Addresses regressions from the reverted merge
of the tbx fault manager for host memory.
Recursive locking of mutex caused deadlock.
To fix, separate tbx fault data from base
cpu fault data, allowing separate mutexes
for each, eliminating recursive locks on
the same mutex.
By separating, we also help ensure that tbx-related
changes don't affect the original cpu fault manager code
paths.
As an added safe guard preventing critical regressions
and avoiding another auto-revert, the tbx fault manager
is hidden behind a new debug flag which is disabled by default.
Related-To: NEO-12268
Signed-off-by: Jack Myers <jack.myers@intel.com>
Addresses regressions from the reverted merge
of the tbx fault manager for host memory.
This fixes attempts by the tbx fault manager
to protect/unprotect host buffer memory, even
if the host ptr was not driver-allocated.
In the case of the smoke test that triggered
the critical regression, clCreateBuffer was
called with the CL_MEM_USE_HOST_PTR flag.
The subsequent `mprotect` calls on the
provided host ptr then failed.
Related-To: NEO-12268
Signed-off-by: Jack Myers <jack.myers@intel.com>
In TBX mode, the host could not write to host buffers after access from device
code due to the lack of a migration mechanism post-initial TBX upload.
Migration is unnecessary with real hardware, but required for TBX.
This patch introduces a new page fault manager type that extends the original
CPU fault manager, enabling automatic migration of host buffers in TBX mode.
Refactoring was necessary to avoid diamond inheritance, achieved by using a
template parameter as the base class for OS-specific fault managers.
Related-To: NEO-12268
Signed-off-by: Jack Myers <jack.myers@intel.com>
Related-To: GSD-9385
In case of indirect allocations, we don't really know
their task count because we can't track their true usage
on GPU.
In case of non-blocking free, don't wait for latestSentTaskCount.
Signed-off-by: Szymon Morek <szymon.morek@intel.com>
- create heapAssigner per root device in memory manager to allow per
device config
Related-To: NEO-7063
Signed-off-by: Mateusz Hoppe <mateusz.hoppe@intel.com>
Enable eviction of CPU side USM allocation for UMD migrations on Windows.
Reverts incorrect auto-revert commit 218de586a4f28b1de3e983b9006e7a99d3a4d10e.
Related-To: NEO-8015
Signed-off-by: Milczarek, Slawomir <slawomir.milczarek@intel.com>
Enable eviction of CPU side USM allocation for UMD migrations on Windows.
Related-To: NEO-8015
Signed-off-by: Milczarek, Slawomir <slawomir.milczarek@intel.com>
By default prefer allocating memory first by KMD, instead of malloc first.
By default prefer not caching allocations on MTL devices. This results
in allocations being handled with non-coherent pat index.
For integrated devices when caching is not preferred do not allow
direct memory access in CPU domain. For map/unmap operations create
a dedicated memory allocation for CPU access, instead of accessing it
directly, reusing the same logic as when mapping/unmapping local memory.
Signed-off-by: Filip Hazubski <filip.hazubski@intel.com>
remove method to removing duplicates from StackVec as the method
implicitly sorted the vector
Related-To: GSD-4692
Signed-off-by: Mateusz Jablonski <mateusz.jablonski@intel.com>
Related-To: LOCI-4176
- Given a Base Pointer passed into Get Peer Allocation, then the base
pointer is used in the map of the new allocation to the virtual memory.
- Enables users to use the same pointer for all devices in Peer To Peer.
- Currently unsupported on reserved memory due to mapped and exec
resiedency of Virtual addresses.
Signed-off-by: Neil R Spruit <neil.r.spruit@intel.com>
Allows the user to use alignments > 64KB in `createUnifiedMemoryAllocation`
So that the restriction in `piextUSMDeviceAlloc` of the DPC++ runtime
could be lifted
Related-To: LOCI-4168
Signed-off-by: Lu, Wenbin <wenbin.lu@intel.com>
