On Fri, Jan 20, 2017 at 10:38:41AM -0600, Andrew F. Davis wrote:
On 01/20/2017 10:25 AM, Tom Rini wrote:
On Fri, Jan 06, 2017 at 01:35:44PM -0600, Andrew F. Davis wrote:
Print statements in SPL depend on lib/common support, so many such statements are ifdef'd, move the check to the common.h header and remove these inline checks.
Signed-off-by: Andrew F. Davis afd@ti.com Reviewed-by: Tom Rini trini@konsulko.com
This patch is a good example of why travis-ci is useful, even if takes a few hours for the cycle to complete (kick it off and check the results in the morning :)). As is, it's broken on PowerPC (where CONFIG_SPL_INIT_MINIMAL is the conditional for puts/printf/etc), mx31pdk and evb-rk3036 (and this is an incomplete list). The ARM targets are harder to just fix as it shows an underlying problem. Today we have no single symbol that means "In SPL I want serial output" (and ditto TPL). We try and rely on SPL_SERIAL_SUPPORT but this misses the case where we use neither TINY_PRINTF nor LIBCOMMON but instead have only puts available in a more raw way. So before we can make the type of change you're doing here we need to introduce a symbol that means "I have output". That will also greatly reduce the logic needed in the tests in common.h for having puts/etc be real or do-while loops. Thanks!
Ahh, I was expecting something like this happening with all the levels of ifdef logic involved. Could we keep this patch and add CONFIG_SPL_INIT_MINIMAL to the check in common.h?
I did that, but it does not address the problem of mx31pdk and rockchip and others.
Removing all these various ifdefs from files is still correct, all this ifdef logic should *not* be around calls to puts/etc in-line in the code, all we need is to figure out the logic for the function's definitions in common.h, which can be cleaned up in a later patch.
Well, thinking about this harder. What would be OK for now would be a v2 (that you've pushed through travis-ci) that just removes the tests as that should be OK. But... I wonder if we don't have those tests due to needing to #if out the prints due to the gcc bug[1] that's only fixed in gcc 6.x and later.
[1]: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=54303