On Wed, Aug 1, 2018 at 2:55 PM Andrew Morton <[email protected]> wrote: > > On Wed, 1 Aug 2018 11:53:31 -0700 Nick Desaulniers <[email protected]> > wrote: > > > As part of the effort to reduce the code duplication between _THIS_IP_ > > and current_text_addr(), let's consolidate callers of > > current_text_addr() to use _THIS_IP_. > > Why not switch everything to current_text_addr()? _THIS_IP_ is ugly ;) > > Several architectures (s390, sparc, sh, ...) do funky things in their > current_text_addr(). Does the generic kernel.h implementation of > current_text_addr() work OK on those architectures?
Good question, one I've been mulling over for a few days. That's definitely something that I've considered. Mostly, it comes down to there being fewer call sites of current_text_addr() than _THIS_IP_, so it's IMO a smaller change that doesn't require coordinating tree-wide changes. What's interesting about current_text_addr() is that while there are many arch specific definitions, there are only 5 call sites in the whole kernel (and 4 of them are arch specific). So unless there's an arch specific reason to prefer inline assembly over higher level C (yes, returning the address of a label in a statement expression is ugly), (it looks like parisc might have a reason, at its lone call site, see: https://lkml.org/lkml/2018/8/1/1678, TBD), then I don't think that current_text_addr() has a raison d'être. I suspect that its use predates the GNU C extensions needed for the current generic C implementation (statement expressions + addr of label) and has just been reimplemented in every new arch's arch/*/include/asm/processor.h simply because the other archs had it. So the plan is to replace the 5 call sites of current_text_addr() with _THIS_IP_ (if possible), then delete all definitions of the arch specific current_text_addr()'s. This can be done on a per tree basis, rather than a treewide patch for converting _THIS_IP_ (who has many more call sites) to current_text_addr(). What do you think? -- Thanks, ~Nick Desaulniers
