Merge tag 'drm-next-2018-08-15' of git://anongit.freedesktop.org/drm/drm
authorLinus Torvalds <torvalds@linux-foundation.org>
Thu, 16 Aug 2018 00:39:07 +0000 (17:39 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Thu, 16 Aug 2018 00:39:07 +0000 (17:39 -0700)
Pull drm updates from Dave Airlie:
 "This is the main drm pull request for 4.19.

  Rob has some new hardware support for new qualcomm hw that I'll send
  along separately. This has the display part of it, the remaining pull
  is for the acceleration engine.

  This also contains a wound-wait/wait-die mutex rework, Peter has acked
  it for merging via my tree.

  Otherwise mostly the usual level of activity. Summary:

  core:
   - Wound-wait/wait-die mutex rework
   - Add writeback connector type
   - Add "content type" property for HDMI
   - Move GEM bo to drm_framebuffer
   - Initial gpu scheduler documentation
   - GPU scheduler fixes for dying processes
   - Console deferred fbcon takeover support
   - Displayport support for CEC tunneling over AUX

  panel:
   - otm8009a panel driver fixes
   - Innolux TV123WAM and G070Y2-L01 panel driver
   - Ilitek ILI9881c panel driver
   - Rocktech RK070ER9427 LCD
   - EDT ETM0700G0EDH6 and EDT ETM0700G0BDH6
   - DLC DLC0700YZG-1
   - BOE HV070WSA-100
   - newhaven, nhd-4.3-480272ef-atxl LCD
   - DataImage SCF0700C48GGU18
   - Sharp LQ035Q7DB03
   - p079zca: Refactor to support multiple panels

  tinydrm:
   - ILI9341 display panel

  New driver:
   - vkms - virtual kms driver to testing.

  i915:
   - Icelake:
        Display enablement
        DSI support
        IRQ support
        Powerwell support
   - GPU reset fixes and improvements
   - Full ppgtt support refactoring
   - PSR fixes and improvements
   - Execlist improvments
   - GuC related fixes

  amdgpu:
   - Initial amdgpu documentation
   - JPEG engine support on VCN
   - CIK uses powerplay by default
   - Move to using core PCIE functionality for gens/lanes
   - DC/Powerplay interface rework
   - Stutter mode support for RV
   - Vega12 Powerplay updates
   - GFXOFF fixes
   - GPUVM fault debugging
   - Vega12 GFXOFF
   - DC improvements
   - DC i2c/aux changes
   - UVD 7.2 fixes
   - Powerplay fixes for Polaris12, CZ/ST
   - command submission bo_list fixes

  amdkfd:
   - Raven support
   - Power management fixes

  udl:
   - Cleanups and fixes

  nouveau:
   - misc fixes and cleanups.

  msm:
   - DPU1 support display controller in sdm845
   - GPU coredump support.

  vmwgfx:
   - Atomic modesetting validation fixes
   - Support for multisample surfaces

  armada:
   - Atomic modesetting support completed.

  exynos:
   - IPPv2 fixes
   - Move g2d to component framework
   - Suspend/resume support cleanups
   - Driver cleanups

  imx:
   - CSI configuration improvements
   - Driver cleanups
   - Use atomic suspend/resume helpers
   - ipu-v3 V4L2 XRGB32/XBGR32 support

  pl111:
   - Add Nomadik LCDC variant

  v3d:
   - GPU scheduler jobs management

  sun4i:
   - R40 display engine support
   - TCON TOP driver

  mediatek:
   - MT2712 SoC support

  rockchip:
   - vop fixes

  omapdrm:
   - Workaround for DRA7 errata i932
   - Fix mm_list locking

  mali-dp:
   - Writeback implementation
        PM improvements
   - Internal error reporting debugfs

  tilcdc:
   - Single fix for deferred probing

  hdlcd:
   - Teardown fixes

  tda998x:
   - Converted to a bridge driver.

  etnaviv:
   - Misc fixes"

* tag 'drm-next-2018-08-15' of git://anongit.freedesktop.org/drm/drm: (1506 commits)
  drm/amdgpu/sriov: give 8s for recover vram under RUNTIME
  drm/scheduler: fix param documentation
  drm/i2c: tda998x: correct PLL divider calculation
  drm/i2c: tda998x: get rid of private fill_modes function
  drm/i2c: tda998x: move mode_valid() to bridge
  drm/i2c: tda998x: register bridge outside of component helper
  drm/i2c: tda998x: cleanup from previous changes
  drm/i2c: tda998x: allocate tda998x_priv inside tda998x_create()
  drm/i2c: tda998x: convert to bridge driver
  drm/scheduler: fix timeout worker setup for out of order job completions
  drm/amd/display: display connected to dp-1 does not light up
  drm/amd/display: update clk for various HDMI color depths
  drm/amd/display: program display clock on cache match
  drm/amd/display: Add NULL check for enabling dp ss
  drm/amd/display: add vbios table check for enabling dp ss
  drm/amd/display: Don't share clk source between DP and HDMI
  drm/amd/display: Fix DP HBR2 Eye Diagram Pattern on Carrizo
  drm/amd/display: Use calculated disp_clk_khz value for dce110
  drm/amd/display: Implement custom degamma lut on dcn
  drm/amd/display: Destroy aux_engines only once
  ...

