public inbox for intel-gfx@lists.freedesktop.org
 help / color / mirror / Atom feed
From: Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com>
To: Chris Wilson <chris@chris-wilson.co.uk>,
	intel-gfx <intel-gfx@lists.freedesktop.org>
Subject: Re: [PATCH 2/4] drm/i915: Handle incomplete Z_FINISH for compressed error states
Date: Tue, 2 Oct 2018 13:20:05 +0100	[thread overview]
Message-ID: <673d9c33-df35-d403-3e61-fbbc98047ba6@linux.intel.com> (raw)
In-Reply-To: <20181001194447.29910-2-chris@chris-wilson.co.uk>


On 01/10/2018 20:44, Chris Wilson wrote:
> The final call to zlib_deflate(Z_FINISH) may require more output
> space to be allocated and so needs to re-invoked. Failure to do so in
> the current code leads to incomplete zlib streams (albeit intact due to
> the use of Z_SYNC_FLUSH) resulting in the occasional short object
> capture.
> 
> Testcase: igt/i915-error-capture.js
> Fixes: 0a97015d45ee ("drm/i915: Compress GPU objects in error state")
> Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
> Cc: <stable@vger.kernel.org> # v4.10+
> ---
>   drivers/gpu/drm/i915/i915_gpu_error.c | 60 +++++++++++++++++++++------
>   1 file changed, 47 insertions(+), 13 deletions(-)
> 
> diff --git a/drivers/gpu/drm/i915/i915_gpu_error.c b/drivers/gpu/drm/i915/i915_gpu_error.c
> index 3d5554f14dfd..ed8c16cbfaa4 100644
> --- a/drivers/gpu/drm/i915/i915_gpu_error.c
> +++ b/drivers/gpu/drm/i915/i915_gpu_error.c
> @@ -237,6 +237,7 @@ static int compress_page(struct compress *c,
>   			 struct drm_i915_error_object *dst)
>   {
>   	struct z_stream_s *zstream = &c->zstream;
> +	int flush = Z_NO_FLUSH;
>   
>   	zstream->next_in = src;
>   	if (c->tmp && i915_memcpy_from_wc(c->tmp, src, PAGE_SIZE))
> @@ -257,8 +258,11 @@ static int compress_page(struct compress *c,
>   			zstream->avail_out = PAGE_SIZE;
>   		}

Looks like the block above, not shown in this diff, could use a check 
and abort if the dst->page_count overgrows the pessimistic allocation of 
the array.

>   
> -		if (zlib_deflate(zstream, Z_SYNC_FLUSH) != Z_OK)
> +		if (zlib_deflate(zstream, flush) != Z_OK)

So this (not always asking for a flush) actually not only fixes the 
flush at the end but improves the API usage and potentially compression 
ratio, correct?

>   			return -EIO;
> +
> +		if (zstream->avail_out)
> +			flush = Z_SYNC_FLUSH;

Hm.. but why this? It will flush only occasionally, when one input page 
did not fit in the available output - but that will depend on 
compressibility so I don't think it has a fixed period. It is not for 
instance for every 4k of compressed output, if that was maybe the goal.

>   	} while (zstream->avail_in);
>   
>   	/* Fallback to uncompressed if we increase size? */
> @@ -268,19 +272,43 @@ static int compress_page(struct compress *c,
>   	return 0;
>   }
>   
> -static void compress_fini(struct compress *c,
> +static int compress_flush(struct compress *c,
>   			  struct drm_i915_error_object *dst)
>   {
>   	struct z_stream_s *zstream = &c->zstream;
> +	unsigned long page;
>   
> -	if (dst) {
> -		zlib_deflate(zstream, Z_FINISH);
> -		dst->unused = zstream->avail_out;
> -	}
> +	do {
> +		switch (zlib_deflate(zstream, Z_FINISH)) {
> +		case Z_OK: /* more space requested */
> +			page = __get_free_page(GFP_ATOMIC | __GFP_NOWARN);
> +			if (!page)
> +				return -ENOMEM;
> +
> +			dst->pages[dst->page_count++] = (void *)page;

I'd put in a check for pages array exhaustion here as well. Or even 
better, compress_page and compress_flush could share this whole block.

> +			zstream->next_out = (void *)page;
> +			zstream->avail_out = PAGE_SIZE;
> +			break;
> +		case Z_STREAM_END:
> +			goto end;
> +		default: /* any error */
> +			return -EIO;
> +		}
> +	} while (1);
> +
> +end:
> +	memset(zstream->next_out, 0, zstream->avail_out);
> +	dst->unused = zstream->avail_out;
> +	return 0;
> +}
> +
> +static void compress_fini(struct compress *c,
> +			  struct drm_i915_error_object *dst)
> +{
> +	struct z_stream_s *zstream = &c->zstream;
>   
>   	zlib_deflateEnd(zstream);
>   	kfree(zstream->workspace);
> -
>   	if (c->tmp)
>   		free_page((unsigned long)c->tmp);
>   }
> @@ -319,6 +347,12 @@ static int compress_page(struct compress *c,
>   	return 0;
>   }
>   
> +static int compress_flush(struct compress *c,
> +			  struct drm_i915_error_object *dst)
> +{
> +	return 0;
> +}
> +
>   static void compress_fini(struct compress *c,
>   			  struct drm_i915_error_object *dst)
>   {
> @@ -951,15 +985,15 @@ i915_error_object_create(struct drm_i915_private *i915,
>   		if (ret)
>   			goto unwind;
>   	}
> -	goto out;
>   
> +	if (compress_flush(&compress, dst)) {
>   unwind:

A bit nasty, jump in conditional block. Could set a boolean and break 
from the above loop. Like "if (failed || compress_flush(...))".

> -	while (dst->page_count--)
> -		free_page((unsigned long)dst->pages[dst->page_count]);
> -	kfree(dst);
> -	dst = NULL;
> +		while (dst->page_count--)
> +			free_page((unsigned long)dst->pages[dst->page_count]);
> +		kfree(dst);
> +		dst = NULL;
> +	}
>   
> -out:
>   	compress_fini(&compress, dst);
>   	ggtt->vm.clear_range(&ggtt->vm, slot, PAGE_SIZE);
>   	return dst;
> 

Regards,

Tvrtko
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

  reply	other threads:[~2018-10-02 12:20 UTC|newest]

Thread overview: 24+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2018-10-01 19:44 [PATCH 1/4] drm/i915: Replace some open-coded i915_map_coherent_type() Chris Wilson
2018-10-01 19:44 ` [PATCH 2/4] drm/i915: Handle incomplete Z_FINISH for compressed error states Chris Wilson
2018-10-02 12:20   ` Tvrtko Ursulin [this message]
2018-10-02 12:24     ` Chris Wilson
2018-10-02 13:13       ` Tvrtko Ursulin
2018-10-02 13:22         ` Chris Wilson
2018-10-02 13:46           ` Tvrtko Ursulin
2018-10-02 13:52             ` Chris Wilson
2018-10-02 14:34               ` Tvrtko Ursulin
2018-10-02 12:36   ` [PATCH v2] " Chris Wilson
2018-10-03  8:24   ` [PATCH v3] " Chris Wilson
2018-10-03  9:04     ` Tvrtko Ursulin
2018-10-03 10:47       ` [Intel-gfx] " Chris Wilson
2018-10-01 19:44 ` [PATCH 3/4] drm/i915: Clear the error PTE just once on finish Chris Wilson
2018-10-02 12:27   ` Tvrtko Ursulin
2018-10-01 19:44 ` [PATCH 4/4] drm/i915: Cache the error string Chris Wilson
2018-10-01 21:39 ` ✗ Fi.CI.SPARSE: warning for series starting with [1/4] drm/i915: Replace some open-coded i915_map_coherent_type() Patchwork
2018-10-01 22:02 ` ✗ Fi.CI.BAT: failure " Patchwork
2018-10-02 11:48 ` [PATCH 1/4] " Tvrtko Ursulin
2018-10-02 13:31 ` ✗ Fi.CI.SPARSE: warning for series starting with [1/4] drm/i915: Replace some open-coded i915_map_coherent_type() (rev2) Patchwork
2018-10-02 13:49 ` ✓ Fi.CI.BAT: success " Patchwork
2018-10-03  2:47 ` ✗ Fi.CI.IGT: failure " Patchwork
2018-10-03  9:56   ` Martin Peres
2018-10-03  9:05 ` ✗ Fi.CI.BAT: failure for series starting with [1/4] drm/i915: Replace some open-coded i915_map_coherent_type() (rev3) Patchwork

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=673d9c33-df35-d403-3e61-fbbc98047ba6@linux.intel.com \
    --to=tvrtko.ursulin@linux.intel.com \
    --cc=chris@chris-wilson.co.uk \
    --cc=intel-gfx@lists.freedesktop.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox