[PATCH 4/5] drm/amdgpu: revert context to stop engine before mode2 reset

[Zhao, Victor]

[AMD Official Use Only - General]

Hi Andrey,

For slow tests I mean the slow hang tests by quark tool.
An example here:
hang_vm_gfx_dispatch_slow.lua - This script runs on a graphics engine using compute engine and has a hacked CS program which is massive and duplicates standard CS program move code hundreds of thousands of times. The effect is a very slowly executing CS program.

It's not a bad job but just need a very long time to finish. I suppose we don’t have a way to stop shader here. And the running apps will be affected when reset is done.


Thanks,
Victor



-----Original Message-----
From: Grodzovsky, Andrey <Andrey.Grodzovsky at amd.com> 
Sent: Tuesday, July 26, 2022 5:20 AM
To: Zhao, Victor <Victor.Zhao at amd.com>; amd-gfx at lists.freedesktop.org
Cc: Deucher, Alexander <Alexander.Deucher at amd.com>; Deng, Emily <Emily.Deng at amd.com>; Koenig, Christian <Christian.Koenig at amd.com>
Subject: Re: [PATCH 4/5] drm/amdgpu: revert context to stop engine before mode2 reset

On 2022-07-22 03:34, Victor Zhao wrote:

> For some hang caused by slow tests, engine cannot be stopped which may 
> cause resume failure after reset. In this case, force halt engine by 
> reverting context addresses


Can you maybe explain a bit more what exactly you mean by slow test and why engine cannot be stopped in this case ?

Andrey


>
> Signed-off-by: Victor Zhao <Victor.Zhao at amd.com>
> ---
>   drivers/gpu/drm/amd/amdgpu/amdgpu_device.c  |  1 +
>   drivers/gpu/drm/amd/amdgpu/amdgpu_gfxhub.h  |  1 +
>   drivers/gpu/drm/amd/amdgpu/gfxhub_v2_1.c    | 36 +++++++++++++++++++++
>   drivers/gpu/drm/amd/amdgpu/sienna_cichlid.c |  2 ++
>   4 files changed, 40 insertions(+)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c 
> b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> index 5498fda8617f..833dc5e224d3 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> @@ -5037,6 +5037,7 @@ static void amdgpu_device_recheck_guilty_jobs(
>   
>   			/* set guilty */
>   			drm_sched_increase_karma(s_job);
> +			amdgpu_reset_prepare_hwcontext(adev, reset_context);
>   retry:
>   			/* do hw reset */
>   			if (amdgpu_sriov_vf(adev)) {
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfxhub.h 
> b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfxhub.h
> index f8036f2b100e..c7b44aeb671b 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfxhub.h
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfxhub.h
> @@ -37,6 +37,7 @@ struct amdgpu_gfxhub_funcs {
>   	void (*utcl2_harvest)(struct amdgpu_device *adev);
>   	void (*mode2_save_regs)(struct amdgpu_device *adev);
>   	void (*mode2_restore_regs)(struct amdgpu_device *adev);
> +	void (*halt)(struct amdgpu_device *adev);
>   };
>   
>   struct amdgpu_gfxhub {
> diff --git a/drivers/gpu/drm/amd/amdgpu/gfxhub_v2_1.c 
> b/drivers/gpu/drm/amd/amdgpu/gfxhub_v2_1.c
> index 51cf8acd2d79..8cf53e039c11 100644
> --- a/drivers/gpu/drm/amd/amdgpu/gfxhub_v2_1.c
> +++ b/drivers/gpu/drm/amd/amdgpu/gfxhub_v2_1.c
> @@ -646,6 +646,41 @@ static void gfxhub_v2_1_restore_regs(struct amdgpu_device *adev)
>   	WREG32_SOC15(GC, 0, mmGCMC_VM_MX_L1_TLB_CNTL, adev->gmc.MC_VM_MX_L1_TLB_CNTL);
>   }
>   
> +static void gfxhub_v2_1_halt(struct amdgpu_device *adev) {
> +	struct amdgpu_vmhub *hub = &adev->vmhub[AMDGPU_GFXHUB_0];
> +	int i;
> +	uint32_t tmp;
> +	int time = 1000;
> +
> +	gfxhub_v2_1_set_fault_enable_default(adev, false);
> +
> +	for (i = 0; i <= 14; i++) {
> +		WREG32_SOC15_OFFSET(GC, 0, mmGCVM_CONTEXT1_PAGE_TABLE_START_ADDR_LO32,
> +				    i * hub->ctx_addr_distance, ~0);
> +		WREG32_SOC15_OFFSET(GC, 0, mmGCVM_CONTEXT1_PAGE_TABLE_START_ADDR_HI32,
> +				    i * hub->ctx_addr_distance, ~0);
> +		WREG32_SOC15_OFFSET(GC, 0, mmGCVM_CONTEXT1_PAGE_TABLE_END_ADDR_LO32,
> +				    i * hub->ctx_addr_distance,
> +				    0);
> +		WREG32_SOC15_OFFSET(GC, 0, mmGCVM_CONTEXT1_PAGE_TABLE_END_ADDR_HI32,
> +				    i * hub->ctx_addr_distance,
> +				    0);
> +	}
> +	tmp = RREG32_SOC15(GC, 0, mmGRBM_STATUS2);
> +	while ((tmp & (GRBM_STATUS2__EA_BUSY_MASK |
> +		      GRBM_STATUS2__EA_LINK_BUSY_MASK)) != 0 &&
> +	       time) {
> +		udelay(100);
> +		time--;
> +		tmp = RREG32_SOC15(GC, 0, mmGRBM_STATUS2);
> +	}
> +
> +	if (!time) {
> +		DRM_WARN("failed to wait for GRBM(EA) idle\n");
> +	}
> +}
> +
>   const struct amdgpu_gfxhub_funcs gfxhub_v2_1_funcs = {
>   	.get_fb_location = gfxhub_v2_1_get_fb_location,
>   	.get_mc_fb_offset = gfxhub_v2_1_get_mc_fb_offset, @@ -658,4 +693,5 
> @@ const struct amdgpu_gfxhub_funcs gfxhub_v2_1_funcs = {
>   	.utcl2_harvest = gfxhub_v2_1_utcl2_harvest,
>   	.mode2_save_regs = gfxhub_v2_1_save_regs,
>   	.mode2_restore_regs = gfxhub_v2_1_restore_regs,
> +	.halt = gfxhub_v2_1_halt,
>   };
> diff --git a/drivers/gpu/drm/amd/amdgpu/sienna_cichlid.c 
> b/drivers/gpu/drm/amd/amdgpu/sienna_cichlid.c
> index 51a5b68f77d3..fead7251292f 100644
> --- a/drivers/gpu/drm/amd/amdgpu/sienna_cichlid.c
> +++ b/drivers/gpu/drm/amd/amdgpu/sienna_cichlid.c
> @@ -97,6 +97,8 @@ sienna_cichlid_mode2_prepare_hwcontext(struct amdgpu_reset_control *reset_ctl,
>   	if (!amdgpu_sriov_vf(adev)) {
>   		if (adev->gfxhub.funcs->mode2_save_regs)
>   			adev->gfxhub.funcs->mode2_save_regs(adev);
> +		if (adev->gfxhub.funcs->halt)
> +			adev->gfxhub.funcs->halt(adev);
>   		r = sienna_cichlid_mode2_suspend_ip(adev);
>   	}
>