[PATCH] powerpc/64s/perf: perf interrupt does not have to get_user_pages to access user memory

[PATCH] powerpc/64s/perf: perf interrupt does not have to get_user_pages to access user memory

[Nicholas Piggin]

read_user_stack_slow that walks user address translation by hand is
only required on hash, because a hash fault can not be serviced from
"NMI" context (to avoid re-entering the hash code) so the user stack
can be mapped into Linux page tables but not accessible by the CPU.

Radix MMU mode does not have this restriction. A page fault failure
would indicate the page is not accessible via get_user_pages either,
so avoid this on radix.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
---
 arch/powerpc/perf/callchain.h    | 2 +-
 arch/powerpc/perf/callchain_64.c | 3 ++-
 2 files changed, 3 insertions(+), 2 deletions(-)

diff --git a/arch/powerpc/perf/callchain.h b/arch/powerpc/perf/callchain.h
index ae24d4a00da6..d6fa6e25234f 100644
--- a/arch/powerpc/perf/callchain.h
+++ b/arch/powerpc/perf/callchain.h
@@ -33,7 +33,7 @@ static inline int __read_user_stack(const void __user *ptr, void *ret,
 
 	rc = copy_from_user_nofault(ret, ptr, size);
 
-	if (IS_ENABLED(CONFIG_PPC64) && rc)
+	if (IS_ENABLED(CONFIG_PPC64) && !radix_enabled() && rc)
 		return read_user_stack_slow(ptr, ret, size);
 
 	return rc;
diff --git a/arch/powerpc/perf/callchain_64.c b/arch/powerpc/perf/callchain_64.c
index fed90e827f3a..0777b04a0c56 100644
--- a/arch/powerpc/perf/callchain_64.c
+++ b/arch/powerpc/perf/callchain_64.c
@@ -21,7 +21,8 @@
 /*
  * On 64-bit we don't want to invoke hash_page on user addresses from
  * interrupt context, so if the access faults, we read the page tables
- * to find which page (if any) is mapped and access it directly.
+ * to find which page (if any) is mapped and access it directly. Radix
+ * has no need for this so it doesn't use read_user_stack_slow.
  */
 int read_user_stack_slow(const void __user *ptr, void *buf, int nb)
 {
-- 
2.23.0

Re: [PATCH] powerpc/64s/perf: perf interrupt does not have to get_user_pages to access user memory

[Michael Ellerman]

On Wed, 11 Nov 2020 22:01:51 +1000, Nicholas Piggin wrote:
> read_user_stack_slow that walks user address translation by hand is
> only required on hash, because a hash fault can not be serviced from
> "NMI" context (to avoid re-entering the hash code) so the user stack
> can be mapped into Linux page tables but not accessible by the CPU.
> 
> Radix MMU mode does not have this restriction. A page fault failure
> would indicate the page is not accessible via get_user_pages either,
> so avoid this on radix.

Applied to powerpc/next.

[1/1] powerpc/64s/perf: perf interrupt does not have to get_user_pages to access user memory
      https://git.kernel.org/powerpc/c/987c426320cce72d1b28f55c8603b239e4f7187c

cheers