880 lines
33 KiB
C
880 lines
33 KiB
C
/*******************************************************************************
|
|
Copyright (c) 2015-2022 NVIDIA 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 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.
|
|
|
|
*******************************************************************************/
|
|
|
|
#include "uvm_range_group.h"
|
|
#include "uvm_va_space.h"
|
|
#include "uvm_va_range.h"
|
|
#include "uvm_va_block.h"
|
|
#include "uvm_tools.h"
|
|
#include "uvm_ioctl.h"
|
|
#include "uvm_types.h"
|
|
#include "uvm_api.h"
|
|
#include "uvm_test.h"
|
|
|
|
static struct kmem_cache *g_uvm_range_group_cache __read_mostly;
|
|
static struct kmem_cache *g_uvm_range_group_range_cache __read_mostly;
|
|
|
|
NV_STATUS uvm_range_group_init(void)
|
|
{
|
|
g_uvm_range_group_cache = NV_KMEM_CACHE_CREATE("uvm_range_group_t", uvm_range_group_t);
|
|
if (!g_uvm_range_group_cache)
|
|
return NV_ERR_NO_MEMORY;
|
|
|
|
g_uvm_range_group_range_cache = NV_KMEM_CACHE_CREATE("uvm_range_group_range_t", uvm_range_group_range_t);
|
|
if (!g_uvm_range_group_range_cache)
|
|
return NV_ERR_NO_MEMORY;
|
|
|
|
return NV_OK;
|
|
}
|
|
|
|
void uvm_range_group_exit(void)
|
|
{
|
|
kmem_cache_destroy_safe(&g_uvm_range_group_cache);
|
|
kmem_cache_destroy_safe(&g_uvm_range_group_range_cache);
|
|
}
|
|
|
|
static uvm_range_group_range_t *range_group_range_create(uvm_range_group_t *range_group,
|
|
NvU64 start,
|
|
NvU64 end)
|
|
{
|
|
uvm_range_group_range_t *rgr = nv_kmem_cache_zalloc(g_uvm_range_group_range_cache, NV_UVM_GFP_FLAGS);
|
|
if (rgr == NULL)
|
|
return NULL;
|
|
|
|
UVM_ASSERT(range_group != NULL);
|
|
INIT_LIST_HEAD(&rgr->range_group_migrated_list_node);
|
|
list_add(&rgr->range_group_list_node, &range_group->ranges);
|
|
rgr->range_group = range_group;
|
|
|
|
rgr->node.start = start;
|
|
rgr->node.end = end;
|
|
|
|
return rgr;
|
|
}
|
|
|
|
static void uvm_range_group_range_destroy(uvm_range_group_range_t *rgr)
|
|
{
|
|
if (rgr == NULL)
|
|
return;
|
|
|
|
list_del(&rgr->range_group_list_node);
|
|
|
|
// The VA space should be locked in write mode when this function is called,
|
|
// so we don't need to acquire the migrated_list lock.
|
|
list_del(&rgr->range_group_migrated_list_node);
|
|
kmem_cache_free(g_uvm_range_group_range_cache, rgr);
|
|
}
|
|
|
|
NV_STATUS uvm_api_create_range_group(UVM_CREATE_RANGE_GROUP_PARAMS *params, struct file *filp)
|
|
{
|
|
uvm_va_space_t *va_space = uvm_va_space_get(filp);
|
|
uvm_range_group_t *range_group = NULL;
|
|
NV_STATUS status = NV_OK;
|
|
int ret;
|
|
|
|
range_group = nv_kmem_cache_zalloc(g_uvm_range_group_cache, NV_UVM_GFP_FLAGS);
|
|
if (!range_group)
|
|
return NV_ERR_NO_MEMORY;
|
|
|
|
range_group->id = atomic64_inc_return(&va_space->range_group_id_counter);
|
|
UVM_ASSERT(range_group->id != UVM_RANGE_GROUP_ID_NONE);
|
|
|
|
atomic_set(&range_group->allow_migration, 1);
|
|
INIT_LIST_HEAD(&range_group->ranges);
|
|
INIT_LIST_HEAD(&range_group->migrated_ranges);
|
|
uvm_spin_lock_init(&range_group->migrated_ranges_lock, UVM_LOCK_ORDER_LEAF);
|
|
|
|
uvm_va_space_down_write(va_space);
|
|
|
|
ret = radix_tree_insert(&va_space->range_groups, range_group->id, range_group);
|
|
status = errno_to_nv_status(ret);
|
|
if (status != NV_OK) {
|
|
kmem_cache_free(g_uvm_range_group_cache, range_group);
|
|
goto done;
|
|
}
|
|
|
|
params->rangeGroupId = range_group->id;
|
|
|
|
done:
|
|
