提交 bab4fee7 编写于 作者: J Junwei Zhang 提交者: Alex Deucher

drm/amdgpu: set vm size and block size by individual gmc by default (v3)

By default, the value is set by individual gmc.
if a specific value is input, it overrides the global value for all

v2: create helper funcs
v3: update gmc9 APU's num_level athough it may be updated in the future.
Signed-off-by: NJunwei Zhang <Jerry.Zhang@amd.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>
上级 e190ed1e
...@@ -1040,35 +1040,31 @@ static bool amdgpu_check_pot_argument(int arg) ...@@ -1040,35 +1040,31 @@ static bool amdgpu_check_pot_argument(int arg)
return (arg & (arg - 1)) == 0; return (arg & (arg - 1)) == 0;
} }
static void amdgpu_get_block_size(struct amdgpu_device *adev) static void amdgpu_check_block_size(struct amdgpu_device *adev)
{ {
/* defines number of bits in page table versus page directory, /* defines number of bits in page table versus page directory,
* a page is 4KB so we have 12 bits offset, minimum 9 bits in the * a page is 4KB so we have 12 bits offset, minimum 9 bits in the
* page table and the remaining bits are in the page directory */ * page table and the remaining bits are in the page directory */
if (amdgpu_vm_block_size == -1) { if (amdgpu_vm_block_size == -1)
return;
/* Total bits covered by PD + PTs */
unsigned bits = ilog2(amdgpu_vm_size) + 18;
/* Make sure the PD is 4K in size up to 8GB address space.
Above that split equal between PD and PTs */
if (amdgpu_vm_size <= 8)
amdgpu_vm_block_size = bits - 9;
else
amdgpu_vm_block_size = (bits + 3) / 2;
} else if (amdgpu_vm_block_size < 9) { if (amdgpu_vm_block_size < 9) {
dev_warn(adev->dev, "VM page table size (%d) too small\n", dev_warn(adev->dev, "VM page table size (%d) too small\n",
amdgpu_vm_block_size); amdgpu_vm_block_size);
amdgpu_vm_block_size = 9; goto def_value;
} }
if (amdgpu_vm_block_size > 24 || if (amdgpu_vm_block_size > 24 ||
(amdgpu_vm_size * 1024) < (1ull << amdgpu_vm_block_size)) { (amdgpu_vm_size * 1024) < (1ull << amdgpu_vm_block_size)) {
dev_warn(adev->dev, "VM page table size (%d) too large\n", dev_warn(adev->dev, "VM page table size (%d) too large\n",
amdgpu_vm_block_size); amdgpu_vm_block_size);
amdgpu_vm_block_size = 9; goto def_value;
} }
return;
def_value:
amdgpu_vm_block_size = -1;
} }
static void amdgpu_check_vm_size(struct amdgpu_device *adev) static void amdgpu_check_vm_size(struct amdgpu_device *adev)
...@@ -1097,8 +1093,7 @@ static void amdgpu_check_vm_size(struct amdgpu_device *adev) ...@@ -1097,8 +1093,7 @@ static void amdgpu_check_vm_size(struct amdgpu_device *adev)
return; return;
def_value: def_value:
amdgpu_vm_size = 8; amdgpu_vm_size = -1;
dev_info(adev->dev, "set default VM size %dGB\n", amdgpu_vm_size);
} }
/** /**
...@@ -1132,7 +1127,7 @@ static void amdgpu_check_arguments(struct amdgpu_device *adev) ...@@ -1132,7 +1127,7 @@ static void amdgpu_check_arguments(struct amdgpu_device *adev)
amdgpu_check_vm_size(adev); amdgpu_check_vm_size(adev);
amdgpu_get_block_size(adev); amdgpu_check_block_size(adev);
if (amdgpu_vram_page_split != -1 && (amdgpu_vram_page_split < 16 || if (amdgpu_vram_page_split != -1 && (amdgpu_vram_page_split < 16 ||
!amdgpu_check_pot_argument(amdgpu_vram_page_split))) { !amdgpu_check_pot_argument(amdgpu_vram_page_split))) {
......
