[RFC PATCH v2 3/3] drm/i915/gvt: Enable vGPU direct flip

Zhang, Tina tina.zhang at intel.com
Wed May 15 05:12:37 UTC 2019



> -----Original Message-----
> From: Zhenyu Wang [mailto:zhenyuw at linux.intel.com]
> Sent: Monday, May 13, 2019 11:38 AM
> To: Zhang, Tina <tina.zhang at intel.com>
> Cc: kraxel at redhat.com; zhenyuw at linux.intel.com; Lv, Zhiyuan
> <zhiyuan.lv at intel.com>; Wang, Zhi A <zhi.a.wang at intel.com>; Tian, Kevin
> <kevin.tian at intel.com>; daniel at ffwll.ch; Kondapally, Kalyan
> <kalyan.kondapally at intel.com>; Yuan, Hang <hang.yuan at intel.com>;
> ville.syrjala at linux.intel.com; intel-gvt-dev at lists.freedesktop.org
> Subject: Re: [RFC PATCH v2 3/3] drm/i915/gvt: Enable vGPU direct flip
> 
> On 2019.05.10 12:15:28 +0800, Tina Zhang wrote:
> > When vGPU does page flip, display model flips the shadow framebuffer
> > to the assigned hardware plane through drm display framework.
> >
> > Signed-off-by: Tina Zhang <tina.zhang at intel.com>
> > ---
> >  drivers/gpu/drm/i915/gvt/dmabuf.c   |   8 +-
> >  drivers/gpu/drm/i915/gvt/dmabuf.h   |   6 +
> >  drivers/gpu/drm/i915/gvt/handlers.c | 332
> > ++++++++++++++++++++++++++++++++++++
> >  3 files changed, 345 insertions(+), 1 deletion(-)
> >
> > diff --git a/drivers/gpu/drm/i915/gvt/dmabuf.c
> > b/drivers/gpu/drm/i915/gvt/dmabuf.c
> > index e9adcd4..b6d9317 100644
> > --- a/drivers/gpu/drm/i915/gvt/dmabuf.c
> > +++ b/drivers/gpu/drm/i915/gvt/dmabuf.c
> > @@ -201,7 +201,7 @@ static bool validate_hotspot(struct
> intel_vgpu_cursor_plane_format *c)
> >  		return false;
> >  }
> >
> > -static int vgpu_get_plane_info(struct drm_device *dev,
> > +int vgpu_get_plane_info(struct drm_device *dev,
> >  		struct intel_vgpu *vgpu,
> >  		struct intel_vgpu_fb_info *info,
> >  		int plane_id)
> > @@ -221,6 +221,8 @@ static int vgpu_get_plane_info(struct drm_device
> *dev,
> >  		info->height = p.height;
> >  		info->stride = p.stride;
> >  		info->drm_format = p.drm_format;
> > +		info->src_x = p.x_offset;
> > +		info->src_y = p.y_offset;
> >
> >  		switch (p.tiled) {
> >  		case PLANE_CTL_TILED_LINEAR:
> > @@ -289,6 +291,8 @@ static int vgpu_get_plane_info(struct drm_device
> *dev,
> >  		return -EFAULT;
> >  	}
> >
> > +	info->vgpu_id = vgpu->id;
> > +
> >  	return 0;
> >  }
> >
> > @@ -406,6 +410,8 @@ int intel_vgpu_query_plane(struct intel_vgpu
> > *vgpu, void *args)
> >
> >  	mutex_unlock(&vgpu->dmabuf_lock);
> >
> > +	fb_info.vgpu_id = vgpu->id;
> > +
> >  	/* Need to allocate a new one*/
> >  	dmabuf_obj = kmalloc(sizeof(struct intel_vgpu_dmabuf_obj),
> GFP_KERNEL);
> >  	if (unlikely(!dmabuf_obj)) {
> > diff --git a/drivers/gpu/drm/i915/gvt/dmabuf.h
> > b/drivers/gpu/drm/i915/gvt/dmabuf.h
> > index cf97eb4..184445b 100644
> > --- a/drivers/gpu/drm/i915/gvt/dmabuf.h
> > +++ b/drivers/gpu/drm/i915/gvt/dmabuf.h
> > @@ -46,6 +46,8 @@ struct intel_vgpu_fb_info {
> >  	__u32 x_hot;    /* horizontal position of cursor hotspot */
> >  	__u32 y_hot;    /* vertical position of cursor hotspot */
> >  	__u32 vgpu_id;