13 files changed:
1  2 
Documentation/devicetree/bindings/display/msm/dsi.txt
Documentation/devicetree/bindings/vendor-prefixes.txt
MAINTAINERS
drivers/gpu/drm/bridge/adv7511/adv7511_drv.c
drivers/gpu/drm/drm_atomic_helper.c
drivers/gpu/drm/i915/i915_pmu.c
drivers/gpu/drm/i915/intel_audio.c
drivers/gpu/drm/i915/intel_lpe_audio.c
drivers/gpu/drm/vc4/vc4_plane.c
drivers/xen/gntdev-dmabuf.c
include/linux/pci.h
kernel/locking/locktorture.c
kernel/printk/printk.c

diff --cc MAINTAINERS
Simple merge
Simple merge
Simple merge
Simple merge
Simple merge
index 589fd92,0000000..cba6b58
mode 100644,000000..100644
--- /dev/null
@@@ -1,857 -1,0 +1,856 @@@
-                                struct device *target_dev,
 +// SPDX-License-Identifier: GPL-2.0
 +
 +/*
 + * Xen dma-buf functionality for gntdev.
 + *
 + * DMA buffer implementation is based on drivers/gpu/drm/drm_prime.c.
 + *
 + * Copyright (c) 2018 Oleksandr Andrushchenko, EPAM Systems Inc.
 + */
 +
 +#include <linux/kernel.h>
 +#include <linux/errno.h>
 +#include <linux/dma-buf.h>
 +#include <linux/slab.h>
 +#include <linux/types.h>
 +#include <linux/uaccess.h>
 +
 +#include <xen/xen.h>
 +#include <xen/grant_table.h>
 +
 +#include "gntdev-common.h"
 +#include "gntdev-dmabuf.h"
 +
 +#ifndef GRANT_INVALID_REF
 +/*
 + * Note on usage of grant reference 0 as invalid grant reference:
 + * grant reference 0 is valid, but never exposed to a driver,
 + * because of the fact it is already in use/reserved by the PV console.
 + */
 +#define GRANT_INVALID_REF     0
 +#endif
 +
 +struct gntdev_dmabuf {
 +      struct gntdev_dmabuf_priv *priv;
 +      struct dma_buf *dmabuf;
 +      struct list_head next;
 +      int fd;
 +
 +      union {
 +              struct {
 +                      /* Exported buffers are reference counted. */
 +                      struct kref refcount;
 +
 +                      struct gntdev_priv *priv;
 +                      struct gntdev_grant_map *map;
 +              } exp;
 +              struct {
 +                      /* Granted references of the imported buffer. */
 +                      grant_ref_t *refs;
 +                      /* Scatter-gather table of the imported buffer. */
 +                      struct sg_table *sgt;
 +                      /* dma-buf attachment of the imported buffer. */
 +                      struct dma_buf_attachment *attach;
 +              } imp;
 +      } u;
 +
 +      /* Number of pages this buffer has. */
 +      int nr_pages;
 +      /* Pages of this buffer. */
 +      struct page **pages;
 +};
 +
 +struct gntdev_dmabuf_wait_obj {
 +      struct list_head next;
 +      struct gntdev_dmabuf *gntdev_dmabuf;
 +      struct completion completion;
 +};
 +
 +struct gntdev_dmabuf_attachment {
 +      struct sg_table *sgt;
 +      enum dma_data_direction dir;
 +};
 +
 +struct gntdev_dmabuf_priv {
 +      /* List of exported DMA buffers. */
 +      struct list_head exp_list;
 +      /* List of wait objects. */
 +      struct list_head exp_wait_list;
 +      /* List of imported DMA buffers. */
 +      struct list_head imp_list;
 +      /* This is the lock which protects dma_buf_xxx lists. */
 +      struct mutex lock;
 +};
 +
 +/* DMA buffer export support. */
 +
 +/* Implementation of wait for exported DMA buffer to be released. */
 +
 +static void dmabuf_exp_release(struct kref *kref);
 +
 +static struct gntdev_dmabuf_wait_obj *
 +dmabuf_exp_wait_obj_new(struct gntdev_dmabuf_priv *priv,
 +                      struct gntdev_dmabuf *gntdev_dmabuf)
 +{
 +      struct gntdev_dmabuf_wait_obj *obj;
 +
 +      obj = kzalloc(sizeof(*obj), GFP_KERNEL);
 +      if (!obj)
 +              return ERR_PTR(-ENOMEM);
 +
 +      init_completion(&obj->completion);
 +      obj->gntdev_dmabuf = gntdev_dmabuf;
 +
 +      mutex_lock(&priv->lock);
 +      list_add(&obj->next, &priv->exp_wait_list);
 +      /* Put our reference and wait for gntdev_dmabuf's release to fire. */
 +      kref_put(&gntdev_dmabuf->u.exp.refcount, dmabuf_exp_release);
 +      mutex_unlock(&priv->lock);
 +      return obj;
 +}
 +
 +static void dmabuf_exp_wait_obj_free(struct gntdev_dmabuf_priv *priv,
 +                                   struct gntdev_dmabuf_wait_obj *obj)
 +{
 +      mutex_lock(&priv->lock);
 +      list_del(&obj->next);
 +      mutex_unlock(&priv->lock);
 +      kfree(obj);
 +}
 +
 +static int dmabuf_exp_wait_obj_wait(struct gntdev_dmabuf_wait_obj *obj,
 +                                  u32 wait_to_ms)
 +{
 +      if (wait_for_completion_timeout(&obj->completion,
 +                      msecs_to_jiffies(wait_to_ms)) <= 0)
 +              return -ETIMEDOUT;
 +
 +      return 0;
 +}
 +
 +static void dmabuf_exp_wait_obj_signal(struct gntdev_dmabuf_priv *priv,
 +                                     struct gntdev_dmabuf *gntdev_dmabuf)
 +{
 +      struct gntdev_dmabuf_wait_obj *obj;
 +
 +      list_for_each_entry(obj, &priv->exp_wait_list, next)
 +              if (obj->gntdev_dmabuf == gntdev_dmabuf) {
 +                      pr_debug("Found gntdev_dmabuf in the wait list, wake\n");
 +                      complete_all(&obj->completion);
 +                      break;
 +              }
 +}
 +
 +static struct gntdev_dmabuf *
 +dmabuf_exp_wait_obj_get_dmabuf(struct gntdev_dmabuf_priv *priv, int fd)
 +{
 +      struct gntdev_dmabuf *gntdev_dmabuf, *ret = ERR_PTR(-ENOENT);
 +
 +      mutex_lock(&priv->lock);
 +      list_for_each_entry(gntdev_dmabuf, &priv->exp_list, next)
 +              if (gntdev_dmabuf->fd == fd) {
 +                      pr_debug("Found gntdev_dmabuf in the wait list\n");
 +                      kref_get(&gntdev_dmabuf->u.exp.refcount);
 +                      ret = gntdev_dmabuf;
 +                      break;
 +              }
 +      mutex_unlock(&priv->lock);
 +      return ret;
 +}
 +
 +static int dmabuf_exp_wait_released(struct gntdev_dmabuf_priv *priv, int fd,
 +                                  int wait_to_ms)
 +{
 +      struct gntdev_dmabuf *gntdev_dmabuf;
 +      struct gntdev_dmabuf_wait_obj *obj;
 +      int ret;
 +
 +      pr_debug("Will wait for dma-buf with fd %d\n", fd);
 +      /*
 +       * Try to find the DMA buffer: if not found means that
 +       * either the buffer has already been released or file descriptor
 +       * provided is wrong.
 +       */
 +      gntdev_dmabuf = dmabuf_exp_wait_obj_get_dmabuf(priv, fd);
 +      if (IS_ERR(gntdev_dmabuf))
 +              return PTR_ERR(gntdev_dmabuf);
 +
 +      /*
 +       * gntdev_dmabuf still exists and is reference count locked by us now,
 +       * so prepare to wait: allocate wait object and add it to the wait list,
 +       * so we can find it on release.
 +       */
 +      obj = dmabuf_exp_wait_obj_new(priv, gntdev_dmabuf);
 +      if (IS_ERR(obj))
 +              return PTR_ERR(obj);
 +
 +      ret = dmabuf_exp_wait_obj_wait(obj, wait_to_ms);
 +      dmabuf_exp_wait_obj_free(priv, obj);
 +      return ret;
 +}
 +
 +/* DMA buffer export support. */
 +
 +static struct sg_table *
 +dmabuf_pages_to_sgt(struct page **pages, unsigned int nr_pages)
 +{
 +      struct sg_table *sgt;
 +      int ret;
 +
 +      sgt = kmalloc(sizeof(*sgt), GFP_KERNEL);
 +      if (!sgt) {
 +              ret = -ENOMEM;
 +              goto out;
 +      }
 +
 +      ret = sg_alloc_table_from_pages(sgt, pages, nr_pages, 0,
 +                                      nr_pages << PAGE_SHIFT,
 +                                      GFP_KERNEL);
 +      if (ret)
 +              goto out;
 +
 +      return sgt;
 +
 +out:
 +      kfree(sgt);
 +      return ERR_PTR(ret);
 +}
 +
 +static int dmabuf_exp_ops_attach(struct dma_buf *dma_buf,
 +                               struct dma_buf_attachment *attach)
 +{
 +      struct gntdev_dmabuf_attachment *gntdev_dmabuf_attach;
 +
 +      gntdev_dmabuf_attach = kzalloc(sizeof(*gntdev_dmabuf_attach),
 +                                     GFP_KERNEL);
 +      if (!gntdev_dmabuf_attach)
 +              return -ENOMEM;
 +
 +      gntdev_dmabuf_attach->dir = DMA_NONE;
 +      attach->priv = gntdev_dmabuf_attach;
 +      return 0;
