Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Fri,  7 Oct 2022 13:11:40 -0700
In-Reply-To: <CAKwvOdkEied8hf6Oid0sGf0ybF2WqrzOvtRiXa=j7Ms-Rc6uBA@mail.gmail.com>
Mime-Version: 1.0
References: <CAKwvOdkEied8hf6Oid0sGf0ybF2WqrzOvtRiXa=j7Ms-Rc6uBA@mail.gmail.com>
Message-ID: <20221007201140.1744961-1-ndesaulniers@google.com>
Subject: [PATCH] fs/select: mark do_select noinline_for_stack for 32b
From:   Nick Desaulniers <ndesaulniers@google.com>
To:     Andrew Morton <akpm@linux-foundation.org>,
        Arnd Bergmann <arnd@arndb.de>
Cc:     Alexander Viro <viro@zeniv.linux.org.uk>,
        Nathan Chancellor <nathan@kernel.org>,
        Tom Rix <trix@redhat.com>, Xiaoming Ni <nixiaoming@huawei.com>,
        Luis Chamberlain <mcgrof@kernel.org>,
        Andi Kleen <ak@linux.intel.com>, linux-fsdevel@vger.kernel.org,
        linux-kernel@vger.kernel.org, llvm@lists.linux.dev,
        Christoph Hellwig <hch@lst.de>,
        Eric Dumazet <edumazet@google.com>,
        "Darrick J. Wong" <darrick.wong@oracle.com>,
        Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
        Nick Desaulniers <ndesaulniers@google.com>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

Effectively a revert of
commit ad312f95d41c ("fs/select: avoid clang stack usage warning")

Various configs can still push the stack useage of core_sys_select()
over the CONFIG_FRAME_WARN threshold (1024B on 32b targets).

  fs/select.c:619:5: error: stack frame size of 1048 bytes in function
  'core_sys_select' [-Werror,-Wframe-larger-than=]

core_sys_select() has a large stack allocation for `stack_fds` where it
tries to do something equivalent to "small string optimization" to
potentially avoid a malloc.

core_sys_select() calls do_select() which has another potentially large
stack allocation, `table`. Both of these values depend on
FRONTEND_STACK_ALLOC.

Mix those two large allocation with register spills which are
exacerbated by various configs and compiler versions and we can just
barely exceed the 1024B limit.

Rather than keep trying to find the right value of MAX_STACK_ALLOC or
FRONTEND_STACK_ALLOC, mark do_select() as noinline_for_stack for 32b
targets.

The intent of FRONTEND_STACK_ALLOC is to help potentially avoid a
dynamic memory allocation. In that spirit, restore the previous
threshold but separate the stack frames for 32b targets.

Link: https://lore.kernel.org/lkml/20221006222124.aabaemy7ofop7ccz@google.com/
Fixes: ad312f95d41c ("fs/select: avoid clang stack usage warning")
Suggested-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Nick Desaulniers <ndesaulniers@google.com>
---
 fs/select.c          | 7 +++++++
 include/linux/poll.h | 4 ----
 2 files changed, 7 insertions(+), 4 deletions(-)

diff --git a/fs/select.c b/fs/select.c
index 0ee55af1a55c..945d04b9cf5a 100644
--- a/fs/select.c
+++ b/fs/select.c
@@ -476,6 +476,13 @@ static inline void wait_key_set(poll_table *wait, unsigned long in,
 		wait->_key |= POLLOUT_SET;
 }
 
+#ifdef CONFIG_64BIT
+#define noinline_for_stack_32b
+#else
+#define noinline_for_stack_32b noinline_for_stack
+#endif
+
+noinline_for_stack_32b
 static int do_select(int n, fd_set_bits *fds, struct timespec64 *end_time)
 {
 	ktime_t expire, *to = NULL;
diff --git a/include/linux/poll.h b/include/linux/poll.h
index a9e0e1c2d1f2..d1ea4f3714a8 100644
--- a/include/linux/poll.h
+++ b/include/linux/poll.h
@@ -14,11 +14,7 @@
 
 /* ~832 bytes of stack space used max in sys_select/sys_poll before allocating
    additional memory. */
-#ifdef __clang__
-#define MAX_STACK_ALLOC 768
-#else
 #define MAX_STACK_ALLOC 832
-#endif
 #define FRONTEND_STACK_ALLOC	256
 #define SELECT_STACK_ALLOC	FRONTEND_STACK_ALLOC
 #define POLL_STACK_ALLOC	FRONTEND_STACK_ALLOC

base-commit: 93ed07a23fd08b8613f64cf0a15d7fbdaca010fd
-- 
2.38.0.rc2.412.g84df46c1b4-goog