> > +	__u32 src_x;
> > +	__u32 src_y;
> >  	struct intel_vgpu_dmabuf_obj *obj;
> >  };
> >
> > @@ -66,4 +68,8 @@ int intel_vgpu_get_dmabuf(struct intel_vgpu *vgpu,
> > unsigned int dmabuf_id);  void intel_vgpu_dmabuf_cleanup(struct
> > intel_vgpu *vgpu);  struct drm_i915_gem_object *vgpu_create_gem(struct
> drm_device *dev,
> >  					    struct intel_vgpu_fb_info *info);
> > +int vgpu_get_plane_info(struct drm_device *dev,
> > +			struct intel_vgpu *vgpu,
> > +			struct intel_vgpu_fb_info *info,
> > +			int plane_id);
> >  #endif
> > diff --git a/drivers/gpu/drm/i915/gvt/handlers.c
> > b/drivers/gpu/drm/i915/gvt/handlers.c
> > index 18f01ee..698a72d 100644
> > --- a/drivers/gpu/drm/i915/gvt/handlers.c
> > +++ b/drivers/gpu/drm/i915/gvt/handlers.c
> > @@ -36,6 +36,8 @@
> >
> >   */
> >
> > +#include <drm/drm_modeset_helper.h>
> > +#include <drm/drm_atomic_uapi.h>
> >  #include "i915_drv.h"
> >  #include "gvt.h"
> >  #include "i915_pvinfo.h"
> > @@ -743,6 +745,334 @@ static int south_chicken2_mmio_write(struct
> intel_vgpu *vgpu,
> >  	return 0;
> >  }
> >
> > +static int vgpu_fb_update(struct drm_device *dev,
> > +			 struct intel_vgpu *vgpu,
> > +			 int plane_id,
> > +			 struct drm_mode_fb_cmd2 *mode_cmd) {
> > +	struct drm_i915_gem_object *obj;
> > +	struct intel_vgpu_fb_info info;
> > +	int ret;
> > +
> > +	ret = vgpu_get_plane_info(dev, vgpu, &info, plane_id);
> > +	if (ret || !info.size || !info.drm_format)
> > +		return ret;
> > +
> > +	if (plane_id == DRM_PLANE_TYPE_PRIMARY)
> > +		obj = vgpu-
> >display.shadow_fbs.shadow_obj[PRIMARY_PLANE];
> > +	else if (plane_id == DRM_PLANE_TYPE_CURSOR)
> > +		obj = vgpu-
> >display.shadow_fbs.shadow_obj[CURSOR_PLANE];
> > +	else
> > +		obj = vgpu->display.shadow_fbs.shadow_obj[SPRITE_PLANE];
> > +
> > +	if (obj == NULL)
> > +		return -ENODEV;
> > +
> > +	memcpy(obj->gvt_info, &info, sizeof(info));
> > +
> > +	mode_cmd->pixel_format = info.drm_format;
> > +	mode_cmd->width = info.width;
> > +	mode_cmd->height = info.height;
> > +	mode_cmd->handles[0] = 0;
> > +	mode_cmd->offsets[0] = 0;
> > +	mode_cmd->pitches[0] = info.stride;
> > +	mode_cmd->flags = DRM_MODE_FB_MODIFIERS;
> > +	mode_cmd->modifier[0] = info.drm_format_mod;
> > +
> > +	return ret;
> > +}
> > +
> > +static int check_vgpu_plane_assignment_status(struct intel_vgpu *vgpu,
> > +					      enum plane_id plane_id,
> > +					      struct drm_crtc **crtc,
> > +					      struct drm_plane **drm_plane,
> > +					      struct drm_plane
> **cur_drm_plane) {
> > +	struct drm_device *dev = &vgpu->gvt->dev_priv->drm;
> > +	struct drm_plane *plane;
> > +	struct intel_plane *intel_plane;
> > +	struct drm_mode_object *obj_fb;
> > +	struct intel_framebuffer *intel_fb;
> > +	struct drm_gem_object *gem_obj;
> > +	struct drm_i915_gem_object *i915_gem_obj;
> > +	struct intel_vgpu_fb_info *fb_info;
> > +	u32 fb_id = 0;
> > +	int ret = 0;
> > +
> > +	*crtc = NULL;
> > +	*drm_plane = NULL;
> > +
> > +	drm_for_each_plane(plane, dev) {
> > +		intel_plane = to_intel_plane(plane);
