x86/mm: Unbreak modules that rely on external PAGE_KERNEL availability
Commit 7744ccdbc16f0 ("x86/mm: Add Secure Memory Encryption (SME)
support") as a side-effect made PAGE_KERNEL all of a sudden unavailable
to modules which can't make use of EXPORT_SYMBOL_GPL() symbols.
This is because once SME is enabled, sme_me_mask (which is introduced as
EXPORT_SYMBOL_GPL) makes its way to PAGE_KERNEL through _PAGE_ENC,
causing imminent build failure for all the modules which make use of all
the EXPORT-SYMBOL()-exported API (such as vmap(), __vmalloc(),
Exporting (as EXPORT_SYMBOL()) interfaces (and having done so for ages)
that take pgprot_t argument, while making it impossible to -- all of a
sudden -- pass PAGE_KERNEL to it, feels rather incosistent.
Restore the original behavior and make it possible to pass PAGE_KERNEL
to all its EXPORT_SYMBOL() consumers.
[ This is all so not wonderful. We shouldn't need that "sme_me_mask"
access at all in all those places that really don't care about that
level of detail, and just want _PAGE_KERNEL or whatever.
We have some similar issues with _PAGE_CACHE_WP and _PAGE_NOCACHE,
both of which hide a "cachemode2protval()" call, and which also ends
up using another EXPORT_SYMBOL(), but at least that only triggers for
the much more rare cases.
Maybe we could move these dynamic page table bits to be generated much
deeper down in the VM layer, instead of hiding them in the macros that
So this all would merit some cleanup. But not today. - Linus ]
Cc: Tom Lendacky <firstname.lastname@example.org>
Signed-off-by: Jiri Kosina <email@example.com>
Despised-by: Thomas Gleixner <firstname.lastname@example.org>
Signed-off-by: Linus Torvalds <email@example.com>
diff --git a/arch/x86/mm/mem_encrypt.c b/arch/x86/mm/mem_encrypt.c
index 16c5f37..0286327 100644
@@ -40,7 +40,7 @@
* section is later cleared.
u64 sme_me_mask __section(.data) = 0;
/* Buffer used for early in-place encryption by BSP, no locking needed */
static char sme_early_buffer[PAGE_SIZE] __aligned(PAGE_SIZE);