diff options
author | Sami Kiminki <skiminki@nvidia.com> | 2015-05-22 13:48:22 -0400 |
---|---|---|
committer | Terje Bergstrom <tbergstrom@nvidia.com> | 2015-09-07 15:37:15 -0400 |
commit | eade809c265ada214c6c47e1ad9db1706c868da0 (patch) | |
tree | a132050d7a4c2193f338e0082d258e1eac958cc1 /drivers/gpu/nvgpu/gk20a/mm_gk20a.h | |
parent | 57034b22ca17b217b683941dcebc0d69587d7d5e (diff) |
gpu: nvgpu: Separate kernel and user GPU VA regions
Separate the kernel and userspace regions in the GPU virtual address
space. Do this by reserving the last part of the GPU VA aperture for
the kernel, and extend GPU VA aperture accordingly for regular address
spaces. This prevents the kernel polluting the userspace-visible GPU
VA regions, and thus, makes the success of fixed-address mapping more
predictable.
Bug 200077571
Change-Id: I63f0e73d4c815a4a9fa4a9ce568709974690ef0f
Signed-off-by: Sami Kiminki <skiminki@nvidia.com>
Reviewed-on: http://git-master/r/747191
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>
Tested-by: Terje Bergstrom <tbergstrom@nvidia.com>
Diffstat (limited to 'drivers/gpu/nvgpu/gk20a/mm_gk20a.h')
-rw-r--r-- | drivers/gpu/nvgpu/gk20a/mm_gk20a.h | 47 |
1 files changed, 21 insertions, 26 deletions
diff --git a/drivers/gpu/nvgpu/gk20a/mm_gk20a.h b/drivers/gpu/nvgpu/gk20a/mm_gk20a.h index 5c6c285a..6786e3c2 100644 --- a/drivers/gpu/nvgpu/gk20a/mm_gk20a.h +++ b/drivers/gpu/nvgpu/gk20a/mm_gk20a.h | |||
@@ -127,9 +127,10 @@ struct gk20a_buffer_state { | |||
127 | }; | 127 | }; |
128 | 128 | ||
129 | enum gmmu_pgsz_gk20a { | 129 | enum gmmu_pgsz_gk20a { |
130 | gmmu_page_size_small = 0, | 130 | gmmu_page_size_small = 0, |
131 | gmmu_page_size_big = 1, | 131 | gmmu_page_size_big = 1, |
132 | gmmu_nr_page_sizes = 2 | 132 | gmmu_page_size_kernel = 2, |
133 | gmmu_nr_page_sizes = 3, | ||
133 | }; | 134 | }; |
134 | 135 | ||
135 | struct gk20a_comptags { | 136 | struct gk20a_comptags { |
@@ -284,8 +285,10 @@ void gk20a_mm_l2_invalidate(struct gk20a *g); | |||
284 | struct mm_gk20a { | 285 | struct mm_gk20a { |
285 | struct gk20a *g; | 286 | struct gk20a *g; |
286 | 287 | ||
288 | /* GPU VA default sizes address spaces for channels */ | ||
287 | struct { | 289 | struct { |
288 | u64 size; | 290 | u64 user_size; /* userspace-visible GPU VA region */ |
291 | u64 kernel_size; /* kernel-only GPU VA region */ | ||
289 | } channel; | 292 | } channel; |
290 | 293 | ||
291 | struct { | 294 | struct { |
@@ -340,26 +343,15 @@ static inline int bar1_aperture_size_mb_gk20a(void) | |||
340 | { | 343 | { |
341 | return 16; /* 16MB is more than enough atm. */ | 344 | return 16; /* 16MB is more than enough atm. */ |
342 | } | 345 | } |
343 | /* max address bits */ | ||
344 | static inline int max_physaddr_bits_gk20a(void) | ||
345 | { | ||
346 | return 40;/*"old" sys physaddr, meaningful? */ | ||
347 | } | ||
348 | static inline int max_vid_physaddr_bits_gk20a(void) | ||
349 | { | ||
350 | /* "vid phys" is asid/smmu phys?, | ||
351 | * i.e. is this the real sys physaddr? */ | ||
352 | return 37; | ||
353 | } | ||
354 | static inline int max_vaddr_bits_gk20a(void) | ||
355 | { | ||
356 | return 40; /* chopped for area? */ | ||
357 | } | ||
358 | 346 | ||
359 | /* | 347 | /*The maximum GPU VA range supported */ |
360 | * Amount of the GVA space we actually use is smaller than the available space. | 348 | #define NV_GMMU_VA_RANGE 38 |
361 | */ | 349 | |
362 | #define NV_GMMU_VA_RANGE 40 | 350 | /* The default userspace-visible GPU VA size */ |
351 | #define NV_MM_DEFAULT_USER_SIZE (1ULL << 37) | ||
352 | |||
353 | /* The default kernel-reserved GPU VA size */ | ||
354 | #define NV_MM_DEFAULT_KERNEL_SIZE (1ULL << 32) | ||
363 | 355 | ||
364 | /* | 356 | /* |
365 | * The bottom 16GB of the space are used for small pages, the remaining high | 357 | * The bottom 16GB of the space are used for small pages, the remaining high |
@@ -370,12 +362,14 @@ static inline u64 __nv_gmmu_va_small_page_limit(void) | |||
370 | return ((u64)SZ_1G * 16); | 362 | return ((u64)SZ_1G * 16); |
371 | } | 363 | } |
372 | 364 | ||
373 | static inline int __nv_gmmu_va_is_upper(struct vm_gk20a *vm, u64 addr) | 365 | static inline int __nv_gmmu_va_is_big_page_region(struct vm_gk20a *vm, u64 addr) |
374 | { | 366 | { |
375 | if (!vm->big_pages) | 367 | if (!vm->big_pages) |
376 | return 0; | 368 | return 0; |
377 | 369 | ||
378 | return addr >= __nv_gmmu_va_small_page_limit(); | 370 | return addr >= vm->vma[gmmu_page_size_big].base && |
371 | addr < vm->vma[gmmu_page_size_big].base + | ||
372 | vm->vma[gmmu_page_size_big].length; | ||
379 | } | 373 | } |
380 | 374 | ||
381 | /* | 375 | /* |
@@ -391,7 +385,7 @@ static inline enum gmmu_pgsz_gk20a __get_pte_size(struct vm_gk20a *vm, | |||
391 | * As a result, even though the allocator supports mixed address spaces | 385 | * As a result, even though the allocator supports mixed address spaces |
392 | * the address spaces must be treated as separate for now. | 386 | * the address spaces must be treated as separate for now. |
393 | */ | 387 | */ |
394 | if (__nv_gmmu_va_is_upper(vm, base)) | 388 | if (__nv_gmmu_va_is_big_page_region(vm, base)) |
395 | return gmmu_page_size_big; | 389 | return gmmu_page_size_big; |
396 | else | 390 | else |
397 | return gmmu_page_size_small; | 391 | return gmmu_page_size_small; |
@@ -617,6 +611,7 @@ int gk20a_init_vm(struct mm_gk20a *mm, | |||
617 | struct vm_gk20a *vm, | 611 | struct vm_gk20a *vm, |
618 | u32 big_page_size, | 612 | u32 big_page_size, |
619 | u64 low_hole, | 613 | u64 low_hole, |
614 | u64 kernel_reserved, | ||
620 | u64 aperture_size, | 615 | u64 aperture_size, |
621 | bool big_pages, | 616 | bool big_pages, |
622 | char *name); | 617 | char *name); |