[PATCH] drm/amdgpu: fix vm size and block size for VMPT (v3)

Zhang, Jerry (Junwei) Jerry.Zhang at amd.com
Wed Apr 5 02:41:32 UTC 2017


On 04/05/2017 10:27 AM, Alex Deucher wrote:
> On Tue, Apr 4, 2017 at 10:11 PM, Zhang, Jerry (Junwei)
> <Jerry.Zhang at amd.com> wrote:
>> On 04/04/2017 03:34 AM, Deucher, Alexander wrote:
>>>>
>>>> -----Original Message-----
>>>> From: Junwei Zhang [mailto:Jerry.Zhang at amd.com]
>>>> Sent: Friday, March 31, 2017 10:44 PM
>>>> To: Deucher, Alexander; Koenig, Christian
>>>> Cc: amd-gfx at lists.freedesktop.org; Zhang, Jerry
>>>> Subject: [PATCH] drm/amdgpu: fix vm size and block size for VMPT (v3)
>>>>
>>>> From: "Zhang, Jerry" <Jerry.Zhang at amd.com>
>>>>
>>>> v2: set both of them in gmc
>>>> v3: move vm size and block size in vm manager
>>>>
>>>> Signed-off-by: Junwei Zhang <Jerry.Zhang at amd.com>
>>>> ---
>>>>   drivers/gpu/drm/amd/amdgpu/amdgpu_device.c |  8 --------
>>>>   drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c     | 22 ++++++++++++---------
>>>> -
>>>>   drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h     |  4 +++-
>>>>   drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c   |  2 +-
>>>>   drivers/gpu/drm/amd/amdgpu/gmc_v6_0.c      | 10 ++++++++--
>>>>   drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c      |  9 +++++++--
>>>>   drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c      |  9 +++++++--
>>>>   drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c      | 21 +++++++++++++--------
>>>>   drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.c    |  2 +-
>>>>   9 files changed, 52 insertions(+), 35 deletions(-)
>>>>
>>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
>>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
>>>> index fbb4afb..1d0c742 100644
>>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
>>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
>>>> @@ -1041,14 +1041,6 @@ static bool amdgpu_check_pot_argument(int arg)
>>>>
>>>>   static void amdgpu_get_block_size(struct amdgpu_device *adev)
>>>>   {
>>>> -     /* from AI, asic starts to support multiple level VMPT */
>>>> -     if (adev->asic_type >= CHIP_VEGA10) {
>>>> -             if (amdgpu_vm_block_size != 9)
>>>> -                     dev_warn(adev->dev,
>>>> -                              "Multi-VMPT limits block size to one
>>>> page!\n");
>>>> -             amdgpu_vm_block_size = 9;
>>>> -             return;
>>>> -     }
>>>>         /* 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
>>>>          * page table and the remaining bits are in the page directory */
>>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
>>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
>>>> index 43adc4b..46d3f1a 100644
>>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
>>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
>>>> @@ -100,13 +100,14 @@ static unsigned amdgpu_vm_num_entries(struct
>>>> amdgpu_device *adev,
>>>>         if (level == 0)
>>>>                 /* For the root directory */
>>>>                 return adev->vm_manager.max_pfn >>
>>>> -                     (amdgpu_vm_block_size * adev-
>>>>> vm_manager.num_level);
>>>> +                     (adev->vm_manager.block_size *
>>>> +                      adev->vm_manager.num_level);
>>>>         else if (level == adev->vm_manager.num_level)
>>>>                 /* For the page tables on the leaves */
>>>> -             return AMDGPU_VM_PTE_COUNT;
>>>> +             return AMDGPU_VM_PTE_COUNT(adev);
>>>>         else
>>>>                 /* Everything in between */
>>>> -             return 1 << amdgpu_vm_block_size;
>>>> +             return 1 << adev->vm_manager.block_size;
>>>>   }
>>>>
>>>>   /**
>>>> @@ -271,7 +272,7 @@ static int amdgpu_vm_alloc_levels(struct
>>>> amdgpu_device *adev,
>>>>                                   unsigned level)