> > +		if (intel_plane->id == plane_id) {
> > +			if (plane->state && plane->state->fb)
> > +				fb_id = plane->state->fb->base.id;
> > +			else if (!plane->state && plane->fb)
> > +				fb_id = plane->fb->base.id;
> > +			else
> > +				continue;
> > +
> > +			obj_fb = idr_find(&dev->mode_config.object_idr,
> fb_id);
> > +			intel_fb = to_intel_framebuffer(obj_to_fb(obj_fb));
> > +			gem_obj = intel_fb->base.obj[0];
> > +			i915_gem_obj = to_intel_bo(gem_obj);
> > +			fb_info = i915_gem_obj->gvt_info;
> > +
> > +			if (i915_gem_object_is_proxy(i915_gem_obj) &&
> > +			    fb_info && fb_info->vgpu_id == vgpu->id &&
> > +			    plane->type == DRM_PLANE_TYPE_PRIMARY) {
> > +				if (plane->state && plane->state->crtc) {
> > +					*crtc = plane->state->crtc;
> > +					*drm_plane = plane;
> > +					*cur_drm_plane =
> > +						plane->state->crtc->cursor;
> > +					break;
> > +				} else if (!plane->state && plane->crtc) {
> > +					*crtc = plane->crtc;
> > +					*drm_plane = plane;
> > +					*cur_drm_plane = plane->crtc-
> >cursor;
> > +					break;
> > +				}
> 
> Instead of walking all planes every time, why not let i915 display to provide a
> helper to get vGPU plane assigned pipe info? which can be got from fb's gem
> object to check its gvt_info.
That could be another option.

> 
> > +			}
> > +		}
> > +	}
> > +
> > +	if (fb_id == 0 || *crtc == NULL || *drm_plane == NULL)
> > +		ret = -1;
> > +
> > +	return ret;
> > +}
> > +
> > +static void __i915_gem_object_reset_page_iter(struct
> > +drm_i915_gem_object *obj) {
> > +	struct radix_tree_iter iter;
> > +	void __rcu **slot;
> > +
> > +	rcu_read_lock();
> > +	radix_tree_for_each_slot(slot, &obj->mm.get_page.radix, &iter, 0)
> > +		radix_tree_delete(&obj->mm.get_page.radix, iter.index);
> > +	rcu_read_unlock();
> > +}
> > +
> > +static struct sg_table *
> > +__i915_gem_object_unset_pages(struct drm_i915_gem_object *obj) {
> 
> Copy i915 internal function is not good..and this is too hacking. I think when
> new display flip, you need to unreference old buffer and create new one to
> reference new backing pages.
OK. Thanks.

BR,
Tina
> 
> > +	struct drm_i915_private *i915 = to_i915(obj->base.dev);
> > +	struct sg_table *pages;
> > +
> > +	pages = fetch_and_zero(&obj->mm.pages);
> > +	if (IS_ERR_OR_NULL(pages))
> > +		return pages;
> > +
> > +	spin_lock(&i915->mm.obj_lock);
> > +	list_del(&obj->mm.link);
> > +	spin_unlock(&i915->mm.obj_lock);
> > +
> > +	if (obj->mm.mapping) {
> > +		void *ptr;
> > +
> > +		ptr = page_mask_bits(obj->mm.mapping);
> > +		if (is_vmalloc_addr(ptr))
> > +			vunmap(ptr);
> > +		else
> > +			kunmap(kmap_to_page(ptr));
> > +
> > +		obj->mm.mapping = NULL;
> > +	}
> > +
> > +	__i915_gem_object_reset_page_iter(obj);
> > +	obj->mm.page_sizes.phys = obj->mm.page_sizes.sg = 0;
> > +
> > +	return pages;
> > +}
> > +
> > +static int handle_vgpu_page_flip(struct intel_vgpu *vgpu, int pipe,
> > +				 enum plane_id plane_id)
> > +{
> > +       /* struct drm_i915_private *dev_priv = vgpu->gvt->dev_priv; */
