On Fri, Oct 04, 2024 at 11:24:28AM -0400, Raymond Mao wrote:
Hi Rasmus,
On Fri, 4 Oct 2024 at 03:50, Rasmus Villemoes rasmus.villemoes@prevas.dk wrote:
Raymond Mao raymond.mao@linaro.org writes:
We don't need an API specially for non-watchdog since sha1_csum_wd supports it by disabling CONFIG_HW_WATCHDOG and CONFIG_WATCHDOG. Set 0x10000 as default chunk size for SHA1.
I have to say I believe this is a step in the wrong direction. Having everybody call a function with that _wd suffix is ugly, as is having them all pass some pre-defined constant. Moreover, nowadays what happens every chunksize bytes isn't restricted to watchdog handling.
So yes, we don't need an API specially for non-watchdog, but why not just make sha1_csum() be the interface to call, and let the "maybe we need to call schedule() once in a while" be an implementation detail of sha1_csum().
The same as with our zlib implementaion; we don't have a separate _wd set of routines, we've just hooked schedule() into the main loop of that inflate algorithm.
The reason I keep the ` _csum_wd()` one is to align to other hash APIs currently we have. For MD5, SHA256, SHA512, we only have `_csum_wd()` but no `_csum()` - SHA1 is a special one due to the historical problem I guess. The minimum refactoring I did here is to unify the interface and make it adaptable to the MbedTLS library I introduced with this series. Yes, I agree some optimizations or refactoring are needed for the hash APIs, but I will prefer to do this in a new series other than increasing the scope of this patch set.
I agree this should be the target of near-future follow-up work rather than gating for the series.