Merge tag 'drm-next-2018-06-06-1' of git://anongit.freedesktop.org/drm/drm
authorLinus Torvalds <torvalds@linux-foundation.org>
Wed, 6 Jun 2018 15:16:33 +0000 (08:16 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Wed, 6 Jun 2018 15:16:33 +0000 (08:16 -0700)
Pull drm updates from Dave Airlie:
 "This starts to support NVIDIA volta hardware with nouveau, and adds
  amdgpu support for the GPU in the Kabylake-G (the intel + radeon
  single package chip), along with some initial Intel icelake enabling.

  Summary:

  New Drivers:
   - v3d - driver for broadcom V3D V3.x+ hardware
   - xen-front - XEN PV display frontend

  core:
   - handle zpos normalization in the core
   - stop looking at legacy pointers in atomic paths
   - improved scheduler documentation
   - improved aspect ratio validation
   - aspect ratio support for 64:27 and 256:135
   - drop unused control node code.

  i915:
   - Icelake (ICL) enabling
   - GuC/HuC refactoring
   - PSR/PSR2 enabling and fixes
   - DPLL management refactoring
   - DP MST fixes
   - NV12 enabling
   - HDCP improvements
   - GEM/Execlist/reset improvements
   - GVT improvements
   - stolen memory first 4k fix

  amdgpu:
   - Vega 20 support
   - VEGAM support (Kabylake-G)
   - preOS scanout buffer reservation
   - power management gfxoff support for raven
   - SR-IOV fixes
   - Vega10 power profiles and clock voltage control
   - scatter/gather display support on CZ/ST

  amdkfd:
   - GFX9 dGPU support
   - userptr memory mapping

  nouveau:
   - major refactoring for Volta GV100 support

  tda998x:
   - HDMI i2c CEC support

  etnaviv:
   - removed unused logging code
   - license text cleanups
   - MMU handling improvements
   - timeout fence fix for 50 days uptime

  tegra:
   - IOMMU support in gr2d/gr3d drivers
   - zpos support

  vc4:
   - syncobj support
   - CTM, plane alpha and async cursor support

  analogix_dp:
   - HPD and aux chan fixes

  sun4i:
   - MIPI DSI support

  tilcdc:
   - clock divider fixes for OMAP-l138 LCDK board

  rcar-du:
   - R8A77965 support
   - dma-buf fences fixes
   - hardware indexed crtc/du group handling
   - generic zplane property support

  atmel-hclcdc:
   - generic zplane property support

  mediatek:
   - use generic video mode function

  exynos:
   - S5PV210 FIMD variant support
   - IPP v2 framework
   - more HW overlays support"

* tag 'drm-next-2018-06-06-1' of git://anongit.freedesktop.org/drm/drm: (1286 commits)
  drm/amdgpu: fix 32-bit build warning
  drm/exynos: fimc: signedness bug in fimc_setup_clocks()
  drm/exynos: scaler: fix static checker warning
  drm/amdgpu: Use dev_info() to report amdkfd is not supported for this ASIC
  drm/amd/display: Remove use of division operator for long longs
  drm/amdgpu: Update GFX info structure to match what vega20 used
  drm/amdgpu/pp: remove duplicate assignment
  drm/sched: add rcu_barrier after entity fini
  drm/amdgpu: move VM BOs on LRU again
  drm/amdgpu: consistenly use VM moved flag
  drm/amdgpu: kmap PDs/PTs in amdgpu_vm_update_directories
  drm/amdgpu: further optimize amdgpu_vm_handle_moved
  drm/amdgpu: cleanup amdgpu_vm_validate_pt_bos v2
  drm/amdgpu: rework VM state machine lock handling v2
  drm/amdgpu: Add runtime VCN PG support
  drm/amdgpu: Enable VCN static PG by default on RV
  drm/amdgpu: Add VCN static PG support on RV
  drm/amdgpu: Enable VCN CG by default on RV
  drm/amdgpu: Add static CG control for VCN on RV
  drm/exynos: Fix default value for zpos plane property
  ...

