                               found = 1;
+                               break;
+                       }
+               }
+               if (!found) {
+                       DRM_DEBUG_KMS("unknown crtc id %d\n", info->mode_crtc.id);
+                       return -EINVAL;
+               }
+               return copy_to_user(out, &ui32, min(size, 4u)) ? -EFAULT : 0;
+       case AMDGPU_INFO_HW_IP_INFO: {
+               struct drm_amdgpu_info_hw_ip ip = {};
+               enum amd_ip_block_type type;
+               uint32_t ring_mask = 0;
+               uint32_t ib_start_alignment = 0;
+               uint32_t ib_size_alignment = 0;
+
+               if (info->query_hw_ip.ip_instance >= AMDGPU_HW_IP_INSTANCE_MAX_COUNT)
+                       return -EINVAL;
+
+               switch (info->query_hw_ip.type) {
+               case AMDGPU_HW_IP_GFX:
+                       type = AMD_IP_BLOCK_TYPE_GFX;
+                       for (i = 0; i < adev->gfx.num_gfx_rings; i++)
+                               ring_mask |= ((adev->gfx.gfx_ring[i].ready ? 1 : 0) << i);
+                       ib_start_alignment = AMDGPU_GPU_PAGE_SIZE;
+                       ib_size_alignment = 8;
+                       break;
+               case AMDGPU_HW_IP_COMPUTE:
+                       type = AMD_IP_BLOCK_TYPE_GFX;
+                       for (i = 0; i < adev->gfx.num_compute_rings; i++)
+                               ring_mask |= ((adev->gfx.compute_ring[i].ready ? 1 : 0) << i);
+                       ib_start_alignment = AMDGPU_GPU_PAGE_SIZE;
+                       ib_size_alignment = 8;
+                       break;
+               case AMDGPU_HW_IP_DMA:
+                       type = AMD_IP_BLOCK_TYPE_SDMA;
+                       for (i = 0; i < adev->sdma.num_instances; i++)
+                               ring_mask |= ((adev->sdma.instance[i].ring.ready ? 1 : 0) << i);
+                       ib_start_alignment = AMDGPU_GPU_PAGE_SIZE;
+                       ib_size_alignment = 1;
+                       break;
+               case AMDGPU_HW_IP_UVD:
+                       type = AMD_IP_BLOCK_TYPE_UVD;
+                       ring_mask = adev->uvd.ring.ready ? 1 : 0;
+                       ib_start_alignment = AMDGPU_GPU_PAGE_SIZE;
+                       ib_size_alignment = 8;
+                       break;
+               case AMDGPU_HW_IP_VCE:
+                       type = AMD_IP_BLOCK_TYPE_VCE;
+                       for (i = 0; i < AMDGPU_MAX_VCE_RINGS; i++)
+                               ring_mask |= ((adev->vce.ring[i].ready ? 1 : 0) << i);
+                       ib_start_alignment = AMDGPU_GPU_PAGE_SIZE;
+                       ib_size_alignment = 8;
+                       break;
+               default:
+                       return -EINVAL;
+               }
+
+               for (i = 0; i < adev->num_ip_blocks; i++) {
+                       if (adev->ip_blocks[i].type == type &&
+                           adev->ip_block_status[i].valid) {
+                               ip.hw_ip_version_major = adev->ip_blocks[i].major;
+                               ip.hw_ip_version_minor = adev->ip_blocks[i].minor;
+                               ip.capabilities_flags = 0;
+                               ip.available_rings = ring_mask;
+                               ip.ib_start_alignment = ib_start_alignment;
+                               ip.ib_size_alignment = ib_size_alignment;
+                               break;
+                       }
+               }
+               return copy_to_user(out, &ip,
+                                   min((size_t)size, sizeof(ip))) ? -EFAULT : 0;
+       }
+       case AMDGPU_INFO_HW_IP_COUNT: {
+               enum amd_ip_block_type type;
+               uint32_t count = 0;
+
+               switch (info->query_hw_ip.type) {
+               case AMDGPU_HW_IP_GFX:
+                       type = AMD_IP_BLOCK_TYPE_GFX;
+                       break;
+               case AMDGPU_HW_IP_COMPUTE:
+                       type = AMD_IP_BLOCK_TYPE_GFX;
+                       break;
+               case AMDGPU_HW_IP_DMA:
+                       type = AMD_IP_BLOCK_TYPE_SDMA;
+                       break;
+               case AMDGPU_HW_IP_UVD:
+                       type = AMD_IP_BLOCK_TYPE_UVD;
+                       break;
+               case AMDGPU_HW_IP_VCE:
+                       type = AMD_IP_BLOCK_TYPE_VCE;
+                       break;
+               default:
+                       return -EINVAL;
+               }
+
+               for (i = 0; i < adev->num_ip_blocks; i++)
+                       if (adev->ip_blocks[i].type == type &&
+                           adev->ip_block_status[i].valid &&
+                           count < AMDGPU_HW_IP_INSTANCE_MAX_COUNT)
+                               count++;
+
+               return copy_to_user(out, &count, min(size, 4u)) ? -EFAULT : 0;
+       }
+       case AMDGPU_INFO_TIMESTAMP:
+               ui64 = amdgpu_asic_get_gpu_clock_counter(adev);
+               return copy_to_user(out, &ui64, min(size, 8u)) ? -EFAULT : 0;
+       case AMDGPU_INFO_FW_VERSION: {
+               struct drm_amdgpu_info_firmware fw_info;
+
+               /* We only support one instance of each IP block right now. */
+               if (info->query_fw.ip_instance != 0)
+                       return -EINVAL;
+
+               switch (info->query_fw.fw_type) {
+               case AMDGPU_INFO_FW_VCE:
+                       fw_info.ver = adev->vce.fw_version;
+                       fw_info.feature = adev->vce.fb_version;
+                       break;
+               case AMDGPU_INFO_FW_UVD:
+                       fw_info.ver = 0;
+                       fw_info.feature = 0;
+                       break;
+               case AMDGPU_INFO_FW_GMC:
+                       fw_info.ver = adev->mc.fw_version;
+                       fw_info.feature = 0;
+                       break;
+               case AMDGPU_INFO_FW_GFX_ME:
+                       fw_info.ver = adev->gfx.me_fw_version;
+                       fw_info.feature = adev->gfx.me_feature_version;
+                       break;
+               case AMDGPU_INFO_FW_GFX_PFP:
+                       fw_info.ver = adev->gfx.pfp_fw_version;
+                       fw_info.feature = adev->gfx.pfp_feature_version;
+                       break;
+               case AMDGPU_INFO_FW_GFX_CE:
+                       fw_info.ver = adev->gfx.ce_fw_version;
+                       fw_info.feature = adev->gfx.ce_feature_version;
+                       break;
+               case AMDGPU_INFO_FW_GFX_RLC:
+                       fw_info.ver = adev->gfx.rlc_fw_version;
+                       fw_info.feature = adev->gfx.rlc_feature_version;
+                       break;
+               case AMDGPU_INFO_FW_GFX_MEC:
+                       if (info->query_fw.index == 0) {
+                               fw_info.ver = adev->gfx.mec_fw_version;
+                               fw_info.feature = adev->gfx.mec_feature_version;
+                       } else if (info->query_fw.index == 1) {
+                               fw_info.ver = adev->gfx.mec2_fw_version;
+                               fw_info.feature = adev->gfx.mec2_feature_version;
+                       } else
+                               return -EINVAL;
+                       break;
+               case AMDGPU_INFO_FW_SMC:
+                       fw_info.ver = adev->pm.fw_version;
+                       fw_info.feature = 0;
+                       break;
+               case AMDGPU_INFO_FW_SDMA:
+                       if (info->query_fw.index >= adev->sdma.num_instances)
