[PATCH] drm/amdgpu: add module parameter to enable/disable bulk move

[Huang Rui]

On Tue, Sep 11, 2018 at 12:49:59PM +0200, Christian König wrote:
> Am 11.09.2018 um 12:24 schrieb Huang Rui:
> >On Tue, Sep 11, 2018 at 06:05:57PM +0800, Christian König wrote:
> >>Am 11.09.2018 um 05:41 schrieb Huang Rui:
> >>
> >>     The bulk moving mechanism still has bug on some corner cases. So disable it by
> >>     default till it is fixed. We can use the module parameter to enable it for
> >>     debugging.
> >>
> >>
> >>Please no. Module parameters are for end users, not developers.
> >>
> >>I really don't want a bug report that this option can crash the system.
> >>
> >How about use a macro such as BULK_ENABLED? Actually, I just want to have a
> >simple way to enable it to continue debugging.
> 
> Yeah, sure that should work.
> 
> But why do we actually need to modifiy
> amdgpu_vm_update_directories() as well? Because of the performance?

Yes, I'm afraid that it will make the performance drop again in some
specific cases.

Thanks,
Ray

> 
> Thanks,
> Christian.
> 
> >
> >#if BULK_ENABLED
> >#else
> >#endif
> >
> >Thanks,
> >Ray
> >
> >>Christian.
> >>
> >>
> >>
> >>     Signed-off-by: Huang Rui <ray.huang at amd.com>
> >>     Cc: Christian König <christian.koenig at amd.com>
> >>     Cc: Tom StDenis <Tom.StDenis at amd.com>
> >>     ---
> >>      drivers/gpu/drm/amd/amdgpu/amdgpu.h     |  1 +
> >>      drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 11 +++++++++++
> >>      drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c  | 11 +++++++++--
> >>      3 files changed, 21 insertions(+), 2 deletions(-)
> >>
> >>     diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
> >>     index e992e0f..75fd7be 100644
> >>     --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
> >>     +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
> >>     @@ -134,6 +134,7 @@ extern int amdgpu_compute_multipipe;
> >>      extern int amdgpu_gpu_recovery;
> >>      extern int amdgpu_emu_mode;
> >>      extern uint amdgpu_smu_memory_pool_size;
> >>     +extern int amdgpu_bulk_move_enabled;
> >>
> >>      #ifdef CONFIG_DRM_AMDGPU_SI
> >>      extern int amdgpu_si_support;
> >>     diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
> >>     index b5c2ccb..d57d16e 100644
> >>     --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
> >>     +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
> >>     @@ -127,6 +127,7 @@ int amdgpu_compute_multipipe = -1;
> >>      int amdgpu_gpu_recovery = -1; /* auto */
> >>      int amdgpu_emu_mode = 0;
> >>      uint amdgpu_smu_memory_pool_size = 0;
> >>     +int amdgpu_bulk_move_enabled = 0;
> >>
> >>      /**
> >>       * DOC: vramlimit (int)
> >>     @@ -532,6 +533,16 @@ MODULE_PARM_DESC(smu_memory_pool_size,
> >>                     "0x1 = 256Mbyte, 0x2 = 512Mbyte, 0x4 = 1 Gbyte, 0x8 = 2GByte");
> >>      module_param_named(smu_memory_pool_size, amdgpu_smu_memory_pool_size, uint, 0444);
> >>
> >>     +/**
> >>     + * DOC: bulk_move_enabled (int)
> >>     + * It is used to enable/disable bulk moving on LRU list mechanism for debugging
> >>     + * and testing. The default is false (disabled).  It will be enabled by default
> >>     + * till all corner cases are hammered out.
> >>     + */
> >>     +MODULE_PARM_DESC(bulk_move_enabled,
> >>     +                "Enable bulk moving on LRU list mechanism (1 = enabled, 0 = disabled (default)");
> >>     +module_param_named(bulk_move_enabled, amdgpu_bulk_move_enabled, uint, 0444);
> >>     +
> >>      #ifdef CONFIG_HSA_AMD
> >>      /**
> >>       * DOC: sched_policy (int)
> >>     diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
> >>     index 9a5b1bb..ab95a9c 100644
> >>     --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
> >>     +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
> >>     @@ -397,7 +397,8 @@ void amdgpu_vm_move_to_lru_tail(struct amdgpu_device *adev,
> >>             }
> >>             spin_unlock(&glob->lru_lock);
> >>
> >>     -       vm->bulk_moveable = true;
> >>     +       if (amdgpu_bulk_move_enabled)
> >>     +               vm->bulk_moveable = true;
> >>      }
> >>
> >>      /**
> >>     @@ -1213,7 +1214,13 @@ int amdgpu_vm_update_directories(struct amdgpu_device *adev,
> >>                     bo_base = list_first_entry(&vm->relocated,
> >>                                                struct amdgpu_vm_bo_base,
> >>                                                vm_status);
> >>     -               amdgpu_vm_bo_idle(bo_base);
> >>     +               if (amdgpu_bulk_move_enabled)
> >>     +                       amdgpu_vm_bo_idle(bo_base);
> >>     +               else {
> >>     +                       /* workaround: don't move PD/PT bos on LRU */
> >>     +                       list_del_init(&bo_base->vm_status);
> >>     +                       bo_base->moved = false;
> >>     +               }
> >>
> >>                     bo = bo_base->bo->parent;
> >>                     if (!bo)
> >>
> >>
> >>     _______________________________________________
> >>     amd-gfx mailing list
> >>     amd-gfx at lists.freedesktop.org
> >>     https://lists.freedesktop.org/mailman/listinfo/amd-gfx
> >>
> >>
> 
> _______________________________________________
> amd-gfx mailing list
> amd-gfx at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/amd-gfx