uvm_va_space_up_write(va_space);
|
|
|
|
return status;
|
|
}
|
|
|
|
static void uvm_range_group_destroy(uvm_va_space_t *va_space, uvm_range_group_t *range_group)
|
|
{
|
|
uvm_range_group_range_t *rgr, *tmp;
|
|
|
|
list_for_each_entry_safe(rgr, tmp, &range_group->ranges, range_group_list_node) {
|
|
UVM_ASSERT(rgr->range_group == range_group);
|
|
|
|
uvm_range_tree_remove(&va_space->range_group_ranges, &rgr->node);
|
|
uvm_range_group_range_destroy(rgr);
|
|
}
|
|
|
|
UVM_ASSERT(list_empty(&range_group->migrated_ranges));
|
|
|
|
kmem_cache_free(g_uvm_range_group_cache, range_group);
|
|
}
|
|
|
|
NV_STATUS uvm_api_destroy_range_group(UVM_DESTROY_RANGE_GROUP_PARAMS *params, struct file *filp)
|
|
{
|
|
uvm_va_space_t *va_space = uvm_va_space_get(filp);
|
|
uvm_range_group_t *range_group = NULL;
|
|
NV_STATUS status = NV_OK;
|
|
|
|
uvm_va_space_down_write(va_space);
|
|
|
|
range_group = radix_tree_delete(&va_space->range_groups, params->rangeGroupId);
|
|
if (!range_group) {
|
|
status = NV_ERR_OBJECT_NOT_FOUND;
|
|
goto done;
|
|
}
|
|
|
|
uvm_range_group_destroy(va_space, range_group);
|
|
|
|
done:
|
|
uvm_va_space_up_write(va_space);
|
|
return status;
|
|
}
|
|
|
|
void uvm_range_group_radix_tree_destroy(uvm_va_space_t *va_space)
|
|
{
|
|
uvm_range_group_t *range_group = NULL;
|
|
struct radix_tree_root *root = &va_space->range_groups;
|
|
NvU64 index = 1;
|
|
|
|
while (radix_tree_gang_lookup(root, (void**)&range_group, index, 1)) {
|
|
UVM_ASSERT(range_group);
|
|
radix_tree_delete(root, range_group->id);
|
|
index = range_group->id + 1;
|
|
uvm_range_group_destroy(va_space, range_group);
|
|
}
|
|
}
|
|
|
|
static NV_STATUS uvm_range_group_va_range_migrate_block_locked(uvm_va_range_t *va_range,
|
|
uvm_va_block_t *va_block,
|
|
uvm_va_block_retry_t *va_block_retry,
|
|
uvm_va_block_context_t *va_block_context,
|
|
uvm_va_block_region_t region,
|
|
uvm_tracker_t *tracker)
|
|
{
|
|
NV_STATUS status;
|
|
NV_STATUS tracker_status;
|
|
uvm_gpu_id_t gpu_id;
|
|
uvm_processor_mask_t *map_mask = &va_block_context->caller_processor_mask;
|
|
uvm_va_policy_t *policy = uvm_va_range_get_policy(va_range);
|
|
|
|
// Set the migration CPU NUMA node from the policy.
|
|
va_block_context->make_resident.dest_nid = policy->preferred_nid;
|
|
|
|
// Unmapping UVM_ID_CPU is guaranteed to never fail
|
|
status = uvm_va_block_unmap(va_block, va_block_context, UVM_ID_CPU, region, NULL, NULL);
|
|
UVM_ASSERT(status == NV_OK);
|
|
|
|
if (uvm_va_policy_is_read_duplicate(uvm_va_range_get_policy(va_range), va_range->va_space)) {
|
|
status = uvm_va_block_make_resident_read_duplicate(va_block,
|
|
va_block_retry,
|
|
va_block_context,
|
|
policy->preferred_location,
|
|
region,
|
|
NULL,
|
|
NULL,
|
|
UVM_MAKE_RESIDENT_CAUSE_API_SET_RANGE_GROUP);
|
|
}
|
|
else {
|
|
status = uvm_va_block_make_resident(va_block,
|
|
va_block_retry,
|
|
va_block_context,
|
|
policy->preferred_location,
|
|
region,
|
|
NULL,
|
|
NULL,
|
|
UVM_MAKE_RESIDENT_CAUSE_API_SET_RANGE_GROUP);
|
|
}
|
|
|
|
if (status != NV_OK)
|
|
return status;
|
|
|
|
// 1- Map all UVM-Lite SetAccessedBy GPUs and the preferred location with
|
|
// RWA permission
|
|
status = uvm_va_block_map_mask(va_block,
|
|
va_block_context,
|
|
&va_range->uvm_lite_gpus,
|
|
region,
|
|
NULL,
|
|
UVM_PROT_READ_WRITE_ATOMIC,
|
|
UvmEventMapRemoteCauseCoherence);
|
|
if (status != NV_OK)
|
|
goto out;
|
|
|
|
// 2- Map faultable SetAccessedBy GPUs.