> > +       /* struct intel_vgpu_plane_info vgpu_plane_info; */
> > +	struct drm_device *drm = &vgpu->gvt->dev_priv->drm;
> > +	struct drm_modeset_acquire_ctx ctx;
> > +	struct drm_crtc *crtc;
> > +	struct drm_plane *plane, *cur_plane;
> > +	struct drm_framebuffer *pri_fb, *cur_fb;
> > +	struct drm_i915_gem_object *pri_obj, *cur_obj;
> > +	struct drm_atomic_state *state;
> > +	struct drm_plane_state *plane_state = NULL, *cur_plane_state =
> NULL;
> > +	struct drm_mode_fb_cmd2 mode_cmd;
> > +	struct intel_vgpu_fb_info *fb_info;
> > +	int ret = 0;
> > +
> > +	if (!vgpu->display.shadow_fbs.enable_direct_flip)
> > +		return ret;
> > +
> > +	drm_modeset_acquire_init(&ctx,
> DRM_MODESET_ACQUIRE_INTERRUPTIBLE);
> > +retry:
> > +	ret = drm_modeset_lock_all_ctx(drm, &ctx);
> > +	if (ret)
> > +		goto fail;
> > +
> > +	/* 1. Check assignment status */
> > +	ret = check_vgpu_plane_assignment_status(vgpu, plane_id, &crtc,
> > +						 &plane, &cur_plane);
> > +	if (ret) {
> > +		DRM_DEBUG_KMS("%s, vgpu plane %d has no assgined
> plane\n",
> > +			      __func__, plane_id);
> > +		ret = -EINVAL;
> > +		goto fail;
> > +	}
> > +	/* 2. Update shadow primary plane */
> > +	if (plane_id == PLANE_PRIMARY) {
> > +		pri_fb = vgpu-
> >display.shadow_fbs.shadow_fb[PRIMARY_PLANE];
> > +		pri_obj = vgpu-
> >display.shadow_fbs.shadow_obj[PRIMARY_PLANE];
> > +
> > +		fb_info = pri_obj->gvt_info;
> > +
> > +		state = drm_atomic_state_alloc(drm);
> > +		if (!state)
> > +			goto fail;
> > +
> > +		state->acquire_ctx = &ctx;
> > +		plane_state = drm_atomic_get_plane_state(state, plane);
> > +		if (IS_ERR(plane_state)) {
> > +			ret = PTR_ERR(plane_state);
> > +			goto out;
> > +		}
> > +
> > +		ret = vgpu_fb_update(drm, vgpu,
> DRM_PLANE_TYPE_PRIMARY, &mode_cmd);
> > +		if (ret) {
> > +			DRM_DEBUG_KMS("vgpu_fb_update failed\n");
> > +			goto out;
> > +		}
> > +
> > +		drm_helper_mode_fill_fb_struct(drm, pri_fb, &mode_cmd);
> > +		ret = drm_atomic_set_crtc_for_plane(plane_state, crtc);
> > +		if (ret != 0)
> > +			goto out;
> > +		if (plane_state->fb != pri_fb)
> > +			drm_atomic_set_fb_for_plane(plane_state, pri_fb);
> > +
> > +		plane_state->src_x = fb_info->src_x;
> > +		plane_state->src_y = fb_info->src_y;
> > +		plane_state->src_w = mode_cmd.width << 16;
> > +		plane_state->src_h = mode_cmd.height << 16;
> > +
> > +		/* Don't allow full modeset */
> > +		state->allow_modeset = false;
> > +
> > +		if (i915_gem_object_has_pinned_pages(pri_obj)) {
> > +			struct sg_table *pages;
> > +			struct i915_vma *vma;
> > +			struct i915_address_space *vm =
> > +						&vgpu->gvt->dev_priv-
> >ggtt.vm;
> > +
> > +			pages = __i915_gem_object_unset_pages(pri_obj);
> > +			if (!IS_ERR(pages))
> > +				pri_obj->ops->put_pages(pri_obj, pages);
> > +			ret = pri_obj->ops->get_pages(pri_obj);
> > +
> > +			/* pri_obj->cache_dirty = true; */
> > +
> > +			vma = i915_vma_instance(pri_obj, vm, NULL);
> > +			if (IS_ERR(vma))
> > +				goto out;
> > +
> > +			vma->ops->set_pages(vma);
> > +
> > +			ret = i915_vma_bind(vma, vma->obj ?
