crypto: x86 - make constants readonly, allow linker to merge them

A lot of asm-optimized routines in arch/x86/crypto/ keep its
constants in .data. This is wrong, they should be on .rodata.

Mnay of these constants are the same in different modules.
For example, 128-bit shuffle mask 0x000102030405060708090A0B0C0D0E0F
exists in at least half a dozen places.

There is a way to let linker merge them and use just one copy.
The rules are as follows: mergeable objects of different sizes
should not share sections. You can't put them all in one .rodata
section, they will lose "mergeability".

GCC puts its mergeable constants in ".rodata.cstSIZE" sections,
or ".rodata.cstSIZE.<object_name>" if -fdata-sections is used.
This patch does the same:

	.section .rodata.cst16.SHUF_MASK, "aM", @progbits, 16

It is important that all data in such section consists of
16-byte elements, not larger ones, and there are no implicit
use of one element from another.

When this is not the case, use non-mergeable section:

	.section .rodata[.VAR_NAME], "a", @progbits

This reduces .data by ~15 kbytes:

    text    data     bss     dec      hex filename
11097415 2705840 2630712 16433967  fac32f vmlinux-prev.o
11112095 2690672 2630712 16433479  fac147 vmlinux.o

Merged objects are visible in System.map:

ffffffff81a28810 r POLY
ffffffff81a28810 r POLY
ffffffff81a28820 r TWOONE
ffffffff81a28820 r TWOONE
ffffffff81a28830 r PSHUFFLE_BYTE_FLIP_MASK <- merged regardless of
ffffffff81a28830 r SHUF_MASK   <------------- the name difference
ffffffff81a28830 r SHUF_MASK
ffffffff81a28830 r SHUF_MASK
..
ffffffff81a28d00 r K512 <- merged three identical 640-byte tables
ffffffff81a28d00 r K512
ffffffff81a28d00 r K512

Use of object names in section name suffixes is not strictly necessary,
but might help if someday link stage will use garbage collection
to eliminate unused sections (ld --gc-sections).

Signed-off-by: Denys Vlasenko <[email protected]>
CC: Herbert Xu <[email protected]>
CC: Josh Poimboeuf <[email protected]>
CC: Xiaodong Liu <[email protected]>
CC: Megha Dey <[email protected]>
CC: [email protected]
CC: [email protected]
CC: [email protected]
Signed-off-by: Herbert Xu <[email protected]>

arch/x86/crypto/aesni-intel_asm.S

@@ -46,28 +46,49 @@
 
 #ifdef __x86_64__
 
-.data
+# constants in mergeable sections, linker can reorder and merge
+.section	.rodata.cst16.gf128mul_x_ble_mask, "aM", @progbits, 16
 .align 16
 .Lgf128mul_x_ble_mask:
 	.octa 0x00000000000000010000000000000087
+.section	.rodata.cst16.POLY, "aM", @progbits, 16
+.align 16
 POLY:   .octa 0xC2000000000000000000000000000001
+.section	.rodata.cst16.TWOONE, "aM", @progbits, 16
+.align 16
 TWOONE: .octa 0x00000001000000000000000000000001
 
-# order of these constants should not change.
-# more specifically, ALL_F should follow SHIFT_MASK,
-# and ZERO should follow ALL_F
-
+.section	.rodata.cst16.SHUF_MASK, "aM", @progbits, 16
+.align 16
 SHUF_MASK:  .octa 0x000102030405060708090A0B0C0D0E0F
+.section	.rodata.cst16.MASK1, "aM", @progbits, 16
+.align 16
 MASK1:      .octa 0x0000000000000000ffffffffffffffff
+.section	.rodata.cst16.MASK2, "aM", @progbits, 16
+.align 16
 MASK2:      .octa 0xffffffffffffffff0000000000000000
-SHIFT_MASK: .octa 0x0f0e0d0c0b0a09080706050403020100
-ALL_F:      .octa 0xffffffffffffffffffffffffffffffff
-ZERO:       .octa 0x00000000000000000000000000000000
+.section	.rodata.cst16.ONE, "aM", @progbits, 16
+.align 16
 ONE:        .octa 0x00000000000000000000000000000001
+.section	.rodata.cst16.F_MIN_MASK, "aM", @progbits, 16
+.align 16
 F_MIN_MASK: .octa 0xf1f2f3f4f5f6f7f8f9fafbfcfdfeff0
+.section	.rodata.cst16.dec, "aM", @progbits, 16
+.align 16
 dec:        .octa 0x1
+.section	.rodata.cst16.enc, "aM", @progbits, 16
+.align 16
 enc:        .octa 0x2
 
+# order of these constants should not change.
+# more specifically, ALL_F should follow SHIFT_MASK,
+# and zero should follow ALL_F
+.section	.rodata, "a", @progbits
+.align 16
+SHIFT_MASK: .octa 0x0f0e0d0c0b0a09080706050403020100
+ALL_F:      .octa 0xffffffffffffffffffffffffffffffff
+            .octa 0x00000000000000000000000000000000
+
 
 .text
 

arch/x86/crypto/aesni-intel_avx-x86_64.S

@@ -122,23 +122,39 @@
 #include <linux/linkage.h>
 #include <asm/inst.h>
 
-.data
+# constants in mergeable sections, linker can reorder and merge
+.section	.rodata.cst16.POLY, "aM", @progbits, 16
 .align 16
-
 POLY:            .octa     0xC2000000000000000000000000000001
+
+.section	.rodata.cst16.POLY2, "aM", @progbits, 16
+.align 16
 POLY2:           .octa     0xC20000000000000000000001C2000000
-TWOONE:          .octa     0x00000001000000000000000000000001
 
-# order of these constants should not change.
-# more specifically, ALL_F should follow SHIFT_MASK, and ZERO should follow ALL_F
+.section	.rodata.cst16.TWOONE, "aM", @progbits, 16
+.align 16
+TWOONE:          .octa     0x00000001000000000000000000000001
 
+.section	.rodata.cst16.SHUF_MASK, "aM", @progbits, 16
+.align 16
 SHUF_MASK:       .octa     0x000102030405060708090A0B0C0D0E0F
-SHIFT_MASK:      .octa     0x0f0e0d0c0b0a09080706050403020100
-ALL_F:           .octa     0xffffffffffffffffffffffffffffffff
-ZERO:            .octa     0x00000000000000000000000000000000
+
+.section	.rodata.cst16.ONE, "aM", @progbits, 16
+.align 16
 ONE:             .octa     0x00000000000000000000000000000001
+
+.section	.rodata.cst16.ONEf, "aM", @progbits, 16
+.align 16
 ONEf:            .octa     0x01000000000000000000000000000000
 
+# order of these constants should not change.
+# more specifically, ALL_F should follow SHIFT_MASK, and zero should follow ALL_F
+.section	.rodata, "a", @progbits
+.align 16
+SHIFT_MASK:      .octa     0x0f0e0d0c0b0a09080706050403020100
+ALL_F:           .octa     0xffffffffffffffffffffffffffffffff
+                 .octa     0x00000000000000000000000000000000
+
 .text
 
 

arch/x86/crypto/camellia-aesni-avx-asm_64.S

@@ -571,7 +571,9 @@ ENDPROC(roundsm16_x4_x5_x6_x7_x0_x1_x2_x3_y4_y5_y6_y7_y0_y1_y2_y3_ab)
 	vmovdqu y6, 14 * 16(rio); \
 	vmovdqu y7, 15 * 16(rio);
 
-.data
+
+/* NB: section is mergeable, all elements must be aligned 16-byte blocks */
+.section	.rodata.cst16, "aM", @progbits, 16
 .align 16
 