|
|
uvm_processor_mask_and(map_mask,
|
|
&uvm_va_range_get_policy(va_range)->accessed_by,
|
|
&va_range->va_space->can_access[uvm_id_value(policy->preferred_location)]);
|
|
uvm_processor_mask_andnot(map_mask, map_mask, &va_range->uvm_lite_gpus);
|
|
|
|
for_each_gpu_id_in_mask(gpu_id, map_mask) {
|
|
status = uvm_va_block_add_mappings(va_block,
|
|
va_block_context,
|
|
gpu_id,
|
|
region,
|
|
NULL,
|
|
UvmEventMapRemoteCausePolicy);
|
|
if (status != NV_OK)
|
|
goto out;
|
|
}
|
|
|
|
out:
|
|
tracker_status = uvm_tracker_add_tracker_safe(tracker, &va_block->tracker);
|
|
|
|
return status == NV_OK ? tracker_status : status;
|
|
}
|
|
|
|
NV_STATUS uvm_range_group_va_range_migrate(uvm_va_range_t *va_range,
|
|
NvU64 start,
|
|
NvU64 end,
|
|
uvm_tracker_t *out_tracker)
|
|
{
|
|
uvm_va_block_t *va_block = NULL;
|
|
size_t i = 0;
|
|
NV_STATUS status = NV_OK;
|
|
uvm_va_block_retry_t va_block_retry;
|
|
uvm_va_block_context_t *va_block_context;
|
|
|
|
// This path is only called for non-migratable range groups so it never
|
|
// creates CPU mappings, meaning no mm is needed.
|
|
va_block_context = uvm_va_block_context_alloc(NULL);
|
|
if (!va_block_context)
|
|
return NV_ERR_NO_MEMORY;
|
|
|
|
uvm_assert_rwsem_locked(&va_range->va_space->lock);
|
|
|
|
// Iterate over blocks, populating them if necessary
|
|
for (i = uvm_va_range_block_index(va_range, start); i <= uvm_va_range_block_index(va_range, end); ++i) {
|
|
uvm_va_block_region_t region;
|
|
status = uvm_va_range_block_create(va_range, i, &va_block);
|
|
if (status != NV_OK)
|
|
break;
|
|
|
|
region = uvm_va_block_region_from_start_end(va_block,
|
|
max(start, va_block->start),
|
|
min(end, va_block->end));
|
|
|
|
status = UVM_VA_BLOCK_LOCK_RETRY(va_block, &va_block_retry,
|
|
uvm_range_group_va_range_migrate_block_locked(va_range,
|
|
va_block,
|
|
&va_block_retry,
|
|
va_block_context,
|
|
region,
|
|
out_tracker));
|
|
if (status != NV_OK)
|
|
break;
|
|
}
|
|
|
|
uvm_va_block_context_free(va_block_context);
|
|
|
|
return status;
|
|
}
|
|
|
|
NV_STATUS uvm_api_set_range_group(UVM_SET_RANGE_GROUP_PARAMS *params, struct file *filp)
|
|
{
|
|
uvm_va_space_t *va_space = uvm_va_space_get(filp);
|
|
uvm_range_group_t *range_group = NULL;
|
|
uvm_va_range_t *va_range, *va_range_last;
|
|
unsigned long long last_address = params->requestedBase + params->length - 1;
|
|
uvm_tracker_t local_tracker;
|
|
NV_STATUS tracker_status;
|
|
NV_STATUS status = NV_OK;
|
|
bool has_va_space_write_lock;
|
|
bool migratable;
|
|
|
|
UVM_ASSERT(va_space);
|
|
|
|
// Check address and length alignment
|
|
if (uvm_api_range_invalid(params->requestedBase, params->length))
|
|
return NV_ERR_INVALID_ADDRESS;
|
|
|
|
uvm_tracker_init(&local_tracker);
|
|
|
|
uvm_va_space_down_write(va_space);
|
|
has_va_space_write_lock = true;
|
|
|
|
// Check that range group exists
|
|
range_group = radix_tree_lookup(&va_space->range_groups, params->rangeGroupId);
|
|
if (!range_group && (params->rangeGroupId != UVM_RANGE_GROUP_ID_NONE)) {
|
|
status = NV_ERR_OBJECT_NOT_FOUND;
|
|
goto done;
|
|
}
|
|
|
|
// If the desired range group is not migratable, any overlapping va_ranges
|
|
// must have a preferred location
|
|
migratable = uvm_range_group_migratable(range_group);
|
|
va_range_last = NULL;
|
|
uvm_for_each_managed_va_range_in_contig(va_range, va_space, params->requestedBase, last_address) {
|
|
va_range_last = va_range;
|
|
if (!migratable && UVM_ID_IS_INVALID(uvm_va_range_get_policy(va_range)->preferred_location)) {
|
|
status = NV_ERR_INVALID_ADDRESS;
|
|
goto done;
|
|
}
|
|
}
|
|
|
|
// Check that we were able to iterate over the entire range without any gaps
|
|
if (!va_range_last || va_range_last->node.end < last_address) {
|
|
status = NV_ERR_INVALID_ADDRESS;
|
|
goto done;
|
|
}
|
|
|
|
status = uvm_range_group_assign_range(va_space, range_group, params->requestedBase, last_address);
|
|
if (status != NV_OK)
|
|
goto done;
|
|
|
|
// Early exit if no need for migration
|
|
if (uvm_range_group_migratable(range_group))
|
|
goto done;
|
|
|
|
uvm_va_space_downgrade_write(va_space);
|
|
has_va_space_write_lock = false;
|
|
|
|
// Already checked for gaps above
|
|
uvm_for_each_va_range_in(va_range, va_space, params->requestedBase, last_address) {
|
|
status = uvm_range_group_va_range_migrate(va_range,
|
|
max(va_range->node.start, params->requestedBase),
|
|
min(va_range->node.end, last_address),
|
|
&local_tracker);
|
|
if (status != NV_OK)
|
|
goto done;
|
|
}
|
|
|
|
done:
|
|
tracker_status = uvm_tracker_wait_deinit(&local_tracker);
|
|
|
|
if (has_va_space_write_lock)
|
|
uvm_va_space_up_write(va_space);
|
|
else
|
|
uvm_va_space_up_read(va_space);
|
|
|
|
return status == NV_OK ? tracker_status : status;
|
|
}
|
|
|
|
static NV_STATUS uvm_range_group_prevent_migration(uvm_range_group_t *range_group,
|
|
uvm_va_space_t *va_space)
|
|
{
|
|
uvm_range_group_range_t *rgr = NULL;
|
|
uvm_va_range_t *va_range;
|
|
uvm_processor_id_t preferred_location;
|
|
uvm_tracker_t local_tracker = UVM_TRACKER_INIT();
|
|
NV_STATUS tracker_status;
|
|
NV_STATUS status = NV_OK;
|
|
|
|
LIST_HEAD(local_migrated_ranges);
|
|
|
|
UVM_ASSERT(range_group);
|
|
UVM_ASSERT(va_space);
|
|
|
|
uvm_assert_rwsem_locked(&va_space->lock);
|
|
|
|
// Move the range group's migrated_ranges list to the local_migrated_ranges
|
|
// list and process it from there.
|
|
uvm_spin_lock(&range_group->migrated_ranges_lock);
|
|
list_replace_init(&range_group->migrated_ranges, &local_migrated_ranges);
|
|
uvm_spin_unlock(&range_group->migrated_ranges_lock);
|
|
|
|
while (true) {
|
|
// Delete each item from the beginning of the list.
|
|
uvm_spin_lock(&range_group->migrated_ranges_lock);
|
|
rgr = list_first_entry_or_null(&local_migrated_ranges,
|
|
uvm_range_group_range_t,
|
|
range_group_migrated_list_node);
|
|
if (rgr)
|
|
list_del_init(&rgr->range_group_migrated_list_node);
|
|
uvm_spin_unlock(&range_group->migrated_ranges_lock);
|
|
|
|
if (!rgr)
|
|
break;
|
|
|
|
uvm_for_each_va_range_in(va_range, va_space, rgr->node.start, rgr->node.end) {
|
|
// VA ranges need to have a preferred location set in order for their
|
|
// range group to be set to non-migratable.
|
|
preferred_location = uvm_va_range_get_policy(va_range)->preferred_location;
|
|
if (UVM_ID_IS_INVALID(preferred_location)) {
|
|
status = NV_ERR_INVALID_ARGUMENT;
|
|
goto done;
|
|
}
|
|
|
|
// If the preferred location is a GPU, check that it's not
|
|
// fault-capable
|
|
if (UVM_ID_IS_GPU(preferred_location) &&
|
|
uvm_processor_mask_test(&va_space->faultable_processors, preferred_location)) {
|
|
status = NV_ERR_INVALID_DEVICE;
|
|
goto done;
|
|
}
|
|
|
|
// Check that all UVM-Lite GPUs are able to access the
|
|
// preferred location
|
|
if (!uvm_processor_mask_subset(&va_range->uvm_lite_gpus,
|
|
&va_space->accessible_from[uvm_id_value(preferred_location)])) {
|
|
status = NV_ERR_INVALID_DEVICE;
|
|
goto done;
|
|
}
|
|
|
|
// Perform the migration of the VA range.
|
|
status = uvm_range_group_va_range_migrate(va_range,
|
|
max(va_range->node.start, rgr->node.start),
|
|
min(va_range->node.end, rgr->node.end),
|
|
&local_tracker);
|
|
if (status != NV_OK)
|
|
goto done;
|
|
}
|
|
}
|
|
|
|
done:
|
|
tracker_status = uvm_tracker_wait_deinit(&local_tracker);
|
|
if (status == NV_OK)
|
|
status = tracker_status;
|
|
|
|
// We may have exited early, in which case rgr may be unprocessed and
|
|
// local_migrated_ranges may not be empty. These entries should be put back
|
|
// on range_group's migrated_ranges list.
|
|
if (status != NV_OK) {
|
|
uvm_spin_lock(&range_group->migrated_ranges_lock);
|
|
if (rgr)
|
|
list_move_tail(&rgr->range_group_migrated_list_node, &range_group->migrated_ranges);
|
|
list_splice_tail(&local_migrated_ranges, &range_group->migrated_ranges);
|
|
uvm_spin_unlock(&range_group->migrated_ranges_lock);
|
|
}
|
|
|
|
return status;
|
|
}
|
|
|
|
static NV_STATUS uvm_range_groups_set_migration_policy(uvm_va_space_t *va_space,
|
|
NvU64 *range_group_ids,
|
|
NvU64 num_group_ids,
|
|
bool allow_migration)
|
|
{
|
|
NV_STATUS status = NV_OK;
|
|
NvU64 i;
|
|
uvm_range_group_t *range_groups[UVM_MAX_RANGE_GROUPS_PER_IOCTL_CALL];
|
|
|
|
UVM_ASSERT(va_space);
|
|
|
|
if (!range_group_ids || num_group_ids == 0 || num_group_ids > UVM_MAX_RANGE_GROUPS_PER_IOCTL_CALL)
|
|
return NV_ERR_INVALID_ARGUMENT;
|
|
|
|
if (!allow_migration) {
|
|
// We take the VA space in write mode to perform range group lookup and
|
|
// change the migratability to false. This will wait for any concurrent
|
|
// fault/migration to finish.
|
|
//
|
|
// TODO: Bug 1878225: Use a finer grain synchronization mechanism to
|
|
// prevent races with concurrent fault handling/migration operations.
|
|
uvm_va_space_down_write(va_space);
|
|
}
|
|
else {
|
|
// It is safe to allow migration with the VA space in read mode since
|
|
// it just flips the value of allow_migration.
|
|
uvm_va_space_down_read(va_space);
|
|
}
|
|
|
|
for (i = 0; i < num_group_ids; ++i) {
|
|
range_groups[i] = radix_tree_lookup(&va_space->range_groups, range_group_ids[i]);
|
|
if (!range_groups[i]) {
|
|
if (!allow_migration)
|
|
uvm_va_space_up_write(va_space);
|
|
else
|
|
uvm_va_space_up_read(va_space);
|
|
|
|
return NV_ERR_OBJECT_NOT_FOUND;
|
|
}
|
|
|
|
atomic_set(&range_groups[i]->allow_migration, !!allow_migration);
|
|
}
|
|
|
|
// If we are allowing migration, we are done.
|
|
if (!allow_migration) {
|
|
// Any fault handler/migration that executes after downgrading the lock
|
|
// mode will see migratability disabled.
|
|
uvm_va_space_downgrade_write(va_space);
|
|
|
|
for (i = 0; i < num_group_ids; ++i) {
|
|
status = uvm_range_group_prevent_migration(range_groups[i], va_space);
|
|
if (status != NV_OK)
|
|
break;
|
|
}
|
|
}
|
|
|
|
uvm_va_space_up_read(va_space);
|
|
return status;
|
|
}
|
|
|
|
static uvm_range_group_range_t *range_group_range_container(uvm_range_tree_node_t *node)
|
|
{
|
|
if (node == NULL)
|
|
return NULL;
|
|
return container_of(node, uvm_range_group_range_t, node);
|
|
}
|
|
|
|
static uvm_range_group_range_t *range_group_range_prev(uvm_va_space_t *va_space, uvm_range_group_range_t *range)
|
|
{
|
|
uvm_range_tree_node_t *node = uvm_range_tree_prev(&va_space->range_group_ranges, &range->node);
|
|
return range_group_range_container(node);
|
|
}
|
|
|
|
static uvm_range_group_range_t *range_group_range_next(uvm_va_space_t *va_space, uvm_range_group_range_t *range)
|
|
{
|
|
uvm_range_tree_node_t *node = uvm_range_tree_next(&va_space->range_group_ranges, &range->node);
|
|
return range_group_range_container(node);
|
|
}
|
|
|
|
NV_STATUS uvm_range_group_assign_range(uvm_va_space_t *va_space, uvm_range_group_t *range_group, NvU64 start, NvU64 end)
|
|
{
|
|
NV_STATUS status;
|
|
uvm_range_group_range_t *rgr;
|
|
uvm_range_group_range_t *temp;
|
|
uvm_range_group_range_t *next;
|
|
uvm_range_group_range_t *prev;
|
|
uvm_range_group_range_t *new_rgr = NULL;
|
|
LIST_HEAD(internal_nodes);
|
|
|
|
uvm_assert_rwsem_locked_write(&va_space->lock);
|
|
|
|
if (range_group != NULL) {
|
|
new_rgr = range_group_range_create(range_group, start, end);
|
|
if (new_rgr == NULL)
|
|
return NV_ERR_NO_MEMORY;
|
|
}
|
|
|
|
uvm_range_group_for_each_range_in(rgr, va_space, start, end) {
|
|
if (rgr->node.start < start && rgr->node.end > end) {
|
|
// The region described by [start, end] lies entirely within rgr
|
|
// and does not sit on rgr's boundary.
|
|
NvU64 orig_end = rgr->node.end;
|
|
|
|
// Check if the rgr is already part of the specified range group and,
|
|
// if so, do nothing.
|
|
if (rgr->range_group == range_group) {
|
|
uvm_range_group_range_destroy(new_rgr);
|
|
return NV_OK;
|
|
}
|
|
|
|
// rgr needs to be split to make room for the new range group range.
|
|
// Do this by first creating a new range group range called "next"
|
|
// for the region of memory just above new_rgr, then shrink rgr
|
|
// down so that it fits just below new_rgr.
|
|
next = range_group_range_create(rgr->range_group, end + 1, orig_end);
|
|
if (next == NULL) {
|
|
uvm_range_group_range_destroy(new_rgr);
|
|
return NV_ERR_NO_MEMORY;
|
|
}
|
|
|
|
uvm_range_tree_shrink_node(&va_space->range_group_ranges, &rgr->node, rgr->node.start, start - 1);
|
|
status = uvm_range_tree_add(&va_space->range_group_ranges, &next->node);
|
|
UVM_ASSERT(status == NV_OK);
|
|
|
|
// Check if we need to add "next" to the range group's migrated list.
|
|
uvm_spin_lock(&rgr->range_group->migrated_ranges_lock);
|
|
if (!list_empty(&rgr->range_group_migrated_list_node))
|
|
list_move_tail(&next->range_group_migrated_list_node, &next->range_group->migrated_ranges);
|
|
uvm_spin_unlock(&rgr->range_group->migrated_ranges_lock);
|
|
}
|
|
else if (rgr->node.start < start) {
|
|
// The region [start, end] overlaps with the end of rgr.
|
|
// Shrink rgr by moving its end downward.
|
|
uvm_range_tree_shrink_node(&va_space->range_group_ranges, &rgr->node, rgr->node.start, start - 1);
|
|
}
|
|
else if (rgr->node.end > end) {
|
|
// The region [start, end] overlaps with the begining of rgr.
|
|
// Shrink rgr by moving its beginning upward.
|
|
uvm_range_tree_shrink_node(&va_space->range_group_ranges, &rgr->node, end + 1, rgr->node.end);
|
|
}
|
|
else {
|
|
// rgr lies entirely within [start, end]. rgr should be destroyed.
|
|
list_move(&rgr->range_group_list_node, &internal_nodes);
|
|
}
|
|
}
|
|
|
|
// Always add new_rgr to range_group's migrated list.
|
|
if (new_rgr) {
|
|
uvm_spin_lock(&new_rgr->range_group->migrated_ranges_lock);
|
|
list_move_tail(&new_rgr->range_group_migrated_list_node, &new_rgr->range_group->migrated_ranges);
|
|
uvm_spin_unlock(&new_rgr->range_group->migrated_ranges_lock);
|
|
}
|
|
|
|
list_for_each_entry_safe(rgr, temp, &internal_nodes, range_group_list_node) {
|
|
uvm_range_tree_remove(&va_space->range_group_ranges, &rgr->node);
|
|
uvm_range_group_range_destroy(rgr);
|
|
}
|
|
|
|
if (range_group == NULL)
|
|
return NV_OK;
|
|
|
|
status = uvm_range_tree_add(&va_space->range_group_ranges, &new_rgr->node);
|
|
UVM_ASSERT(status == NV_OK);
|
|
|
|
prev = range_group_range_prev(va_space, new_rgr);
|
|
if (prev != NULL && prev->node.end + 1 == new_rgr->node.start && prev->range_group == new_rgr->range_group) {
|
|
uvm_range_tree_merge_prev(&va_space->range_group_ranges, &new_rgr->node);
|
|
uvm_range_group_range_destroy(prev);
|
|
}
|
|
|
|
next = range_group_range_next(va_space, new_rgr);
|
|
if (next != NULL && next->node.start - 1 == new_rgr->node.end && next->range_group == new_rgr->range_group) {
|
|
uvm_range_tree_merge_next(&va_space->range_group_ranges, &new_rgr->node);
|
|
uvm_range_group_range_destroy(next);
|
|
}
|
|
|
|
return NV_OK;
|
|
}
|
|
|
|
bool uvm_range_group_address_migratable(uvm_va_space_t *va_space, NvU64 address)
|
|
{
|
|
uvm_range_group_range_t *rgr = uvm_range_group_range_find(va_space, address);
|
|
return rgr == NULL || uvm_range_group_migratable(rgr->range_group);
|
|
}
|
|
|
|
