[PATCH v7 03/24] drm/gpuvm: Introduce drm_gpuvm_madvise_ops_create
Himal Prasad Ghimiray
himal.prasad.ghimiray at intel.com
Thu Aug 14 10:59:29 UTC 2025
This ops is used to iterate over GPUVA's in the user-provided range
and split the existing sparse VMA's if the start or end of the input
range lies within it. The operations can create up to 2 REMAPS and 2 MAPs.
The primary use case is for drivers to assign attributes to GPU VAs in
the specified range without performing unmaps or merging mappings,
supporting fine-grained control over sparse va's.
v2
- use drm_gpuvm_sm_map_ops_create with flags instead of defining new
ops_create (Danilo)
- Add doc (Danilo)
v3
- Fix doc
- Fix unmapping check
v4
- Fix mapping for non madvise ops
v5
- Fix mapping (Matthew Brost)
- Rebase on top of struct changes
v6
- flag moved to map_req
v7 (Danilo)
- Use different functions
- Add kernel-doc
- Modify op_unmap_cb and op_map_cb to handle madvise and NULL ptr
- use gem_obj check in single place
Cc: Danilo Krummrich <dakr at kernel.org>
Cc: Matthew Brost <matthew.brost at intel.com>
Cc: Boris Brezillon <bbrezillon at kernel.org>
Cc: <dri-devel at lists.freedesktop.org>
Signed-off-by: Himal Prasad Ghimiray<himal.prasad.ghimiray at intel.com>
Reviewed-by: Matthew Brost <matthew.brost at intel.com> #v6
---
drivers/gpu/drm/drm_gpuvm.c | 225 ++++++++++++++++++++++++++++++------
include/drm/drm_gpuvm.h | 3 +
2 files changed, 191 insertions(+), 37 deletions(-)
diff --git a/drivers/gpu/drm/drm_gpuvm.c b/drivers/gpu/drm/drm_gpuvm.c
index 6c18cec70f11..d6bea8a4fffd 100644
--- a/drivers/gpu/drm/drm_gpuvm.c
+++ b/drivers/gpu/drm/drm_gpuvm.c
@@ -420,6 +420,71 @@
* new: |-----------|-----| (b.bo_offset=m,a.bo_offset=n+2)
*/
+/**
+ * DOC: Madvise Logic - Splitting and Traversal
+ *
+ * This logic handles GPU VA range updates by generating remap and map operations
+ * without performing unmaps or merging existing mappings.
+ *
+ * 1) The requested range lies entirely within a single drm_gpuva. The logic splits
+ * the existing mapping at the start and end boundaries and inserts a new map.
+ *
+ * ::
+ * a start end b
+ * pre: |-----------------------|
+ * drm_gpuva1
+ *
+ * a start end b
+ * new: |-----|=========|-------|
+ * remap map remap
+ *
+ * one REMAP and one MAP : Same behaviour as SPLIT and MERGE
+ *
+ * 2) The requested range spans multiple drm_gpuva regions. The logic traverses
+ * across boundaries, remapping the start and end segments, and inserting two
+ * map operations to cover the full range.
+ *
+ * :: a start b c end d
+ * pre: |------------------|--------------|------------------|
+ * drm_gpuva1 drm_gpuva2 drm_gpuva3
+ *
+ * a start b c end d
+ * new: |-------|==========|--------------|========|---------|
+ * remap1 map1 drm_gpuva2 map2 remap2
+ *
+ * two REMAPS and two MAPS
+ *
+ * 3) Either start or end lies within a drm_gpuva. A single remap and map operation
+ * are generated to update the affected portion.
+ *
+ *
+ * :: a/start b c end d
+ * pre: |------------------|--------------|------------------|
+ * drm_gpuva1 drm_gpuva2 drm_gpuva3
+ *
+ * a/start b c end d
+ * new: |------------------|--------------|========|---------|
+ * drm_gpuva1 drm_gpuva2 map1 remap1
+ *
+ * :: a start b c/end d
+ * pre: |------------------|--------------|------------------|
+ * drm_gpuva1 drm_gpuva2 drm_gpuva3
+ *
+ * a start b c/end d
+ * new: |-------|==========|--------------|------------------|
+ * remap1 map1 drm_gpuva2 drm_gpuva3
+ *
+ * one REMAP and one MAP
+ *
+ * 4) Both start and end align with existing drm_gpuva boundaries. No operations
+ * are needed as the range is already covered.
+ *
+ * 5) No existing drm_gpuvas. No operations.
+ *
+ * Unlike drm_gpuvm_sm_map_ops_create, this logic avoids unmaps and merging,
+ * focusing solely on remap and map operations for efficient traversal and update.
+ */
+
/**
* DOC: Locking
*
@@ -2063,6 +2128,9 @@ op_map_cb(const struct drm_gpuvm_ops *fn, void *priv,
{
struct drm_gpuva_op op = {};
+ if (!req)
+ return 0;
+
op.op = DRM_GPUVA_OP_MAP;
op.map.va.addr = req->map.va.addr;
op.map.va.range = req->map.va.range;
@@ -2092,10 +2160,13 @@ op_remap_cb(const struct drm_gpuvm_ops *fn, void *priv,
static int
op_unmap_cb(const struct drm_gpuvm_ops *fn, void *priv,
- struct drm_gpuva *va, bool merge)
+ struct drm_gpuva *va, bool merge, bool madvise)
{
struct drm_gpuva_op op = {};
+ if (madvise)
+ return 0;
+
op.op = DRM_GPUVA_OP_UNMAP;
op.unmap.va = va;
op.unmap.keep = merge;
@@ -2106,11 +2177,12 @@ op_unmap_cb(const struct drm_gpuvm_ops *fn, void *priv,
static int
__drm_gpuvm_sm_map(struct drm_gpuvm *gpuvm,
const struct drm_gpuvm_ops *ops, void *priv,
- const struct drm_gpuvm_map_req *req)
+ const struct drm_gpuvm_map_req *req,
+ bool madvise)
{
struct drm_gem_object *req_obj = req->map.gem.obj;
+ const struct drm_gpuvm_map_req *op_map = madvise ? NULL : req;
struct drm_gpuva *va, *next;
-
u64 req_offset = req->map.gem.offset;
u64 req_range = req->map.va.range;
u64 req_addr = req->map.va.addr;
@@ -2128,19 +2200,22 @@ __drm_gpuvm_sm_map(struct drm_gpuvm *gpuvm,
u64 end = addr + range;
bool merge = !!va->gem.obj;
+ if (madvise && obj)
+ continue;
+
if (addr == req_addr) {
merge &= obj == req_obj &&
offset == req_offset;
if (end == req_end) {
- ret = op_unmap_cb(ops, priv, va, merge);
+ ret = op_unmap_cb(ops, priv, va, merge, madvise);
if (ret)
return ret;
break;
}
if (end < req_end) {
- ret = op_unmap_cb(ops, priv, va, merge);
+ ret = op_unmap_cb(ops, priv, va, merge, madvise);
if (ret)
return ret;
continue;
@@ -2161,6 +2236,9 @@ __drm_gpuvm_sm_map(struct drm_gpuvm *gpuvm,
ret = op_remap_cb(ops, priv, NULL, &n, &u);
if (ret)
return ret;
+
+ if (madvise)
+ op_map = req;
break;
}
} else if (addr < req_addr) {
@@ -2181,6 +2259,9 @@ __drm_gpuvm_sm_map(struct drm_gpuvm *gpuvm,
ret = op_remap_cb(ops, priv, &p, NULL, &u);
if (ret)
return ret;
+
+ if (madvise)
+ op_map = req;
break;
}
@@ -2188,6 +2269,18 @@ __drm_gpuvm_sm_map(struct drm_gpuvm *gpuvm,
ret = op_remap_cb(ops, priv, &p, NULL, &u);
if (ret)
return ret;
+
+ if (madvise) {
+ struct drm_gpuvm_map_req map_req = {
+ .map.va.addr = req_addr,
+ .map.va.range = end - req_addr,
+ };
+
+ ret = op_map_cb(ops, priv, &map_req);
+ if (ret)
+ return ret;
+ }
+
continue;
}
@@ -2203,6 +2296,9 @@ __drm_gpuvm_sm_map(struct drm_gpuvm *gpuvm,
ret = op_remap_cb(ops, priv, &p, &n, &u);
if (ret)
return ret;
+
+ if (madvise)