 +}
 +
 +static void dmabuf_exp_ops_detach(struct dma_buf *dma_buf,
 +                                struct dma_buf_attachment *attach)
 +{
 +      struct gntdev_dmabuf_attachment *gntdev_dmabuf_attach = attach->priv;
 +
 +      if (gntdev_dmabuf_attach) {
 +              struct sg_table *sgt = gntdev_dmabuf_attach->sgt;
 +
 +              if (sgt) {
 +                      if (gntdev_dmabuf_attach->dir != DMA_NONE)
 +                              dma_unmap_sg_attrs(attach->dev, sgt->sgl,
 +                                                 sgt->nents,
 +                                                 gntdev_dmabuf_attach->dir,
 +                                                 DMA_ATTR_SKIP_CPU_SYNC);
 +                      sg_free_table(sgt);
 +              }
 +
 +              kfree(sgt);
 +              kfree(gntdev_dmabuf_attach);
 +              attach->priv = NULL;
 +      }
 +}
 +
 +static struct sg_table *
 +dmabuf_exp_ops_map_dma_buf(struct dma_buf_attachment *attach,
 +                         enum dma_data_direction dir)
 +{
 +      struct gntdev_dmabuf_attachment *gntdev_dmabuf_attach = attach->priv;
 +      struct gntdev_dmabuf *gntdev_dmabuf = attach->dmabuf->priv;
 +      struct sg_table *sgt;
 +
 +      pr_debug("Mapping %d pages for dev %p\n", gntdev_dmabuf->nr_pages,
 +               attach->dev);
 +
 +      if (dir == DMA_NONE || !gntdev_dmabuf_attach)
 +              return ERR_PTR(-EINVAL);
 +
 +      /* Return the cached mapping when possible. */
 +      if (gntdev_dmabuf_attach->dir == dir)
 +              return gntdev_dmabuf_attach->sgt;
 +
 +      /*
 +       * Two mappings with different directions for the same attachment are
 +       * not allowed.
 +       */
 +      if (gntdev_dmabuf_attach->dir != DMA_NONE)
 +              return ERR_PTR(-EBUSY);
 +
 +      sgt = dmabuf_pages_to_sgt(gntdev_dmabuf->pages,
 +                                gntdev_dmabuf->nr_pages);
 +      if (!IS_ERR(sgt)) {
 +              if (!dma_map_sg_attrs(attach->dev, sgt->sgl, sgt->nents, dir,
 +                                    DMA_ATTR_SKIP_CPU_SYNC)) {
 +                      sg_free_table(sgt);
 +                      kfree(sgt);
 +                      sgt = ERR_PTR(-ENOMEM);
 +              } else {
 +                      gntdev_dmabuf_attach->sgt = sgt;
 +                      gntdev_dmabuf_attach->dir = dir;
 +              }
 +      }
 +      if (IS_ERR(sgt))
 +              pr_debug("Failed to map sg table for dev %p\n", attach->dev);
 +      return sgt;
 +}
 +
 +static void dmabuf_exp_ops_unmap_dma_buf(struct dma_buf_attachment *attach,
 +                                       struct sg_table *sgt,
 +                                       enum dma_data_direction dir)
 +{
 +      /* Not implemented. The unmap is done at dmabuf_exp_ops_detach(). */
 +}
 +
 +static void dmabuf_exp_release(struct kref *kref)
 +{
 +      struct gntdev_dmabuf *gntdev_dmabuf =
 +              container_of(kref, struct gntdev_dmabuf, u.exp.refcount);
 +
 +      dmabuf_exp_wait_obj_signal(gntdev_dmabuf->priv, gntdev_dmabuf);
 +      list_del(&gntdev_dmabuf->next);
 +      kfree(gntdev_dmabuf);
 +}
 +
 +static void dmabuf_exp_remove_map(struct gntdev_priv *priv,
 +                                struct gntdev_grant_map *map)
 +{
 +      mutex_lock(&priv->lock);
 +      list_del(&map->next);
 +      gntdev_put_map(NULL /* already removed */, map);
 +      mutex_unlock(&priv->lock);
 +}
 +
 +static void dmabuf_exp_ops_release(struct dma_buf *dma_buf)
 +{
 +      struct gntdev_dmabuf *gntdev_dmabuf = dma_buf->priv;
 +      struct gntdev_dmabuf_priv *priv = gntdev_dmabuf->priv;
 +
 +      dmabuf_exp_remove_map(gntdev_dmabuf->u.exp.priv,
 +                            gntdev_dmabuf->u.exp.map);
 +      mutex_lock(&priv->lock);
 +      kref_put(&gntdev_dmabuf->u.exp.refcount, dmabuf_exp_release);
 +      mutex_unlock(&priv->lock);
 +}
 +
 +static void *dmabuf_exp_ops_kmap(struct dma_buf *dma_buf,
 +                               unsigned long page_num)
