diff mbox

[2/2] drm/i915: fully apply WaSkipStolenMemoryFirstPage

Message ID 1481745339-17324-1-git-send-email-paulo.r.zanoni@intel.com (mailing list archive)
State New, archived
Headers show

Commit Message

Zanoni, Paulo R Dec. 14, 2016, 7:55 p.m. UTC
Don't even tell the mm allocator to handle the first page of stolen on
the affected platforms. This means that we won't inherit the FB in
case the BIOS decides to put it at the start of stolen. But the BIOS
should not be putting it at the start of stolen since it's going to
get corrupted. I suppose the bug here is that some pixels at the very
top of the screen will be corrupted, so it's not exactly easy to
notice.

We have confirmation that the first page of stolen does actually get
corrupted, so I really think we should do this in order to avoid any
possible future headaches, even if that means losing BIOS framebuffer
inheritance. Let's not use the HW in a way it's not supposed to be
used.

v2: don't even put the first page on the mm (Chris).
v3: drm_mm_init() takes size instead of end as argument (Ville).

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=94605
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
---
 drivers/gpu/drm/i915/i915_gem_gtt.h    |  1 +
 drivers/gpu/drm/i915/i915_gem_stolen.c | 34 +++++++++++++---------------------
 drivers/gpu/drm/i915/intel_fbc.c       |  6 +++---
 3 files changed, 17 insertions(+), 24 deletions(-)

