[U-Boot] [U-Boot, 1/1] arm: mach-omap2: Fix secure file generation

[Andrew F. Davis]

On 12/09/2016 03:30 PM, Tom Rini wrote:
> On Fri, Dec 09, 2016 at 02:24:32PM -0600, Andrew F. Davis wrote:
>> On 12/09/2016 02:10 PM, Tom Rini wrote:
>>> On Fri, Dec 09, 2016 at 02:05:29PM -0600, Andrew F. Davis wrote:
>>>> On 12/09/2016 01:59 PM, Tom Rini wrote:
>>>>> On Thu, Dec 08, 2016 at 04:48:07PM -0600, Andrew F. Davis wrote:
>>>>>
>>>>>> When TI_SECURE_DEV_PKG is not defined we warn that the file '*_HS' was
>>>>>> not generated but generate an unsigned one anyway. When TI_SECURE_DEV_PKG
>>>>>> is exported and the user re-builds, make will detect this file as
>>>>>> unchangedand and so assume it does not need to be re-generated. This
>>>>>> causes it to pack unsigned files. Fix this by not generating these
>>>>>> fake unsigned *_HS files.
>>>>>>
>>>>>> Signed-off-by: Andrew F. Davis <afd at ti.com>
>>>>>> Reviewed-by: Tom Rini <trini at konsulko.com>
>>>>>> ---
>>>>>>  arch/arm/mach-omap2/config_secure.mk | 4 ++--
>>>>>>  1 file changed, 2 insertions(+), 2 deletions(-)
>>>>>>
>>>>>> diff --git a/arch/arm/mach-omap2/config_secure.mk b/arch/arm/mach-omap2/config_secure.mk
>>>>>> index 1122439..33c7059 100644
>>>>>> --- a/arch/arm/mach-omap2/config_secure.mk
>>>>>> +++ b/arch/arm/mach-omap2/config_secure.mk
>>>>>> @@ -35,12 +35,12 @@ cmd_omapsecureimg = $(TI_SECURE_DEV_PKG)/scripts/secure-binary-image.sh \
>>>>>>  else
>>>>>>  cmd_omapsecureimg = echo "WARNING:" \
>>>>>>  	"$(TI_SECURE_DEV_PKG)/scripts/secure-binary-image.sh not found." \
>>>>>> -	"$@ was NOT created!"; cp $< $@
>>>>>> +	"$@ was NOT created!";
>>>>>>  endif
>>>>>>  else
>>>>>>  cmd_omapsecureimg = echo "WARNING: TI_SECURE_DEV_PKG environment" \
>>>>>>  	"variable must be defined for TI secure devices." \
>>>>>> -	"$@ was NOT created!"; cp $< $@
>>>>>> +	"$@ was NOT created!";
>>>>>>  endif
>>>>>>  endif
>>>>>
>>>>> OK, but now that I build test this (without the tools present) this is a
>>>>> NAK.  The root problem is that if we don't make that dummy file we then:
>>>>>        arm:  +   am57xx_hs_evm
>>>>> +(am57xx_hs_evm) ./tools/mkimage: Can't open u-boot-nodtb_HS.bin: No such file or directory
>>>>> +(am57xx_hs_evm) ./tools/mkimage: failed to build FIT
>>>>> +(am57xx_hs_evm) make[1]: *** [u-boot_HS.img] Error 1
>>>>> +(am57xx_hs_evm) make: *** [sub-make] Error 2
>>>>
>>>> Is this not okay? build *should* fail if TI_SECURE_DEV_PKG is not
>>>> defined. You cannot sign images that *need* to be signed to work on this
>>>> platform, making a fake un-bootable image instead of failing is a hack
>>>> and it confuses the make system when you do put the signing tool in-place.
>>>
>>> Well, I suppose this is a valid question.  I run into it failing as I
>>> (and travis-ci) build all ARM targets.  Maybe we can have the build not
>>> happen (and echo a Warning) and then not invoke mkimage later on if the
>>> env isn't right?
>>
>> For test building you can export TI_SECURE_DEV_PKG to point to a dummy
>> signing tool which just runs cp $1 $2. For real world building this tool
>> is needed just as much as the compiler, if you don't have it you will
>> not build working images, build needs to fail here.
> 
> Hmmm, OK.  But can we not automate that based on TI_SECURE_DEV_PKG being
> unset?
> 

That is what we already do, if TI_SECURE_DEV_PKG is unset, build should
fail, but right now it fakes a successful build, most likely to keep the
auto-validation happy as it does not have the signing tool.

The only other thing I can think of is to always try to sign the images,
even when they have not changed on disk since the last build. Would this
be acceptable?