]> asedeno.scripts.mit.edu Git - linux.git/commitdiff
media: vicodec: avoid clang frame size warning
authorArnd Bergmann <arnd@arndb.de>
Fri, 22 Feb 2019 14:50:03 +0000 (09:50 -0500)
committerMauro Carvalho Chehab <mchehab+samsung@kernel.org>
Tue, 19 Mar 2019 18:00:32 +0000 (14:00 -0400)
Clang-9 makes some different inlining decisions compared to gcc, which
leads to a warning about a possible stack overflow problem when building
with CONFIG_KASAN, including when setting asan-stack=0, which avoids
most other frame overflow warnings:

drivers/media/platform/vicodec/codec-fwht.c:673:12: error: stack frame size of 2224 bytes in function 'encode_plane'

Manually adding noinline_for_stack annotations in those functions
called by encode_plane() or decode_plane() that require a significant
amount of kernel stack makes this impossible to happen with any
compiler.

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl>
Signed-off-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>
drivers/media/platform/vicodec/codec-fwht.c

index d1d6085da9f1de33c7a2665bc45c3b2543bd79cc..cf469a1191aa7e4bce2522159c4243e4bd97c5b2 100644 (file)
@@ -46,8 +46,12 @@ static const uint8_t zigzag[64] = {
        63,
 };
 
-
-static int rlc(const s16 *in, __be16 *output, int blocktype)
+/*
+ * noinline_for_stack to work around
+ * https://bugs.llvm.org/show_bug.cgi?id=38809
+ */
+static int noinline_for_stack
+rlc(const s16 *in, __be16 *output, int blocktype)
 {
        s16 block[8 * 8];
        s16 *wp = block;
@@ -106,8 +110,8 @@ static int rlc(const s16 *in, __be16 *output, int blocktype)
  * This function will worst-case increase rlc_in by 65*2 bytes:
  * one s16 value for the header and 8 * 8 coefficients of type s16.
  */
-static u16 derlc(const __be16 **rlc_in, s16 *dwht_out,
-                const __be16 *end_of_input)
+static noinline_for_stack u16
+derlc(const __be16 **rlc_in, s16 *dwht_out, const __be16 *end_of_input)
 {
        /* header */
        const __be16 *input = *rlc_in;
@@ -240,8 +244,9 @@ static void dequantize_inter(s16 *coeff)
                        *coeff <<= *quant;
 }
 
-static void fwht(const u8 *block, s16 *output_block, unsigned int stride,
-                unsigned int input_step, bool intra)
+static void noinline_for_stack fwht(const u8 *block, s16 *output_block,
+                                   unsigned int stride,
+                                   unsigned int input_step, bool intra)
 {
        /* we'll need more than 8 bits for the transformed coefficients */
        s32 workspace1[8], workspace2[8];
@@ -373,7 +378,8 @@ static void fwht(const u8 *block, s16 *output_block, unsigned int stride,
  * Furthermore values can be negative... This is just a version that
  * works with 16 signed data
  */
-static void fwht16(const s16 *block, s16 *output_block, int stride, int intra)
+static void noinline_for_stack
+fwht16(const s16 *block, s16 *output_block, int stride, int intra)
 {
        /* we'll need more than 8 bits for the transformed coefficients */
        s32 workspace1[8], workspace2[8];
@@ -456,7 +462,8 @@ static void fwht16(const s16 *block, s16 *output_block, int stride, int intra)
        }
 }
 
-static void ifwht(const s16 *block, s16 *output_block, int intra)
+static noinline_for_stack void
+ifwht(const s16 *block, s16 *output_block, int intra)
 {
        /*
         * we'll need more than 8 bits for the transformed coefficients
@@ -604,9 +611,9 @@ static int var_inter(const s16 *old, const s16 *new)
        return ret;
 }
 
-static int decide_blocktype(const u8 *cur, const u8 *reference,
-                           s16 *deltablock, unsigned int stride,
-                           unsigned int input_step)
+static noinline_for_stack int
+decide_blocktype(const u8 *cur, const u8 *reference, s16 *deltablock,
+                unsigned int stride, unsigned int input_step)
 {
        s16 tmp[64];
        s16 old[64];