Re: [PATCH 01/16] drm/i915: Keep a small stash of preallocated WC pages

[Joonas Lahtinen <joonas.lahtinen@linux.intel.com>]

On ke, 2017-07-26 at 14:25 +0100, Chris Wilson wrote:
> We use WC pages for coherent writes into the ppGTT on !llc
> architectuures. However, to create a WC page requires a stop_machine(),
> i.e. is very slow. To compensate we currently keep a per-vm cache of
> recently freed pages, but we still see the slow startup of new contexts.
> We can amoritize that cost slightly by allocating WC pages in small
> batches (PAGEVEC_SIZE == 14) and since creating a WC page implies a
> stop_machine() there is no penalty for keeping that stash global.
> 
> Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

<SNIP>

> +++ b/drivers/gpu/drm/i915/i915_drv.h
> @@ -1464,6 +1464,9 @@ struct i915_gem_mm {
>  	struct llist_head free_list;
>  	struct work_struct free_work;
>  
> +	/** Small stash of WC pages */

Please briefly add the "why".

> +	struct pagevec wc_stash;
> +
>  	/** Usable portion of the GTT for GEM */
>  	dma_addr_t stolen_base; /* limited to low memory (32-bit) */
>  
> diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.c b/drivers/gpu/drm/i915/i915_gem_gtt.c
> index 10aa7762d9a6..ad4e48435853 100644
> --- a/drivers/gpu/drm/i915/i915_gem_gtt.c
> +++ b/drivers/gpu/drm/i915/i915_gem_gtt.c
> @@ -351,39 +351,85 @@ static gen6_pte_t iris_pte_encode(dma_addr_t addr,
>  
>  static struct page *vm_alloc_page(struct i915_address_space *vm, gfp_t gfp)
>  {
> -	struct page *page;
> +	struct pagevec *pvec = &vm->free_pages;
>  
>  	if (I915_SELFTEST_ONLY(should_fail(&vm->fault_attr, 1)))
>  		i915_gem_shrink_all(vm->i915);
>  
> -	if (vm->free_pages.nr)
> -		return vm->free_pages.pages[--vm->free_pages.nr];
> +	if (likely(pvec->nr))
> +		return pvec->pages[--pvec->nr];
> +
> +	if (!vm->pt_kmap_wc)
> +		return alloc_page(gfp);
> +
> +	lockdep_assert_held(&vm->i915->drm.struct_mutex);

Lift this to the beginning of the func, should trigger easier.
 
> -static void vm_free_pages_release(struct i915_address_space *vm)
> +static void vm_free_pages_release(struct i915_address_space *vm,
> +				  bool immediate)
>  {
> -	GEM_BUG_ON(!pagevec_count(&vm->free_pages));
> +	struct pagevec *pvec = &vm->free_pages;
> +
> +	GEM_BUG_ON(!pagevec_count(pvec));
> +
> +	if (vm->pt_kmap_wc) {
> +		struct pagevec *stash = &vm->i915->mm.wc_stash;
>  
> -	if (vm->pt_kmap_wc)
> -		set_pages_array_wb(vm->free_pages.pages,
> -				   pagevec_count(&vm->free_pages));
> +		/* When we use WC, first fill up the global stash and then
> +		 * only if full immediately free the overflow.
> +		 */
> +
> +		lockdep_assert_held(&vm->i915->drm.struct_mutex);

I'd again lift this up to the beginning.

> @@ -447,12 +493,31 @@ static void fill_page_dma_32(struct i915_address_space *vm,
>  static int
>  setup_scratch_page(struct i915_address_space *vm, gfp_t gfp)
>  {
> -	return __setup_page_dma(vm, &vm->scratch_page, gfp | __GFP_ZERO);
> +	struct page *page;
> +	dma_addr_t addr;
> +
> +	page = alloc_page(gfp | __GFP_ZERO);
> +	if (unlikely(!page))
> +		return -ENOMEM;
> +
> +	addr = dma_map_page(vm->dma, page, 0, PAGE_SIZE,
> +			    PCI_DMA_BIDIRECTIONAL);
> +	if (unlikely(dma_mapping_error(vm->dma, addr))) {
> +		__free_page(page);
> +		return -ENOMEM;
> +	}
> +
> +	vm->scratch_page.page = page;
> +	vm->scratch_page.daddr = addr;
> +	return 0;

Unrelated hunk, please split.

>  }
>  
>  static void cleanup_scratch_page(struct i915_address_space *vm)
>  {
> -	cleanup_page_dma(vm, &vm->scratch_page);
> +	struct i915_page_dma *p = &vm->scratch_page;
> +
> +	dma_unmap_page(vm->dma, p->daddr, PAGE_SIZE, PCI_DMA_BIDIRECTIONAL);
> +	__free_page(p->page);

Ditto.

>  }
>  
>  static struct i915_page_table *alloc_pt(struct i915_address_space *vm)
> @@ -1332,18 +1397,18 @@ static int gen8_ppgtt_init(struct i915_hw_ppgtt *ppgtt)
>  		1ULL << 48 :
>  		1ULL << 32;
>  
> -	ret = gen8_init_scratch(&ppgtt->base);
> -	if (ret) {
> -		ppgtt->base.total = 0;
> -		return ret;
> -	}
> -
>  	/* There are only few exceptions for gen >=6. chv and bxt.
>  	 * And we are not sure about the latter so play safe for now.
>  	 */
>  	if (IS_CHERRYVIEW(dev_priv) || IS_BROXTON(dev_priv))
>  		ppgtt->base.pt_kmap_wc = true;
>  
> +	ret = gen8_init_scratch(&ppgtt->base);
> +	if (ret) {
> +		ppgtt->base.total = 0;
> +		return ret;
> +	}
> +

More unrelated hunks.

Regards, Joonas
-- 
Joonas Lahtinen
Open Source Technology Center
Intel Corporation
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx