linux-toradex.git/include/linux/unaligned, branch v4.12-rc4

include/linux/unaligned: force inlining of byteswap operations

2016-03-17T22:09:34+00:00

Sometimes gcc mysteriously doesn't inline
very small functions we expect to be inlined. See

    https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66122

With this .config:
http://busybox.net/~vda/kernel_config_OPTIMIZE_INLINING_and_Os,
the following functions get deinlined many times.
Examples of disassembly:

 (24 copies, 108 calls):
       66 8b 07                mov    (%rdi),%ax
       55                      push   %rbp
       48 89 e5                mov    %rsp,%rbp
       86 e0                   xchg   %ah,%al
       5d                      pop    %rbp
       c3                      retq

 (25 copies, 181 calls):
       8b 07                   mov    (%rdi),%eax
       55                      push   %rbp
       48 89 e5                mov    %rsp,%rbp
       0f c8                   bswap  %eax
       5d                      pop    %rbp
       c3                      retq

 (23 copies, 94 calls):
       48 8b 07                mov    (%rdi),%rax
       55                      push   %rbp
       48 89 e5                mov    %rsp,%rbp
       48 0f c8                bswap  %rax
       5d                      pop    %rbp
       c3                      retq

 (2 copies, 11 calls):
       89 f8                   mov    %edi,%eax
       55                      push   %rbp
       c1 ef 08                shr    $0x8,%edi
       c1 e0 08                shl    $0x8,%eax
       09 c7                   or     %eax,%edi
       48 89 e5                mov    %rsp,%rbp
       66 89 3e                mov    %di,(%rsi)

 (8 copies, 43 calls):
       55                      push   %rbp
       0f cf                   bswap  %edi
       89 3e                   mov    %edi,(%rsi)
       48 89 e5                mov    %rsp,%rbp
       5d                      pop    %rbp
       c3                      retq

 (26 copies, 157 calls):
       55                      push   %rbp
       48 0f cf                bswap  %rdi
       48 89 3e                mov    %rdi,(%rsi)
       48 89 e5                mov    %rsp,%rbp
       5d                      pop    %rbp
       c3                      retq

This patch fixes this via s/inline/__always_inline/.

It only affects arches with efficient unaligned access insns, such as x86.
(arched which lack such ops do not include linux/unaligned/access_ok.h)

Code size decrease after the patch is ~8.5k:

    text     data      bss       dec     hex filename
92197848 20826112 36417536 149441496 8e84bd8 vmlinux
92189231 20826144 36417536 149432911 8e82a4f vmlinux6_unaligned_be_after

Signed-off-by: Denys Vlasenko 
Acked-by: Ingo Molnar 
Cc: Thomas Graf 
Cc: Peter Zijlstra 
Cc: David Rientjes 
Cc: Arnd Bergmann 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds

include/linux/unaligned/packed_struct.h: use __packed

2011-01-13T16:03:08+00:00

Cc: Will Newton 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds

include/linux/unaligned: pack the whole struct rather than just the field

2010-12-23T03:43:34+00:00

The current packed struct implementation of unaligned access adds the
packed attribute only to the field within the unaligned struct rather than
to the struct as a whole.  This is not sufficient to enforce proper
behaviour on architectures with a default struct alignment of more than
one byte.

For example, the current implementation of __get_unaligned_cpu16 when
compiled for arm with gcc -O1 -mstructure-size-boundary=32 assumes the
struct is on a 4 byte boundary so performs the load of the 16bit packed
field as if it were on a 4 byte boundary:

__get_unaligned_cpu16:
        ldrh    r0, [r0, #0]
        bx      lr

Moving the packed attribute to the struct rather than the field causes the
proper unaligned access code to be generated:

__get_unaligned_cpu16:
	ldrb	r3, [r0, #0]	@ zero_extendqisi2
	ldrb	r0, [r0, #1]	@ zero_extendqisi2
	orr	r0, r3, r0, asl #8
	bx	lr

Signed-off-by: Will Newton 
Cc: Arnd Bergmann 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds

include/linux/unaligned/{l,b}e_byteshift.h: fix usage for compressed kernels

2009-09-24T14:21:05+00:00

When unaligned accesses are required for uncompressing a kernel (such as
for LZO decompression on ARM in a patch that follows), including
 causes issues as it brings in a lot of things that are
not available in the decompression environment.

linux/kernel.h brings at least:
extern int console_printk[];
extern const char hex_asc[];
which causes errors at link-time as they are not available when
compiling the pre-boot environement. There are also a few others:

  arch/arm/boot/compressed/misc.o: In function `valid_user_regs':
   arch/arm/include/asm/ptrace.h:158: undefined reference to `elf_hwcap'
  arch/arm/boot/compressed/misc.o: In function `console_silent':
   include/linux/kernel.h:292: undefined reference to `console_printk'
  arch/arm/boot/compressed/misc.o: In function `console_verbose':
   include/linux/kernel.h:297: undefined reference to `console_printk'
  arch/arm/boot/compressed/misc.o: In function `pack_hex_byte':
   include/linux/kernel.h:360: undefined reference to `hex_asc'
  arch/arm/boot/compressed/misc.o: In function `hweight_long':
   include/linux/bitops.h:45: undefined reference to `hweight32'
  arch/arm/boot/compressed/misc.o: In function `__cmpxchg_local_generic':
   include/asm-generic/cmpxchg-local.h:21: undefined reference to `wrong_size_cmpxchg'
   include/asm-generic/cmpxchg-local.h:42: undefined reference to `wrong_size_cmpxchg'
  arch/arm/boot/compressed/misc.o: In function `__xchg':
   arch/arm/include/asm/system.h:309: undefined reference to `__bad_xchg'

However, those files apparently use nothing from , all
they need is the declaration of types such as u32 or u64, so
 should be enough

Signed-off-by: Albin Tonnerre 
Cc: Sam Ravnborg 
Cc: Russell King 
Cc: Ingo Molnar 
Cc: Thomas Gleixner 
Cc: "H. Peter Anvin" 
Cc: Phillip Lougher 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds

kernel: add common infrastructure for unaligned access

2008-04-29T15:06:27+00:00

Create a linux/unaligned directory similar in spirit to the linux/byteorder
folder to hold generic implementations collected from various arches.

Currently there are five implementations:
1) packed_struct.h: C-struct based, from asm-generic/unaligned.h
2) le_byteshift.h: Open coded byte-swapping, heavily based on asm-arm
3) be_byteshift.h: Open coded byte-swapping, heavily based on asm-arm
4) memmove.h: taken from multiple implementations in tree
5) access_ok.h: taken from x86 and others, unaligned access is ok.

All of the new implementations checks for sizes not equal to 1,2,4,8
and will fail to link.

API additions:

get_unaligned_{le16|le32|le64|be16|be32|be64}(p) which is meant to replace
code of the form:
le16_to_cpu(get_unaligned((__le16 *)p));

put_unaligned_{le16|le32|le64|be16|be32|be64}(val, pointer) which is meant to
replace code of the form:
put_unaligned(cpu_to_le16(val), (__le16 *)p);

The headers that arches should include from their asm/unaligned.h:

access_ok.h : Wrappers of the byteswapping functions in asm/byteorder

Choose a particular implementation for little-endian access:
le_byteshift.h
le_memmove.h (arch must be LE)
le_struct.h (arch must be LE)

Choose a particular implementation for big-endian access:
be_byteshift.h
be_memmove.h (arch must be BE)
be_struct.h (arch must be BE)

After including as needed from the above, include unaligned/generic.h and
define your arch's get/put_unaligned as (for LE):

Signed-off-by: Harvey Harrison 
Cc: 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds