static int intel_fbdev_set_par(struct fb_info *info)
{
struct drm_fb_helper *fb_helper = info->par;
struct intel_fbdev *ifbdev =
container_of(fb_helper, struct intel_fbdev, helper);
int ret;
ret = drm_fb_helper_set_par(info);
if (ret == 0) {
/*
* FIXME: fbdev presumes that all callbacks also work from
* atomic contexts and relies on that for emergency oops
* printing. KMS totally doesn't do that and the locking here is
* by far not the only place this goes wrong. Ignore this for
* now until we solve this for real.
*/
mutex_lock(&fb_helper->dev->struct_mutex);
ret = i915_gem_object_set_to_gtt_domain(ifbdev->fb->obj,
true);
mutex_unlock(&fb_helper->dev->struct_mutex);
}
return ret;
}
static int gtt_set(struct drm_i915_gem_object *obj,
unsigned long offset,
u32 v)
{
struct i915_vma *vma;
u32 __iomem *map;
int err;
err = i915_gem_object_set_to_gtt_domain(obj, true);
if (err)
return err;
vma = i915_gem_object_ggtt_pin(obj, NULL, 0, 0, PIN_MAPPABLE);
if (IS_ERR(vma))
return PTR_ERR(vma);
map = i915_vma_pin_iomap(vma);
i915_vma_unpin(vma);
if (IS_ERR(map))
return PTR_ERR(map);
iowrite32(v, &map[offset / sizeof(*map)]);
i915_vma_unpin_iomap(vma);
return 0;
}
static int gpu_set(struct drm_i915_gem_object *obj,
unsigned long offset,
u32 v)
{
struct drm_i915_private *i915 = to_i915(obj->base.dev);
struct drm_i915_gem_request *rq;
struct i915_vma *vma;
u32 *cs;
int err;
err = i915_gem_object_set_to_gtt_domain(obj, true);
if (err)
return err;
vma = i915_gem_object_ggtt_pin(obj, NULL, 0, 0, 0);
if (IS_ERR(vma))
return PTR_ERR(vma);
rq = i915_gem_request_alloc(i915->engine[RCS], i915->kernel_context);
if (IS_ERR(rq)) {
i915_vma_unpin(vma);
return PTR_ERR(rq);
}
cs = intel_ring_begin(rq, 4);
if (IS_ERR(cs)) {
__i915_add_request(rq, false);
i915_vma_unpin(vma);
return PTR_ERR(cs);
}
if (INTEL_GEN(i915) >= 8) {
*cs++ = MI_STORE_DWORD_IMM_GEN4 | 1 << 22;
*cs++ = lower_32_bits(i915_ggtt_offset(vma) + offset);
*cs++ = upper_32_bits(i915_ggtt_offset(vma) + offset);
*cs++ = v;
} else if (INTEL_GEN(i915) >= 4) {
*cs++ = MI_STORE_DWORD_IMM_GEN4 | 1 << 22;
*cs++ = 0;
*cs++ = i915_ggtt_offset(vma) + offset;
*cs++ = v;
} else {
*cs++ = MI_STORE_DWORD_IMM | 1 << 22;
*cs++ = i915_ggtt_offset(vma) + offset;
*cs++ = v;
*cs++ = MI_NOOP;
}
intel_ring_advance(rq, cs);
i915_vma_move_to_active(vma, rq, EXEC_OBJECT_WRITE);
i915_vma_unpin(vma);
reservation_object_lock(obj->resv, NULL);
reservation_object_add_excl_fence(obj->resv, &rq->fence);
reservation_object_unlock(obj->resv);
__i915_add_request(rq, true);
return 0;
}
static int i915_gem_end_cpu_access(struct dma_buf *dma_buf, enum dma_data_direction direction)
{
struct drm_i915_gem_object *obj = dma_buf_to_obj(dma_buf);
struct drm_device *dev = obj->base.dev;
int ret;
ret = i915_mutex_lock_interruptible(dev);
if (ret)
return ret;
ret = i915_gem_object_set_to_gtt_domain(obj, false);
mutex_unlock(&dev->struct_mutex);
return ret;
}
static void i915_gem_end_cpu_access(struct dma_buf *dma_buf, enum dma_data_direction direction)
{
struct drm_i915_gem_object *obj = dma_buf_to_obj(dma_buf);
struct drm_device *dev = obj->base.dev;
struct drm_i915_private *dev_priv = to_i915(dev);
bool was_interruptible;
int ret;
mutex_lock(&dev->struct_mutex);
was_interruptible = dev_priv->mm.interruptible;
dev_priv->mm.interruptible = false;
ret = i915_gem_object_set_to_gtt_domain(obj, false);
dev_priv->mm.interruptible = was_interruptible;
mutex_unlock(&dev->struct_mutex);
if (unlikely(ret))
DRM_ERROR("unable to flush buffer following CPU access; rendering may be corrupt\n");
}
static int render_state_setup(struct render_state *so)
{
const struct intel_renderstate_rodata *rodata = so->rodata;