 +{
 +      /* Not implemented. */
 +      return NULL;
 +}
 +
 +static void dmabuf_exp_ops_kunmap(struct dma_buf *dma_buf,
 +                                unsigned long page_num, void *addr)
 +{
 +      /* Not implemented. */
 +}
 +
 +static int dmabuf_exp_ops_mmap(struct dma_buf *dma_buf,
 +                             struct vm_area_struct *vma)
 +{
 +      /* Not implemented. */
 +      return 0;
 +}
 +
 +static const struct dma_buf_ops dmabuf_exp_ops =  {
 +      .attach = dmabuf_exp_ops_attach,
 +      .detach = dmabuf_exp_ops_detach,
 +      .map_dma_buf = dmabuf_exp_ops_map_dma_buf,
 +      .unmap_dma_buf = dmabuf_exp_ops_unmap_dma_buf,
 +      .release = dmabuf_exp_ops_release,
 +      .map = dmabuf_exp_ops_kmap,
 +      .unmap = dmabuf_exp_ops_kunmap,
 +      .mmap = dmabuf_exp_ops_mmap,
 +};
 +
 +struct gntdev_dmabuf_export_args {
 +      struct gntdev_priv *priv;
 +      struct gntdev_grant_map *map;
 +      struct gntdev_dmabuf_priv *dmabuf_priv;
 +      struct device *dev;
 +      int count;
 +      struct page **pages;
 +      u32 fd;
 +};
 +
 +static int dmabuf_exp_from_pages(struct gntdev_dmabuf_export_args *args)
 +{
 +      DEFINE_DMA_BUF_EXPORT_INFO(exp_info);
 +      struct gntdev_dmabuf *gntdev_dmabuf;
 +      int ret;
 +
 +      gntdev_dmabuf = kzalloc(sizeof(*gntdev_dmabuf), GFP_KERNEL);
 +      if (!gntdev_dmabuf)
 +              return -ENOMEM;
 +
 +      kref_init(&gntdev_dmabuf->u.exp.refcount);
 +
 +      gntdev_dmabuf->priv = args->dmabuf_priv;
 +      gntdev_dmabuf->nr_pages = args->count;
 +      gntdev_dmabuf->pages = args->pages;
 +      gntdev_dmabuf->u.exp.priv = args->priv;
 +      gntdev_dmabuf->u.exp.map = args->map;
 +
 +      exp_info.exp_name = KBUILD_MODNAME;
 +      if (args->dev->driver && args->dev->driver->owner)
 +              exp_info.owner = args->dev->driver->owner;
 +      else
 +              exp_info.owner = THIS_MODULE;
 +      exp_info.ops = &dmabuf_exp_ops;
 +      exp_info.size = args->count << PAGE_SHIFT;
 +      exp_info.flags = O_RDWR;
 +      exp_info.priv = gntdev_dmabuf;
 +
 +      gntdev_dmabuf->dmabuf = dma_buf_export(&exp_info);
 +      if (IS_ERR(gntdev_dmabuf->dmabuf)) {
 +              ret = PTR_ERR(gntdev_dmabuf->dmabuf);
 +              gntdev_dmabuf->dmabuf = NULL;
 +              goto fail;
 +      }
 +
 +      ret = dma_buf_fd(gntdev_dmabuf->dmabuf, O_CLOEXEC);
 +      if (ret < 0)
 +              goto fail;
 +
 +      gntdev_dmabuf->fd = ret;
 +      args->fd = ret;
 +
 +      pr_debug("Exporting DMA buffer with fd %d\n", ret);
 +
 +      mutex_lock(&args->dmabuf_priv->lock);
 +      list_add(&gntdev_dmabuf->next, &args->dmabuf_priv->exp_list);
 +      mutex_unlock(&args->dmabuf_priv->lock);
 +      return 0;
 +
 +fail:
 +      if (gntdev_dmabuf->dmabuf)
 +              dma_buf_put(gntdev_dmabuf->dmabuf);
 +      kfree(gntdev_dmabuf);
 +      return ret;
 +}
 +
 +static struct gntdev_grant_map *
 +dmabuf_exp_alloc_backing_storage(struct gntdev_priv *priv, int dmabuf_flags,
 +                               int count)
 +{
 +      struct gntdev_grant_map *map;
 +
 +      if (unlikely(count <= 0))
 +              return ERR_PTR(-EINVAL);
 +
 +      if ((dmabuf_flags & GNTDEV_DMA_FLAG_WC) &&
 +          (dmabuf_flags & GNTDEV_DMA_FLAG_COHERENT)) {
 +              pr_debug("Wrong dma-buf flags: 0x%x\n", dmabuf_flags);
 +              return ERR_PTR(-EINVAL);
 +      }
 +
 +      map = gntdev_alloc_map(priv, count, dmabuf_flags);
 +      if (!map)
 +              return ERR_PTR(-ENOMEM);
 +
 +      if (unlikely(gntdev_account_mapped_pages(count))) {
 +              pr_debug("can't map %d pages: over limit\n", count);
 +              gntdev_put_map(NULL, map);
 +              return ERR_PTR(-ENOMEM);
 +      }
 +      return map;
 +}
 +
 +static int dmabuf_exp_from_refs(struct gntdev_priv *priv, int flags,
 +                              int count, u32 domid, u32 *refs, u32 *fd)
 +{
 +      struct gntdev_grant_map *map;
 +      struct gntdev_dmabuf_export_args args;
 +      int i, ret;
 +
 +      map = dmabuf_exp_alloc_backing_storage(priv, flags, count);
 +      if (IS_ERR(map))
 +              return PTR_ERR(map);
 +
 +      for (i = 0; i < count; i++) {
 +              map->grants[i].domid = domid;
 +              map->grants[i].ref = refs[i];
 +      }
 +
 +      mutex_lock(&priv->lock);
 +      gntdev_add_map(priv, map);
 +      mutex_unlock(&priv->lock);
 +
 +      map->flags |= GNTMAP_host_map;
 +#if defined(CONFIG_X86)
 +      map->flags |= GNTMAP_device_map;
 +#endif
 +
 +      ret = gntdev_map_grant_pages(map);
 +      if (ret < 0)
 +              goto out;
 +
 +      args.priv = priv;
 +      args.map = map;
 +      args.dev = priv->dma_dev;
 +      args.dmabuf_priv = priv->dmabuf_priv;
 +      args.count = map->count;
 +      args.pages = map->pages;
 +      args.fd = -1; /* Shut up unnecessary gcc warning for i386 */
 +
 +      ret = dmabuf_exp_from_pages(&args);
 +      if (ret < 0)
 +              goto out;
 +
 +      *fd = args.fd;
 +      return 0;
 +
 +out:
 +      dmabuf_exp_remove_map(priv, map);
 +      return ret;
 +}
 +
 +/* DMA buffer import support. */
 +
 +static int
 +dmabuf_imp_grant_foreign_access(struct page **pages, u32 *refs,
 +                              int count, int domid)
 +{
 +      grant_ref_t priv_gref_head;
 +      int i, ret;
 +
 +      ret = gnttab_alloc_grant_references(count, &priv_gref_head);
 +      if (ret < 0) {
 +              pr_debug("Cannot allocate grant references, ret %d\n", ret);
 +              return ret;
 +      }
 +
 +      for (i = 0; i < count; i++) {
 +              int cur_ref;
 +
 +              cur_ref = gnttab_claim_grant_reference(&priv_gref_head);
 +              if (cur_ref < 0) {
 +                      ret = cur_ref;
 +                      pr_debug("Cannot claim grant reference, ret %d\n", ret);
 +                      goto out;
 +              }
 +
 +              gnttab_grant_foreign_access_ref(cur_ref, domid,
 +                                              xen_page_to_gfn(pages[i]), 0);
 +              refs[i] = cur_ref;
 +      }
 +
 +      return 0;
 +
 +out:
 +      gnttab_free_grant_references(priv_gref_head);
 +      return ret;
 +}
 +
 +static void dmabuf_imp_end_foreign_access(u32 *refs, int count)
 +{
 +      int i;
 +
 +      for (i = 0; i < count; i++)
 +              if (refs[i] != GRANT_INVALID_REF)
 +                      gnttab_end_foreign_access(refs[i], 0, 0UL);
 +}
 +
 +static void dmabuf_imp_free_storage(struct gntdev_dmabuf *gntdev_dmabuf)
 +{
 +      kfree(gntdev_dmabuf->pages);
 +      kfree(gntdev_dmabuf->u.imp.refs);
 +      kfree(gntdev_dmabuf);
 +}
 +
 +static struct gntdev_dmabuf *dmabuf_imp_alloc_storage(int count)
 +{
 +      struct gntdev_dmabuf *gntdev_dmabuf;
 +      int i;
 +
 +      gntdev_dmabuf = kzalloc(sizeof(*gntdev_dmabuf), GFP_KERNEL);
 +      if (!gntdev_dmabuf)
 +              goto fail_no_free;
 +
 +      gntdev_dmabuf->u.imp.refs = kcalloc(count,
 +                                          sizeof(gntdev_dmabuf->u.imp.refs[0]),
 +                                          GFP_KERNEL);
 +      if (!gntdev_dmabuf->u.imp.refs)
 +              goto fail;
 +
 +      gntdev_dmabuf->pages = kcalloc(count,
 +                                     sizeof(gntdev_dmabuf->pages[0]),
 +                                     GFP_KERNEL);
 +      if (!gntdev_dmabuf->pages)
 +              goto fail;
 +
 +      gntdev_dmabuf->nr_pages = count;
 +
 +      for (i = 0; i < count; i++)
 +              gntdev_dmabuf->u.imp.refs[i] = GRANT_INVALID_REF;
 +
 +      return gntdev_dmabuf;
 +
 +fail:
 +      dmabuf_imp_free_storage(gntdev_dmabuf);
 +fail_no_free:
 +      return ERR_PTR(-ENOMEM);
 +}
 +
 +static struct gntdev_dmabuf *
 +dmabuf_imp_to_refs(struct gntdev_dmabuf_priv *priv, struct device *dev,
 +                 int fd, int count, int domid)
 +{
 +      struct gntdev_dmabuf *gntdev_dmabuf, *ret;
 +      struct dma_buf *dma_buf;
 +      struct dma_buf_attachment *attach;
 +      struct sg_table *sgt;