> > +					    pri_obj->cache_level : 0,
> > +					    PIN_UPDATE);
> > +			if (ret)
> > +				goto out;
> > +		}
> > +
> > +		/* 3. Update shadow cursor plane */
> > +		if (cur_plane) {
> > +			ret = vgpu_fb_update(drm, vgpu,
> DRM_PLANE_TYPE_CURSOR,
> > +					     &mode_cmd);
> > +			if (ret) {
> > +				DRM_DEBUG_KMS("cursor vgpu_fb_update
> failed\n");
> > +				goto flip;
> > +			}
> > +			cur_fb = vgpu-
> >display.shadow_fbs.shadow_fb[CURSOR_PLANE];
> > +			cur_obj = vgpu-
> >display.shadow_fbs.shadow_obj[CURSOR_PLANE];
> > +			cur_plane_state =
> drm_atomic_get_plane_state(state,
> > +								     cur_plane);
> > +			if (IS_ERR(cur_plane_state)) {
> > +				ret = PTR_ERR(cur_plane_state);
> > +				goto flip;
> > +			}
> > +
> > +			fb_info = cur_obj->gvt_info;
> > +			drm_helper_mode_fill_fb_struct(drm, cur_fb,
> &mode_cmd);
> > +			ret =
> drm_atomic_set_crtc_for_plane(cur_plane_state,
> > +							    crtc);
> > +			if (ret != 0)
> > +				goto flip;
> > +			if (cur_plane_state->fb != cur_fb)
> > +
> 	drm_atomic_set_fb_for_plane(cur_plane_state,
> > +							    cur_fb);
> > +
> > +			cur_plane_state->crtc_x = fb_info->x_pos;
> > +			cur_plane_state->crtc_y = fb_info->y_pos;
> > +			cur_plane_state->crtc_w = mode_cmd.width;
> > +			cur_plane_state->crtc_h = mode_cmd.height;
> > +			cur_plane_state->src_x = 0;
> > +			cur_plane_state->src_y = 0;
> > +			cur_plane_state->src_w = mode_cmd.width << 16;
> > +			cur_plane_state->src_h = mode_cmd.height << 16;
> > +
> > +			if (i915_gem_object_has_pinned_pages(cur_obj)) {
> > +				struct sg_table *pages;
> > +				struct i915_vma *vma;
> > +				struct i915_address_space *vm =
> > +					&vgpu->gvt->dev_priv->ggtt.vm;
> > +
> > +				pages =
> __i915_gem_object_unset_pages(cur_obj);
> > +				if (!IS_ERR(pages))
> > +					cur_obj->ops->put_pages(cur_obj,
> pages);
> > +				ret = cur_obj->ops->get_pages(cur_obj);
> > +
> > +				vma = i915_vma_instance(cur_obj, vm, NULL);
> > +				if (IS_ERR(vma))
> > +					goto out;
> > +
> > +				vma->ops->set_pages(vma);
> > +
> > +				ret = i915_vma_bind(vma, vma->obj ?
> > +						    cur_obj->cache_level : 0,
> > +						    PIN_UPDATE);
> > +				if (ret)
> > +					goto out;
> > +			}
> > +		}
> > +flip:
> > +
> > +		/* 4. Commit drm_framebuffer */
> > +		ret = drm_atomic_commit(state);
> > +		if (plane_state->fb != pri_fb)
> > +			drm_framebuffer_put(pri_fb);
> > +		if (cur_plane_state && cur_plane_state->fb != cur_fb)
> > +			drm_framebuffer_put(cur_fb);
> > +	} else {
> > +		DRM_DEBUG_KMS("Assigning sprite plane hasn't been
> supported\n");
> > +	}
> > +
> > +out:
> > +	drm_atomic_state_put(state);
> > +fail:
> > +	if (ret == -EDEADLK) {
> > +	       ret = drm_modeset_backoff(&ctx);
> > +	       if (!ret)
> > +		       goto retry;
> > +	}
> > +
> > +	drm_modeset_drop_locks(&ctx);
> > +	drm_modeset_acquire_fini(&ctx);
> > +
> > +	return ret;
> > +}
> > +
> >  #define DSPSURF_TO_PIPE(offset) \
> >  	calc_index(offset, _DSPASURF, _DSPBSURF, 0, DSPSURF(PIPE_C))
> >
> > @@ -758,6 +1088,8 @@ static int pri_surf_mmio_write(struct intel_vgpu
> > *vgpu, unsigned int offset,
> >
> >  	vgpu_vreg_t(vgpu, PIPE_FLIPCOUNT_G4X(pipe))++;
> >
> > +	handle_vgpu_page_flip(vgpu, pipe, PLANE_PRIMARY);
> > +
> >  	if (vgpu_vreg_t(vgpu, DSPCNTR(pipe)) & PLANE_CTL_ASYNC_FLIP)
> >  		intel_vgpu_trigger_virtual_event(vgpu, event);
> >  	else
> > --
> > 2.7.4
> >
> > _______________________________________________
> > intel-gvt-dev mailing list
> > intel-gvt-dev at lists.freedesktop.org
> > https://lists.freedesktop.org/mailman/listinfo/intel-gvt-dev
> 
> --
> Open Source Technology Center, Intel ltd.
> 
> $gpg --keyserver wwwkeys.pgp.net --recv-keys 4D781827


More information about the intel-gvt-dev mailing list