1  2 
MAINTAINERS
drivers/gpu/drm/amd/amdgpu/amdgpu_acp.c
drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c
drivers/gpu/drm/drm_dp_helper.c
drivers/gpu/drm/drm_drv.c
drivers/gpu/drm/vc4/vc4_drv.c
drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
drivers/gpu/drm/xen/xen_drm_front.c

diff --cc MAINTAINERS
Simple merge
index 27579443cdc5a72935fb7f9dd05682be5b637cea,1ce10bc2d37bd1c2628c458d727f475a8723f967..f9b9ab90558c92c223050d5c47a8547cb1d1ca06
@@@ -4577,14 -4650,11 +4650,13 @@@ static int dm_update_crtcs_state(struc
                /* TODO This hack should go away */
                if (aconnector && enable) {
                        // Make sure fake sink is created in plug-in scenario
 -                      new_con_state = drm_atomic_get_connector_state(state,
 +                      drm_new_conn_state = drm_atomic_get_new_connector_state(state,
 +                                                                  &aconnector->base);
 +                      drm_old_conn_state = drm_atomic_get_old_connector_state(state,
                                                                    &aconnector->base);
  
 -                      if (IS_ERR(new_con_state)) {
 -                              ret = PTR_ERR_OR_ZERO(new_con_state);
 +                      if (IS_ERR(drm_new_conn_state)) {
 +                              ret = PTR_ERR_OR_ZERO(drm_new_conn_state);
                                break;
                        }
  
