amd/amdkfd/kfd_flat_memory.c

51  * mapped into this pointer space as apertures.
53  * based on what apertures the request falls in.
114  * A 64b pointer is compared to the apertures that are defined (Base/Limit), in
129  * memory object cannot straddle multiple apertures in the 64b address
153  * apertures to set the Mtype field for S_LOAD/FLAT_* ops which is input to the
157  * address space that doesn’t hit any defined base/limit apertures.
158  * The following diagram is a complete picture of the gfxip7.x SUA apertures.
167  * In all cases (for SUA and DUA apertures discussed later), aperture base
177  * In no case is a payload that straddles multiple apertures expected to work.
183  * or a payload straddles multiple apertures.  The S_LOAD instruction
186  * Apertures cannot overlap.
212  * Private memories are mapped as apertures into the address space,
214  * to the LDS or Scratch memory when it falls into one of these apertures.
215  * Like the SUA apertures, the Shared/Private apertures are 64KB aligned and
217  * the Shared/Private apertures are always placed in a limited selection of
219  * Shared/Private apertures can be placed anywhere in the 32b space
223  * HSA64 Apertures for FLAT_* vector ops
225  * For HSA64 SUA mode, the Shared and Private apertures are always placed
234  * HSA32 Apertures for FLAT_* vector ops
236  * In HSA32 mode, the Private and Shared apertures can be placed anywhere
238  * the apertures). If the base address of the apertures are non-zero
239  * (ie apertures exists), the size is always 64KB.
242  * GPUVM Apertures for FLAT_* vector ops
244  * In GPUVM mode, the Shared/Private apertures are specified identically
297 /* On GFXv9 the LDS and scratch apertures are programmed independently
388 		 * For 64 bit process apertures will be statically reserved in  in kfd_init_apertures()
390 		 * amdkfd doesn't currently support apertures for 32 bit process  in kfd_init_apertures()