unsigned int i = 0, reloc_index = 0;
struct page *page;
u32 *d;
int ret;
ret = i915_gem_object_set_to_cpu_domain(so->obj, true);
if (ret)
return ret;
page = i915_gem_object_get_dirty_page(so->obj, 0);
d = kmap(page);
while (i < rodata->batch_items) {
u32 s = rodata->batch[i];
if (i * 4 == rodata->reloc[reloc_index]) {
u64 r = s + so->ggtt_offset;
s = lower_32_bits(r);
if (so->gen >= 8) {
if (i + 1 >= rodata->batch_items ||
rodata->batch[i + 1] != 0) {
ret = -EINVAL;
goto err_out;
}
d[i++] = s;
s = upper_32_bits(r);
}
reloc_index++;
}
d[i++] = s;
}
while (i % CACHELINE_DWORDS)
OUT_BATCH(d, i, MI_NOOP);
so->aux_batch_offset = i * sizeof(u32);
OUT_BATCH(d, i, MI_BATCH_BUFFER_END);
so->aux_batch_size = (i * sizeof(u32)) - so->aux_batch_offset;
/*
* Since we are sending length, we need to strictly conform to
* all requirements. For Gen2 this must be a multiple of 8.
*/
so->aux_batch_size = ALIGN(so->aux_batch_size, 8);
kunmap(page);
ret = i915_gem_object_set_to_gtt_domain(so->obj, false);
if (ret)
return ret;
if (rodata->reloc[reloc_index] != -1) {
DRM_ERROR("only %d relocs resolved\n", reloc_index);
return -EINVAL;
}
return 0;
err_out:
kunmap(page);
return ret;
}
rodata->batch[i + 1] != 0)
return -EINVAL;
d[i] = s;
i++;
s = (goffset & 0xffffffff00000000ull) >> 32;
}
reloc_index++;
}
d[i] = s;
i++;
}
ret = i915_gem_object_set_to_gtt_domain(so->obj, false);
if (ret)
return ret;
if (rodata->reloc_items != reloc_index) {
DRM_ERROR("not all relocs resolved, %d out of %d\n",
reloc_index, rodata->reloc_items);
return -EINVAL;
}
so->len = rodata->batch_items * 4;
return 0;
}
int i915_gem_render_state_init(struct intel_ring_buffer *ring)
static int do_switch(struct drm_i915_gem_object *from_obj,
struct i915_hw_context *to,
u32 seqno)
{
struct intel_ring_buffer *ring = NULL;
u32 hw_flags = 0;
int ret;
BUG_ON(to == NULL);
BUG_ON(from_obj != NULL && from_obj->pin_count == 0);
ret = i915_gem_object_pin(to->obj, CONTEXT_ALIGN, false);
if (ret)
return ret;
/* Clear this page out of any CPU caches for coherent swap-in/out. Note
* that thanks to write = false in this call and us not setting any gpu
* write domains when putting a context object onto the active list
* (when switching away from it), this won't block.
* XXX: We need a real interface to do this instead of trickery. */
ret = i915_gem_object_set_to_gtt_domain(to->obj, false);
if (ret) {
i915_gem_object_unpin(to->obj);
return ret;
}
if (!to->obj->has_global_gtt_mapping)
i915_gem_gtt_bind_object(to->obj, to->obj->cache_level);
if (!to->is_initialized || is_default_context(to))
hw_flags |= MI_RESTORE_INHIBIT;
else if (WARN_ON_ONCE(from_obj == to->obj)) /* not yet expected */
hw_flags |= MI_FORCE_RESTORE;
ring = to->ring;
ret = mi_set_context(ring, to, hw_flags);
if (ret) {
i915_gem_object_unpin(to->obj);
return ret;
}
/* The backing object for the context is done after switching to the
* *next* context. Therefore we cannot retire the previous context until
* the next context has already started running. In fact, the below code
* is a bit suboptimal because the retiring can occur simply after the
* MI_SET_CONTEXT instead of when the next seqno has completed.
*/
if (from_obj != NULL) {
from_obj->base.read_domains = I915_GEM_DOMAIN_INSTRUCTION;
i915_gem_object_move_to_active(from_obj, ring, seqno);
/* As long as MI_SET_CONTEXT is serializing, ie. it flushes the
* whole damn pipeline, we don't need to explicitly mark the
* object dirty. The only exception is that the context must be
* correct in case the object gets swapped out. Ideally we'd be
* able to defer doing this until we know the object would be
* swapped, but there is no way to do that yet.
*/
from_obj->dirty = 1;
BUG_ON(from_obj->ring != to->ring);
i915_gem_object_unpin(from_obj);
drm_gem_object_unreference(&from_obj->base);
}
drm_gem_object_reference(&to->obj->base);
ring->last_context_obj = to->obj;
to->is_initialized = true;
return 0;
}
static int do_switch(struct i915_hw_context *to)
{
struct intel_ring_buffer *ring = to->ring;
struct i915_hw_context *from = ring->last_context;
u32 hw_flags = 0;
int ret, i;
BUG_ON(from != NULL && from->obj != NULL && from->obj->pin_count == 0);
if (from == to && !to->remap_slice)
return 0;
ret = i915_gem_obj_ggtt_pin(to->obj, CONTEXT_ALIGN, false, false);
if (ret)
return ret;
/*
* Pin can switch back to the default context if we end up calling into
* evict_everything - as a last ditch gtt defrag effort that also
* switches to the default context. Hence we need to reload from here.
*/
from = ring->last_context;
/*
* Clear this page out of any CPU caches for coherent swap-in/out. Note
* that thanks to write = false in this call and us not setting any gpu
* write domains when putting a context object onto the active list
* (when switching away from it), this won't block.
*
* XXX: We need a real interface to do this instead of trickery.
*/
ret = i915_gem_object_set_to_gtt_domain(to->obj, false);
if (ret) {
i915_gem_object_unpin(to->obj);
return ret;
}
if (!to->obj->has_global_gtt_mapping)
i915_gem_gtt_bind_object(to->obj, to->obj->cache_level);
if (!to->is_initialized || is_default_context(to))
hw_flags |= MI_RESTORE_INHIBIT;
ret = mi_set_context(ring, to, hw_flags);
if (ret) {
i915_gem_object_unpin(to->obj);
return ret;
}
for (i = 0; i < MAX_L3_SLICES; i++) {
if (!(to->remap_slice & (1<<i)))
continue;
ret = i915_gem_l3_remap(ring, i);
/* If it failed, try again next round */
if (ret)
DRM_DEBUG_DRIVER("L3 remapping failed\n");
else
to->remap_slice &= ~(1<<i);
}
/* The backing object for the context is done after switching to the
* *next* context. Therefore we cannot retire the previous context until
* the next context has already started running. In fact, the below code
* is a bit suboptimal because the retiring can occur simply after the
* MI_SET_CONTEXT instead of when the next seqno has completed.
*/
if (from != NULL) {
from->obj->base.read_domains = I915_GEM_DOMAIN_INSTRUCTION;
i915_vma_move_to_active(i915_gem_obj_to_ggtt(from->obj), ring);
/* As long as MI_SET_CONTEXT is serializing, ie. it flushes the
* whole damn pipeline, we don't need to explicitly mark the
* object dirty. The only exception is that the context must be
* correct in case the object gets swapped out. Ideally we'd be
* able to defer doing this until we know the object would be
* swapped, but there is no way to do that yet.
*/
from->obj->dirty = 1;
BUG_ON(from->obj->ring != ring);
/* obj is kept alive until the next request by its active ref */
i915_gem_object_unpin(from->obj);
i915_gem_context_unreference(from);
}
i915_gem_context_reference(to);
ring->last_context = to;
to->is_initialized = true;
return 0;
}
static int render_state_setup(struct render_state *so)
{
struct drm_device *dev = so->vma->vm->dev;
const struct intel_renderstate_rodata *rodata = so->rodata;
const bool has_64bit_reloc = INTEL_GEN(dev) >= 8;
unsigned int i = 0, reloc_index = 0;
struct page *page;
u32 *d;
int ret;
ret = i915_gem_object_set_to_cpu_domain(so->vma->obj, true);
if (ret)
return ret;
page = i915_gem_object_get_dirty_page(so->vma->obj, 0);
d = kmap(page);
while (i < rodata->batch_items) {
u32 s = rodata->batch[i];
if (i * 4 == rodata->reloc[reloc_index]) {
u64 r = s + so->vma->node.start;
s = lower_32_bits(r);
if (has_64bit_reloc) {
if (i + 1 >= rodata->batch_items ||
rodata->batch[i + 1] != 0) {
ret = -EINVAL;
goto err_out;
}
d[i++] = s;
s = upper_32_bits(r);
}
reloc_index++;
}
d[i++] = s;
}
while (i % CACHELINE_DWORDS)
OUT_BATCH(d, i, MI_NOOP);
so->aux_batch_offset = i * sizeof(u32);
if (HAS_POOLED_EU(dev)) {
/*
* We always program 3x6 pool config but depending upon which
* subslice is disabled HW drops down to appropriate config
* shown below.