 +      struct sg_page_iter sg_iter;
 +      int i;
 +
 +      dma_buf = dma_buf_get(fd);
 +      if (IS_ERR(dma_buf))
 +              return ERR_CAST(dma_buf);
 +
 +      gntdev_dmabuf = dmabuf_imp_alloc_storage(count);
 +      if (IS_ERR(gntdev_dmabuf)) {
 +              ret = gntdev_dmabuf;
 +              goto fail_put;
 +      }
 +
 +      gntdev_dmabuf->priv = priv;
 +      gntdev_dmabuf->fd = fd;
 +
 +      attach = dma_buf_attach(dma_buf, dev);
 +      if (IS_ERR(attach)) {
 +              ret = ERR_CAST(attach);
 +              goto fail_free_obj;
 +      }
 +
 +      gntdev_dmabuf->u.imp.attach = attach;
 +
 +      sgt = dma_buf_map_attachment(attach, DMA_BIDIRECTIONAL);
 +      if (IS_ERR(sgt)) {
 +              ret = ERR_CAST(sgt);
 +              goto fail_detach;
 +      }
 +
 +      /* Check number of pages that imported buffer has. */
 +      if (attach->dmabuf->size != gntdev_dmabuf->nr_pages << PAGE_SHIFT) {
 +              ret = ERR_PTR(-EINVAL);
 +              pr_debug("DMA buffer has %zu pages, user-space expects %d\n",
 +                       attach->dmabuf->size, gntdev_dmabuf->nr_pages);
 +              goto fail_unmap;
 +      }
 +
 +      gntdev_dmabuf->u.imp.sgt = sgt;
 +
 +      /* Now convert sgt to array of pages and check for page validity. */
 +      i = 0;
 +      for_each_sg_page(sgt->sgl, &sg_iter, sgt->nents, 0) {
 +              struct page *page = sg_page_iter_page(&sg_iter);
 +              /*
 +               * Check if page is valid: this can happen if we are given
 +               * a page from VRAM or other resources which are not backed
 +               * by a struct page.
 +               */
 +              if (!pfn_valid(page_to_pfn(page))) {
 +                      ret = ERR_PTR(-EINVAL);
 +                      goto fail_unmap;
 +              }
 +
 +              gntdev_dmabuf->pages[i++] = page;
 +      }
 +
 +      ret = ERR_PTR(dmabuf_imp_grant_foreign_access(gntdev_dmabuf->pages,
 +                                                    gntdev_dmabuf->u.imp.refs,
 +                                                    count, domid));
 +      if (IS_ERR(ret))
 +              goto fail_end_access;
 +
 +      pr_debug("Imported DMA buffer with fd %d\n", fd);
 +
 +      mutex_lock(&priv->lock);
 +      list_add(&gntdev_dmabuf->next, &priv->imp_list);
 +      mutex_unlock(&priv->lock);
 +
 +      return gntdev_dmabuf;
 +
 +fail_end_access:
 +      dmabuf_imp_end_foreign_access(gntdev_dmabuf->u.imp.refs, count);
 +fail_unmap:
 +      dma_buf_unmap_attachment(attach, sgt, DMA_BIDIRECTIONAL);
 +fail_detach:
 +      dma_buf_detach(dma_buf, attach);
 +fail_free_obj:
 +      dmabuf_imp_free_storage(gntdev_dmabuf);
 +fail_put:
 +      dma_buf_put(dma_buf);
 +      return ret;
 +}
 +
 +/*
 + * Find the hyper dma-buf by its file descriptor and remove
 + * it from the buffer's list.
 + */
 +static struct gntdev_dmabuf *
 +dmabuf_imp_find_unlink(struct gntdev_dmabuf_priv *priv, int fd)
 +{
 +      struct gntdev_dmabuf *q, *gntdev_dmabuf, *ret = ERR_PTR(-ENOENT);
 +
 +      mutex_lock(&priv->lock);
 +      list_for_each_entry_safe(gntdev_dmabuf, q, &priv->imp_list, next) {
 +              if (gntdev_dmabuf->fd == fd) {
 +                      pr_debug("Found gntdev_dmabuf in the import list\n");
 +                      ret = gntdev_dmabuf;
 +                      list_del(&gntdev_dmabuf->next);
 +                      break;
 +              }
 +      }
 +      mutex_unlock(&priv->lock);
 +      return ret;
 +}
 +
 +static int dmabuf_imp_release(struct gntdev_dmabuf_priv *priv, u32 fd)
 +{
 +      struct gntdev_dmabuf *gntdev_dmabuf;
 +      struct dma_buf_attachment *attach;
 +      struct dma_buf *dma_buf;
 +
 +      gntdev_dmabuf = dmabuf_imp_find_unlink(priv, fd);
 +      if (IS_ERR(gntdev_dmabuf))
 +              return PTR_ERR(gntdev_dmabuf);
 +
 +      pr_debug("Releasing DMA buffer with fd %d\n", fd);
 +
 +      dmabuf_imp_end_foreign_access(gntdev_dmabuf->u.imp.refs,
 +                                    gntdev_dmabuf->nr_pages);
