diff options
| author |   Heiko Carstens <hca@linux.ibm.com> | 2025-02-11 20:19:29 +0100 |
|---|---|---|
| committer |   Vasily Gorbik <gor@linux.ibm.com> | 2025-03-04 17:18:03 +0100 |
| commit | 88e87cb7b8f8afcf1776d5b8a3ddd17fe8d2a32f (patch) | |
| tree | 599444e250ca73b57bd710742836bf24fc6e0a6e | |
| parent | s390/uaccess: Define INLINE_COPY_FROM_USER and INLINE_COPY_TO_USER (diff) | |
| download | wireguard-linux-88e87cb7b8f8afcf1776d5b8a3ddd17fe8d2a32f.tar.xz wireguard-linux-88e87cb7b8f8afcf1776d5b8a3ddd17fe8d2a32f.zip | |
s390/uaccess: Optimize raw_copy_from_user() / raw_copy_to_user() for constant sizes
Avoid that the compiler generates an mvcos loop for constant sizes
smaller than 4096 bytes. The mvcos instruction copies between zero and
4096 bytes (effective length) with one operation. Therefore it is not
necessary to implement a loop for sizes smaller or equal to 4096
bytes.
This reduces the kernel text size by ~50kb (defconfig, gcc 14.2.0):
add/remove: 4/5 grow/shrink: 6/471 up/down: 2294/-51700 (-49406)
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
| -rw-r--r-- | arch/s390/include/asm/uaccess.h | 4 |
1 files changed, 4 insertions, 0 deletions
diff --git a/arch/s390/include/asm/uaccess.h b/arch/s390/include/asm/uaccess.h index 333a2afc0d6f..b5712d690a93 100644 --- a/arch/s390/include/asm/uaccess.h +++ b/arch/s390/include/asm/uaccess.h @@ -49,6 +49,8 @@ raw_copy_from_user(void *to, const void __user *from, unsigned long size) : CC_OUT(cc, cc), [size] "+d" (size), [to] "=Q" (*(char *)to) : [spec] "I" (0x81), [from] "Q" (*(const char __user *)from) : CC_CLOBBER_LIST("memory", "0")); + if (__builtin_constant_p(osize) && osize <= 4096) + return osize - size; if (likely(CC_TRANSFORM(cc) == 0)) return osize - size; size -= 4096; @@ -75,6 +77,8 @@ raw_copy_to_user(void __user *to, const void *from, unsigned long size) : CC_OUT(cc, cc), [size] "+d" (size), [to] "=Q" (*(char __user *)to) : [spec] "I" (0x81), [from] "Q" (*(const char *)from) : CC_CLOBBER_LIST("memory", "0")); + if (__builtin_constant_p(osize) && osize <= 4096) + return osize - size; if (likely(CC_TRANSFORM(cc) == 0)) return osize - size; size -= 4096; |