+ op_map = req;
break;
}
} else if (addr > req_addr) {
@@ -2211,16 +2307,18 @@ __drm_gpuvm_sm_map(struct drm_gpuvm *gpuvm,
(addr - req_addr);
if (end == req_end) {
- ret = op_unmap_cb(ops, priv, va, merge);
+ ret = op_unmap_cb(ops, priv, va, merge, madvise);
if (ret)
return ret;
+
break;
}
if (end < req_end) {
- ret = op_unmap_cb(ops, priv, va, merge);
+ ret = op_unmap_cb(ops, priv, va, merge, madvise);
if (ret)
return ret;
+
continue;
}
@@ -2239,12 +2337,20 @@ __drm_gpuvm_sm_map(struct drm_gpuvm *gpuvm,
ret = op_remap_cb(ops, priv, NULL, &n, &u);
if (ret)
return ret;
+
+ if (madvise) {
+ struct drm_gpuvm_map_req map_req = {
+ .map.va.addr = addr,
+ .map.va.range = req_end - addr,
+ };
+
+ return op_map_cb(ops, priv, &map_req);
+ }
break;
}
}
}
-
- return op_map_cb(ops, priv, req);
+ return op_map_cb(ops, priv, op_map);
}
static int
@@ -2296,7 +2402,7 @@ __drm_gpuvm_sm_unmap(struct drm_gpuvm *gpuvm,
if (ret)
return ret;
} else {
- ret = op_unmap_cb(ops, priv, va, false);
+ ret = op_unmap_cb(ops, priv, va, false, false);
if (ret)
return ret;
}
@@ -2345,7 +2451,7 @@ drm_gpuvm_sm_map(struct drm_gpuvm *gpuvm, void *priv,
ops->sm_step_unmap)))
return -EINVAL;
- return __drm_gpuvm_sm_map(gpuvm, ops, priv, req);
+ return __drm_gpuvm_sm_map(gpuvm, ops, priv, req, false);
}
EXPORT_SYMBOL_GPL(drm_gpuvm_sm_map);
@@ -2483,7 +2589,7 @@ drm_gpuvm_sm_map_exec_lock(struct drm_gpuvm *gpuvm,
return ret;
}
- return __drm_gpuvm_sm_map(gpuvm, &lock_ops, exec, req);
+ return __drm_gpuvm_sm_map(gpuvm, &lock_ops, exec, req, false);
}
EXPORT_SYMBOL_GPL(drm_gpuvm_sm_map_exec_lock);
@@ -2602,6 +2708,38 @@ static const struct drm_gpuvm_ops gpuvm_list_ops = {
.sm_step_unmap = drm_gpuva_sm_step,
};
+static struct drm_gpuva_ops *
+__drm_gpuvm_sm_map_ops_create(struct drm_gpuvm *gpuvm,
+ const struct drm_gpuvm_map_req *req,
+ bool madvise)
+{
+ struct drm_gpuva_ops *ops;
+ struct {
+ struct drm_gpuvm *vm;
+ struct drm_gpuva_ops *ops;
+ } args;
+ int ret;
+
+ ops = kzalloc(sizeof(*ops), GFP_KERNEL);
+ if (unlikely(!ops))
+ return ERR_PTR(-ENOMEM);
+
+ INIT_LIST_HEAD(&ops->list);
+
+ args.vm = gpuvm;
+ args.ops = ops;
+
+ ret = __drm_gpuvm_sm_map(gpuvm, &gpuvm_list_ops, &args, req, madvise);
+ if (ret)
+ goto err_free_ops;
+
+ return ops;
+
+err_free_ops:
+ drm_gpuva_ops_free(gpuvm, ops);
+ return ERR_PTR(ret);
+}
+
/**
* drm_gpuvm_sm_map_ops_create() - creates the &drm_gpuva_ops to split and merge
* @gpuvm: the &drm_gpuvm representing the GPU VA space
@@ -2635,34 +2773,47 @@ struct drm_gpuva_ops *
drm_gpuvm_sm_map_ops_create(struct drm_gpuvm *gpuvm,
const struct drm_gpuvm_map_req *req)
{
- struct drm_gpuva_ops *ops;
- struct {
- struct drm_gpuvm *vm;
- struct drm_gpuva_ops *ops;
- } args;
- int ret;
-
- ops = kzalloc(sizeof(*ops), GFP_KERNEL);
- if (unlikely(!ops))
- return ERR_PTR(-ENOMEM);
-
- INIT_LIST_HEAD(&ops->list);
-
- args.vm = gpuvm;
- args.ops = ops;
-
- ret = __drm_gpuvm_sm_map(gpuvm, &gpuvm_list_ops, &args, req);
- if (ret)
- goto err_free_ops;
-
- return ops;
-
-err_free_ops:
- drm_gpuva_ops_free(gpuvm, ops);
- return ERR_PTR(ret);
+ return __drm_gpuvm_sm_map_ops_create(gpuvm, req, false);
}
EXPORT_SYMBOL_GPL(drm_gpuvm_sm_map_ops_create);
+/**
+ * drm_gpuvm_madvise_ops_create() - creates the &drm_gpuva_ops to split
+ * @gpuvm: the &drm_gpuvm representing the GPU VA space
+ * @req: map request arguments
+ *
+ * This function creates a list of operations to perform splitting
+ * of existent mapping(s) at start or end, based on the request map.
+ *
+ * The list can be iterated with &drm_gpuva_for_each_op and must be processed
+ * in the given order. It can contain map and remap operations, but it
+ * also can be empty if no operation is required, e.g. if the requested mapping
+ * already exists is the exact same way.
+ *
+ * There will be no unmap operations, a maximum of two remap operations and two
+ * map operations. The two map operations correspond to: one from start to the
+ * end of drm_gpuvaX, and another from the start of drm_gpuvaY to end.
+ *
+ * Note that before calling this function again with another mapping request it
+ * is necessary to update the &drm_gpuvm's view of the GPU VA space. The
+ * previously obtained operations must be either processed or abandoned. To
+ * update the &drm_gpuvm's view of the GPU VA space drm_gpuva_insert(),
+ * drm_gpuva_destroy_locked() and/or drm_gpuva_destroy_unlocked() should be
+ * used.
+ *
+ * After the caller finished processing the returned &drm_gpuva_ops, they must
+ * be freed with &drm_gpuva_ops_free.
+ *
+ * Returns: a pointer to the &drm_gpuva_ops on success, an ERR_PTR on failure
+ */
+struct drm_gpuva_ops *
+drm_gpuvm_madvise_ops_create(struct drm_gpuvm *gpuvm,
+ const struct drm_gpuvm_map_req *req)
+{
+ return __drm_gpuvm_sm_map_ops_create(gpuvm, req, true);
+}
+EXPORT_SYMBOL_GPL(drm_gpuvm_madvise_ops_create);
+
/**
* drm_gpuvm_sm_unmap_ops_create() - creates the &drm_gpuva_ops to split on
* unmap
diff --git a/include/drm/drm_gpuvm.h b/include/drm/drm_gpuvm.h
index 05347ac6cc73..4a22b9d848f7 100644
--- a/include/drm/drm_gpuvm.h
+++ b/include/drm/drm_gpuvm.h
@@ -1062,6 +1062,9 @@ struct drm_gpuvm_map_req {
struct drm_gpuva_ops *
drm_gpuvm_sm_map_ops_create(struct drm_gpuvm *gpuvm,
const struct drm_gpuvm_map_req *req);
+struct drm_gpuva_ops *
+drm_gpuvm_madvise_ops_create(struct drm_gpuvm *gpuvm,
+ const struct drm_gpuvm_map_req *req);
struct drm_gpuva_ops *
drm_gpuvm_sm_unmap_ops_create(struct drm_gpuvm *gpuvm,
--
2.34.1
More information about the Intel-xe
mailing list