[Intel-gfx] [PATCH 10/27] drm/i915/icl: Enhanced execution list support

[Daniele Ceraolo Spurio]

On 09/01/18 15:28, Paulo Zanoni wrote:
> From: Thomas Daniel <thomas.daniel at intel.com>
> 
> Supports two-element submission using the new enhanced execlist mechanism
> 

This could use a few lines to describe enhanced execlist. Something like:

"Enhanced Execlists is an upgraded version of execlists which supports 
up to 8 ports. The lrcs to be submitted are written to a submit queue, 
which is then loaded on the HW. When writing to the ELSP register, the 
lrcs are written cyclically in the queue from position 0 to position 7. 
Alternatively, it is possible to write directly in the individual 
positions of the queue using the ELSQ registers. To be able to re-use 
all the existing code we're using the latter method and we're currently 
limiting ourself to only using 2 elements"

> v2: Rebase.
> v3: Switch from !IS_GEN11 to GEN < 11 (Daniele Ceraolo Spurio).
> v4: Use the elsq registers instead of elsp. (Daniele Ceraolo Spurio)
> 
> Signed-off-by: Thomas Daniel <thomas.daniel at intel.com>
> Signed-off-by: Rodrigo Vivi <rodrigo.vivi at intel.com>
> ---
>   drivers/gpu/drm/i915/intel_lrc.c | 21 ++++++++++++++++++++-
>   drivers/gpu/drm/i915/intel_lrc.h |  3 +++
>   2 files changed, 23 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c
> index de41ad2d5fbc..3c6f587fa903 100644
> --- a/drivers/gpu/drm/i915/intel_lrc.c
> +++ b/drivers/gpu/drm/i915/intel_lrc.c
> @@ -430,9 +430,18 @@ static inline void elsp_write(u64 desc, u32 __iomem *elsp)
>   
>   static void execlists_submit_ports(struct intel_engine_cs *engine)
>   {
> +	struct drm_i915_private *dev_priv = engine->i915;
>   	struct execlist_port *port = engine->execlists.port;
> +	u32 __iomem *elsq =
> +		engine->i915->regs + i915_mmio_reg_offset(RING_ELSQ(engine));

Should we cache this, like we do for execlists.elsp?

>   	unsigned int n;
>   
> +	/*
> +	 * Gen11+ note: the submit queue is not cleared after being submitted
> +	 * to the HW so we need to make sure we always clean it up. This is
> +	 * currently ensured by the fact that we always write the same number
> +	 * of elsq entries, keep this in mind before changing the loop below.
> +	 */
>   	for (n = execlists_num_ports(&engine->execlists); n--; ) {
>   		struct drm_i915_gem_request *rq;
>   		unsigned int count;
> @@ -456,8 +465,18 @@ static void execlists_submit_ports(struct intel_engine_cs *engine)
>   			desc = 0;
>   		}
>   
> -		elsp_write(desc, engine->execlists.elsp);
> +		if (INTEL_GEN(engine->i915) >= 11) {
> +			writel(lower_32_bits(desc), elsq + n * 2);
> +			writel(upper_32_bits(desc), elsq + n * 2 + 1);
> +		} else {
> +			elsp_write(desc, engine->execlists.elsp);
> +		}
>   	}
> +
> +	/* for gen11+ we need to manually load the submit queue */
> +	if (INTEL_GEN(engine->i915) >= 11)
> +		I915_WRITE_FW(RING_ELCR(engine), ELCR_LOAD);
> +
>   	execlists_clear_active(&engine->execlists, EXECLISTS_ACTIVE_HWACK);
>   }
>   
> diff --git a/drivers/gpu/drm/i915/intel_lrc.h b/drivers/gpu/drm/i915/intel_lrc.h
> index 6d4f9b995a11..cb00e1dd6ed2 100644
> --- a/drivers/gpu/drm/i915/intel_lrc.h
> +++ b/drivers/gpu/drm/i915/intel_lrc.h
> @@ -38,6 +38,9 @@
>   #define	  CTX_CTRL_ENGINE_CTX_RESTORE_INHIBIT	(1 << 0)
>   #define   CTX_CTRL_RS_CTX_ENABLE                (1 << 1)
>   #define RING_CONTEXT_STATUS_BUF_BASE(engine)	_MMIO((engine)->mmio_base + 0x370)
> +#define RING_ELSQ(engine)			_MMIO((engine)->mmio_base + 0x510)
> +#define RING_ELCR(engine)			_MMIO((engine)->mmio_base + 0x550)

Do we need to add the new regs to the forcewake domain identification in 
logical_ring_setup? They should be in the same well as ELSP so maybe 
just that one is enough.

> +#define	  ELCR_LOAD				(1 << 0) >   #define RING_CONTEXT_STATUS_BUF_LO(engine, i) 
_MMIO((engine)->mmio_base + 0x370 + (i) * 8)
>   #define RING_CONTEXT_STATUS_BUF_HI(engine, i)	_MMIO((engine)->mmio_base + 0x370 + (i) * 8 + 4)
>   #define RING_CONTEXT_STATUS_PTR(engine)		_MMIO((engine)->mmio_base + 0x3a0)
> 

Side note, we do have pre-emption enabled in the features but this patch 
does not update inject_preempt_context to use the new way to submit to 
the HW, so I guess that's probably break.

Daniele