...@@ -86,7 +86,7 @@ int amdgpu_runtime_pm = -1; ...@@ -86,7 +86,7 @@ int amdgpu_runtime_pm = -1;
unsigned amdgpu_ip_block_mask = 0xffffffff; unsigned amdgpu_ip_block_mask = 0xffffffff;
int amdgpu_bapm = -1; int amdgpu_bapm = -1;
int amdgpu_deep_color = 0; int amdgpu_deep_color = 0;
int amdgpu_vm_size = 64; int amdgpu_vm_size = -1;
int amdgpu_vm_block_size = -1; int amdgpu_vm_block_size = -1;
int amdgpu_vm_fault_stop = 0; int amdgpu_vm_fault_stop = 0;
int amdgpu_vm_debug = 0; int amdgpu_vm_debug = 0;
......
...@@ -2064,6 +2064,44 @@ void amdgpu_vm_bo_invalidate(struct amdgpu_device *adev, ...@@ -2064,6 +2064,44 @@ void amdgpu_vm_bo_invalidate(struct amdgpu_device *adev,
} }
} }
static uint32_t amdgpu_vm_get_block_size(uint64_t vm_size)
{
/* Total bits covered by PD + PTs */
unsigned bits = ilog2(vm_size) + 18;
/* Make sure the PD is 4K in size up to 8GB address space.
Above that split equal between PD and PTs */
if (vm_size <= 8)
return (bits - 9);
else
return ((bits + 3) / 2);
}
/**
* amdgpu_vm_adjust_size - adjust vm size and block size
*
* @adev: amdgpu_device pointer
* @vm_size: the default vm size if it's set auto
*/
void amdgpu_vm_adjust_size(struct amdgpu_device *adev, uint64_t vm_size)
{
/* adjust vm size firstly */
if (amdgpu_vm_size == -1)
adev->vm_manager.vm_size = vm_size;
else
adev->vm_manager.vm_size = amdgpu_vm_size;
/* block size depends on vm size */
if (amdgpu_vm_block_size == -1)
adev->vm_manager.block_size =
amdgpu_vm_get_block_size(adev->vm_manager.vm_size);
else
adev->vm_manager.block_size = amdgpu_vm_block_size;
DRM_INFO("vm size is %llu GB, block size is %u-bit\n",
adev->vm_manager.vm_size, adev->vm_manager.block_size);
}
/** /**
* amdgpu_vm_init - initialize a vm instance * amdgpu_vm_init - initialize a vm instance
* *
......
...@@ -234,5 +234,6 @@ int amdgpu_vm_bo_clear_mappings(struct amdgpu_device *adev, ...@@ -234,5 +234,6 @@ int amdgpu_vm_bo_clear_mappings(struct amdgpu_device *adev,
uint64_t saddr, uint64_t size); uint64_t saddr, uint64_t size);
void amdgpu_vm_bo_rmv(struct amdgpu_device *adev, void amdgpu_vm_bo_rmv(struct amdgpu_device *adev,
struct amdgpu_bo_va *bo_va); struct amdgpu_bo_va *bo_va);
void amdgpu_vm_adjust_size(struct amdgpu_device *adev, uint64_t vm_size);
#endif #endif
...@@ -849,13 +849,9 @@ static int gmc_v6_0_sw_init(void *handle) ...@@ -849,13 +849,9 @@ static int gmc_v6_0_sw_init(void *handle)
if (r) if (r)
return r; return r;
adev->vm_manager.vm_size = amdgpu_vm_size; amdgpu_vm_adjust_size(adev, 64);
adev->vm_manager.block_size = amdgpu_vm_block_size;
adev->vm_manager.max_pfn = adev->vm_manager.vm_size << 18; adev->vm_manager.max_pfn = adev->vm_manager.vm_size << 18;
DRM_INFO("vm size is %llu GB, block size is %d-bit\n",
adev->vm_manager.vm_size, adev->vm_manager.block_size);
adev->mc.mc_mask = 0xffffffffffULL; adev->mc.mc_mask = 0xffffffffffULL;
adev->need_dma32 = false; adev->need_dma32 = false;
......
...@@ -1003,13 +1003,9 @@ static int gmc_v7_0_sw_init(void *handle) ...@@ -1003,13 +1003,9 @@ static int gmc_v7_0_sw_init(void *handle)
* Currently set to 4GB ((1 << 20) 4k pages). * Currently set to 4GB ((1 << 20) 4k pages).
* Max GPUVM size for cayman and SI is 40 bits. * Max GPUVM size for cayman and SI is 40 bits.
*/ */
adev->vm_manager.vm_size = amdgpu_vm_size; amdgpu_vm_adjust_size(adev, 64);
adev->vm_manager.block_size = amdgpu_vm_block_size;
adev->vm_manager.max_pfn = adev->vm_manager.vm_size << 18; adev->vm_manager.max_pfn = adev->vm_manager.vm_size << 18;
DRM_INFO("vm size is %llu GB, block size is %d-bit\n",
adev->vm_manager.vm_size, adev->vm_manager.block_size);
/* Set the internal MC address mask /* Set the internal MC address mask
* This is the max address of the GPU's * This is the max address of the GPU's
* internal address space. * internal address space.
......
...@@ -1087,13 +1087,9 @@ static int gmc_v8_0_sw_init(void *handle) ...@@ -1087,13 +1087,9 @@ static int gmc_v8_0_sw_init(void *handle)
* Currently set to 4GB ((1 << 20) 4k pages). * Currently set to 4GB ((1 << 20) 4k pages).
* Max GPUVM size for cayman and SI is 40 bits. * Max GPUVM size for cayman and SI is 40 bits.
*/ */
adev->vm_manager.vm_size = amdgpu_vm_size; amdgpu_vm_adjust_size(adev, 64);
adev->vm_manager.block_size = amdgpu_vm_block_size;
adev->vm_manager.max_pfn = adev->vm_manager.vm_size << 18; adev->vm_manager.max_pfn = adev->vm_manager.vm_size << 18;
DRM_INFO("vm size is %llu GB, block size is %d-bit\n",
adev->vm_manager.vm_size, adev->vm_manager.block_size);
/* Set the internal MC address mask /* Set the internal MC address mask
* This is the max address of the GPU's * This is the max address of the GPU's
* internal address space. * internal address space.
......
...@@ -520,6 +520,11 @@ static int gmc_v9_0_vm_init(struct amdgpu_device *adev) ...@@ -520,6 +520,11 @@ static int gmc_v9_0_vm_init(struct amdgpu_device *adev)
* amdkfd will use VMIDs 8-15 * amdkfd will use VMIDs 8-15
*/ */
adev->vm_manager.num_ids = AMDGPU_NUM_OF_VMIDS; adev->vm_manager.num_ids = AMDGPU_NUM_OF_VMIDS;
/* TODO: fix num_level for APU when updating vm size and block size */
if (adev->flags & AMD_IS_APU)
adev->vm_manager.num_level = 1;
else
adev->vm_manager.num_level = 3; adev->vm_manager.num_level = 3;
amdgpu_vm_manager_init(adev); amdgpu_vm_manager_init(adev);
...@@ -552,8 +557,7 @@ static int gmc_v9_0_sw_init(void *handle) ...@@ -552,8 +557,7 @@ static int gmc_v9_0_sw_init(void *handle)
if (adev->flags & AMD_IS_APU) { if (adev->flags & AMD_IS_APU) {
adev->mc.vram_type = AMDGPU_VRAM_TYPE_UNKNOWN; adev->mc.vram_type = AMDGPU_VRAM_TYPE_UNKNOWN;
adev->vm_manager.vm_size = amdgpu_vm_size; amdgpu_vm_adjust_size(adev, 64);
adev->vm_manager.block_size = amdgpu_vm_block_size;
} else { } else {
/* XXX Don't know how to get VRAM type yet. */ /* XXX Don't know how to get VRAM type yet. */
adev->mc.vram_type = AMDGPU_VRAM_TYPE_HBM; adev->mc.vram_type = AMDGPU_VRAM_TYPE_HBM;
...@@ -564,11 +568,11 @@ static int gmc_v9_0_sw_init(void *handle) ...@@ -564,11 +568,11 @@ static int gmc_v9_0_sw_init(void *handle)
*/ */
adev->vm_manager.vm_size = 1U << 18; adev->vm_manager.vm_size = 1U << 18;
adev->vm_manager.block_size = 9; adev->vm_manager.block_size = 9;
DRM_INFO("vm size is %llu GB, block size is %u-bit\n",
adev->vm_manager.vm_size,
adev->vm_manager.block_size);
} }
DRM_INFO("vm size is %llu GB, block size is %d-bit\n",
adev->vm_manager.vm_size, adev->vm_manager.block_size);
/* This interrupt is VMC page fault.*/ /* This interrupt is VMC page fault.*/
r = amdgpu_irq_add_id(adev, AMDGPU_IH_CLIENTID_VMC, 0, r = amdgpu_irq_add_id(adev, AMDGPU_IH_CLIENTID_VMC, 0,
&adev->mc.vm_fault); &adev->mc.vm_fault);
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册