[PATCH 10/10] vmwgfx: Implement GMR2

[Konrad Rzeszutek Wilk]

On Wed, Aug 31, 2011 at 09:42:55AM +0200, Thomas Hellstrom wrote:
> Guest Memory Regions 2 is a way to bind pages to the GPU, but using
> the FIFO instead of an io-submitted descriptor chain.
> 
> Signed-off-by: Thomas Hellstrom <thellstrom at vmware.com>
> Reviewed-by: Jakob Bornecantz <jakob at vmware.com>
> ---
>  drivers/gpu/drm/vmwgfx/vmwgfx_gmr.c |   81 ++++++++++++++++++++++++++++++++++-
>  1 files changed, 80 insertions(+), 1 deletions(-)
> 
> diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_gmr.c b/drivers/gpu/drm/vmwgfx/vmwgfx_gmr.c
> index de0c594..f4e7763 100644
> --- a/drivers/gpu/drm/vmwgfx/vmwgfx_gmr.c
> +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_gmr.c
> @@ -1,6 +1,6 @@
>  /**************************************************************************
>   *
> - * Copyright © 2009 VMware, Inc., Palo Alto, CA., USA
> + * Copyright © 2009-2011 VMware, Inc., Palo Alto, CA., USA
>   * All Rights Reserved.
>   *
>   * Permission is hereby granted, free of charge, to any person obtaining a
> @@ -29,6 +29,77 @@
>  #include "drmP.h"
>  #include "ttm/ttm_bo_driver.h"
>  
> +#define VMW_PPN_SIZE sizeof(unsigned long)
> +
> +static int vmw_gmr2_bind(struct vmw_private *dev_priv,
> +			 struct page *pages[],
> +			 unsigned long num_pages,
> +			 int gmr_id)
> +{
> +	SVGAFifoCmdDefineGMR2 define_cmd;
> +	SVGAFifoCmdRemapGMR2 remap_cmd;
> +	uint32_t define_size = sizeof(define_cmd) + 4;
> +	uint32_t remap_size = VMW_PPN_SIZE * num_pages + sizeof(remap_cmd) + 4;

Why the +4? Is it just as a precaution in case you over-ran?

> +	uint32_t *cmd;
> +	uint32_t *cmd_orig;
> +	uint32_t i;
> +
> +	cmd_orig = cmd = vmw_fifo_reserve(dev_priv, define_size + remap_size);
> +	if (unlikely(cmd == NULL))
> +		return -ENOMEM;
> +
> +	define_cmd.gmrId = gmr_id;
> +	define_cmd.numPages = num_pages;
> +
> +	remap_cmd.gmrId = gmr_id;
> +	remap_cmd.flags = (VMW_PPN_SIZE > sizeof(*cmd)) ?
> +		SVGA_REMAP_GMR2_PPN64 : SVGA_REMAP_GMR2_PPN32;
> +	remap_cmd.offsetPages = 0;
> +	remap_cmd.numPages = num_pages;
> +
> +	*cmd++ = SVGA_CMD_DEFINE_GMR2;
> +	memcpy(cmd, &define_cmd, sizeof(define_cmd));
> +	cmd += sizeof(define_cmd) / sizeof(uint32);

ALIGN macro? Thought I would have thought you would
want:
 cmd += sizeof(define_cmd) - sizeof(uint32);

since you stuck a SVGA_CMD_DEFINE_GMR2 and then copied
the define_cmd right afterwards?

> +
> +	*cmd++ = SVGA_CMD_REMAP_GMR2;
> +	memcpy(cmd, &remap_cmd, sizeof(remap_cmd));
> +	cmd += sizeof(remap_cmd) / sizeof(uint32);
> +
> +	for (i = 0; i < num_pages; ++i) {
> +		if (VMW_PPN_SIZE > 4)
> +			*cmd = page_to_pfn(*pages++);
> +		else
> +			*((uint64_t *)cmd) = page_to_pfn(*pages++);
> +
> +		cmd += VMW_PPN_SIZE / sizeof(*cmd);
> +	}
> +
> +	vmw_fifo_commit(dev_priv, define_size + remap_size);
> +
> +	return 0;
> +}
> +
> +static void vmw_gmr2_unbind(struct vmw_private *dev_priv,
> +			    int gmr_id)
> +{
> +	SVGAFifoCmdDefineGMR2 define_cmd;
> +	uint32_t define_size = sizeof(define_cmd) + 4;
> +	uint32_t *cmd;
> +
> +	cmd = vmw_fifo_reserve(dev_priv, define_size);
> +	if (unlikely(cmd == NULL)) {
> +		DRM_ERROR("GMR2 unbind failed.\n");
> +		return;
> +	}
> +	define_cmd.gmrId = gmr_id;
> +	define_cmd.numPages = 0;
> +
> +	*cmd++ = SVGA_CMD_DEFINE_GMR2;

Ah, that is what the +4 is for. Would it be cleaner
to just do 'sizeof(SVGA_CMD_DEFINE_GMR2)' ?


> +	memcpy(cmd, &define_cmd, sizeof(define_cmd));
> +
> +	vmw_fifo_commit(dev_priv, define_size);
> +}
> +
>  /**
>   * FIXME: Adjust to the ttm lowmem / highmem storage to minimize
>   * the number of used descriptors.
> @@ -170,6 +241,9 @@ int vmw_gmr_bind(struct vmw_private *dev_priv,
>  	struct list_head desc_pages;
>  	int ret;
>  
> +	if (likely(dev_priv->capabilities & SVGA_CAP_GMR2))
> +		return vmw_gmr2_bind(dev_priv, pages, num_pages, gmr_id);
> +
>  	if (unlikely(!(dev_priv->capabilities & SVGA_CAP_GMR)))
>  		return -EINVAL;
>  
> @@ -192,6 +266,11 @@ int vmw_gmr_bind(struct vmw_private *dev_priv,
>  
>  void vmw_gmr_unbind(struct vmw_private *dev_priv, int gmr_id)
>  {
> +	if (likely(dev_priv->capabilities & SVGA_CAP_GMR2)) {
> +		vmw_gmr2_unbind(dev_priv, gmr_id);
> +		return;
> +	}
> +
>  	mutex_lock(&dev_priv->hw_mutex);
>  	vmw_write(dev_priv, SVGA_REG_GMR_ID, gmr_id);
>  	wmb();
> -- 
> 1.7.4.4
> 
> _______________________________________________
> dri-devel mailing list
> dri-devel at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/dri-devel