bool uvm_range_group_any_migratable(uvm_va_space_t *va_space, NvU64 start, NvU64 end)
|
|
{
|
|
uvm_range_group_range_iter_t iter;
|
|
uvm_range_group_for_all_ranges_in(&iter, va_space, start, end) {
|
|
if (iter.migratable)
|
|
return true;
|
|
}
|
|
|
|
return false;
|
|
}
|
|
|
|
bool uvm_range_group_all_migratable(uvm_va_space_t *va_space, NvU64 start, NvU64 end)
|
|
{
|
|
uvm_range_group_range_t *rgr;
|
|
uvm_range_group_for_each_range_in(rgr, va_space, start, end) {
|
|
if (!uvm_range_group_migratable(rgr->range_group))
|
|
return false;
|
|
}
|
|
|
|
return true;
|
|
}
|
|
|
|
uvm_range_group_range_t *uvm_range_group_range_find(uvm_va_space_t *va_space, NvU64 addr)
|
|
{
|
|
uvm_range_tree_node_t *node;
|
|
uvm_assert_rwsem_locked(&va_space->lock);
|
|
|
|
node = uvm_range_tree_find(&va_space->range_group_ranges, addr);
|
|
return range_group_range_container(node);
|
|
}
|
|
|
|
uvm_range_group_range_t *uvm_range_group_range_iter_first(uvm_va_space_t *va_space, NvU64 start, NvU64 end)
|
|
{
|
|
uvm_range_tree_node_t *node = uvm_range_tree_iter_first(&va_space->range_group_ranges, start, end);
|
|
return range_group_range_container(node);
|
|
}
|
|
|
|
uvm_range_group_range_t *uvm_range_group_range_iter_next(uvm_va_space_t *va_space,
|
|
uvm_range_group_range_t *range,
|
|
NvU64 end)
|
|
{
|
|
uvm_range_tree_node_t *node = uvm_range_tree_iter_next(&va_space->range_group_ranges, &range->node, end);
|
|
return range_group_range_container(node);
|
|
}
|
|
|
|
static void range_group_range_iter_advance(uvm_range_group_range_iter_t *iter, NvU64 end)
|
|
{
|
|
if (iter->node == NULL) {
|
|
iter->end = end;
|
|
}
|
|
else {
|
|
iter->is_current = iter->start >= iter->node->node.start;
|
|
if (iter->is_current)
|
|
iter->end = min(iter->node->node.end, end);
|
|
else
|
|
iter->end = min(iter->node->node.start - 1, end);
|
|
}
|
|
iter->migratable = iter->node == NULL || !iter->is_current || uvm_range_group_migratable(iter->node->range_group);
|
|
}
|
|
|
|
|
|
void uvm_range_group_range_iter_all_first(uvm_va_space_t *va_space,
|
|
NvU64 start,
|
|
NvU64 end,
|
|
uvm_range_group_range_iter_t *iter)
|
|
{
|
|
iter->valid = true;
|
|
iter->start = start;
|
|
iter->node = uvm_range_group_range_iter_first(va_space, start, end);
|
|
|
|
range_group_range_iter_advance(iter, end);
|
|
}
|
|
|
|
bool uvm_range_group_range_iter_all_next(uvm_va_space_t *va_space,
|
|
uvm_range_group_range_iter_t *iter,
|
|
NvU64 end)
|
|
{
|
|
iter->valid = iter->end < end;
|
|
if (!iter->valid)
|
|
return false;
|
|
|
|
iter->start = iter->end + 1;
|
|
if (iter->is_current)
|
|
iter->node = uvm_range_group_range_iter_next(va_space, iter->node, end);
|
|
|
|
range_group_range_iter_advance(iter, end);
|
|
return true;
|
|
}
|
|
|
|
void uvm_range_group_range_migratability_iter_first(uvm_va_space_t *va_space,
|
|
NvU64 start,
|
|
NvU64 end,
|
|
uvm_range_group_range_iter_t *iter)
|
|
{
|
|
uvm_range_group_range_iter_t next;
|
|
|
|
uvm_range_group_range_iter_all_first(va_space, start, end, iter);
|
|
next = *iter;
|
|
|
|
while (uvm_range_group_range_iter_all_next(va_space, &next, end) && next.migratable == iter->migratable)
|
|
*iter = next;
|
|
|
|
iter->start = start;
|
|
}
|
|
|
|
void uvm_range_group_range_migratability_iter_next(uvm_va_space_t *va_space,
|
|
uvm_range_group_range_iter_t *iter,
|
|
NvU64 end)
|
|
{
|
|
uvm_range_group_range_iter_t next;
|
|
NvU64 start;
|
|
|
|
if (!uvm_range_group_range_iter_all_next(va_space, iter, end))
|
|
return;
|
|
|
|
start = iter->start;
|
|
next = *iter;