>>>>   {
>>>>         unsigned shift = (adev->vm_manager.num_level - level) *
>>>> -             amdgpu_vm_block_size;
>>>> +             adev->vm_manager.block_size;
>>>>         unsigned pt_idx, from, to;
>>>>         int r;
>>>>
>>>> @@ -970,7 +971,7 @@ static struct amdgpu_bo *amdgpu_vm_get_pt(struct
>>>> amdgpu_pte_update_params *p,
>>>>         unsigned idx, level = p->adev->vm_manager.num_level;
>>>>
>>>>         while (entry->entries) {
>>>> -             idx = addr >> (amdgpu_vm_block_size * level--);
>>>> +             idx = addr >> (p->adev->vm_manager.block_size * level--);
>>>>                 idx %= amdgpu_bo_size(entry->bo) / 8;
>>>>                 entry = &entry->entries[idx];
>>>>         }
>>>> @@ -997,7 +998,8 @@ static void amdgpu_vm_update_ptes(struct
>>>> amdgpu_pte_update_params *params,
>>>>                                   uint64_t start, uint64_t end,
>>>>                                   uint64_t dst, uint64_t flags)
>>>>   {
>>>> -     const uint64_t mask = AMDGPU_VM_PTE_COUNT - 1;
>>>> +     struct amdgpu_device *adev = params->adev;
>>>> +     const uint64_t mask = AMDGPU_VM_PTE_COUNT(adev) - 1;
>>>>
>>>>         uint64_t cur_pe_start, cur_nptes, cur_dst;
>>>>         uint64_t addr; /* next GPU address to be updated */
>>>> @@ -1021,7 +1023,7 @@ static void amdgpu_vm_update_ptes(struct
>>>> amdgpu_pte_update_params *params,
>>>>         if ((addr & ~mask) == (end & ~mask))
>>>>                 nptes = end - addr;
>>>>         else
>>>> -             nptes = AMDGPU_VM_PTE_COUNT - (addr & mask);
>>>> +             nptes = AMDGPU_VM_PTE_COUNT(adev) - (addr & mask);
>>>>
>>>>         cur_pe_start = amdgpu_bo_gpu_offset(pt);
>>>>         cur_pe_start += (addr & mask) * 8;
>>>> @@ -1049,7 +1051,7 @@ static void amdgpu_vm_update_ptes(struct
>>>> amdgpu_pte_update_params *params,
>>>>                 if ((addr & ~mask) == (end & ~mask))
>>>>                         nptes = end - addr;
>>>>                 else
>>>> -                     nptes = AMDGPU_VM_PTE_COUNT - (addr & mask);
>>>> +                     nptes = AMDGPU_VM_PTE_COUNT(adev) - (addr &
>>>> mask);
>>>>
>>>>                 next_pe_start = amdgpu_bo_gpu_offset(pt);
>>>>                 next_pe_start += (addr & mask) * 8;
>>>> @@ -1196,7 +1198,7 @@ static int amdgpu_vm_bo_update_mapping(struct
>>>> amdgpu_device *adev,
>>>>          * reserve space for one command every (1 << BLOCK_SIZE)
>>>>          *  entries or 2k dwords (whatever is smaller)
>>>>          */
>>>> -     ncmds = (nptes >> min(amdgpu_vm_block_size, 11)) + 1;
>>>> +     ncmds = (nptes >> min(adev->vm_manager.block_size, 11)) + 1;
>>>>
>>>>         /* padding, etc. */
>>>>         ndw = 64;
>>>> @@ -2067,7 +2069,7 @@ void amdgpu_vm_bo_invalidate(struct
>>>> amdgpu_device *adev,
>>>>   int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm)
>>>>   {
>>>>         const unsigned align = min(AMDGPU_VM_PTB_ALIGN_SIZE,
>>>> -             AMDGPU_VM_PTE_COUNT * 8);
>>>> +             AMDGPU_VM_PTE_COUNT(adev) * 8);
>>>>         unsigned ring_instance;
>>>>         struct amdgpu_ring *ring;
>>>>         struct amd_sched_rq *rq;
>>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
>>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
>>>> index 102b1f7..1242264 100644
>>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
>>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
>>>> @@ -45,7 +45,7 @@
>>>>   #define AMDGPU_VM_MAX_UPDATE_SIZE    0x3FFFF
>>>>
>>>>   /* number of entries in page table */
>>>> -#define AMDGPU_VM_PTE_COUNT (1 << amdgpu_vm_block_size)
>>>> +#define AMDGPU_VM_PTE_COUNT(adev) (1 << adev-
>>>>> vm_manager.block_size)
>>>
>>>
>>> For safety, put parens around adev.  E.g.,
>>> #define AMDGPU_VM_PTE_COUNT(adev) (1 << (adev)->vm_manager.block_size)
>>
>>
>> Yeah, thanks to point it out.
>>
>>>
>>> With that fixed:
>>> Reviewed-by: Alex Deucher <alexander.deucher at amd.com>
>>>
>>> As a follow on patch, it would be nice to add some reasonable clamping in
>>> the
>>> gmc files so we can still override the default settings and not cause too
>>> much
>>> problem with multiple asics in the same system.
>>
>>
>> Could you elaborate it a bit more, about "clamping"?
>
> e.g., setting amdgpu_vm_block_size to -1 (auto) bt default and setting
> a reasonable default for each gmc if it's set to 0, or overriding if
> it's not set to -1.  That way we can have different defaults on each
> family and still override for testing/debugging.

Thanks for your explanation.
Now I got the "clamping" :)

Jerry

>
> Alex
>
>
>>
>> Jerry
>>
>>>
>>> Alex
>>>
>>>>
>>>>   /* PTBs (Page Table Blocks) need to be aligned to 32K */
>>>>   #define AMDGPU_VM_PTB_ALIGN_SIZE   32768
>>>> @@ -155,6 +155,8 @@ struct amdgpu_vm_manager {
>>>>
>>>>         uint64_t                                max_pfn;
>>>>         uint32_t                                num_level;
>>>> +     uint64_t                                vm_size;
>>>> +     uint32_t                                block_size;
>>>>         /* vram base address for page table entry  */
>>>>         u64                                     vram_base_offset;
>>>>         /* is vm enabled? */
>>>> diff --git a/drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c
>>>> b/drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c
>>>> index 30ef312..9223c2b 100644
>>>> --- a/drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c
>>>> +++ b/drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c
>>>> @@ -222,7 +222,7 @@ int gfxhub_v1_0_gart_enable(struct amdgpu_device
>>>> *adev)
>>>>
>>>>         EXECUTE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
>>>>                 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
>>>>                                 PAGE_TABLE_BLOCK_SIZE,
>>>> -                                 amdgpu_vm_block_size - 9);
>>>> +                             adev->vm_manager.block_size - 9);
>>>>                 WREG32(SOC15_REG_OFFSET(GC, 0,
>>>> mmVM_CONTEXT1_CNTL) + i, tmp);
>>>>                 WREG32(SOC15_REG_OFFSET(GC, 0,
>>>> mmVM_CONTEXT1_PAGE_TABLE_START_ADDR_LO32) + i*2, 0);
>>>>                 WREG32(SOC15_REG_OFFSET(GC, 0,
>>>> mmVM_CONTEXT1_PAGE_TABLE_START_ADDR_HI32) + i*2, 0);
>>>> diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v6_0.c
>>>> b/drivers/gpu/drm/amd/amdgpu/gmc_v6_0.c
>>>> index d958660..30d5c42 100644
>>>> --- a/drivers/gpu/drm/amd/amdgpu/gmc_v6_0.c
>>>> +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v6_0.c
>>>> @@ -543,7 +543,8 @@ static int gmc_v6_0_gart_enable(struct
>>>> amdgpu_device *adev)
>>>>         WREG32(mmVM_CONTEXT1_CNTL,
>>>>                VM_CONTEXT1_CNTL__ENABLE_CONTEXT_MASK |
>>>>                (1UL << VM_CONTEXT1_CNTL__PAGE_TABLE_DEPTH__SHIFT) |
>>>> -            ((amdgpu_vm_block_size - 9) <<
>>>> VM_CONTEXT1_CNTL__PAGE_TABLE_BLOCK_SIZE__SHIFT));
>>>> +            ((adev->vm_manager.block_size - 9)
>>>> +            << VM_CONTEXT1_CNTL__PAGE_TABLE_BLOCK_SIZE__SHIFT));
>>>>         if (amdgpu_vm_fault_stop ==
>>>> AMDGPU_VM_FAULT_STOP_ALWAYS)
>>>>                 gmc_v6_0_set_fault_enable_default(adev, false);
>>>>         else
>>>> @@ -848,7 +849,12 @@ static int gmc_v6_0_sw_init(void *handle)
>>>>         if (r)