+                               return -EINVAL;
+                       fw_info.ver = adev->sdma.instance[info->query_fw.index].fw_version;
+                       fw_info.feature = adev->sdma.instance[info->query_fw.index].feature_version;
+                       break;
+               default:
+                       return -EINVAL;
+               }
+               return copy_to_user(out, &fw_info,
+                                   min((size_t)size, sizeof(fw_info))) ? -EFAULT : 0;
+       }
+       case AMDGPU_INFO_NUM_BYTES_MOVED:
+               ui64 = atomic64_read(&adev->num_bytes_moved);
+               return copy_to_user(out, &ui64, min(size, 8u)) ? -EFAULT : 0;
+       case AMDGPU_INFO_VRAM_USAGE:
+               ui64 = atomic64_read(&adev->vram_usage);
+               return copy_to_user(out, &ui64, min(size, 8u)) ? -EFAULT : 0;
+       case AMDGPU_INFO_VIS_VRAM_USAGE:
+               ui64 = atomic64_read(&adev->vram_vis_usage);
+               return copy_to_user(out, &ui64, min(size, 8u)) ? -EFAULT : 0;
+       case AMDGPU_INFO_GTT_USAGE:
+               ui64 = atomic64_read(&adev->gtt_usage);
+               return copy_to_user(out, &ui64, min(size, 8u)) ? -EFAULT : 0;
+       case AMDGPU_INFO_GDS_CONFIG: {
+               struct drm_amdgpu_info_gds gds_info;
+
+               memset(&gds_info, 0, sizeof(gds_info));
+               gds_info.gds_gfx_partition_size = adev->gds.mem.gfx_partition_size >> AMDGPU_GDS_SHIFT;
+               gds_info.compute_partition_size = adev->gds.mem.cs_partition_size >> AMDGPU_GDS_SHIFT;
+               gds_info.gds_total_size = adev->gds.mem.total_size >> AMDGPU_GDS_SHIFT;
+               gds_info.gws_per_gfx_partition = adev->gds.gws.gfx_partition_size >> AMDGPU_GWS_SHIFT;
+               gds_info.gws_per_compute_partition = adev->gds.gws.cs_partition_size >> AMDGPU_GWS_SHIFT;
+               gds_info.oa_per_gfx_partition = adev->gds.oa.gfx_partition_size >> AMDGPU_OA_SHIFT;
+               gds_info.oa_per_compute_partition = adev->gds.oa.cs_partition_size >> AMDGPU_OA_SHIFT;
+               return copy_to_user(out, &gds_info,
+                                   min((size_t)size, sizeof(gds_info))) ? -EFAULT : 0;
+       }
+       case AMDGPU_INFO_VRAM_GTT: {
+               struct drm_amdgpu_info_vram_gtt vram_gtt;
+
+               vram_gtt.vram_size = adev->mc.real_vram_size;
+               vram_gtt.vram_cpu_accessible_size = adev->mc.visible_vram_size;
+               vram_gtt.vram_cpu_accessible_size -= adev->vram_pin_size;
+               vram_gtt.gtt_size  = adev->mc.gtt_size;
+               vram_gtt.gtt_size -= adev->gart_pin_size;
+               return copy_to_user(out, &vram_gtt,
+                                   min((size_t)size, sizeof(vram_gtt))) ? -EFAULT : 0;
+       }
+       case AMDGPU_INFO_READ_MMR_REG: {
+               unsigned n, alloc_size;
+               uint32_t *regs;
+               unsigned se_num = (info->read_mmr_reg.instance >>
+                                  AMDGPU_INFO_MMR_SE_INDEX_SHIFT) &
+                                 AMDGPU_INFO_MMR_SE_INDEX_MASK;
+               unsigned sh_num = (info->read_mmr_reg.instance >>
+                                  AMDGPU_INFO_MMR_SH_INDEX_SHIFT) &
+                                 AMDGPU_INFO_MMR_SH_INDEX_MASK;
+
+               /* set full masks if the userspace set all bits
+                * in the bitfields */
+               if (se_num == AMDGPU_INFO_MMR_SE_INDEX_MASK)
+                       se_num = 0xffffffff;
+               if (sh_num == AMDGPU_INFO_MMR_SH_INDEX_MASK)
+                       sh_num = 0xffffffff;
+
+               regs = kmalloc_array(info->read_mmr_reg.count, sizeof(*regs), GFP_KERNEL);
+               if (!regs)
+                       return -ENOMEM;
+               alloc_size = info->read_mmr_reg.count * sizeof(*regs);
+