Simple merge
Simple merge
Simple merge
Simple merge
index 0000000000000000000000000000000000000000,3345ac71b391fb42049fda3e950863feaa2cdabe..b3786c1a4e80fdedb4bc1da0a4c046184d4ff250
mode 000000,100644..100644
--- /dev/null
@@@ -1,0 -1,840 +1,839 @@@
 -      dev->bus->force_dma = true;
+ // SPDX-License-Identifier: GPL-2.0 OR MIT
+ /*
+  *  Xen para-virtual DRM device
+  *
+  * Copyright (C) 2016-2018 EPAM Systems Inc.
+  *
+  * Author: Oleksandr Andrushchenko <oleksandr_andrushchenko@epam.com>
+  */
+ #include <drm/drmP.h>
+ #include <drm/drm_atomic_helper.h>
+ #include <drm/drm_crtc_helper.h>
+ #include <drm/drm_gem.h>
+ #include <linux/of_device.h>
+ #include <xen/platform_pci.h>
+ #include <xen/xen.h>
+ #include <xen/xenbus.h>
+ #include <xen/interface/io/displif.h>
+ #include "xen_drm_front.h"
+ #include "xen_drm_front_cfg.h"
+ #include "xen_drm_front_evtchnl.h"
+ #include "xen_drm_front_gem.h"
+ #include "xen_drm_front_kms.h"
+ #include "xen_drm_front_shbuf.h"
+ struct xen_drm_front_dbuf {
+       struct list_head list;
+       u64 dbuf_cookie;
+       u64 fb_cookie;
+       struct xen_drm_front_shbuf *shbuf;
+ };
+ static int dbuf_add_to_list(struct xen_drm_front_info *front_info,
+                           struct xen_drm_front_shbuf *shbuf, u64 dbuf_cookie)
+ {
+       struct xen_drm_front_dbuf *dbuf;
+       dbuf = kzalloc(sizeof(*dbuf), GFP_KERNEL);
+       if (!dbuf)
+               return -ENOMEM;
+       dbuf->dbuf_cookie = dbuf_cookie;
+       dbuf->shbuf = shbuf;
+       list_add(&dbuf->list, &front_info->dbuf_list);
+       return 0;
+ }
+ static struct xen_drm_front_dbuf *dbuf_get(struct list_head *dbuf_list,
+                                          u64 dbuf_cookie)
+ {
+       struct xen_drm_front_dbuf *buf, *q;
+       list_for_each_entry_safe(buf, q, dbuf_list, list)
+               if (buf->dbuf_cookie == dbuf_cookie)
+                       return buf;
+       return NULL;
+ }
+ static void dbuf_flush_fb(struct list_head *dbuf_list, u64 fb_cookie)
+ {
+       struct xen_drm_front_dbuf *buf, *q;
+       list_for_each_entry_safe(buf, q, dbuf_list, list)
+               if (buf->fb_cookie == fb_cookie)
+                       xen_drm_front_shbuf_flush(buf->shbuf);
+ }
+ static void dbuf_free(struct list_head *dbuf_list, u64 dbuf_cookie)
+ {
+       struct xen_drm_front_dbuf *buf, *q;
+       list_for_each_entry_safe(buf, q, dbuf_list, list)
+               if (buf->dbuf_cookie == dbuf_cookie) {
+                       list_del(&buf->list);
+                       xen_drm_front_shbuf_unmap(buf->shbuf);
+                       xen_drm_front_shbuf_free(buf->shbuf);
+                       kfree(buf);
+                       break;
+               }
+ }
+ static void dbuf_free_all(struct list_head *dbuf_list)
+ {
+       struct xen_drm_front_dbuf *buf, *q;
+       list_for_each_entry_safe(buf, q, dbuf_list, list) {
+               list_del(&buf->list);
+               xen_drm_front_shbuf_unmap(buf->shbuf);
+               xen_drm_front_shbuf_free(buf->shbuf);
+               kfree(buf);
+       }
+ }
+ static struct xendispl_req *
+ be_prepare_req(struct xen_drm_front_evtchnl *evtchnl, u8 operation)
+ {
+       struct xendispl_req *req;
+       req = RING_GET_REQUEST(&evtchnl->u.req.ring,
+                              evtchnl->u.req.ring.req_prod_pvt);
+       req->operation = operation;
+       req->id = evtchnl->evt_next_id++;
+       evtchnl->evt_id = req->id;
+       return req;
+ }
+ static int be_stream_do_io(struct xen_drm_front_evtchnl *evtchnl,
+                          struct xendispl_req *req)
+ {
+       reinit_completion(&evtchnl->u.req.completion);
+       if (unlikely(evtchnl->state != EVTCHNL_STATE_CONNECTED))
+               return -EIO;
+       xen_drm_front_evtchnl_flush(evtchnl);
+       return 0;
+ }
+ static int be_stream_wait_io(struct xen_drm_front_evtchnl *evtchnl)
+ {
+       if (wait_for_completion_timeout(&evtchnl->u.req.completion,
+                       msecs_to_jiffies(XEN_DRM_FRONT_WAIT_BACK_MS)) <= 0)
+               return -ETIMEDOUT;
+       return evtchnl->u.req.resp_status;
+ }
+ int xen_drm_front_mode_set(struct xen_drm_front_drm_pipeline *pipeline,
+                          u32 x, u32 y, u32 width, u32 height,
+                          u32 bpp, u64 fb_cookie)
+ {
+       struct xen_drm_front_evtchnl *evtchnl;
+       struct xen_drm_front_info *front_info;
+       struct xendispl_req *req;
+       unsigned long flags;
+       int ret;
+       front_info = pipeline->drm_info->front_info;
+       evtchnl = &front_info->evt_pairs[pipeline->index].req;
+       if (unlikely(!evtchnl))
+               return -EIO;
