linux_dsm_epyc7002/drivers/gpu/drm/i915/i915_vma.h

387 lines
11 KiB
C
Raw Normal View History

/*
* Copyright © 2016 Intel Corporation
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice (including the next
* paragraph) shall be included in all copies or substantial portions of the
* Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
* IN THE SOFTWARE.
*
*/
#ifndef __I915_VMA_H__
#define __I915_VMA_H__
#include <linux/io-mapping.h>
#include <linux/rbtree.h>
#include <drm/drm_mm.h>
#include "gem/i915_gem_object.h"
#include "i915_gem_gtt.h"
#include "i915_gem_fence_reg.h"
#include "i915_active.h"
#include "i915_request.h"
#include "i915_vma_types.h"
struct i915_vma *
i915_vma_instance(struct drm_i915_gem_object *obj,
struct i915_address_space *vm,
const struct i915_ggtt_view *view);
void i915_vma_unpin_and_release(struct i915_vma **p_vma, unsigned int flags);
#define I915_VMA_RELEASE_MAP BIT(0)
static inline bool i915_vma_is_active(const struct i915_vma *vma)
{
return !i915_active_is_idle(&vma->active);
}
drm/i915: Pull i915_vma_pin under the vm->mutex Replace the struct_mutex requirement for pinning the i915_vma with the local vm->mutex instead. Note that the vm->mutex is tainted by the shrinker (we require unbinding from inside fs-reclaim) and so we cannot allocate while holding that mutex. Instead we have to preallocate workers to do allocate and apply the PTE updates after we have we reserved their slot in the drm_mm (using fences to order the PTE writes with the GPU work and with later unbind). In adding the asynchronous vma binding, one subtle requirement is to avoid coupling the binding fence into the backing object->resv. That is the asynchronous binding only applies to the vma timeline itself and not to the pages as that is a more global timeline (the binding of one vma does not need to be ordered with another vma, nor does the implicit GEM fencing depend on a vma, only on writes to the backing store). Keeping the vma binding distinct from the backing store timelines is verified by a number of async gem_exec_fence and gem_exec_schedule tests. The way we do this is quite simple, we keep the fence for the vma binding separate and only wait on it as required, and never add it to the obj->resv itself. Another consequence in reducing the locking around the vma is the destruction of the vma is no longer globally serialised by struct_mutex. A natural solution would be to add a kref to i915_vma, but that requires decoupling the reference cycles, possibly by introducing a new i915_mm_pages object that is own by both obj->mm and vma->pages. However, we have not taken that route due to the overshadowing lmem/ttm discussions, and instead play a series of complicated games with trylocks to (hopefully) ensure that only one destruction path is called! v2: Add some commentary, and some helpers to reduce patch churn. Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20191004134015.13204-4-chris@chris-wilson.co.uk
2019-10-04 20:39:58 +07:00
int __must_check __i915_vma_move_to_active(struct i915_vma *vma,
struct i915_request *rq);
int __must_check i915_vma_move_to_active(struct i915_vma *vma,
struct i915_request *rq,
unsigned int flags);
#define __i915_vma_flags(v) ((unsigned long *)&(v)->flags.counter)
static inline bool i915_vma_is_ggtt(const struct i915_vma *vma)
{
return test_bit(I915_VMA_GGTT_BIT, __i915_vma_flags(vma));
}
static inline bool i915_vma_has_ggtt_write(const struct i915_vma *vma)
{
return test_bit(I915_VMA_GGTT_WRITE_BIT, __i915_vma_flags(vma));
}
static inline void i915_vma_set_ggtt_write(struct i915_vma *vma)
{
GEM_BUG_ON(!i915_vma_is_ggtt(vma));
set_bit(I915_VMA_GGTT_WRITE_BIT, __i915_vma_flags(vma));
}
static inline bool i915_vma_unset_ggtt_write(struct i915_vma *vma)
{
return test_and_clear_bit(I915_VMA_GGTT_WRITE_BIT,
__i915_vma_flags(vma));
}
void i915_vma_flush_writes(struct i915_vma *vma);
static inline bool i915_vma_is_map_and_fenceable(const struct i915_vma *vma)
{
return test_bit(I915_VMA_CAN_FENCE_BIT, __i915_vma_flags(vma));
}
static inline bool i915_vma_set_userfault(struct i915_vma *vma)
{
GEM_BUG_ON(!i915_vma_is_map_and_fenceable(vma));
return test_and_set_bit(I915_VMA_USERFAULT_BIT, __i915_vma_flags(vma));
}
static inline void i915_vma_unset_userfault(struct i915_vma *vma)
{
return clear_bit(I915_VMA_USERFAULT_BIT, __i915_vma_flags(vma));
}
static inline bool i915_vma_has_userfault(const struct i915_vma *vma)
{
return test_bit(I915_VMA_USERFAULT_BIT, __i915_vma_flags(vma));
}
static inline bool i915_vma_is_closed(const struct i915_vma *vma)
{
return !list_empty(&vma->closed_link);
}
static inline u32 i915_ggtt_offset(const struct i915_vma *vma)
{
GEM_BUG_ON(!i915_vma_is_ggtt(vma));
GEM_BUG_ON(!drm_mm_node_allocated(&vma->node));
GEM_BUG_ON(upper_32_bits(vma->node.start));
GEM_BUG_ON(upper_32_bits(vma->node.start + vma->node.size - 1));
return lower_32_bits(vma->node.start);
}
drm/i915/guc: Move the pin bias value from GuC to GGTT Removing the pin bias from GuC allows us to not check for GuC every time we pin a context, which fixes the assertion error on unresolved GuC platform default in mock contexts selftest. It also seems that we were using uninitialized WOPCM variables when setting the GuC pin bias. The pin bias has to be set after the WOPCM, but before the call to i915_gem_contexts_init where the first contexts are pinned. v2: This also makes it so that there's no need to set GuC variables from within the WOPCM init function or to move the WOPCM init, while keeping the correct initialization order. Also for mock tests the pin bias is left at 0 and we make sure that the pin bias with GuC will not be smaller than without GuC. v3: Avoid unused i915 in intel_guc_ggtt_offset if debug is disabled. v4: Squash with WOPCM init reordering. Moved the i915_ggtt_pin_bias helper to this patch, and made some functions use it instead of directly dereferencing i915->ggtt. v5: Since we now don't use wopcm.guc.base for the pin bias there's no need to validate it. It also has already been verified in WOPCM init. v6: Deleted the now unnecessarily introduced includes from previous versions. Dropped naming changes from dev_priv to i915 for better patch readability. v7: Changed some comments to make more sense in the context they're in. v8: Moved and renamed the function which now returns the wopcm.guc.size to intel_guc.c:intel_guc_reserved_gtt_size to avoid any possible confusion with the pin_bias in ggtt, which should be used for pinning. Fixed patch not applying or the most recent upstream. Fixes: f7dc0157e4b5 ("drm/i915/uc: Fetch GuC/HuC firmwares from guc/huc specific init") Testcase: igt/drv_selftest/mock_contexts #GuC Signed-off-by: Jakub Bartmiński <jakub.bartminski@intel.com> Cc: Chris Wilson <chris@chris-wilson.co.uk> Cc: Michał Winiarski <michal.winiarski@intel.com> Cc: Michal Wajdeczko <michal.wajdeczko@intel.com> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Link: https://patchwork.freedesktop.org/patch/msgid/20180727141148.30874-3-jakub.bartminski@intel.com
2018-07-27 21:11:45 +07:00
static inline u32 i915_ggtt_pin_bias(struct i915_vma *vma)
{
return i915_vm_to_ggtt(vma->vm)->pin_bias;
}
static inline struct i915_vma *i915_vma_get(struct i915_vma *vma)
{
i915_gem_object_get(vma->obj);
return vma;
}
static inline struct i915_vma *i915_vma_tryget(struct i915_vma *vma)
{
if (likely(kref_get_unless_zero(&vma->obj->base.refcount)))
return vma;
return NULL;
}
static inline void i915_vma_put(struct i915_vma *vma)
{
i915_gem_object_put(vma->obj);
}
static __always_inline ptrdiff_t ptrdiff(const void *a, const void *b)
{
return a - b;
}
static inline long
i915_vma_compare(struct i915_vma *vma,
struct i915_address_space *vm,
const struct i915_ggtt_view *view)
{
ptrdiff_t cmp;
GEM_BUG_ON(view && !i915_is_ggtt(vm));
cmp = ptrdiff(vma->vm, vm);
if (cmp)
return cmp;
BUILD_BUG_ON(I915_GGTT_VIEW_NORMAL != 0);
cmp = vma->ggtt_view.type;
if (!view)
return cmp;
cmp -= view->type;
if (cmp)
return cmp;
assert_i915_gem_gtt_types();
/* ggtt_view.type also encodes its size so that we both distinguish
* different views using it as a "type" and also use a compact (no
* accessing of uninitialised padding bytes) memcmp without storing
* an extra parameter or adding more code.
*
* To ensure that the memcmp is valid for all branches of the union,
* even though the code looks like it is just comparing one branch,
* we assert above that all branches have the same address, and that
* each branch has a unique type/size.
*/
BUILD_BUG_ON(I915_GGTT_VIEW_NORMAL >= I915_GGTT_VIEW_PARTIAL);
BUILD_BUG_ON(I915_GGTT_VIEW_PARTIAL >= I915_GGTT_VIEW_ROTATED);
BUILD_BUG_ON(I915_GGTT_VIEW_ROTATED >= I915_GGTT_VIEW_REMAPPED);
BUILD_BUG_ON(offsetof(typeof(*view), rotated) !=
offsetof(typeof(*view), partial));
BUILD_BUG_ON(offsetof(typeof(*view), rotated) !=
offsetof(typeof(*view), remapped));
return memcmp(&vma->ggtt_view.partial, &view->partial, view->type);
}
drm/i915: Pull i915_vma_pin under the vm->mutex Replace the struct_mutex requirement for pinning the i915_vma with the local vm->mutex instead. Note that the vm->mutex is tainted by the shrinker (we require unbinding from inside fs-reclaim) and so we cannot allocate while holding that mutex. Instead we have to preallocate workers to do allocate and apply the PTE updates after we have we reserved their slot in the drm_mm (using fences to order the PTE writes with the GPU work and with later unbind). In adding the asynchronous vma binding, one subtle requirement is to avoid coupling the binding fence into the backing object->resv. That is the asynchronous binding only applies to the vma timeline itself and not to the pages as that is a more global timeline (the binding of one vma does not need to be ordered with another vma, nor does the implicit GEM fencing depend on a vma, only on writes to the backing store). Keeping the vma binding distinct from the backing store timelines is verified by a number of async gem_exec_fence and gem_exec_schedule tests. The way we do this is quite simple, we keep the fence for the vma binding separate and only wait on it as required, and never add it to the obj->resv itself. Another consequence in reducing the locking around the vma is the destruction of the vma is no longer globally serialised by struct_mutex. A natural solution would be to add a kref to i915_vma, but that requires decoupling the reference cycles, possibly by introducing a new i915_mm_pages object that is own by both obj->mm and vma->pages. However, we have not taken that route due to the overshadowing lmem/ttm discussions, and instead play a series of complicated games with trylocks to (hopefully) ensure that only one destruction path is called! v2: Add some commentary, and some helpers to reduce patch churn. Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20191004134015.13204-4-chris@chris-wilson.co.uk
2019-10-04 20:39:58 +07:00
struct i915_vma_work *i915_vma_work(void);
int i915_vma_bind(struct i915_vma *vma,
enum i915_cache_level cache_level,
u32 flags,
struct i915_vma_work *work);
bool i915_gem_valid_gtt_space(struct i915_vma *vma, unsigned long color);
bool i915_vma_misplaced(const struct i915_vma *vma,
u64 size, u64 alignment, u64 flags);
void __i915_vma_set_map_and_fenceable(struct i915_vma *vma);
void i915_vma_revoke_mmap(struct i915_vma *vma);
drm/i915: Pull i915_vma_pin under the vm->mutex Replace the struct_mutex requirement for pinning the i915_vma with the local vm->mutex instead. Note that the vm->mutex is tainted by the shrinker (we require unbinding from inside fs-reclaim) and so we cannot allocate while holding that mutex. Instead we have to preallocate workers to do allocate and apply the PTE updates after we have we reserved their slot in the drm_mm (using fences to order the PTE writes with the GPU work and with later unbind). In adding the asynchronous vma binding, one subtle requirement is to avoid coupling the binding fence into the backing object->resv. That is the asynchronous binding only applies to the vma timeline itself and not to the pages as that is a more global timeline (the binding of one vma does not need to be ordered with another vma, nor does the implicit GEM fencing depend on a vma, only on writes to the backing store). Keeping the vma binding distinct from the backing store timelines is verified by a number of async gem_exec_fence and gem_exec_schedule tests. The way we do this is quite simple, we keep the fence for the vma binding separate and only wait on it as required, and never add it to the obj->resv itself. Another consequence in reducing the locking around the vma is the destruction of the vma is no longer globally serialised by struct_mutex. A natural solution would be to add a kref to i915_vma, but that requires decoupling the reference cycles, possibly by introducing a new i915_mm_pages object that is own by both obj->mm and vma->pages. However, we have not taken that route due to the overshadowing lmem/ttm discussions, and instead play a series of complicated games with trylocks to (hopefully) ensure that only one destruction path is called! v2: Add some commentary, and some helpers to reduce patch churn. Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20191004134015.13204-4-chris@chris-wilson.co.uk
2019-10-04 20:39:58 +07:00
int __i915_vma_unbind(struct i915_vma *vma);
int __must_check i915_vma_unbind(struct i915_vma *vma);
void i915_vma_unlink_ctx(struct i915_vma *vma);
void i915_vma_close(struct i915_vma *vma);
drm/i915: Lazily unbind vma on close When userspace is passing around swapbuffers using DRI, we frequently have to open and close the same object in the foreign address space. This shows itself as the same object being rebound at roughly 30fps (with a second object also being rebound at 30fps), which involves us having to rewrite the page tables and maintain the drm_mm range manager every time. However, since the object still exists and it is only the local handle that disappears, if we are lazy and do not unbind the VMA immediately when the local user closes the object but defer it until the GPU is idle, then we can reuse the same VMA binding. We still have to be careful to mark the handle and lookup tables as closed to maintain the uABI, just allowing the underlying VMA to be resurrected if the user is able to access the same object from the same context again. If the object itself is destroyed (neither userspace keeping a handle to it), the VMA will be reaped immediately as usual. In the future, this will be even more useful as instantiating a new VMA for use on the GPU will become heavier. A nuisance indeed, so nip it in the bud. v2: s/__i915_vma_final_close/i915_vma_destroy/ etc. v3: Leave a hint as to why we deferred the unbind on close. Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20180503195115.22309-1-chris@chris-wilson.co.uk
2018-05-04 02:51:14 +07:00
void i915_vma_reopen(struct i915_vma *vma);
static inline struct i915_vma *__i915_vma_get(struct i915_vma *vma)
{
if (kref_get_unless_zero(&vma->ref))
return vma;
return NULL;
}
void i915_vma_release(struct kref *ref);
static inline void __i915_vma_put(struct i915_vma *vma)
{
kref_put(&vma->ref, i915_vma_release);
}
#define assert_vma_held(vma) dma_resv_assert_held((vma)->resv)
static inline void i915_vma_lock(struct i915_vma *vma)
{
dma_resv_lock(vma->resv, NULL);
}
static inline void i915_vma_unlock(struct i915_vma *vma)
{
dma_resv_unlock(vma->resv);
}
drm/i915: Pull i915_vma_pin under the vm->mutex Replace the struct_mutex requirement for pinning the i915_vma with the local vm->mutex instead. Note that the vm->mutex is tainted by the shrinker (we require unbinding from inside fs-reclaim) and so we cannot allocate while holding that mutex. Instead we have to preallocate workers to do allocate and apply the PTE updates after we have we reserved their slot in the drm_mm (using fences to order the PTE writes with the GPU work and with later unbind). In adding the asynchronous vma binding, one subtle requirement is to avoid coupling the binding fence into the backing object->resv. That is the asynchronous binding only applies to the vma timeline itself and not to the pages as that is a more global timeline (the binding of one vma does not need to be ordered with another vma, nor does the implicit GEM fencing depend on a vma, only on writes to the backing store). Keeping the vma binding distinct from the backing store timelines is verified by a number of async gem_exec_fence and gem_exec_schedule tests. The way we do this is quite simple, we keep the fence for the vma binding separate and only wait on it as required, and never add it to the obj->resv itself. Another consequence in reducing the locking around the vma is the destruction of the vma is no longer globally serialised by struct_mutex. A natural solution would be to add a kref to i915_vma, but that requires decoupling the reference cycles, possibly by introducing a new i915_mm_pages object that is own by both obj->mm and vma->pages. However, we have not taken that route due to the overshadowing lmem/ttm discussions, and instead play a series of complicated games with trylocks to (hopefully) ensure that only one destruction path is called! v2: Add some commentary, and some helpers to reduce patch churn. Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20191004134015.13204-4-chris@chris-wilson.co.uk
2019-10-04 20:39:58 +07:00
int __must_check
i915_vma_pin(struct i915_vma *vma, u64 size, u64 alignment, u64 flags);
int i915_ggtt_pin(struct i915_vma *vma, u32 align, unsigned int flags);
static inline int i915_vma_pin_count(const struct i915_vma *vma)
{
return atomic_read(&vma->flags) & I915_VMA_PIN_MASK;
}
static inline bool i915_vma_is_pinned(const struct i915_vma *vma)
{
return i915_vma_pin_count(vma);
}
static inline void __i915_vma_pin(struct i915_vma *vma)
{
atomic_inc(&vma->flags);
drm/i915: Pull i915_vma_pin under the vm->mutex Replace the struct_mutex requirement for pinning the i915_vma with the local vm->mutex instead. Note that the vm->mutex is tainted by the shrinker (we require unbinding from inside fs-reclaim) and so we cannot allocate while holding that mutex. Instead we have to preallocate workers to do allocate and apply the PTE updates after we have we reserved their slot in the drm_mm (using fences to order the PTE writes with the GPU work and with later unbind). In adding the asynchronous vma binding, one subtle requirement is to avoid coupling the binding fence into the backing object->resv. That is the asynchronous binding only applies to the vma timeline itself and not to the pages as that is a more global timeline (the binding of one vma does not need to be ordered with another vma, nor does the implicit GEM fencing depend on a vma, only on writes to the backing store). Keeping the vma binding distinct from the backing store timelines is verified by a number of async gem_exec_fence and gem_exec_schedule tests. The way we do this is quite simple, we keep the fence for the vma binding separate and only wait on it as required, and never add it to the obj->resv itself. Another consequence in reducing the locking around the vma is the destruction of the vma is no longer globally serialised by struct_mutex. A natural solution would be to add a kref to i915_vma, but that requires decoupling the reference cycles, possibly by introducing a new i915_mm_pages object that is own by both obj->mm and vma->pages. However, we have not taken that route due to the overshadowing lmem/ttm discussions, and instead play a series of complicated games with trylocks to (hopefully) ensure that only one destruction path is called! v2: Add some commentary, and some helpers to reduce patch churn. Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20191004134015.13204-4-chris@chris-wilson.co.uk
2019-10-04 20:39:58 +07:00
GEM_BUG_ON(!i915_vma_is_pinned(vma));
}
static inline void __i915_vma_unpin(struct i915_vma *vma)
{
drm/i915: Pull i915_vma_pin under the vm->mutex Replace the struct_mutex requirement for pinning the i915_vma with the local vm->mutex instead. Note that the vm->mutex is tainted by the shrinker (we require unbinding from inside fs-reclaim) and so we cannot allocate while holding that mutex. Instead we have to preallocate workers to do allocate and apply the PTE updates after we have we reserved their slot in the drm_mm (using fences to order the PTE writes with the GPU work and with later unbind). In adding the asynchronous vma binding, one subtle requirement is to avoid coupling the binding fence into the backing object->resv. That is the asynchronous binding only applies to the vma timeline itself and not to the pages as that is a more global timeline (the binding of one vma does not need to be ordered with another vma, nor does the implicit GEM fencing depend on a vma, only on writes to the backing store). Keeping the vma binding distinct from the backing store timelines is verified by a number of async gem_exec_fence and gem_exec_schedule tests. The way we do this is quite simple, we keep the fence for the vma binding separate and only wait on it as required, and never add it to the obj->resv itself. Another consequence in reducing the locking around the vma is the destruction of the vma is no longer globally serialised by struct_mutex. A natural solution would be to add a kref to i915_vma, but that requires decoupling the reference cycles, possibly by introducing a new i915_mm_pages object that is own by both obj->mm and vma->pages. However, we have not taken that route due to the overshadowing lmem/ttm discussions, and instead play a series of complicated games with trylocks to (hopefully) ensure that only one destruction path is called! v2: Add some commentary, and some helpers to reduce patch churn. Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20191004134015.13204-4-chris@chris-wilson.co.uk
2019-10-04 20:39:58 +07:00
GEM_BUG_ON(!i915_vma_is_pinned(vma));
atomic_dec(&vma->flags);
}
static inline void i915_vma_unpin(struct i915_vma *vma)
{
GEM_BUG_ON(!drm_mm_node_allocated(&vma->node));
__i915_vma_unpin(vma);
}
static inline bool i915_vma_is_bound(const struct i915_vma *vma,
unsigned int where)
{
return atomic_read(&vma->flags) & where;
}
static inline bool i915_node_color_differs(const struct drm_mm_node *node,
unsigned long color)
{
return drm_mm_node_allocated(node) && node->color != color;
}
/**
* i915_vma_pin_iomap - calls ioremap_wc to map the GGTT VMA via the aperture
* @vma: VMA to iomap
*
* The passed in VMA has to be pinned in the global GTT mappable region.
* An extra pinning of the VMA is acquired for the return iomapping,
* the caller must call i915_vma_unpin_iomap to relinquish the pinning
* after the iomapping is no longer required.
*
* Returns a valid iomapped pointer or ERR_PTR.
*/
void __iomem *i915_vma_pin_iomap(struct i915_vma *vma);
#define IO_ERR_PTR(x) ((void __iomem *)ERR_PTR(x))
/**
* i915_vma_unpin_iomap - unpins the mapping returned from i915_vma_iomap
* @vma: VMA to unpin
*
* Unpins the previously iomapped VMA from i915_vma_pin_iomap().
*
drm/i915: Pull i915_vma_pin under the vm->mutex Replace the struct_mutex requirement for pinning the i915_vma with the local vm->mutex instead. Note that the vm->mutex is tainted by the shrinker (we require unbinding from inside fs-reclaim) and so we cannot allocate while holding that mutex. Instead we have to preallocate workers to do allocate and apply the PTE updates after we have we reserved their slot in the drm_mm (using fences to order the PTE writes with the GPU work and with later unbind). In adding the asynchronous vma binding, one subtle requirement is to avoid coupling the binding fence into the backing object->resv. That is the asynchronous binding only applies to the vma timeline itself and not to the pages as that is a more global timeline (the binding of one vma does not need to be ordered with another vma, nor does the implicit GEM fencing depend on a vma, only on writes to the backing store). Keeping the vma binding distinct from the backing store timelines is verified by a number of async gem_exec_fence and gem_exec_schedule tests. The way we do this is quite simple, we keep the fence for the vma binding separate and only wait on it as required, and never add it to the obj->resv itself. Another consequence in reducing the locking around the vma is the destruction of the vma is no longer globally serialised by struct_mutex. A natural solution would be to add a kref to i915_vma, but that requires decoupling the reference cycles, possibly by introducing a new i915_mm_pages object that is own by both obj->mm and vma->pages. However, we have not taken that route due to the overshadowing lmem/ttm discussions, and instead play a series of complicated games with trylocks to (hopefully) ensure that only one destruction path is called! v2: Add some commentary, and some helpers to reduce patch churn. Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20191004134015.13204-4-chris@chris-wilson.co.uk
2019-10-04 20:39:58 +07:00
* This function is only valid to be called on a VMA previously
* iomapped by the caller with i915_vma_pin_iomap().
*/
void i915_vma_unpin_iomap(struct i915_vma *vma);
static inline struct page *i915_vma_first_page(struct i915_vma *vma)
{
GEM_BUG_ON(!vma->pages);
return sg_page(vma->pages->sgl);
}
/**
* i915_vma_pin_fence - pin fencing state
* @vma: vma to pin fencing for
*
* This pins the fencing state (whether tiled or untiled) to make sure the
* vma (and its object) is ready to be used as a scanout target. Fencing
* status must be synchronize first by calling i915_vma_get_fence():
*
* The resulting fence pin reference must be released again with
* i915_vma_unpin_fence().
*
* Returns:
*
* True if the vma has a fence, false otherwise.
*/
int __must_check i915_vma_pin_fence(struct i915_vma *vma);
int __must_check i915_vma_revoke_fence(struct i915_vma *vma);
drm/i915: Pull i915_vma_pin under the vm->mutex Replace the struct_mutex requirement for pinning the i915_vma with the local vm->mutex instead. Note that the vm->mutex is tainted by the shrinker (we require unbinding from inside fs-reclaim) and so we cannot allocate while holding that mutex. Instead we have to preallocate workers to do allocate and apply the PTE updates after we have we reserved their slot in the drm_mm (using fences to order the PTE writes with the GPU work and with later unbind). In adding the asynchronous vma binding, one subtle requirement is to avoid coupling the binding fence into the backing object->resv. That is the asynchronous binding only applies to the vma timeline itself and not to the pages as that is a more global timeline (the binding of one vma does not need to be ordered with another vma, nor does the implicit GEM fencing depend on a vma, only on writes to the backing store). Keeping the vma binding distinct from the backing store timelines is verified by a number of async gem_exec_fence and gem_exec_schedule tests. The way we do this is quite simple, we keep the fence for the vma binding separate and only wait on it as required, and never add it to the obj->resv itself. Another consequence in reducing the locking around the vma is the destruction of the vma is no longer globally serialised by struct_mutex. A natural solution would be to add a kref to i915_vma, but that requires decoupling the reference cycles, possibly by introducing a new i915_mm_pages object that is own by both obj->mm and vma->pages. However, we have not taken that route due to the overshadowing lmem/ttm discussions, and instead play a series of complicated games with trylocks to (hopefully) ensure that only one destruction path is called! v2: Add some commentary, and some helpers to reduce patch churn. Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20191004134015.13204-4-chris@chris-wilson.co.uk
2019-10-04 20:39:58 +07:00
int __i915_vma_pin_fence(struct i915_vma *vma);
static inline void __i915_vma_unpin_fence(struct i915_vma *vma)
{
GEM_BUG_ON(atomic_read(&vma->fence->pin_count) <= 0);
atomic_dec(&vma->fence->pin_count);
}
/**
* i915_vma_unpin_fence - unpin fencing state
* @vma: vma to unpin fencing for
*
* This releases the fence pin reference acquired through
* i915_vma_pin_fence. It will handle both objects with and without an
* attached fence correctly, callers do not need to distinguish this.
*/
static inline void
i915_vma_unpin_fence(struct i915_vma *vma)
{
if (vma->fence)
__i915_vma_unpin_fence(vma);
}
void i915_vma_parked(struct intel_gt *gt);
drm/i915: Lazily unbind vma on close When userspace is passing around swapbuffers using DRI, we frequently have to open and close the same object in the foreign address space. This shows itself as the same object being rebound at roughly 30fps (with a second object also being rebound at 30fps), which involves us having to rewrite the page tables and maintain the drm_mm range manager every time. However, since the object still exists and it is only the local handle that disappears, if we are lazy and do not unbind the VMA immediately when the local user closes the object but defer it until the GPU is idle, then we can reuse the same VMA binding. We still have to be careful to mark the handle and lookup tables as closed to maintain the uABI, just allowing the underlying VMA to be resurrected if the user is able to access the same object from the same context again. If the object itself is destroyed (neither userspace keeping a handle to it), the VMA will be reaped immediately as usual. In the future, this will be even more useful as instantiating a new VMA for use on the GPU will become heavier. A nuisance indeed, so nip it in the bud. v2: s/__i915_vma_final_close/i915_vma_destroy/ etc. v3: Leave a hint as to why we deferred the unbind on close. Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20180503195115.22309-1-chris@chris-wilson.co.uk
2018-05-04 02:51:14 +07:00
#define for_each_until(cond) if (cond) break; else
/**
* for_each_ggtt_vma - Iterate over the GGTT VMA belonging to an object.
* @V: the #i915_vma iterator
* @OBJ: the #drm_i915_gem_object
*
* GGTT VMA are placed at the being of the object's vma_list, see
* vma_create(), so we can stop our walk as soon as we see a ppgtt VMA,
* or the list is empty ofc.
*/
#define for_each_ggtt_vma(V, OBJ) \
list_for_each_entry(V, &(OBJ)->vma.list, obj_link) \
for_each_until(!i915_vma_is_ggtt(V))
struct i915_vma *i915_vma_alloc(void);
void i915_vma_free(struct i915_vma *vma);
struct i915_vma *i915_vma_make_unshrinkable(struct i915_vma *vma);
void i915_vma_make_shrinkable(struct i915_vma *vma);
void i915_vma_make_purgeable(struct i915_vma *vma);
int i915_vma_wait_for_bind(struct i915_vma *vma);
drm/i915: Pull i915_vma_pin under the vm->mutex Replace the struct_mutex requirement for pinning the i915_vma with the local vm->mutex instead. Note that the vm->mutex is tainted by the shrinker (we require unbinding from inside fs-reclaim) and so we cannot allocate while holding that mutex. Instead we have to preallocate workers to do allocate and apply the PTE updates after we have we reserved their slot in the drm_mm (using fences to order the PTE writes with the GPU work and with later unbind). In adding the asynchronous vma binding, one subtle requirement is to avoid coupling the binding fence into the backing object->resv. That is the asynchronous binding only applies to the vma timeline itself and not to the pages as that is a more global timeline (the binding of one vma does not need to be ordered with another vma, nor does the implicit GEM fencing depend on a vma, only on writes to the backing store). Keeping the vma binding distinct from the backing store timelines is verified by a number of async gem_exec_fence and gem_exec_schedule tests. The way we do this is quite simple, we keep the fence for the vma binding separate and only wait on it as required, and never add it to the obj->resv itself. Another consequence in reducing the locking around the vma is the destruction of the vma is no longer globally serialised by struct_mutex. A natural solution would be to add a kref to i915_vma, but that requires decoupling the reference cycles, possibly by introducing a new i915_mm_pages object that is own by both obj->mm and vma->pages. However, we have not taken that route due to the overshadowing lmem/ttm discussions, and instead play a series of complicated games with trylocks to (hopefully) ensure that only one destruction path is called! v2: Add some commentary, and some helpers to reduce patch churn. Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20191004134015.13204-4-chris@chris-wilson.co.uk
2019-10-04 20:39:58 +07:00
static inline int i915_vma_sync(struct i915_vma *vma)
{
/* Wait for the asynchronous bindings and pending GPU reads */
return i915_active_wait(&vma->active);
}
#endif