Comments

Daniel Vetter Dec. 15, 2016, 7:54 a.m. UTC | #1
On Wed, Dec 14, 2016 at 05:55:39PM -0200, Paulo Zanoni wrote:
> Don't even tell the mm allocator to handle the first page of stolen on
> the affected platforms. This means that we won't inherit the FB in
> case the BIOS decides to put it at the start of stolen. But the BIOS
> should not be putting it at the start of stolen since it's going to
> get corrupted. I suppose the bug here is that some pixels at the very
> top of the screen will be corrupted, so it's not exactly easy to
> notice.
> 
> We have confirmation that the first page of stolen does actually get
> corrupted, so I really think we should do this in order to avoid any
> possible future headaches, even if that means losing BIOS framebuffer
> inheritance. Let's not use the HW in a way it's not supposed to be
> used.
> 
> v2: don't even put the first page on the mm (Chris).
> v3: drm_mm_init() takes size instead of end as argument (Ville).
> 
> Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=94605
> Cc: Chris Wilson <chris@chris-wilson.co.uk>
> Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
> ---
>  drivers/gpu/drm/i915/i915_gem_gtt.h    |  1 +
>  drivers/gpu/drm/i915/i915_gem_stolen.c | 34 +++++++++++++---------------------
>  drivers/gpu/drm/i915/intel_fbc.c       |  6 +++---
>  3 files changed, 17 insertions(+), 24 deletions(-)
> 
> diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.h b/drivers/gpu/drm/i915/i915_gem_gtt.h
> index 8965bbb..aefc968 100644
> --- a/drivers/gpu/drm/i915/i915_gem_gtt.h
> +++ b/drivers/gpu/drm/i915/i915_gem_gtt.h
> @@ -316,6 +316,7 @@ struct i915_ggtt {
>  	struct io_mapping mappable;	/* Mapping to our CPU mappable region */
>  
>  	size_t stolen_size;		/* Total size of stolen memory */
> +	size_t stolen_usable_start;	/* First page may be unusable. */
>  	size_t stolen_usable_size;	/* Total size minus BIOS reserved */
>  	size_t stolen_reserved_base;
>  	size_t stolen_reserved_size;
> diff --git a/drivers/gpu/drm/i915/i915_gem_stolen.c b/drivers/gpu/drm/i915/i915_gem_stolen.c
> index b1c8897..56d7e0b 100644
> --- a/drivers/gpu/drm/i915/i915_gem_stolen.c
> +++ b/drivers/gpu/drm/i915/i915_gem_stolen.c
> @@ -54,12 +54,6 @@ int i915_gem_stolen_insert_node_in_range(struct drm_i915_private *dev_priv,
>  	if (!drm_mm_initialized(&dev_priv->mm.stolen))
>  		return -ENODEV;
>  
> -	/* See the comment at the drm_mm_init() call for more about this check.
> -	 * WaSkipStolenMemoryFirstPage:bdw+ (incomplete)
> -	 */
> -	if (start < 4096 && INTEL_GEN(dev_priv) >= 8)
> -		start = 4096;
> -
>  	mutex_lock(&dev_priv->mm.stolen_lock);
>  	ret = drm_mm_insert_node_in_range(&dev_priv->mm.stolen, node, size,
>  					  alignment, start, end,
> @@ -74,10 +68,11 @@ int i915_gem_stolen_insert_node(struct drm_i915_private *dev_priv,
>  				unsigned alignment)
>  {
>  	struct i915_ggtt *ggtt = &dev_priv->ggtt;
> +	u64 start = ggtt->stolen_usable_start;
> +	u64 end = start + ggtt->stolen_usable_size;
>  
>  	return i915_gem_stolen_insert_node_in_range(dev_priv, node, size,
> -						    alignment, 0,
> -						    ggtt->stolen_usable_size);
> +						    alignment, start, end);
>  }
>  
>  void i915_gem_stolen_remove_node(struct drm_i915_private *dev_priv,
> @@ -489,20 +484,17 @@ int i915_gem_init_stolen(struct drm_i915_private *dev_priv)
>  		      ggtt->stolen_size >> 10,
>  		      (ggtt->stolen_size - reserved_total) >> 10);
>  
> -	ggtt->stolen_usable_size = ggtt->stolen_size - reserved_total;
> +	ggtt->stolen_usable_start = 0;
> +	/* WaSkipStolenMemoryFirstPage:bdw+ */
> +	if (INTEL_GEN(dev_priv) >= 8)
> +		ggtt->stolen_usable_start = 4096;
>  
> -	/*
> -	 * Basic memrange allocator for stolen space.
> -	 *
> -	 * TODO: Notice that some platforms require us to not use the first page
> -	 * of the stolen memory but their BIOSes may still put the framebuffer
> -	 * on the first page. So we don't reserve this page for now because of
> -	 * that. Our current solution is to just prevent new nodes from being
> -	 * inserted on the first page - see the check we have at
> -	 * i915_gem_stolen_insert_node_in_range(). We may want to fix the fbcon
> -	 * problem later.
> -	 */
> -	drm_mm_init(&dev_priv->mm.stolen, 0, ggtt->stolen_usable_size);
> +	ggtt->stolen_usable_size = ggtt->stolen_size - reserved_total -
> +				   ggtt->stolen_usable_start;
> +
> +	/* Basic memrange allocator for stolen space. */
> +	drm_mm_init(&dev_priv->mm.stolen, ggtt->stolen_usable_start,
> +		    ggtt->stolen_usable_size);
>  
>  	return 0;
>  }
> diff --git a/drivers/gpu/drm/i915/intel_fbc.c b/drivers/gpu/drm/i915/intel_fbc.c
> index 62f215b..534fcb3c 100644
> --- a/drivers/gpu/drm/i915/intel_fbc.c
> +++ b/drivers/gpu/drm/i915/intel_fbc.c
> @@ -528,7 +528,7 @@ static int find_compression_threshold(struct drm_i915_private *dev_priv,
>  	struct i915_ggtt *ggtt = &dev_priv->ggtt;
>  	int compression_threshold = 1;
>  	int ret;
> -	u64 end;
> +	u64 start = ggtt->stolen_usable_start, end;
>  
>  	/* The FBC hardware for BDW/SKL doesn't have access to the stolen
>  	 * reserved range size, so it always assumes the maximum (8mb) is used.
> @@ -538,7 +538,7 @@ static int find_compression_threshold(struct drm_i915_private *dev_priv,
>  	    IS_SKYLAKE(dev_priv) || IS_KABYLAKE(dev_priv))
>  		end = ggtt->stolen_size - 8 * 1024 * 1024;
>  	else
> -		end = ggtt->stolen_usable_size;
> +		end = start + ggtt->stolen_usable_size;

Funny mix&match of size end end we have :( We might want to switch
drm_mm_init to start/end for consistency with all the _in_range functions.

Let's see whether I'll regret this too ...

Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
>  
>  	/* HACK: This code depends on what we will do in *_enable_fbc. If that
>  	 * code changes, this code needs to change as well.
> @@ -549,7 +549,7 @@ static int find_compression_threshold(struct drm_i915_private *dev_priv,
>  
>  	/* Try to over-allocate to reduce reallocations and fragmentation. */
>  	ret = i915_gem_stolen_insert_node_in_range(dev_priv, node, size <<= 1,
> -						   4096, 0, end);
> +						   4096, start, end);
>  	if (ret == 0)
>  		return compression_threshold;
>  
> -- 
> 2.7.4
> 
> _______________________________________________
> Intel-gfx mailing list
> Intel-gfx@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/intel-gfx
Chris Wilson Dec. 15, 2016, 8:17 a.m. UTC | #2
On Wed, Dec 14, 2016 at 05:55:39PM -0200, Paulo Zanoni wrote:
> Don't even tell the mm allocator to handle the first page of stolen on
> the affected platforms. This means that we won't inherit the FB in
> case the BIOS decides to put it at the start of stolen. But the BIOS
> should not be putting it at the start of stolen since it's going to
> get corrupted. I suppose the bug here is that some pixels at the very
> top of the screen will be corrupted, so it's not exactly easy to
> notice.
> 
> We have confirmation that the first page of stolen does actually get
> corrupted, so I really think we should do this in order to avoid any
> possible future headaches, even if that means losing BIOS framebuffer
> inheritance. Let's not use the HW in a way it's not supposed to be
> used.
> 
> v2: don't even put the first page on the mm (Chris).
> v3: drm_mm_init() takes size instead of end as argument (Ville).
> 
> Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=94605
> Cc: Chris Wilson <chris@chris-wilson.co.uk>
> Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
> ---
>  drivers/gpu/drm/i915/i915_gem_gtt.h    |  1 +
>  drivers/gpu/drm/i915/i915_gem_stolen.c | 34 +++++++++++++---------------------
>  drivers/gpu/drm/i915/intel_fbc.c       |  6 +++---
>  3 files changed, 17 insertions(+), 24 deletions(-)
> 
> diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.h b/drivers/gpu/drm/i915/i915_gem_gtt.h
> index 8965bbb..aefc968 100644
> --- a/drivers/gpu/drm/i915/i915_gem_gtt.h
> +++ b/drivers/gpu/drm/i915/i915_gem_gtt.h
> @@ -316,6 +316,7 @@ struct i915_ggtt {
>  	struct io_mapping mappable;	/* Mapping to our CPU mappable region */
>  

/* Stolen memory is segmented in hardware with different portions
 * offlimits to certain functions.
 *
 * The drm_mm is initialised to the total accessible range, as found
 * from the PCI config. On Broadwell+, this is further restricted to
 * avoid the first page! The upper end of stolen memory is reserved for
 * hardware functions (such as intermediate encodings and compression)
 * and similarly removed from the accessible range.
 */

>  	size_t stolen_size;		/* Total size of stolen memory */
> +	size_t stolen_usable_start;	/* First page may be unusable. */
>  	size_t stolen_usable_size;	/* Total size minus BIOS reserved */
>  	size_t stolen_reserved_base;
>  	size_t stolen_reserved_size;

Aside: Why are these hw values using size_t? 

> diff --git a/drivers/gpu/drm/i915/i915_gem_stolen.c b/drivers/gpu/drm/i915/i915_gem_stolen.c
> index b1c8897..56d7e0b 100644
> --- a/drivers/gpu/drm/i915/i915_gem_stolen.c
> +++ b/drivers/gpu/drm/i915/i915_gem_stolen.c
> @@ -54,12 +54,6 @@ int i915_gem_stolen_insert_node_in_range(struct drm_i915_private *dev_priv,
>  	if (!drm_mm_initialized(&dev_priv->mm.stolen))
>  		return -ENODEV;
>  
> -	/* See the comment at the drm_mm_init() call for more about this check.
> -	 * WaSkipStolenMemoryFirstPage:bdw+ (incomplete)
> -	 */
> -	if (start < 4096 && INTEL_GEN(dev_priv) >= 8)
> -		start = 4096;
> -
>  	mutex_lock(&dev_priv->mm.stolen_lock);
>  	ret = drm_mm_insert_node_in_range(&dev_priv->mm.stolen, node, size,
>  					  alignment, start, end,
> @@ -74,10 +68,11 @@ int i915_gem_stolen_insert_node(struct drm_i915_private *dev_priv,
>  				unsigned alignment)
>  {
>  	struct i915_ggtt *ggtt = &dev_priv->ggtt;
> +	u64 start = ggtt->stolen_usable_start;
> +	u64 end = start + ggtt->stolen_usable_size;
>  
>  	return i915_gem_stolen_insert_node_in_range(dev_priv, node, size,
> -						    alignment, 0,
> -						    ggtt->stolen_usable_size);

? The range is already restricted, you cannot get anything before
stolen_usable_start, or anything after usable_size. Just pass 0,
U64_MAX.

