[Intel-gfx] [PATCH 1/2 v2 addendum v2] drm/i915: Allow parsing of variable size child device entries from VBT

[Jani Nikula]

On Fri, 14 Aug 2015, Jani Nikula <jani.nikula at linux.intel.com> wrote:
> On Thu, 13 Aug 2015, David Weinehall <david.weinehall at linux.intel.com> wrote:
>> On Wed, Aug 12, 2015 at 05:19:35PM +0300, Jani Nikula wrote:
>>> On Wed, 12 Aug 2015, David Weinehall <david.weinehall at linux.intel.com> wrote:
>>> > Some more fixup is needed; the bits from Antti's patch
>>> > that actually expanded the struct to fully fit the newer
>>> > versions of the child_device_config was part of the second
>>> > patch; since that patch hasn't been merged yet we need this bit:
>>> >
>>> > This applies on top of the patch you already merged
>>> > (the Iboost patch will need corresponding adjustment to
>>> >  remove the changes I split out):
>>> >
>>> > Expand common_child_dev_config to be able to fit all information
>>> > defined by the latest VBT specification.
>>> >
>>> > Signed-off-by: David Weinehall <david.weinehall at linux.intel.com>
>>> > CC: Antti Koskipaa <antti.koskipaa at linux.intel.com>
>>> > ---
>>> >  intel_bios.c |    7 ++++++-
>>> >  intel_bios.h |    4 ++++
>>> >  2 files changed, 10 insertions(+), 1 deletion(-)
>>> >
>>> > diff --git a/drivers/gpu/drm/i915/intel_bios.c b/drivers/gpu/drm/i915/intel_bios.c
>>> > index 990acc20771a..40e2cc4e7419 100644
>>> > --- a/drivers/gpu/drm/i915/intel_bios.c
>>> > +++ b/drivers/gpu/drm/i915/intel_bios.c
>>> > @@ -1038,6 +1038,10 @@ parse_device_mapping(struct drm_i915_private *dev_priv,
>>> >  		DRM_DEBUG_KMS("No general definition block is found, no devices defined.\n");
>>> >  		return;
>>> >  	}
>>> > +	/* Remember to keep this in sync with child_device_config;
>>> > +	 * whenever a new feature is added to BDB that causes that
>>> > +	 * struct to grow this needs to be updated too
>>> > +	 */
>>> >  	if (bdb->version < 195) {
>>> >  		expected_size = 33;
>>> >  	} else if (bdb->version == 195) {
>>> > @@ -1051,7 +1055,8 @@ parse_device_mapping(struct drm_i915_private *dev_priv,
>>> >  	}
>>> >  
>>> >  	if (expected_size > sizeof(*p_child)) {
>>> > -		DRM_ERROR("child_device_config cannot fit in p_child\n");
>>> > +		DRM_ERROR("child_device_config (size %u) cannot fit in p_child (size %u); bdb->version: %u\n",
>>> > +			  expected_size, sizeof(*p_child), bdb->version);
>>> 
>>> drivers/gpu/drm/i915/intel_bios.c: In function ‘parse_device_mapping’:
>>> drivers/gpu/drm/i915/intel_bios.c:1058:3: warning: format ‘%u’ expects argument of type ‘unsigned int’, but argument 3 has type ‘long unsigned int’ [-Wformat=]
>>>    DRM_ERROR("child_device_config (size %u) cannot fit in p_child (size %u); bdb->version: %u\n",
>>>    ^
>>>   CC [M]  drivers/gpu/drm/i915/intel_fifo_underrun.o
>>
>> Well spotted.  Or rather, stupid of me to forget that sizeof yields
>> size_t as type, thus necessitating %zu as conversion specifier.
>>
>> Fixed version:
>>
>> Expand common_child_dev_config to be able to fit all information
>> defined by the latest VBT specification.
>>
>> v2: Use proper conversion specifier for size_t (%zu)
>>
>> Signed-off-by: David Weinehall <david.weinehall at linux.intel.com>
>> CC: Antti Koskipaa <antti.koskipaa at linux.intel.com>
>>
>>  intel_bios.c |    7 ++++++-
>>  intel_bios.h |    4 ++++
>>  2 files changed, 10 insertions(+), 1 deletion(-)
>>
>> diff --git a/drivers/gpu/drm/i915/intel_bios.c b/drivers/gpu/drm/i915/intel_bios.c
>> index 990acc20771a..5673ed53797b 100644
>> --- a/drivers/gpu/drm/i915/intel_bios.c
>> +++ b/drivers/gpu/drm/i915/intel_bios.c
>> @@ -1038,6 +1038,10 @@ parse_device_mapping(struct drm_i915_private *dev_priv,
>>  		DRM_DEBUG_KMS("No general definition block is found, no devices defined.\n");
>>  		return;
>>  	}
>> +	/* Remember to keep this in sync with child_device_config;
>> +	 * whenever a new feature is added to BDB that causes that
>> +	 * struct to grow this needs to be updated too
>> +	 */
>>  	if (bdb->version < 195) {
>>  		expected_size = 33;
>>  	} else if (bdb->version == 195) {
>> @@ -1051,7 +1055,8 @@ parse_device_mapping(struct drm_i915_private *dev_priv,
>>  	}
>>  
>>  	if (expected_size > sizeof(*p_child)) {
>> -		DRM_ERROR("child_device_config cannot fit in p_child\n");
>> +		DRM_ERROR("child_device_config (size %u) cannot fit in p_child (size %zu); bdb->version: %u\n",
>> +			  expected_size, sizeof(*p_child), bdb->version);
>>  		return;
>>  	}
>>  
>> diff --git a/drivers/gpu/drm/i915/intel_bios.h b/drivers/gpu/drm/i915/intel_bios.h
>> index f7ad6a585129..71cb96f77870 100644
>> --- a/drivers/gpu/drm/i915/intel_bios.h
>> +++ b/drivers/gpu/drm/i915/intel_bios.h
>> @@ -239,6 +239,10 @@ struct common_child_dev_config {
>>  	u8 not_common2[2];
>>  	u8 ddc_pin;
>>  	u16 edid_ptr;
>> +	u8 obsolete;
>> +	u8 flags_1;
>> +	u8 not_common3[13];
>> +	u8 iboost_level;
>
> How does this impact the if (p_defs->child_dev_size != sizeof(*p_child))
> check in parse_sdvo_device_mapping()?

Ugh. What a mess the whole child device parsing is.

What's really broken in the current code is *assuming* some expected
size for future versions, *and* then refusing to parse if the child dev
size has grown. We should just parse *up to* min(VBT reported child dev
size, the biggest child dev we know about). The current code keeps us
doing busywork whenever someone decides to add a field at the end.

Side note, we have *two* functions and loops iterating child
devices. One for new stuff and another for old stuff. Yay.

BR,
Jani.


>
> BR,
> Jani.
>
>>  } __packed;
>>  
>>  /* This field changes depending on the BDB version, so the most reliable way to
>
> -- 
> Jani Nikula, Intel Open Source Technology Center

-- 
Jani Nikula, Intel Open Source Technology Center