+       mutex_lock(&evtchnl->u.req.req_io_lock);
+       spin_lock_irqsave(&front_info->io_lock, flags);
+       req = be_prepare_req(evtchnl, XENDISPL_OP_SET_CONFIG);
+       req->op.set_config.x = x;
+       req->op.set_config.y = y;
+       req->op.set_config.width = width;
+       req->op.set_config.height = height;
+       req->op.set_config.bpp = bpp;
+       req->op.set_config.fb_cookie = fb_cookie;
+       ret = be_stream_do_io(evtchnl, req);
+       spin_unlock_irqrestore(&front_info->io_lock, flags);
+       if (ret == 0)
+               ret = be_stream_wait_io(evtchnl);
+       mutex_unlock(&evtchnl->u.req.req_io_lock);
+       return ret;
+ }
+ int xen_drm_front_dbuf_create(struct xen_drm_front_info *front_info,
+                             u64 dbuf_cookie, u32 width, u32 height,
+                             u32 bpp, u64 size, struct page **pages)
+ {
+       struct xen_drm_front_evtchnl *evtchnl;
+       struct xen_drm_front_shbuf *shbuf;
+       struct xendispl_req *req;
+       struct xen_drm_front_shbuf_cfg buf_cfg;
+       unsigned long flags;
+       int ret;
+       evtchnl = &front_info->evt_pairs[GENERIC_OP_EVT_CHNL].req;
+       if (unlikely(!evtchnl))
+               return -EIO;
+       memset(&buf_cfg, 0, sizeof(buf_cfg));
+       buf_cfg.xb_dev = front_info->xb_dev;
+       buf_cfg.pages = pages;
+       buf_cfg.size = size;
+       buf_cfg.be_alloc = front_info->cfg.be_alloc;
+       shbuf = xen_drm_front_shbuf_alloc(&buf_cfg);
+       if (IS_ERR(shbuf))
+               return PTR_ERR(shbuf);
+       ret = dbuf_add_to_list(front_info, shbuf, dbuf_cookie);
+       if (ret < 0) {
+               xen_drm_front_shbuf_free(shbuf);
+               return ret;
+       }
+       mutex_lock(&evtchnl->u.req.req_io_lock);
+       spin_lock_irqsave(&front_info->io_lock, flags);
+       req = be_prepare_req(evtchnl, XENDISPL_OP_DBUF_CREATE);
+       req->op.dbuf_create.gref_directory =
+                       xen_drm_front_shbuf_get_dir_start(shbuf);
+       req->op.dbuf_create.buffer_sz = size;
+       req->op.dbuf_create.dbuf_cookie = dbuf_cookie;
+       req->op.dbuf_create.width = width;
+       req->op.dbuf_create.height = height;
+       req->op.dbuf_create.bpp = bpp;
+       if (buf_cfg.be_alloc)
+               req->op.dbuf_create.flags |= XENDISPL_DBUF_FLG_REQ_ALLOC;
+       ret = be_stream_do_io(evtchnl, req);
+       spin_unlock_irqrestore(&front_info->io_lock, flags);
+       if (ret < 0)
+               goto fail;
+       ret = be_stream_wait_io(evtchnl);
+       if (ret < 0)
+               goto fail;
+       ret = xen_drm_front_shbuf_map(shbuf);
+       if (ret < 0)
+               goto fail;
+       mutex_unlock(&evtchnl->u.req.req_io_lock);
+       return 0;
+ fail:
+       mutex_unlock(&evtchnl->u.req.req_io_lock);
+       dbuf_free(&front_info->dbuf_list, dbuf_cookie);
+       return ret;
+ }
+ static int xen_drm_front_dbuf_destroy(struct xen_drm_front_info *front_info,
+                                     u64 dbuf_cookie)
+ {
+       struct xen_drm_front_evtchnl *evtchnl;
+       struct xendispl_req *req;
+       unsigned long flags;
+       bool be_alloc;
+       int ret;
+       evtchnl = &front_info->evt_pairs[GENERIC_OP_EVT_CHNL].req;
+       if (unlikely(!evtchnl))
+               return -EIO;
+       be_alloc = front_info->cfg.be_alloc;
+       /*
+        * For the backend allocated buffer release references now, so backend
+        * can free the buffer.
+        */
+       if (be_alloc)
+               dbuf_free(&front_info->dbuf_list, dbuf_cookie);
+       mutex_lock(&evtchnl->u.req.req_io_lock);
+       spin_lock_irqsave(&front_info->io_lock, flags);
+       req = be_prepare_req(evtchnl, XENDISPL_OP_DBUF_DESTROY);
+       req->op.dbuf_destroy.dbuf_cookie = dbuf_cookie;
+       ret = be_stream_do_io(evtchnl, req);
+       spin_unlock_irqrestore(&front_info->io_lock, flags);
+       if (ret == 0)
+               ret = be_stream_wait_io(evtchnl);
+       /*
+        * Do this regardless of communication status with the backend:
+        * if we cannot remove remote resources remove what we can locally.
+        */
+       if (!be_alloc)
+               dbuf_free(&front_info->dbuf_list, dbuf_cookie);
+       mutex_unlock(&evtchnl->u.req.req_io_lock);
+       return ret;
+ }
+ int xen_drm_front_fb_attach(struct xen_drm_front_info *front_info,
+                           u64 dbuf_cookie, u64 fb_cookie, u32 width,
+                           u32 height, u32 pixel_format)
+ {
+       struct xen_drm_front_evtchnl *evtchnl;
+       struct xen_drm_front_dbuf *buf;
+       struct xendispl_req *req;
+       unsigned long flags;
+       int ret;
+       evtchnl = &front_info->evt_pairs[GENERIC_OP_EVT_CHNL].req;
+       if (unlikely(!evtchnl))
+               return -EIO;
+       buf = dbuf_get(&front_info->dbuf_list, dbuf_cookie);
+       if (!buf)
+               return -EINVAL;
+       buf->fb_cookie = fb_cookie;
+       mutex_lock(&evtchnl->u.req.req_io_lock);
+       spin_lock_irqsave(&front_info->io_lock, flags);
+       req = be_prepare_req(evtchnl, XENDISPL_OP_FB_ATTACH);
+       req->op.fb_attach.dbuf_cookie = dbuf_cookie;
+       req->op.fb_attach.fb_cookie = fb_cookie;
+       req->op.fb_attach.width = width;
+       req->op.fb_attach.height = height;
+       req->op.fb_attach.pixel_format = pixel_format;
+       ret = be_stream_do_io(evtchnl, req);
+       spin_unlock_irqrestore(&front_info->io_lock, flags);
+       if (ret == 0)
+               ret = be_stream_wait_io(evtchnl);
+       mutex_unlock(&evtchnl->u.req.req_io_lock);
+       return ret;
+ }
+ int xen_drm_front_fb_detach(struct xen_drm_front_info *front_info,
+                           u64 fb_cookie)
+ {
+       struct xen_drm_front_evtchnl *evtchnl;
+       struct xendispl_req *req;
+       unsigned long flags;
+       int ret;
+       evtchnl = &front_info->evt_pairs[GENERIC_OP_EVT_CHNL].req;
+       if (unlikely(!evtchnl))
+               return -EIO;
+       mutex_lock(&evtchnl->u.req.req_io_lock);
+       spin_lock_irqsave(&front_info->io_lock, flags);
+       req = be_prepare_req(evtchnl, XENDISPL_OP_FB_DETACH);
+       req->op.fb_detach.fb_cookie = fb_cookie;
+       ret = be_stream_do_io(evtchnl, req);
+       spin_unlock_irqrestore(&front_info->io_lock, flags);
+       if (ret == 0)
+               ret = be_stream_wait_io(evtchnl);
+       mutex_unlock(&evtchnl->u.req.req_io_lock);
+       return ret;
+ }
+ int xen_drm_front_page_flip(struct xen_drm_front_info *front_info,
+                           int conn_idx, u64 fb_cookie)
+ {
+       struct xen_drm_front_evtchnl *evtchnl;
+       struct xendispl_req *req;
+       unsigned long flags;
+       int ret;
+       if (unlikely(conn_idx >= front_info->num_evt_pairs))
+               return -EINVAL;
+       dbuf_flush_fb(&front_info->dbuf_list, fb_cookie);
+       evtchnl = &front_info->evt_pairs[conn_idx].req;
+       mutex_lock(&evtchnl->u.req.req_io_lock);
+       spin_lock_irqsave(&front_info->io_lock, flags);
+       req = be_prepare_req(evtchnl, XENDISPL_OP_PG_FLIP);
+       req->op.pg_flip.fb_cookie = fb_cookie;
+       ret = be_stream_do_io(evtchnl, req);
+       spin_unlock_irqrestore(&front_info->io_lock, flags);
+       if (ret == 0)
+               ret = be_stream_wait_io(evtchnl);
+       mutex_unlock(&evtchnl->u.req.req_io_lock);
+       return ret;
+ }
+ void xen_drm_front_on_frame_done(struct xen_drm_front_info *front_info,
+                                int conn_idx, u64 fb_cookie)
+ {
+       struct xen_drm_front_drm_info *drm_info = front_info->drm_info;
+       if (unlikely(conn_idx >= front_info->cfg.num_connectors))
+               return;
+       xen_drm_front_kms_on_frame_done(&drm_info->pipeline[conn_idx],
+                                       fb_cookie);
+ }
+ static int xen_drm_drv_dumb_create(struct drm_file *filp,
+                                  struct drm_device *dev,
+                                  struct drm_mode_create_dumb *args)
+ {
+       struct xen_drm_front_drm_info *drm_info = dev->dev_private;
+       struct drm_gem_object *obj;
+       int ret;
+       /*
+        * Dumb creation is a two stage process: first we create a fully
+        * constructed GEM object which is communicated to the backend, and
+        * only after that we can create GEM's handle. This is done so,
+        * because of the possible races: once you create a handle it becomes
+        * immediately visible to user-space, so the latter can try accessing
+        * object without pages etc.
+        * For details also see drm_gem_handle_create
+        */
+       args->pitch = DIV_ROUND_UP(args->width * args->bpp, 8);
+       args->size = args->pitch * args->height;
+       obj = xen_drm_front_gem_create(dev, args->size);
+       if (IS_ERR_OR_NULL(obj)) {
+               ret = PTR_ERR(obj);
+               goto fail;
+       }
+       ret = xen_drm_front_dbuf_create(drm_info->front_info,
+                                       xen_drm_front_dbuf_to_cookie(obj),
+                                       args->width, args->height, args->bpp,
+                                       args->size,
+                                       xen_drm_front_gem_get_pages(obj));
+       if (ret)
+               goto fail_backend;
+       /* This is the tail of GEM object creation */
+       ret = drm_gem_handle_create(filp, obj, &args->handle);
+       if (ret)
+               goto fail_handle;
+       /* Drop reference from allocate - handle holds it now */
+       drm_gem_object_put_unlocked(obj);
+       return 0;
+ fail_handle:
+       xen_drm_front_dbuf_destroy(drm_info->front_info,
+                                  xen_drm_front_dbuf_to_cookie(obj));
+ fail_backend:
+       /* drop reference from allocate */
+       drm_gem_object_put_unlocked(obj);
+ fail:
+       DRM_ERROR("Failed to create dumb buffer: %d\n", ret);
+       return ret;
+ }
+ static void xen_drm_drv_free_object_unlocked(struct drm_gem_object *obj)
+ {
+       struct xen_drm_front_drm_info *drm_info = obj->dev->dev_private;
+       int idx;
+       if (drm_dev_enter(obj->dev, &idx)) {
+               xen_drm_front_dbuf_destroy(drm_info->front_info,
+                                          xen_drm_front_dbuf_to_cookie(obj));
+               drm_dev_exit(idx);
+       } else {
+               dbuf_free(&drm_info->front_info->dbuf_list,
+                         xen_drm_front_dbuf_to_cookie(obj));
+       }
+       xen_drm_front_gem_free_object_unlocked(obj);
+ }
+ static void xen_drm_drv_release(struct drm_device *dev)
+ {
+       struct xen_drm_front_drm_info *drm_info = dev->dev_private;
+       struct xen_drm_front_info *front_info = drm_info->front_info;
+       xen_drm_front_kms_fini(drm_info);
+       drm_atomic_helper_shutdown(dev);
+       drm_mode_config_cleanup(dev);
+       drm_dev_fini(dev);
+       kfree(dev);
+       if (front_info->cfg.be_alloc)
+               xenbus_switch_state(front_info->xb_dev,
+                                   XenbusStateInitialising);
+       kfree(drm_info);
+ }
+ static const struct file_operations xen_drm_dev_fops = {
+       .owner          = THIS_MODULE,
+       .open           = drm_open,
+       .release        = drm_release,
+       .unlocked_ioctl = drm_ioctl,
+ #ifdef CONFIG_COMPAT
+       .compat_ioctl   = drm_compat_ioctl,
+ #endif
+       .poll           = drm_poll,
+       .read           = drm_read,
+       .llseek         = no_llseek,
+       .mmap           = xen_drm_front_gem_mmap,
+ };
+ static const struct vm_operations_struct xen_drm_drv_vm_ops = {
+       .open           = drm_gem_vm_open,
+       .close          = drm_gem_vm_close,
+ };
+ static struct drm_driver xen_drm_driver = {
+       .driver_features           = DRIVER_GEM | DRIVER_MODESET |
+                                    DRIVER_PRIME | DRIVER_ATOMIC,
+       .release                   = xen_drm_drv_release,
+       .gem_vm_ops                = &xen_drm_drv_vm_ops,
+       .gem_free_object_unlocked  = xen_drm_drv_free_object_unlocked,
+       .prime_handle_to_fd        = drm_gem_prime_handle_to_fd,
+       .prime_fd_to_handle        = drm_gem_prime_fd_to_handle,
+       .gem_prime_import          = drm_gem_prime_import,
+       .gem_prime_export          = drm_gem_prime_export,
+       .gem_prime_import_sg_table = xen_drm_front_gem_import_sg_table,
+       .gem_prime_get_sg_table    = xen_drm_front_gem_get_sg_table,
+       .gem_prime_vmap            = xen_drm_front_gem_prime_vmap,
+       .gem_prime_vunmap          = xen_drm_front_gem_prime_vunmap,
+       .gem_prime_mmap            = xen_drm_front_gem_prime_mmap,
+       .dumb_create               = xen_drm_drv_dumb_create,
+       .fops                      = &xen_drm_dev_fops,
+       .name                      = "xendrm-du",
+       .desc                      = "Xen PV DRM Display Unit",
+       .date                      = "20180221",
+       .major                     = 1,
+       .minor                     = 0,
+ };
+ static int xen_drm_drv_init(struct xen_drm_front_info *front_info)
+ {
+       struct device *dev = &front_info->xb_dev->dev;
+       struct xen_drm_front_drm_info *drm_info;
+       struct drm_device *drm_dev;
+       int ret;
+       DRM_INFO("Creating %s\n", xen_drm_driver.desc);
+       drm_info = kzalloc(sizeof(*drm_info), GFP_KERNEL);
+       if (!drm_info) {
+               ret = -ENOMEM;
+               goto fail;
+       }
+       drm_info->front_info = front_info;
+       front_info->drm_info = drm_info;
+       drm_dev = drm_dev_alloc(&xen_drm_driver, dev);
+       if (IS_ERR(drm_dev)) {
+               ret = PTR_ERR(drm_dev);
+               goto fail;
+       }
+       drm_info->drm_dev = drm_dev;
+       drm_dev->dev_private = drm_info;
+       ret = xen_drm_front_kms_init(drm_info);
+       if (ret) {
+               DRM_ERROR("Failed to initialize DRM/KMS, ret %d\n", ret);
+               goto fail_modeset;
+       }
+       ret = drm_dev_register(drm_dev, 0);
+       if (ret)
+               goto fail_register;
+       DRM_INFO("Initialized %s %d.%d.%d %s on minor %d\n",
+                xen_drm_driver.name, xen_drm_driver.major,
+                xen_drm_driver.minor, xen_drm_driver.patchlevel,
+                xen_drm_driver.date, drm_dev->primary->index);
+       return 0;
+ fail_register:
+       drm_dev_unregister(drm_dev);
+ fail_modeset:
+       drm_kms_helper_poll_fini(drm_dev);
+       drm_mode_config_cleanup(drm_dev);
+ fail:
+       kfree(drm_info);
+       return ret;
+ }
+ static void xen_drm_drv_fini(struct xen_drm_front_info *front_info)
+ {
+       struct xen_drm_front_drm_info *drm_info = front_info->drm_info;
+       struct drm_device *dev;
+       if (!drm_info)
+               return;
+       dev = drm_info->drm_dev;
+       if (!dev)
+               return;
+       /* Nothing to do if device is already unplugged */
+       if (drm_dev_is_unplugged(dev))
+               return;
+       drm_kms_helper_poll_fini(dev);
+       drm_dev_unplug(dev);
+       front_info->drm_info = NULL;
+       xen_drm_front_evtchnl_free_all(front_info);
+       dbuf_free_all(&front_info->dbuf_list);
+       /*
+        * If we are not using backend allocated buffers, then tell the
+        * backend we are ready to (re)initialize. Otherwise, wait for
+        * drm_driver.release.
+        */
+       if (!front_info->cfg.be_alloc)
+               xenbus_switch_state(front_info->xb_dev,
+                                   XenbusStateInitialising);
+ }
+ static int displback_initwait(struct xen_drm_front_info *front_info)
+ {
+       struct xen_drm_front_cfg *cfg = &front_info->cfg;
+       int ret;
+       cfg->front_info = front_info;
+       ret = xen_drm_front_cfg_card(front_info, cfg);
+       if (ret < 0)
+               return ret;
+       DRM_INFO("Have %d conector(s)\n", cfg->num_connectors);
+       /* Create event channels for all connectors and publish */
+       ret = xen_drm_front_evtchnl_create_all(front_info);
+       if (ret < 0)
+               return ret;
+       return xen_drm_front_evtchnl_publish_all(front_info);
+ }
+ static int displback_connect(struct xen_drm_front_info *front_info)
+ {
+       xen_drm_front_evtchnl_set_state(front_info, EVTCHNL_STATE_CONNECTED);
+       return xen_drm_drv_init(front_info);
+ }
+ static void displback_disconnect(struct xen_drm_front_info *front_info)
+ {
+       if (!front_info->drm_info)
+               return;
+       /* Tell the backend to wait until we release the DRM driver. */
+       xenbus_switch_state(front_info->xb_dev, XenbusStateReconfiguring);
+       xen_drm_drv_fini(front_info);
+ }
+ static void displback_changed(struct xenbus_device *xb_dev,
+                             enum xenbus_state backend_state)
+ {
+       struct xen_drm_front_info *front_info = dev_get_drvdata(&xb_dev->dev);
+       int ret;
+       DRM_DEBUG("Backend state is %s, front is %s\n",
+                 xenbus_strstate(backend_state),
+                 xenbus_strstate(xb_dev->state));
+       switch (backend_state) {
+       case XenbusStateReconfiguring:
+               /* fall through */
+       case XenbusStateReconfigured:
+               /* fall through */
+       case XenbusStateInitialised:
+               break;
+       case XenbusStateInitialising:
+               if (xb_dev->state == XenbusStateReconfiguring)
+                       break;
+               /* recovering after backend unexpected closure */
+               displback_disconnect(front_info);
+               break;
+       case XenbusStateInitWait:
+               if (xb_dev->state == XenbusStateReconfiguring)
+                       break;
+               /* recovering after backend unexpected closure */
+               displback_disconnect(front_info);
+               if (xb_dev->state != XenbusStateInitialising)
+                       break;
+               ret = displback_initwait(front_info);
+               if (ret < 0)
+                       xenbus_dev_fatal(xb_dev, ret, "initializing frontend");
+               else
+                       xenbus_switch_state(xb_dev, XenbusStateInitialised);
+               break;
+       case XenbusStateConnected:
+               if (xb_dev->state != XenbusStateInitialised)
+                       break;
+               ret = displback_connect(front_info);
+               if (ret < 0) {
+                       displback_disconnect(front_info);
+                       xenbus_dev_fatal(xb_dev, ret, "connecting backend");
+               } else {
+                       xenbus_switch_state(xb_dev, XenbusStateConnected);
+               }
+               break;
+       case XenbusStateClosing:
+               /*
+                * in this state backend starts freeing resources,
+                * so let it go into closed state, so we can also
+                * remove ours
+                */
+               break;
+       case XenbusStateUnknown:
+               /* fall through */
+       case XenbusStateClosed:
+               if (xb_dev->state == XenbusStateClosed)
+                       break;
+               displback_disconnect(front_info);
+               break;
+       }
+ }
+ static int xen_drv_probe(struct xenbus_device *xb_dev,
+                        const struct xenbus_device_id *id)
+ {
+       struct xen_drm_front_info *front_info;
+       struct device *dev = &xb_dev->dev;
+       int ret;
+       /*
+        * The device is not spawn from a device tree, so arch_setup_dma_ops
+        * is not called, thus leaving the device with dummy DMA ops.
+        * This makes the device return error on PRIME buffer import, which
+        * is not correct: to fix this call of_dma_configure() with a NULL
+        * node to set default DMA ops.
+        */
 -      ret = of_dma_configure(dev, NULL);
+       dev->coherent_dma_mask = DMA_BIT_MASK(32);
++      ret = of_dma_configure(dev, NULL, true);
+       if (ret < 0) {
+               DRM_ERROR("Cannot setup DMA ops, ret %d", ret);
+               return ret;
+       }
+       front_info = devm_kzalloc(&xb_dev->dev,
+                                 sizeof(*front_info), GFP_KERNEL);
+       if (!front_info)
+               return -ENOMEM;
+       front_info->xb_dev = xb_dev;
+       spin_lock_init(&front_info->io_lock);
+       INIT_LIST_HEAD(&front_info->dbuf_list);
+       dev_set_drvdata(&xb_dev->dev, front_info);
+       return xenbus_switch_state(xb_dev, XenbusStateInitialising);
+ }
+ static int xen_drv_remove(struct xenbus_device *dev)
+ {
+       struct xen_drm_front_info *front_info = dev_get_drvdata(&dev->dev);
+       int to = 100;
+       xenbus_switch_state(dev, XenbusStateClosing);
+       /*
+        * On driver removal it is disconnected from XenBus,
+        * so no backend state change events come via .otherend_changed
+        * callback. This prevents us from exiting gracefully, e.g.
+        * signaling the backend to free event channels, waiting for its
+        * state to change to XenbusStateClosed and cleaning at our end.
+        * Normally when front driver removed backend will finally go into
+        * XenbusStateInitWait state.
+        *
+        * Workaround: read backend's state manually and wait with time-out.
+        */
+       while ((xenbus_read_unsigned(front_info->xb_dev->otherend, "state",
+                                    XenbusStateUnknown) != XenbusStateInitWait) &&
+                                    --to)
+               msleep(10);
+       if (!to) {
+               unsigned int state;
+               state = xenbus_read_unsigned(front_info->xb_dev->otherend,
+                                            "state", XenbusStateUnknown);
+               DRM_ERROR("Backend state is %s while removing driver\n",
+                         xenbus_strstate(state));
+       }
+       xen_drm_drv_fini(front_info);
+       xenbus_frontend_closed(dev);
+       return 0;
+ }
+ static const struct xenbus_device_id xen_driver_ids[] = {
+       { XENDISPL_DRIVER_NAME },
+       { "" }
+ };
+ static struct xenbus_driver xen_driver = {
+       .ids = xen_driver_ids,
+       .probe = xen_drv_probe,
+       .remove = xen_drv_remove,
+       .otherend_changed = displback_changed,
+ };
+ static int __init xen_drv_init(void)
+ {
+       /* At the moment we only support case with XEN_PAGE_SIZE == PAGE_SIZE */
+       if (XEN_PAGE_SIZE != PAGE_SIZE) {
+               DRM_ERROR(XENDISPL_DRIVER_NAME ": different kernel and Xen page sizes are not supported: XEN_PAGE_SIZE (%lu) != PAGE_SIZE (%lu)\n",
+                         XEN_PAGE_SIZE, PAGE_SIZE);
+               return -ENODEV;
+       }
+       if (!xen_domain())
+               return -ENODEV;
+       if (!xen_has_pv_devices())
+               return -ENODEV;
+       DRM_INFO("Registering XEN PV " XENDISPL_DRIVER_NAME "\n");
+       return xenbus_register_frontend(&xen_driver);
+ }
+ static void __exit xen_drv_fini(void)
+ {
+       DRM_INFO("Unregistering XEN PV " XENDISPL_DRIVER_NAME "\n");
+       xenbus_unregister_driver(&xen_driver);
+ }
+ module_init(xen_drv_init);
+ module_exit(xen_drv_fini);
+ MODULE_DESCRIPTION("Xen para-virtualized display device frontend");
+ MODULE_LICENSE("GPL");
+ MODULE_ALIAS("xen:" XENDISPL_DRIVER_NAME);