>>>>                 return r;
>>>>
>>>> -     adev->vm_manager.max_pfn = amdgpu_vm_size << 18;
>>>> +     adev->vm_manager.vm_size = amdgpu_vm_size;
>>>> +     adev->vm_manager.block_size = amdgpu_vm_block_size;
>>>> +     adev->vm_manager.max_pfn = adev->vm_manager.vm_size << 18;
>>>> +
>>>> +     DRM_INFO("vm size is %d GB, block size is %d-bit\n",
>>>> +             adev->vm_manager.vm_size, adev-
>>>>> vm_manager.block_size);
>>>>
>>>>         adev->mc.mc_mask = 0xffffffffffULL;
>>>>
>>>> diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c
>>>> b/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c
>>>> index 0c0a601..9427c7d 100644
>>>> --- a/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c
>>>> +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c
>>>> @@ -639,7 +639,7 @@ static int gmc_v7_0_gart_enable(struct
>>>> amdgpu_device *adev)
>>>>         tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
>>>> ENABLE_CONTEXT, 1);
>>>>         tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
>>>> PAGE_TABLE_DEPTH, 1);
>>>>         tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
>>>> PAGE_TABLE_BLOCK_SIZE,
>>>> -                         amdgpu_vm_block_size - 9);
>>>> +                         adev->vm_manager.block_size - 9);
>>>>         WREG32(mmVM_CONTEXT1_CNTL, tmp);
>>>>         if (amdgpu_vm_fault_stop ==
>>>> AMDGPU_VM_FAULT_STOP_ALWAYS)
>>>>                 gmc_v7_0_set_fault_enable_default(adev, false);
>>>> @@ -998,7 +998,12 @@ static int gmc_v7_0_sw_init(void *handle)
>>>>          * Currently set to 4GB ((1 << 20) 4k pages).
>>>>          * Max GPUVM size for cayman and SI is 40 bits.
>>>>          */
>>>> -     adev->vm_manager.max_pfn = amdgpu_vm_size << 18;
>>>> +     adev->vm_manager.vm_size = amdgpu_vm_size;
>>>> +     adev->vm_manager.block_size = amdgpu_vm_block_size;
>>>> +     adev->vm_manager.max_pfn = adev->vm_manager.vm_size << 18;
>>>> +
>>>> +     DRM_INFO("vm size is %d GB, block size is %d-bit\n",
>>>> +             adev->vm_manager.vm_size, adev-
>>>>> vm_manager.block_size);
>>>>
>>>>         /* Set the internal MC address mask
>>>>          * This is the max address of the GPU's
>>>> diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c
>>>> b/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c
>>>> index d19d1c5..16742be 100644
>>>> --- a/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c
>>>> +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c
>>>> @@ -848,7 +848,7 @@ static int gmc_v8_0_gart_enable(struct
>>>> amdgpu_device *adev)
>>>>         tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
>>>> WRITE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
>>>>         tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
>>>> EXECUTE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
>>>>         tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
>>>> PAGE_TABLE_BLOCK_SIZE,
>>>> -                         amdgpu_vm_block_size - 9);
>>>> +                         adev->vm_manager.block_size - 9);
>>>>         WREG32(mmVM_CONTEXT1_CNTL, tmp);
>>>>         if (amdgpu_vm_fault_stop ==
>>>> AMDGPU_VM_FAULT_STOP_ALWAYS)
>>>>                 gmc_v8_0_set_fault_enable_default(adev, false);
>>>> @@ -1082,7 +1082,12 @@ static int gmc_v8_0_sw_init(void *handle)
>>>>          * Currently set to 4GB ((1 << 20) 4k pages).
>>>>          * Max GPUVM size for cayman and SI is 40 bits.
