Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.38 as permitted sender) client-ip=23.128.96.38;
Date:   Mon, 18 Sep 2023 17:32:41 +0200
Mime-Version: 1.0
Message-ID: <20230918153241.2942764-1-glider@google.com>
Subject: [PATCH v1] mm: make __GFP_BITS_SHIFT independent of CONFIG_LOCKDEP
From:   Alexander Potapenko <glider@google.com>
To:     glider@google.com, dvyukov@google.com, elver@google.com,
        akpm@linux-foundation.org, mingo@kernel.org, linux-mm@kvack.org
Cc:     linux-kernel@vger.kernel.org, kasan-dev@googlegroups.com
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

The list of GFP flags is sparse already (see 0x200u and flags defined
based on CONFIG_KASAN_HW_TAGS), so it makes little sense to save one
bit at the end of it if CONFIG_LOCKDEP is undefined.

The dependence of __GFP_BITS_SHIFT on CONFIG_LOCKDEP also does not let us
add new flags after ___GFP_NOLOCKDEP.

Signed-off-by: Alexander Potapenko <glider@google.com>

---

An alternative to this patch would be putting all ___GFP_ flags into
enum, so that we they occupy as few bits as possible.
My understanding is that this is doable, because tools/perf/builtin-kmem.c
does not depend on the flag values, but maybe I am missing something?
---
 include/linux/gfp_types.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/include/linux/gfp_types.h b/include/linux/gfp_types.h
index 6583a58670c57..bb8d3b31e3b6d 100644
--- a/include/linux/gfp_types.h
+++ b/include/linux/gfp_types.h
@@ -249,7 +249,7 @@ typedef unsigned int __bitwise gfp_t;
 #define __GFP_NOLOCKDEP ((__force gfp_t)___GFP_NOLOCKDEP)
 
 /* Room for N __GFP_FOO bits */
-#define __GFP_BITS_SHIFT (26 + IS_ENABLED(CONFIG_LOCKDEP))
+#define __GFP_BITS_SHIFT (27)
 #define __GFP_BITS_MASK ((__force gfp_t)((1 << __GFP_BITS_SHIFT) - 1))
 
 /**
-- 
2.42.0.459.ge4e396fd5e-goog