 #define SHUFB_BYTES(idx) \
@@ -711,6 +713,7 @@ ENDPROC(roundsm16_x4_x5_x6_x7_x0_x1_x2_x3_y4_y5_y6_y7_y0_y1_y2_y3_ab)
 	.byte 0x08, 0x05, 0x02, 0x0f, 0x0c, 0x09, 0x06, 0x03
 
 /* 4-bit mask */
+.section	.rodata.cst4.L0f0f0f0f, "aM", @progbits, 4
 .align 4
 .L0f0f0f0f:
 	.long 0x0f0f0f0f

arch/x86/crypto/camellia-aesni-avx2-asm_64.S

@@ -610,20 +610,25 @@ ENDPROC(roundsm32_x4_x5_x6_x7_x0_x1_x2_x3_y4_y5_y6_y7_y0_y1_y2_y3_ab)
 	vmovdqu y6, 14 * 32(rio); \
 	vmovdqu y7, 15 * 32(rio);
 
-.data
-.align 32
 
+.section	.rodata.cst32.shufb_16x16b, "aM", @progbits, 32
+.align 32
 #define SHUFB_BYTES(idx) \
 	0 + (idx), 4 + (idx), 8 + (idx), 12 + (idx)
-
 .Lshufb_16x16b:
 	.byte SHUFB_BYTES(0), SHUFB_BYTES(1), SHUFB_BYTES(2), SHUFB_BYTES(3)
 	.byte SHUFB_BYTES(0), SHUFB_BYTES(1), SHUFB_BYTES(2), SHUFB_BYTES(3)
 
+.section	.rodata.cst32.pack_bswap, "aM", @progbits, 32
+.align 32
 .Lpack_bswap:
 	.long 0x00010203, 0x04050607, 0x80808080, 0x80808080
 	.long 0x00010203, 0x04050607, 0x80808080, 0x80808080
 
+/* NB: section is mergeable, all elements must be aligned 16-byte blocks */
+.section	.rodata.cst16, "aM", @progbits, 16
+.align 16
+
 /* For CTR-mode IV byteswap */
 .Lbswap128_mask:
 	.byte 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0
@@ -750,6 +755,7 @@ ENDPROC(roundsm32_x4_x5_x6_x7_x0_x1_x2_x3_y4_y5_y6_y7_y0_y1_y2_y3_ab)
 	.byte 0x00, 0x0d, 0x0a, 0x07, 0x04, 0x01, 0x0e, 0x0b
 	.byte 0x08, 0x05, 0x02, 0x0f, 0x0c, 0x09, 0x06, 0x03
 
+.section	.rodata.cst4.L0f0f0f0f, "aM", @progbits, 4
 .align 4
 /* 4-bit mask */
 .L0f0f0f0f:

arch/x86/crypto/cast5-avx-x86_64-asm_64.S

@@ -195,19 +195,29 @@
 	vpshufb rmask,	x0, x0;           \
 	vpshufb rmask,	x1, x1;
 
-.data
-
+.section	.rodata.cst16.bswap_mask, "aM", @progbits, 16
 .align 16
 .Lbswap_mask:
 	.byte 3, 2, 1, 0, 7, 6, 5, 4, 11, 10, 9, 8, 15, 14, 13, 12
+.section	.rodata.cst16.bswap128_mask, "aM", @progbits, 16
+.align 16
 .Lbswap128_mask:
 	.byte 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0
+.section	.rodata.cst16.bswap_iv_mask, "aM", @progbits, 16
+.align 16
 .Lbswap_iv_mask:
 	.byte 7, 6, 5, 4, 3, 2, 1, 0, 7, 6, 5, 4, 3, 2, 1, 0
+
+.section	.rodata.cst4.16_mask, "aM", @progbits, 4
+.align 4
 .L16_mask:
 	.byte 16, 16, 16, 16
+.section	.rodata.cst4.32_mask, "aM", @progbits, 4
+.align 4
 .L32_mask:
 	.byte 32, 0, 0, 0
+.section	.rodata.cst4.first_mask, "aM", @progbits, 4
+.align 4
 .Lfirst_mask:
 	.byte 0x1f, 0, 0, 0
 

arch/x86/crypto/cast6-avx-x86_64-asm_64.S

@@ -225,8 +225,7 @@
 	vpshufb rmask,		x2, x2;       \
 	vpshufb rmask,		x3, x3;
 
-.data
-
+.section	.rodata.cst16, "aM", @progbits, 16
 .align 16
 .Lxts_gf128mul_and_shl1_mask:
 	.byte 0x87, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0
@@ -244,10 +243,19 @@
 	.byte 12, 13, 14, 15, 8, 9, 10, 11, 7, 6, 5, 4, 3, 2, 1, 0
 .Lrkr_dec_QBAR_QBAR_QBAR_QBAR:
 	.byte 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0
+
+.section	.rodata.cst4.L16_mask, "aM", @progbits, 4
+.align 4
 .L16_mask:
 	.byte 16, 16, 16, 16
+
+.section	.rodata.cst4.L32_mask, "aM", @progbits, 4
+.align 4
 .L32_mask:
 	.byte 32, 0, 0, 0
+
+.section	.rodata.cst4.first_mask, "aM", @progbits, 4
+.align 4
 .Lfirst_mask:
 	.byte 0x1f, 0, 0, 0
 

arch/x86/crypto/chacha20-avx2-x86_64.S

@@ -11,13 +11,18 @@
 
 #include <linux/linkage.h>
 
-.data
+.section	.rodata.cst32.ROT8, "aM", @progbits, 32
 .align 32
-
 ROT8:	.octa 0x0e0d0c0f0a09080b0605040702010003
 	.octa 0x0e0d0c0f0a09080b0605040702010003
+
+.section	.rodata.cst32.ROT16, "aM", @progbits, 32
+.align 32
 ROT16:	.octa 0x0d0c0f0e09080b0a0504070601000302
 	.octa 0x0d0c0f0e09080b0a0504070601000302
+
+.section	.rodata.cst32.CTRINC, "aM", @progbits, 32
+.align 32
 CTRINC:	.octa 0x00000003000000020000000100000000
 	.octa 0x00000007000000060000000500000004
 

arch/x86/crypto/chacha20-ssse3-x86_64.S

@@ -11,11 +11,14 @@
 
 #include <linux/linkage.h>
 
-.data
+.section	.rodata.cst16.ROT8, "aM", @progbits, 16
 .align 16
-
 ROT8:	.octa 0x0e0d0c0f0a09080b0605040702010003
+.section	.rodata.cst16.ROT16, "aM", @progbits, 16
+.align 16
 ROT16:	.octa 0x0d0c0f0e09080b0a0504070601000302
+.section	.rodata.cst16.CTRINC, "aM", @progbits, 16
+.align 16
 CTRINC:	.octa 0x00000003000000020000000100000000
 
 .text

arch/x86/crypto/crct10dif-pcl-asm_64.S

@@ -554,12 +554,11 @@ _only_less_than_2:
 
 ENDPROC(crc_t10dif_pcl)
 
-.data
-
+.section	.rodata, "a", @progbits
+.align 16
 # precomputed constants
 # these constants are precomputed from the poly:
 # 0x8bb70000 (0x8bb7 scaled to 32 bits)
-.align 16
 # Q = 0x18BB70000
 # rk1 = 2^(32*3) mod Q << 32
 # rk2 = 2^(32*5) mod Q << 32
@@ -613,14 +612,23 @@ rk20:
 
 
 
+.section	.rodata.cst16.mask1, "aM", @progbits, 16
+.align 16
 mask1:
 .octa 0x80808080808080808080808080808080
+
+.section	.rodata.cst16.mask2, "aM", @progbits, 16
+.align 16
 mask2:
 .octa 0x00000000FFFFFFFFFFFFFFFFFFFFFFFF
 
+.section	.rodata.cst16.SHUF_MASK, "aM", @progbits, 16
+.align 16
 SHUF_MASK:
 .octa 0x000102030405060708090A0B0C0D0E0F
 
+.section	.rodata.cst32.pshufb_shf_table, "aM", @progbits, 32
+.align 32
 pshufb_shf_table:
 # use these values for shift constants for the pshufb instruction
 # different alignments result in values as shown:

arch/x86/crypto/des3_ede-asm_64.S

@@ -537,7 +537,7 @@ ENTRY(des3_ede_x86_64_crypt_blk_3way)
 	ret;
 ENDPROC(des3_ede_x86_64_crypt_blk_3way)
 
-.data
+.section	.rodata, "a", @progbits
 .align 16
 .L_s1:
 	.quad 0x0010100001010400, 0x0000000000000000

arch/x86/crypto/ghash-clmulni-intel_asm.S

@@ -20,8 +20,7 @@
 #include <asm/inst.h>
 #include <asm/frame.h>
 
-.data
-
+.section	.rodata.cst16.bswap_mask, "aM", @progbits, 16
 .align 16
 .Lbswap_mask:
 	.octa 0x000102030405060708090a0b0c0d0e0f

arch/x86/crypto/poly1305-avx2-x86_64.S

@@ -11,11 +11,13 @@
 
 #include <linux/linkage.h>
 
-.data
+.section	.rodata.cst32.ANMASK, "aM", @progbits, 32
 .align 32
-
 ANMASK:	.octa 0x0000000003ffffff0000000003ffffff
 	.octa 0x0000000003ffffff0000000003ffffff
+
+.section	.rodata.cst32.ORMASK, "aM", @progbits, 32
+.align 32
 ORMASK:	.octa 0x00000000010000000000000001000000
 	.octa 0x00000000010000000000000001000000
 

arch/x86/crypto/poly1305-sse2-x86_64.S

@@ -11,10 +11,12 @@
 
 #include <linux/linkage.h>
 
-.data
+.section	.rodata.cst16.ANMASK, "aM", @progbits, 16
 .align 16
-
 ANMASK:	.octa 0x0000000003ffffff0000000003ffffff
+
+.section	.rodata.cst16.ORMASK, "aM", @progbits, 16
+.align 16
 ORMASK:	.octa 0x00000000010000000000000001000000
 
 .text

arch/x86/crypto/serpent-avx-x86_64-asm_64.S

@@ -29,11 +29,12 @@
 
 .file "serpent-avx-x86_64-asm_64.S"
 
-.data
+.section	.rodata.cst16.bswap128_mask, "aM", @progbits, 16
 .align 16
-
 .Lbswap128_mask:
 	.byte 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0
+.section	.rodata.cst16.xts_gf128mul_and_shl1_mask, "aM", @progbits, 16
+.align 16
 .Lxts_gf128mul_and_shl1_mask:
 	.byte 0x87, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0
 

arch/x86/crypto/serpent-avx2-asm_64.S

@@ -20,13 +20,18 @@
 
 .file "serpent-avx2-asm_64.S"
 
-.data
+.section	.rodata.cst16.bswap128_mask, "aM", @progbits, 16
 .align 16
-
 .Lbswap128_mask:
 	.byte 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0
+
+.section	.rodata.cst16.xts_gf128mul_and_shl1_mask_0, "aM", @progbits, 16
+.align 16
 .Lxts_gf128mul_and_shl1_mask_0:
 	.byte 0x87, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0
+
+.section	.rodata.cst16.xts_gf128mul_and_shl1_mask_1, "aM", @progbits, 16
+.align 16
 .Lxts_gf128mul_and_shl1_mask_1:
 	.byte 0x0e, 1, 0, 0, 0, 0, 0, 0, 2, 0, 0, 0, 0, 0, 0, 0