*
* In the below table 2x6 config always refers to
* fused-down version, native 2x6 is not available and can
* be ignored
*
* SNo subslices config eu pool configuration
* -----------------------------------------------------------
* 1 3 subslices enabled (3x6) - 0x00777000 (9+9)
* 2 ss0 disabled (2x6) - 0x00777000 (3+9)
* 3 ss1 disabled (2x6) - 0x00770000 (6+6)
* 4 ss2 disabled (2x6) - 0x00007000 (9+3)
*/
u32 eu_pool_config = 0x00777000;
OUT_BATCH(d, i, GEN9_MEDIA_POOL_STATE);
OUT_BATCH(d, i, GEN9_MEDIA_POOL_ENABLE);
OUT_BATCH(d, i, eu_pool_config);
OUT_BATCH(d, i, 0);
OUT_BATCH(d, i, 0);
OUT_BATCH(d, i, 0);
}
OUT_BATCH(d, i, MI_BATCH_BUFFER_END);
so->aux_batch_size = (i * sizeof(u32)) - so->aux_batch_offset;
/*
* Since we are sending length, we need to strictly conform to
* all requirements. For Gen2 this must be a multiple of 8.
*/
so->aux_batch_size = ALIGN(so->aux_batch_size, 8);
kunmap(page);
ret = i915_gem_object_set_to_gtt_domain(so->vma->obj, false);
if (ret)
return ret;
if (rodata->reloc[reloc_index] != -1) {
DRM_ERROR("only %d relocs resolved\n", reloc_index);
return -EINVAL;
}
return 0;
err_out:
kunmap(page);
return ret;
}
static int render_state_setup(struct intel_render_state *so,
struct drm_i915_private *i915)
{
const struct intel_renderstate_rodata *rodata = so->rodata;
struct drm_i915_gem_object *obj = so->vma->obj;
unsigned int i = 0, reloc_index = 0;
unsigned int needs_clflush;
u32 *d;
int ret;
ret = i915_gem_obj_prepare_shmem_write(obj, &needs_clflush);
if (ret)
return ret;
d = kmap_atomic(i915_gem_object_get_dirty_page(obj, 0));
while (i < rodata->batch_items) {
u32 s = rodata->batch[i];
if (i * 4 == rodata->reloc[reloc_index]) {
u64 r = s + so->vma->node.start;
s = lower_32_bits(r);
if (HAS_64BIT_RELOC(i915)) {
if (i + 1 >= rodata->batch_items ||
rodata->batch[i + 1] != 0)
goto err;
d[i++] = s;
s = upper_32_bits(r);
}
reloc_index++;
}
d[i++] = s;
}
if (rodata->reloc[reloc_index] != -1) {
DRM_ERROR("only %d relocs resolved\n", reloc_index);
goto err;
}
so->batch_offset = so->vma->node.start;
so->batch_size = rodata->batch_items * sizeof(u32);
while (i % CACHELINE_DWORDS)
OUT_BATCH(d, i, MI_NOOP);
so->aux_offset = i * sizeof(u32);
if (HAS_POOLED_EU(i915)) {
/*
* We always program 3x6 pool config but depending upon which
* subslice is disabled HW drops down to appropriate config
* shown below.
*
* In the below table 2x6 config always refers to
* fused-down version, native 2x6 is not available and can
* be ignored
*
* SNo subslices config eu pool configuration
* -----------------------------------------------------------
* 1 3 subslices enabled (3x6) - 0x00777000 (9+9)
* 2 ss0 disabled (2x6) - 0x00777000 (3+9)
* 3 ss1 disabled (2x6) - 0x00770000 (6+6)
* 4 ss2 disabled (2x6) - 0x00007000 (9+3)
*/
u32 eu_pool_config = 0x00777000;
OUT_BATCH(d, i, GEN9_MEDIA_POOL_STATE);
OUT_BATCH(d, i, GEN9_MEDIA_POOL_ENABLE);
OUT_BATCH(d, i, eu_pool_config);
OUT_BATCH(d, i, 0);
OUT_BATCH(d, i, 0);
OUT_BATCH(d, i, 0);
}
OUT_BATCH(d, i, MI_BATCH_BUFFER_END);
so->aux_size = i * sizeof(u32) - so->aux_offset;
so->aux_offset += so->batch_offset;
/*
* Since we are sending length, we need to strictly conform to
* all requirements. For Gen2 this must be a multiple of 8.
*/
so->aux_size = ALIGN(so->aux_size, 8);
if (needs_clflush)
drm_clflush_virt_range(d, i * sizeof(u32));
kunmap_atomic(d);
ret = i915_gem_object_set_to_gtt_domain(obj, false);
out:
i915_gem_obj_finish_shmem_access(obj);
return ret;
err:
kunmap_atomic(d);
ret = -EINVAL;
goto out;
}