|
|
while (uvm_range_group_range_iter_all_next(va_space, &next, end) && next.migratable == iter->migratable)
|
|
*iter = next;
|
|
|
|
iter->start = start;
|
|
}
|
|
|
|
void uvm_range_group_migratable_page_mask(uvm_va_block_t *va_block,
|
|
uvm_va_block_region_t region,
|
|
uvm_page_mask_t *mask_out)
|
|
{
|
|
uvm_range_group_range_iter_t iter;
|
|
uvm_va_space_t *va_space = uvm_va_block_get_va_space(va_block);
|
|
|
|
uvm_page_mask_zero(mask_out);
|
|
|
|
uvm_range_group_for_each_migratability_in(&iter,
|
|
va_space,
|
|
uvm_va_block_region_start(va_block, region),
|
|
uvm_va_block_region_end(va_block, region)) {
|
|
if (iter.migratable)
|
|
uvm_page_mask_region_fill(mask_out, uvm_va_block_region_from_start_end(va_block, iter.start, iter.end));
|
|
}
|
|
}
|
|
|
|
NV_STATUS uvm_api_prevent_migration_range_groups(UVM_PREVENT_MIGRATION_RANGE_GROUPS_PARAMS *params, struct file *filp)
|
|
{
|
|
NV_STATUS status;
|
|
uvm_va_space_t *va_space = uvm_va_space_get(filp);
|
|
|
|
status = uvm_range_groups_set_migration_policy(va_space, params->rangeGroupIds, params->numGroupIds, false);
|
|
if (status == NV_OK)
|
|
uvm_tools_flush_events();
|
|
|
|
return status;
|
|
}
|
|
|
|
NV_STATUS uvm_api_allow_migration_range_groups(UVM_ALLOW_MIGRATION_RANGE_GROUPS_PARAMS *params, struct file *filp)
|
|
{
|
|
uvm_va_space_t *va_space = uvm_va_space_get(filp);
|
|
|
|
return uvm_range_groups_set_migration_policy(va_space, params->rangeGroupIds, params->numGroupIds, true);
|
|
}
|
|
|
|
NV_STATUS uvm_test_range_group_range_info(UVM_TEST_RANGE_GROUP_RANGE_INFO_PARAMS *params, struct file *filp)
|
|
{
|
|
uvm_range_group_range_t *rgr;
|
|
uvm_va_space_t *va_space = uvm_va_space_get(filp);
|
|
|
|
uvm_va_space_down_read(va_space);
|
|
|
|
rgr = uvm_range_group_range_iter_first(va_space, params->lookup_address, ULLONG_MAX);
|
|
|
|
params->range_group_present = rgr != NULL && rgr->node.start <= params->lookup_address;
|
|
if (params->range_group_present) {
|
|
params->range_group_range_start = rgr->node.start;
|
|
params->range_group_range_end = rgr->node.end;
|
|
params->range_group_id = rgr->range_group->id;
|
|
}
|
|
else {
|
|
uvm_range_group_range_t *prev;
|
|
uvm_range_group_range_t *next = rgr;
|
|
|
|
if (next) {
|
|
params->range_group_range_end = next->node.start - 1;
|
|
prev = range_group_range_prev(va_space, next);
|
|
}
|
|
else {
|
|
params->range_group_range_end = ULLONG_MAX;
|
|
prev = list_last_entry_or_null(&va_space->range_group_ranges.head, uvm_range_group_range_t, node.list);
|
|
}
|
|
|
|
if (prev)
|
|
params->range_group_range_start = prev->node.end + 1;
|
|
else
|
|
params->range_group_range_start = 0;
|
|
|
|
params->range_group_id = UVM_RANGE_GROUP_ID_NONE;
|
|
}
|
|
|
|
uvm_va_space_up_read(va_space);
|
|
|
|
return NV_OK;
|
|
}
|
|
|
|
NV_STATUS uvm_test_range_group_range_count(UVM_TEST_RANGE_GROUP_RANGE_COUNT_PARAMS *params, struct file *filp)
|
|
{
|
|
uvm_range_group_range_t *rgr;
|
|
uvm_range_group_t *range_group;
|
|
uvm_va_space_t *va_space = uvm_va_space_get(filp);
|
|
|
|
params->count = 0;
|
|
uvm_va_space_down_read(va_space);
|
|
|
|
range_group = radix_tree_lookup(&va_space->range_groups, params->rangeGroupId);
|
|
if (range_group == NULL) {
|
|
uvm_va_space_up_read(va_space);
|
|
return NV_ERR_OBJECT_NOT_FOUND;
|
|
}
|
|
|
|
list_for_each_entry(rgr, &range_group->ranges, range_group_list_node) {
|
|
UVM_ASSERT(rgr->range_group == range_group);
|
|
params->count++;
|
|
}
|
|
|
|
uvm_va_space_up_read(va_space);
|
|
|
|
return NV_OK;
|
|
}
|