> +						    alignment, start, end);
>  }
>  
>  void i915_gem_stolen_remove_node(struct drm_i915_private *dev_priv,
> @@ -489,20 +484,17 @@ int i915_gem_init_stolen(struct drm_i915_private *dev_priv)
>  		      ggtt->stolen_size >> 10,
>  		      (ggtt->stolen_size - reserved_total) >> 10);
>  
> -	ggtt->stolen_usable_size = ggtt->stolen_size - reserved_total;
> +	ggtt->stolen_usable_start = 0;
> +	/* WaSkipStolenMemoryFirstPage:bdw+ */
> +	if (INTEL_GEN(dev_priv) >= 8)
> +		ggtt->stolen_usable_start = 4096;
>  
> -	/*
> -	 * Basic memrange allocator for stolen space.
> -	 *
> -	 * TODO: Notice that some platforms require us to not use the first page
> -	 * of the stolen memory but their BIOSes may still put the framebuffer
> -	 * on the first page. So we don't reserve this page for now because of
> -	 * that. Our current solution is to just prevent new nodes from being
> -	 * inserted on the first page - see the check we have at
> -	 * i915_gem_stolen_insert_node_in_range(). We may want to fix the fbcon
> -	 * problem later.
> -	 */
> -	drm_mm_init(&dev_priv->mm.stolen, 0, ggtt->stolen_usable_size);
> +	ggtt->stolen_usable_size = ggtt->stolen_size - reserved_total -
> +				   ggtt->stolen_usable_start;
> +
> +	/* Basic memrange allocator for stolen space. */
> +	drm_mm_init(&dev_priv->mm.stolen, ggtt->stolen_usable_start,
> +		    ggtt->stolen_usable_size);
>  
>  	return 0;
>  }
> diff --git a/drivers/gpu/drm/i915/intel_fbc.c b/drivers/gpu/drm/i915/intel_fbc.c
> index 62f215b..534fcb3c 100644
> --- a/drivers/gpu/drm/i915/intel_fbc.c
> +++ b/drivers/gpu/drm/i915/intel_fbc.c
> @@ -528,7 +528,7 @@ static int find_compression_threshold(struct drm_i915_private *dev_priv,
>  	struct i915_ggtt *ggtt = &dev_priv->ggtt;
>  	int compression_threshold = 1;
>  	int ret;
> -	u64 end;
> +	u64 start = ggtt->stolen_usable_start, end;
>  
>  	/* The FBC hardware for BDW/SKL doesn't have access to the stolen
>  	 * reserved range size, so it always assumes the maximum (8mb) is used.
> @@ -538,7 +538,7 @@ static int find_compression_threshold(struct drm_i915_private *dev_priv,
>  	    IS_SKYLAKE(dev_priv) || IS_KABYLAKE(dev_priv))
>  		end = ggtt->stolen_size - 8 * 1024 * 1024;
>  	else
> -		end = ggtt->stolen_usable_size;
> +		end = start + ggtt->stolen_usable_size;

end = U64_MAX;
No need to set start.
-Chris
diff mbox

Patch

diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.h b/drivers/gpu/drm/i915/i915_gem_gtt.h
index 8965bbb..aefc968 100644
--- a/drivers/gpu/drm/i915/i915_gem_gtt.h
+++ b/drivers/gpu/drm/i915/i915_gem_gtt.h
@@ -316,6 +316,7 @@  struct i915_ggtt {
 	struct io_mapping mappable;	/* Mapping to our CPU mappable region */
 
 	size_t stolen_size;		/* Total size of stolen memory */
+	size_t stolen_usable_start;	/* First page may be unusable. */
 	size_t stolen_usable_size;	/* Total size minus BIOS reserved */
 	size_t stolen_reserved_base;
 	size_t stolen_reserved_size;
diff --git a/drivers/gpu/drm/i915/i915_gem_stolen.c b/drivers/gpu/drm/i915/i915_gem_stolen.c
index b1c8897..56d7e0b 100644
--- a/drivers/gpu/drm/i915/i915_gem_stolen.c
+++ b/drivers/gpu/drm/i915/i915_gem_stolen.c
@@ -54,12 +54,6 @@  int i915_gem_stolen_insert_node_in_range(struct drm_i915_private *dev_priv,
 	if (!drm_mm_initialized(&dev_priv->mm.stolen))
 		return -ENODEV;
 
-	/* See the comment at the drm_mm_init() call for more about this check.
-	 * WaSkipStolenMemoryFirstPage:bdw+ (incomplete)
-	 */
-	if (start < 4096 && INTEL_GEN(dev_priv) >= 8)
-		start = 4096;
-
 	mutex_lock(&dev_priv->mm.stolen_lock);
 	ret = drm_mm_insert_node_in_range(&dev_priv->mm.stolen, node, size,
 					  alignment, start, end,
@@ -74,10 +68,11 @@  int i915_gem_stolen_insert_node(struct drm_i915_private *dev_priv,
 				unsigned alignment)
 {
 	struct i915_ggtt *ggtt = &dev_priv->ggtt;
+	u64 start = ggtt->stolen_usable_start;
+	u64 end = start + ggtt->stolen_usable_size;
 
 	return i915_gem_stolen_insert_node_in_range(dev_priv, node, size,
-						    alignment, 0,
-						    ggtt->stolen_usable_size);
+						    alignment, start, end);
 }
 
 void i915_gem_stolen_remove_node(struct drm_i915_private *dev_priv,
@@ -489,20 +484,17 @@  int i915_gem_init_stolen(struct drm_i915_private *dev_priv)
 		      ggtt->stolen_size >> 10,
 		      (ggtt->stolen_size - reserved_total) >> 10);
 
-	ggtt->stolen_usable_size = ggtt->stolen_size - reserved_total;
+	ggtt->stolen_usable_start = 0;
+	/* WaSkipStolenMemoryFirstPage:bdw+ */
+	if (INTEL_GEN(dev_priv) >= 8)
+		ggtt->stolen_usable_start = 4096;
 
-	/*
-	 * Basic memrange allocator for stolen space.
-	 *
-	 * TODO: Notice that some platforms require us to not use the first page
-	 * of the stolen memory but their BIOSes may still put the framebuffer
-	 * on the first page. So we don't reserve this page for now because of
-	 * that. Our current solution is to just prevent new nodes from being
-	 * inserted on the first page - see the check we have at
-	 * i915_gem_stolen_insert_node_in_range(). We may want to fix the fbcon
-	 * problem later.
-	 */
-	drm_mm_init(&dev_priv->mm.stolen, 0, ggtt->stolen_usable_size);
+	ggtt->stolen_usable_size = ggtt->stolen_size - reserved_total -
+				   ggtt->stolen_usable_start;
+
+	/* Basic memrange allocator for stolen space. */
+	drm_mm_init(&dev_priv->mm.stolen, ggtt->stolen_usable_start,
+		    ggtt->stolen_usable_size);
 
 	return 0;
 }
diff --git a/drivers/gpu/drm/i915/intel_fbc.c b/drivers/gpu/drm/i915/intel_fbc.c
index 62f215b..534fcb3c 100644
--- a/drivers/gpu/drm/i915/intel_fbc.c
+++ b/drivers/gpu/drm/i915/intel_fbc.c
@@ -528,7 +528,7 @@  static int find_compression_threshold(struct drm_i915_private *dev_priv,
 	struct i915_ggtt *ggtt = &dev_priv->ggtt;
 	int compression_threshold = 1;
 	int ret;
-	u64 end;
+	u64 start = ggtt->stolen_usable_start, end;
 
 	/* The FBC hardware for BDW/SKL doesn't have access to the stolen
 	 * reserved range size, so it always assumes the maximum (8mb) is used.
@@ -538,7 +538,7 @@  static int find_compression_threshold(struct drm_i915_private *dev_priv,
 	    IS_SKYLAKE(dev_priv) || IS_KABYLAKE(dev_priv))
 		end = ggtt->stolen_size - 8 * 1024 * 1024;
 	else
-		end = ggtt->stolen_usable_size;
+		end = start + ggtt->stolen_usable_size;
 
 	/* HACK: This code depends on what we will do in *_enable_fbc. If that
 	 * code changes, this code needs to change as well.
@@ -549,7 +549,7 @@  static int find_compression_threshold(struct drm_i915_private *dev_priv,
 
 	/* Try to over-allocate to reduce reallocations and fragmentation. */
 	ret = i915_gem_stolen_insert_node_in_range(dev_priv, node, size <<= 1,
-						   4096, 0, end);
+						   4096, start, end);
 	if (ret == 0)
 		return compression_threshold;