[Intel-gfx] [PATCH v4 1/6] drm/i915/gen8: Add infrastructure to initialize WA batch buffers

[Chris Wilson]

On Tue, Jun 16, 2015 at 08:25:20PM +0100, Arun Siluvery wrote:
> +static int gen8_init_indirectctx_bb(struct intel_engine_cs *ring,
> +				    uint32_t offset,
> +				    uint32_t *num_dwords)
> +{
> +	uint32_t index;
> +	struct page *page;
> +	uint32_t *cmd;
> +
> +	page = i915_gem_object_get_page(ring->wa_ctx.obj, 0);
> +	cmd = kmap_atomic(page);
> +
> +	index = offset;
> +
> +	/* FIXME: fill one cacheline with NOOPs.
> +	 * Replace these instructions with WA
> +	 */
> +	while (index < (offset + 16))
> +		cmd[index++] = MI_NOOP;
> +
> +	/*
> +	 * MI_BATCH_BUFFER_END is not required in Indirect ctx BB because
> +	 * execution depends on the length specified in terms of cache lines
> +	 * in the register CTX_RCS_INDIRECT_CTX
> +	 */
> +
> +	kunmap_atomic(cmd);
> +
> +	if (index > (PAGE_SIZE / sizeof(uint32_t)))
> +		return -EINVAL;

Check before you GPF!

You just overran the buffer and corrupted memory, if you didn't succeed
in trapping a segfault.

To be generic, align to the cacheline then check you have enough room
for your own data.
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre