On 17/12/2025 11:01, Ryan Roberts wrote:

On 17/12/2025 09:26, Kevin Brodsky wrote:

quoted

On 17/12/2025 05:14, Andrew Morton wrote:

quoted

On Mon, 15 Dec 2025 15:03:23 +0000 Kevin Brodsky [off-list ref] wrote:

quoted

Add basic KUnit tests for the generic aspects of the lazy MMU mode:
ensure that it appears active when it should, depending on how
enable/disable and pause/resume pairs are nested.

I needed this for powerpc allmodconfig;

--- a/arch/powerpc/mm/book3s64/hash_tlb.c~mm-add-basic-tests-for-lazy_mmu-fix
+++ a/arch/powerpc/mm/book3s64/hash_tlb.c

@@ -30,6 +30,7 @@
 #include <trace/events/thp.h>
 
 DEFINE_PER_CPU(struct ppc64_tlb_batch, ppc64_tlb_batch);
+EXPORT_SYMBOL_GPL(ppc64_tlb_batch);
 
 /*
  * A linux PTE was changed and the corresponding hash table entry

@@ -154,6 +155,7 @@ void __flush_tlb_pending(struct ppc64_tl
 		flush_hash_range(i, local);
 	batch->index = 0;
 }
+EXPORT_SYMBOL_GPL(__flush_tlb_pending);
 
 void hash__tlb_flush(struct mmu_gather *tlb)
 {

_

Oh indeed I hadn't considered that arch_{enter,leave}_lazy_mmu_mode()
refer to those symbols on powerpc... Maybe a bit overkill to export
those just for a test module, but I'm not sure there's a good
alternative. Forcing LAZY_MMU_MODE_KUNIT_TEST=y is ugly as it would also
force KUNIT=y. Alternatively we could depend on !PPC, not pretty either.

Does EXPORT_SYMBOL_IF_KUNIT() help?

It most certainly would, I didn't know about that one, thanks!

- Kevin