+               for (i = 0; i < info->read_mmr_reg.count; i++)
+                       if (amdgpu_asic_read_register(adev, se_num, sh_num,
+                                                     info->read_mmr_reg.dword_offset + i,
+                                                     &regs[i])) {
+                               DRM_DEBUG_KMS("unallowed offset %#x\n",
+                                             info->read_mmr_reg.dword_offset + i);
+                               kfree(regs);
+                               return -EFAULT;
+                       }
+               n = copy_to_user(out, regs, min(size, alloc_size));
+               kfree(regs);
+               return n ? -EFAULT : 0;
+       }
+       case AMDGPU_INFO_DEV_INFO: {
+               struct drm_amdgpu_info_device dev_info = {};
+               struct amdgpu_cu_info cu_info;
+
+               dev_info.device_id = dev->pdev->device;
+               dev_info.chip_rev = adev->rev_id;
+               dev_info.external_rev = adev->external_rev_id;
+               dev_info.pci_rev = dev->pdev->revision;
+               dev_info.family = adev->family;
+               dev_info.num_shader_engines = adev->gfx.config.max_shader_engines;
+               dev_info.num_shader_arrays_per_engine = adev->gfx.config.max_sh_per_se;
+               /* return all clocks in KHz */
+               dev_info.gpu_counter_freq = amdgpu_asic_get_xclk(adev) * 10;
+               if (adev->pm.dpm_enabled) {
+                       dev_info.max_engine_clock =
+                               adev->pm.dpm.dyn_state.max_clock_voltage_on_ac.sclk * 10;
+                       dev_info.max_memory_clock =
+                               adev->pm.dpm.dyn_state.max_clock_voltage_on_ac.mclk * 10;
+               } else {
+                       dev_info.max_engine_clock = adev->pm.default_sclk * 10;
+                       dev_info.max_memory_clock = adev->pm.default_mclk * 10;
+               }
+               dev_info.enabled_rb_pipes_mask = adev->gfx.config.backend_enable_mask;
+               dev_info.num_rb_pipes = adev->gfx.config.max_backends_per_se *
+                                       adev->gfx.config.max_shader_engines;
+               dev_info.num_hw_gfx_contexts = adev->gfx.config.max_hw_contexts;
+               dev_info._pad = 0;
+               dev_info.ids_flags = 0;
+               if (adev->flags & AMD_IS_APU)
+                       dev_info.ids_flags |= AMDGPU_IDS_FLAGS_FUSION;
+               dev_info.virtual_address_offset = AMDGPU_VA_RESERVED_SIZE;
+               dev_info.virtual_address_max = (uint64_t)adev->vm_manager.max_pfn * AMDGPU_GPU_PAGE_SIZE;
+               dev_info.virtual_address_alignment = max((int)PAGE_SIZE, AMDGPU_GPU_PAGE_SIZE);
+               dev_info.pte_fragment_size = (1 << AMDGPU_LOG2_PAGES_PER_FRAG) *
+                                            AMDGPU_GPU_PAGE_SIZE;
+               dev_info.gart_page_size = AMDGPU_GPU_PAGE_SIZE;
+
+               amdgpu_asic_get_cu_info(adev, &cu_info);
+               dev_info.cu_active_number = cu_info.number;
+               dev_info.cu_ao_mask = cu_info.ao_cu_mask;
+               dev_info.ce_ram_size = adev->gfx.ce_ram_size;
+               memcpy(&dev_info.cu_bitmap[0], &cu_info.bitmap[0], sizeof(cu_info.bitmap));
+               dev_info.vram_type = adev->mc.vram_type;
+               dev_info.vram_bit_width = adev->mc.vram_width;
+               dev_info.vce_harvest_config = adev->vce.harvest_config;
+
+               return copy_to_user(out, &dev_info,
+                                   min((size_t)size, sizeof(dev_info))) ? -EFAULT : 0;
+       }
+       default:
+               DRM_DEBUG_KMS("Invalid request %d\n", info->query);
+               return -EINVAL;
+       }
+       return 0;
+}
+
+
+/*
+ * Outdated mess for old drm with Xorg being in charge (void function now).
+ */
+/**
+ * amdgpu_driver_lastclose_kms - drm callback for last close
+ *
+ * @dev: drm dev pointer
+ *
+ * Switch vga_switcheroo state after last close (all asics).
+ */
+void amdgpu_driver_lastclose_kms(struct drm_device *dev)
+{
+       struct amdgpu_device *adev = dev->dev_private;
+
+       amdgpu_fbdev_restore_mode(adev);
+       vga_switcheroo_process_delayed_switch();
+}
+
+/**
+ * amdgpu_driver_open_kms - drm callback for open
+ *
+ * @dev: drm dev pointer
+ * @file_priv: drm file
+ *
+ * On device open, init vm on cayman+ (all asics).
+ * Returns 0 on success, error on failure.
+ */
+int amdgpu_driver_open_kms(struct drm_device *dev, struct drm_file *file_priv)
+{
+       struct amdgpu_device *adev = dev->dev_private;
+       struct amdgpu_fpriv *fpriv;
+       int r;
+
+       file_priv->driver_priv = NULL;
+
+       r = pm_runtime_get_sync(dev->dev);
+       if (r < 0)
+               return r;
+
+       fpriv = kzalloc(sizeof(*fpriv), GFP_KERNEL);
+       if (unlikely(!fpriv))
+               return -ENOMEM;
+
+       r = amdgpu_vm_init(adev, &fpriv->vm);
+       if (r)
+               goto error_free;
+
+       mutex_init(&fpriv->bo_list_lock);
+       idr_init(&fpriv->bo_list_handles);
+
+       amdgpu_ctx_mgr_init(&fpriv->ctx_mgr);
+
+       file_priv->driver_priv = fpriv;
+
+       pm_runtime_mark_last_busy(dev->dev);
+       pm_runtime_put_autosuspend(dev->dev);
+       return 0;
+
+error_free:
+       kfree(fpriv);
+
+       return r;
+}
+
+/**
+ * amdgpu_driver_postclose_kms - drm callback for post close
+ *
+ * @dev: drm dev pointer
+ * @file_priv: drm file
+ *
+ * On device post close, tear down vm on cayman+ (all asics).
+ */
+void amdgpu_driver_postclose_kms(struct drm_device *dev,
+                                struct drm_file *file_priv)
+{
+       struct amdgpu_device *adev = dev->dev_private;
+       struct amdgpu_fpriv *fpriv = file_priv->driver_priv;
+       struct amdgpu_bo_list *list;
+       int handle;
+
+       if (!fpriv)
+               return;
+
+       amdgpu_ctx_mgr_fini(&fpriv->ctx_mgr);
+
+       amdgpu_vm_fini(adev, &fpriv->vm);
+
+       idr_for_each_entry(&fpriv->bo_list_handles, list, handle)
+               amdgpu_bo_list_free(list);
+
+       idr_destroy(&fpriv->bo_list_handles);
+       mutex_destroy(&fpriv->bo_list_lock);
+
+       kfree(fpriv);
+       file_priv->driver_priv = NULL;
+}
+
+/**
+ * amdgpu_driver_preclose_kms - drm callback for pre close
+ *
+ * @dev: drm dev pointer
+ * @file_priv: drm file
+ *
+ * On device pre close, tear down hyperz and cmask filps on r1xx-r5xx
+ * (all asics).
+ */
+void amdgpu_driver_preclose_kms(struct drm_device *dev,
+                               struct drm_file *file_priv)
+{
+       struct amdgpu_device *adev = dev->dev_private;
+
+       amdgpu_uvd_free_handles(adev, file_priv);
+       amdgpu_vce_free_handles(adev, file_priv);
+}
+
+/*
+ * VBlank related functions.
+ */
+/**
+ * amdgpu_get_vblank_counter_kms - get frame count
+ *
+ * @dev: drm dev pointer
+ * @pipe: crtc to get the frame count from
+ *
+ * Gets the frame count on the requested crtc (all asics).
+ * Returns frame count on success, -EINVAL on failure.
+ */
+u32 amdgpu_get_vblank_counter_kms(struct drm_device *dev, unsigned int pipe)
+{
+       struct amdgpu_device *adev = dev->dev_private;
+       int vpos, hpos, stat;
+       u32 count;
+
+       if (pipe >= adev->mode_info.num_crtc) {
+               DRM_ERROR("Invalid crtc %u\n", pipe);
+               return -EINVAL;
+       }
+
+       /* The hw increments its frame counter at start of vsync, not at start
+        * of vblank, as is required by DRM core vblank counter handling.
+        * Cook the hw count here to make it appear to the caller as if it
+        * incremented at start of vblank. We measure distance to start of
+        * vblank in vpos. vpos therefore will be >= 0 between start of vblank
+        * and start of vsync, so vpos >= 0 means to bump the hw frame counter
+        * result by 1 to give the proper appearance to caller.
+        */
+       if (adev->mode_info.crtcs[pipe]) {
+               /* Repeat readout if needed to provide stable result if
+                * we cross start of vsync during the queries.
+                */
+               do {
+                       count = amdgpu_display_vblank_get_counter(adev, pipe);
+                       /* Ask amdgpu_get_crtc_scanoutpos to return vpos as
+                        * distance to start of vblank, instead of regular
+                        * vertical scanout pos.
+                        */
+                       stat = amdgpu_get_crtc_scanoutpos(
+                               dev, pipe, GET_DISTANCE_TO_VBLANKSTART,
+                               &vpos, &hpos, NULL, NULL,
+                               &adev->mode_info.crtcs[pipe]->base.hwmode);
+               } while (count != amdgpu_display_vblank_get_counter(adev, pipe));
+
+               if (((stat & (DRM_SCANOUTPOS_VALID | DRM_SCANOUTPOS_ACCURATE)) !=
+                   (DRM_SCANOUTPOS_VALID | DRM_SCANOUTPOS_ACCURATE))) {
+                       DRM_DEBUG_VBL("Query failed! stat %d\n", stat);
+               } else {
+                       DRM_DEBUG_VBL("crtc %d: dist from vblank start %d\n",
+                                     pipe, vpos);
+
+                       /* Bump counter if we are at >= leading edge of vblank,
+                        * but before vsync where vpos would turn negative and
+                        * the hw counter really increments.
+                        */
+                       if (vpos >= 0)
+                               count++;
+               }
+       } else {
+               /* Fallback to use value as is. */
+               count = amdgpu_display_vblank_get_counter(adev, pipe);
+               DRM_DEBUG_VBL("NULL mode info! Returned count may be wrong.\n");
+       }
+
+       return count;
+}
+
+/**
+ * amdgpu_enable_vblank_kms - enable vblank interrupt
+ *
+ * @dev: drm dev pointer
+ * @pipe: crtc to enable vblank interrupt for
+ *
+ * Enable the interrupt on the requested crtc (all asics).
+ * Returns 0 on success, -EINVAL on failure.
+ */
+int amdgpu_enable_vblank_kms(struct drm_device *dev, unsigned int pipe)
+{
+       struct amdgpu_device *adev = dev->dev_private;
+       int idx = amdgpu_crtc_idx_to_irq_type(adev, pipe);
+
+       return amdgpu_irq_get(adev, &adev->crtc_irq, idx);
+}
+
+/**
+ * amdgpu_disable_vblank_kms - disable vblank interrupt
+ *
+ * @dev: drm dev pointer
+ * @pipe: crtc to disable vblank interrupt for
+ *
+ * Disable the interrupt on the requested crtc (all asics).
+ */
+void amdgpu_disable_vblank_kms(struct drm_device *dev, unsigned int pipe)
+{
+       struct amdgpu_device *adev = dev->dev_private;
+       int idx = amdgpu_crtc_idx_to_irq_type(adev, pipe);
+
+       amdgpu_irq_put(adev, &adev->crtc_irq, idx);
+}
+
+/**
+ * amdgpu_get_vblank_timestamp_kms - get vblank timestamp
+ *
+ * @dev: drm dev pointer
+ * @crtc: crtc to get the timestamp for
+ * @max_error: max error
+ * @vblank_time: time value
+ * @flags: flags passed to the driver
+ *
+ * Gets the timestamp on the requested crtc based on the
+ * scanout position.  (all asics).
+ * Returns postive status flags on success, negative error on failure.
+ */
+int amdgpu_get_vblank_timestamp_kms(struct drm_device *dev, unsigned int pipe,
+                                   int *max_error,
+                                   struct timeval *vblank_time,
+                                   unsigned flags)
+{
+       struct drm_crtc *crtc;
+       struct amdgpu_device *adev = dev->dev_private;
+
+       if (pipe >= dev->num_crtcs) {
+               DRM_ERROR("Invalid crtc %u\n", pipe);
+               return -EINVAL;
+       }
+
+       /* Get associated drm_crtc: */
+       crtc = &adev->mode_info.crtcs[pipe]->base;
+
+       /* Helper routine in DRM core does all the work: */
+       return drm_calc_vbltimestamp_from_scanoutpos(dev, pipe, max_error,
+                                                    vblank_time, flags,
+                                                    &crtc->hwmode);
+}
+
+const struct drm_ioctl_desc amdgpu_ioctls_kms[] = {
+       DRM_IOCTL_DEF_DRV(AMDGPU_GEM_CREATE, amdgpu_gem_create_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
+       DRM_IOCTL_DEF_DRV(AMDGPU_CTX, amdgpu_ctx_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
+       DRM_IOCTL_DEF_DRV(AMDGPU_BO_LIST, amdgpu_bo_list_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
+       /* KMS */
+       DRM_IOCTL_DEF_DRV(AMDGPU_GEM_MMAP, amdgpu_gem_mmap_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
+       DRM_IOCTL_DEF_DRV(AMDGPU_GEM_WAIT_IDLE, amdgpu_gem_wait_idle_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
+       DRM_IOCTL_DEF_DRV(AMDGPU_CS, amdgpu_cs_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
+       DRM_IOCTL_DEF_DRV(AMDGPU_INFO, amdgpu_info_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
+       DRM_IOCTL_DEF_DRV(AMDGPU_WAIT_CS, amdgpu_cs_wait_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
+       DRM_IOCTL_DEF_DRV(AMDGPU_GEM_METADATA, amdgpu_gem_metadata_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
+       DRM_IOCTL_DEF_DRV(AMDGPU_GEM_VA, amdgpu_gem_va_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
+       DRM_IOCTL_DEF_DRV(AMDGPU_GEM_OP, amdgpu_gem_op_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
+       DRM_IOCTL_DEF_DRV(AMDGPU_GEM_USERPTR, amdgpu_gem_userptr_ioctl, DRM_AUTH|DRM_RENDER_ALLOW),
+};
+int amdgpu_max_kms_ioctl = ARRAY_SIZE(amdgpu_ioctls_kms);