On Thu, Sep 18, 2025 at 10:49:28PM -0500, Serge E. Hallyn wrote: > On Thu, Sep 18, 2025 at 10:30:18PM +0300, Jarkko Sakkinen wrote: > > The current shenanigans for duration calculation introduce too much > > complexity for a trivial problem, and further the code is hard to patch and > > maintain. > > > > Address these issues with a flat look-up table, which is easy to understand > > and patch. If leaf driver specific patching is required in future, it is > > easy enough to make a copy of this table during driver initialization and > > add the chip parameter back. > > > > 'chip->duration' is retained for TPM 1.x. > > > > As the first entry for this new behavior address TCG spec update mentioned > > in this issue: > > > > https://github.com/raspberrypi/linux/issues/7054 > > > > Therefore, for TPM_SelfTest the duration is set to 3000 ms. > > > > This does not categorize a as bug, given that this is introduced to the > > spec after the feature was originally made. > > > > Cc: Frédéric Jouen <[email protected]> > > Signed-off-by: Jarkko Sakkinen <[email protected]> > > fwiw (which shouldn't be much) looks good to me, but two questions, > one here and one below. > > First, it looks like in the existing code it is possible for a tpm2 > chip to set its own timeouts and then set the TPM_CHIP_FLAG_HAVE_TIMEOUTS > flag to avoid using the defaults, but I don't see anything using that > in-tree. Is it possible that there are out of tree drivers that will be > sabotaged here? Or am I misunderstanding that completely?
Good questions, and I can brief a bit about the context of the pre-existing art and this change. This complexity was formed in 2014 when I originally developed TPM2 support and the only available testing plaform was early Intel PTT with a flakky version of TPM2 support (e.g., no localities). Since then we haven't had per leaf-driver divergence. Further, I think that this type of layout is actually a better fit if we ever need to quirks for command durations for a particular device, as then we can migrate to "copy and patch" semantics i.e., have a copy of this map in the chip structure. As per out-of-tree drivers, it's unfortunate reality of out-of-tree drivers :-) However, this will definitely add some extra work, when backporting fixes (not overwhelmingly much). BR, Jarkko