>>>>          */
>>>> -     adev->vm_manager.max_pfn = amdgpu_vm_size << 18;
>>>> +     adev->vm_manager.vm_size = amdgpu_vm_size;
>>>> +     adev->vm_manager.block_size = amdgpu_vm_block_size;
>>>> +     adev->vm_manager.max_pfn = adev->vm_manager.vm_size << 18;
>>>> +
>>>> +     DRM_INFO("vm size is %d GB, block size is %d-bit\n",
>>>> +             adev->vm_manager.vm_size, adev-
>>>>> vm_manager.block_size);
>>>>
>>>>         /* Set the internal MC address mask
>>>>          * This is the max address of the GPU's
>>>> diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
>>>> b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
>>>> index df69aae..2180edb 100644
>>>> --- a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
>>>> +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
>>>> @@ -543,11 +543,23 @@ static int gmc_v9_0_sw_init(void *handle)
>>>>
>>>>         if (adev->flags & AMD_IS_APU) {
>>>>                 adev->mc.vram_type = AMDGPU_VRAM_TYPE_UNKNOWN;
>>>> +             adev->vm_manager.vm_size = amdgpu_vm_size;
>>>> +             adev->vm_manager.block_size = amdgpu_vm_block_size;
>>>>         } else {
>>>>                 /* XXX Don't know how to get VRAM type yet. */
>>>>                 adev->mc.vram_type = AMDGPU_VRAM_TYPE_HBM;
>>>> +             /*
>>>> +              * To fulfill 4-level page support,
>>>> +              * vm size is 256TB (48bit), maximum size of Vega10,
>>>> +              * block size 512 (9bit)
>>>> +              */
>>>> +             adev->vm_manager.vm_size = 1U << 18;
>>>> +             adev->vm_manager.block_size = 9;
>>>>         }
>>>>
>>>> +     DRM_INFO("vm size is %d GB, block size is %d-bit\n",
>>>> +             adev->vm_manager.vm_size, adev-
>>>>> vm_manager.block_size);
>>>> +
>>>>         /* This interrupt is VMC page fault.*/
>>>>         r = amdgpu_irq_add_id(adev, AMDGPU_IH_CLIENTID_VMC, 0,
>>>>                                 &adev->mc.vm_fault);
>>>> @@ -557,14 +569,7 @@ static int gmc_v9_0_sw_init(void *handle)
>>>>         if (r)
>>>>                 return r;
>>>>
>>>> -     /* Because of four level VMPTs, vm size is at least 512GB.
>>>> -      * The maximum size is 256TB (48bit).
>>>> -      */
>>>> -     if (amdgpu_vm_size < 512) {
>>>> -             DRM_WARN("VM size is at least 512GB!\n");
>>>> -             amdgpu_vm_size = 512;
>>>> -     }
>>>> -     adev->vm_manager.max_pfn = (uint64_t)amdgpu_vm_size << 18;
>>>> +     adev->vm_manager.max_pfn = adev->vm_manager.vm_size << 18;
>>>>
>>>>         /* Set the internal MC address mask
>>>>          * This is the max address of the GPU's
>>>> diff --git a/drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.c
>>>> b/drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.c
>>>> index 266a0f4..2cc1d58 100644
>>>> --- a/drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.c
>>>> +++ b/drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.c
>>>> @@ -242,7 +242,7 @@ int mmhub_v1_0_gart_enable(struct amdgpu_device
>>>> *adev)
>>>>
>>>>         EXECUTE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
>>>>                 tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
>>>>                                 PAGE_TABLE_BLOCK_SIZE,
>>>> -                             amdgpu_vm_block_size - 9);
>>>> +                             adev->vm_manager.block_size - 9);
>>>>                 WREG32(SOC15_REG_OFFSET(MMHUB, 0,
>>>> mmVM_CONTEXT1_CNTL) + i, tmp);
>>>>                 WREG32(SOC15_REG_OFFSET(MMHUB, 0,
>>>> mmVM_CONTEXT1_PAGE_TABLE_START_ADDR_LO32) + i*2, 0);
>>>>                 WREG32(SOC15_REG_OFFSET(MMHUB, 0,
>>>> mmVM_CONTEXT1_PAGE_TABLE_START_ADDR_HI32) + i*2, 0);
>>>> --
>>>> 1.9.1
>>>
>>>
>> _______________________________________________
>> amd-gfx mailing list
>> amd-gfx at lists.freedesktop.org
>> https://lists.freedesktop.org/mailman/listinfo/amd-gfx


More information about the amd-gfx mailing list