 +
 +      attach = gntdev_dmabuf->u.imp.attach;
 +
 +      if (gntdev_dmabuf->u.imp.sgt)
 +              dma_buf_unmap_attachment(attach, gntdev_dmabuf->u.imp.sgt,
 +                                       DMA_BIDIRECTIONAL);
 +      dma_buf = attach->dmabuf;
 +      dma_buf_detach(attach->dmabuf, attach);
 +      dma_buf_put(dma_buf);
 +
 +      dmabuf_imp_free_storage(gntdev_dmabuf);
 +      return 0;
 +}
 +
 +/* DMA buffer IOCTL support. */
 +
 +long gntdev_ioctl_dmabuf_exp_from_refs(struct gntdev_priv *priv, int use_ptemod,
 +                                     struct ioctl_gntdev_dmabuf_exp_from_refs __user *u)
 +{
 +      struct ioctl_gntdev_dmabuf_exp_from_refs op;
 +      u32 *refs;
 +      long ret;
 +
 +      if (use_ptemod) {
 +              pr_debug("Cannot provide dma-buf: use_ptemode %d\n",
 +                       use_ptemod);
 +              return -EINVAL;
 +      }
 +
 +      if (copy_from_user(&op, u, sizeof(op)) != 0)
 +              return -EFAULT;
 +
 +      if (unlikely(op.count <= 0))
 +              return -EINVAL;
 +
 +      refs = kcalloc(op.count, sizeof(*refs), GFP_KERNEL);
 +      if (!refs)
 +              return -ENOMEM;
 +
 +      if (copy_from_user(refs, u->refs, sizeof(*refs) * op.count) != 0) {
 +              ret = -EFAULT;
 +              goto out;
 +      }
 +
 +      ret = dmabuf_exp_from_refs(priv, op.flags, op.count,
 +                                 op.domid, refs, &op.fd);
 +      if (ret)
 +              goto out;
 +
 +      if (copy_to_user(u, &op, sizeof(op)) != 0)
 +              ret = -EFAULT;
 +
 +out:
 +      kfree(refs);
 +      return ret;
 +}
 +
 +long gntdev_ioctl_dmabuf_exp_wait_released(struct gntdev_priv *priv,
 +                                         struct ioctl_gntdev_dmabuf_exp_wait_released __user *u)
 +{
 +      struct ioctl_gntdev_dmabuf_exp_wait_released op;
 +
 +      if (copy_from_user(&op, u, sizeof(op)) != 0)
 +              return -EFAULT;
 +
 +      return dmabuf_exp_wait_released(priv->dmabuf_priv, op.fd,
 +                                      op.wait_to_ms);
 +}
 +
 +long gntdev_ioctl_dmabuf_imp_to_refs(struct gntdev_priv *priv,
 +                                   struct ioctl_gntdev_dmabuf_imp_to_refs __user *u)
 +{
 +      struct ioctl_gntdev_dmabuf_imp_to_refs op;
 +      struct gntdev_dmabuf *gntdev_dmabuf;
 +      long ret;
 +
 +      if (copy_from_user(&op, u, sizeof(op)) != 0)
 +              return -EFAULT;
 +
 +      if (unlikely(op.count <= 0))
 +              return -EINVAL;
 +
 +      gntdev_dmabuf = dmabuf_imp_to_refs(priv->dmabuf_priv,
 +                                         priv->dma_dev, op.fd,
 +                                         op.count, op.domid);
 +      if (IS_ERR(gntdev_dmabuf))
 +              return PTR_ERR(gntdev_dmabuf);
 +
 +      if (copy_to_user(u->refs, gntdev_dmabuf->u.imp.refs,
 +                       sizeof(*u->refs) * op.count) != 0) {
 +              ret = -EFAULT;
 +              goto out_release;
 +      }
 +      return 0;
 +
 +out_release:
 +      dmabuf_imp_release(priv->dmabuf_priv, op.fd);
 +      return ret;
 +}
 +
 +long gntdev_ioctl_dmabuf_imp_release(struct gntdev_priv *priv,
 +                                   struct ioctl_gntdev_dmabuf_imp_release __user *u)
 +{
 +      struct ioctl_gntdev_dmabuf_imp_release op;
 +
 +      if (copy_from_user(&op, u, sizeof(op)) != 0)
 +              return -EFAULT;
 +
 +      return dmabuf_imp_release(priv->dmabuf_priv, op.fd);
 +}
 +
 +struct gntdev_dmabuf_priv *gntdev_dmabuf_init(void)
 +{
 +      struct gntdev_dmabuf_priv *priv;
 +
 +      priv = kzalloc(sizeof(*priv), GFP_KERNEL);
 +      if (!priv)
 +              return ERR_PTR(-ENOMEM);
 +
 +      mutex_init(&priv->lock);
 +      INIT_LIST_HEAD(&priv->exp_list);
 +      INIT_LIST_HEAD(&priv->exp_wait_list);
 +      INIT_LIST_HEAD(&priv->imp_list);
 +
 +      return priv;
 +}
 +
 +void gntdev_dmabuf_fini(struct gntdev_dmabuf_priv *priv)
 +{
 +      kfree(priv);
 +}
Simple merge
Simple merge
Simple merge