[Intel-gfx] [PATCH 33/33] drm/i915: Compress GPU objects in error state

[Joonas Lahtinen]

On su, 2016-08-07 at 15:45 +0100, Chris Wilson wrote:
> @@ -309,12 +310,30 @@ void i915_error_printf(struct drm_i915_error_state_buf *e, const char *f, ...)
>  	va_end(args);
>  }
>  
> +static bool
> +ascii85_encode(u32 in, char *out)


base64 is more de facto and I bet userland "expects" it too.

I'd also throw the routines under lib/

> @@ -326,13 +345,23 @@ static void print_error_obj(struct drm_i915_error_state_buf *m,
>  			   lower_32_bits(obj->gtt_offset));
>  	}
>  
> -	for (page = offset = 0; page < obj->page_count; page++) {
> -		for (elt = 0; elt < PAGE_SIZE/4; elt++) {
> -			err_printf(m, "%08x :  %08x\n", offset,
> -				   obj->pages[page][elt]);
> -			offset += 4;
> +	err_puts(m, ":"); /* indicate compressed data */

I'd also keep the the uncompressed option, because somebody might be
trying to make a micro-kernel without extra algorithms options. A
config setting could be justified.

> +	for (page = 0; page < obj->page_count; page++) {
> +		int i, len;
> +
> +		len = PAGE_SIZE;
> +		if (page == obj->page_count - 1)
> +			len -= obj->unused;
> +		len = (len + 3) / 4;
> +
> +		for (i = 0; i < len; i++) {
> +			if (ascii85_encode(obj->pages[page][i], out))
> +				err_puts(m, out);
> +			else
> +				err_puts(m, "z");

I think the encode function could take ranges, you do not very often
care about encoding/decoding single character.

>  		}
>  	}
> +	err_puts(m, "\n");
>  }
>  
>  int i915_error_state_to_str(struct drm_i915_error_state_buf *m,
> @@ -593,17 +622,37 @@ static void i915_error_state_free(struct kref *error_ref)
>  	kfree(error);
>  }
>  
> -static int compress_page(void *src, struct drm_i915_error_object *dst)
> +static int compress_page(struct z_stream_s *zstream,
> +			 void *src,
> +			 struct drm_i915_error_object *dst)
>  {
> -	unsigned long page;
> +	zstream->next_in = src;
> +	zstream->avail_in = PAGE_SIZE;
>  
> -	page = __get_free_page(GFP_ATOMIC | __GFP_NOWARN);
> -	if (!page)
> -		return -ENOMEM;
> +	do {
> +		if (zstream->avail_out == 0) {
> +			unsigned long page;
> +
> +			page = __get_free_page(GFP_ATOMIC | __GFP_NOWARN);
> +			if (!page)
> +				return -ENOMEM;
> +
> +			dst->pages[dst->page_count++] = (void *)page;

Why is not dst->pages of different type?

> +
> +			zstream->next_out = (void *)page;
> +			zstream->avail_out = PAGE_SIZE;
> +		}
>  
> -	dst->pages[dst->page_count++] = (void *)page;
> +		if (zlib_deflate(zstream, Z_SYNC_FLUSH) != Z_OK)
> +			return -EIO;
> +
> +#if 0
> +		/* XXX fallback to uncompressed if we increases size? */
> +		if (zstream->total_out > zstream->total_in)
> +			return -E2BIG;
> +#endif

Not something we would merge. FIXME: or TODO: comment should be enough,
or make it DRM_INFO and we can act if we get reports?

> +	} while (zstream->avail_in);
>  
> -	memcpy((void *)page, src, PAGE_SIZE);

// The function name has been so descriptive previously :P

> @@ -622,6 +672,7 @@ i915_error_object_create(struct drm_i915_private *i915,
>  		return NULL;
>  
>  	num_pages = min_t(u64, vma->size, vma->obj->base.size) >> PAGE_SHIFT;
> +	num_pages = DIV_ROUND_UP(10 * num_pages, 8); /* worstcase zlib growth */

This kind of calculation could be made into a zlib function?

> @@ -629,6 +680,18 @@ i915_error_object_create(struct drm_i915_private *i915,
>  
>  	dst->gtt_offset = vma->node.start;
>  	dst->page_count = 0;
> +	dst->unused = 0;
> +
> +	memset(&zstream, 0, sizeof(zstream));
> +	zstream.workspace = kmalloc(zlib_deflate_workspacesize(MAX_WBITS,
> +							       MAX_MEM_LEVEL),
> +				    GFP_ATOMIC | __GFP_NOWARN);

Wouldn't look better with an intermediate variable?

Regards, Joonas
-- 
Joonas Lahtinen
Open Source Technology Center
Intel Corporation