ArmPkg: Fix InternalMemCopyMem()

This function crashed when regions overalapped. The condition to
optimize the copy of overlapped regions was not correct.

ArmPkg: Add comments to InternalMemCopyMem()

These comments explain the flow of this assembly function.

git-svn-id: https://edk2.svn.sourceforge.net/svnroot/edk2/trunk/edk2@11486 6f19259b-4bc3-4df7-8a09-765794883524

diff --git a/ArmPkg/Library/BaseMemoryLibStm/Arm/CopyMem.S b/ArmPkg/Library/BaseMemoryLibStm/Arm/CopyMem.S
index 1298f94e8dd3695c7a0b26e6099aed1bced7c2b4..4e0122bdecbc9d2282941712ee7f4050c2ee63db 100755 (executable)
--- a/ArmPkg/Library/BaseMemoryLibStm/Arm/CopyMem.S
+++ b/ArmPkg/Library/BaseMemoryLibStm/Arm/CopyMem.S
@@ -43,70 +43,129 @@ GCC_ASM_EXPORT(InternalMemCopyMem)
 \r
 ASM_PFX(InternalMemCopyMem):\r
   stmfd  sp!, {r4-r11, lr}\r
-  tst  r0, #3\r
+  // Save the input parameters in extra registers (r11 = destination, r14 = source, r12 = length)\r
   mov  r11, r0\r
   mov  r10, r0\r
-  mov  ip, r2\r
-  mov  lr, r1\r
+  mov  r12, r2\r
+  mov  r14, r1\r
+  \r
+memcopy_check_overlapped:\r
+  cmp  r11, r1\r
+  // If (dest < source)\r
+  bcc  memcopy_check_optim_default\r
+  // If (dest <= source). But with the previous condition ->  If (dest == source)\r
+  bls  memcopy_end\r
+\r
+  // If (source + length < dest)\r
+  rsb  r3, r1, r11\r
+  cmp  r12, r3\r
+  bcc  memcopy_check_optim_default\r
+\r
+  // If (length == 0)  \r
+  cmp  r12, #0\r
+  beq  memcopy_end\r
+  \r
+  b     memcopy_check_optim_overlap\r
+\r
+memcopy_check_optim_default:\r
+  // Check if we can use an optimized path ((length >= 32) && destination word-aligned && source word-aligned) for the memcopy (optimized path if r0 == 1)\r
+  tst  r0, #0xF\r
   movne  r0, #0\r
-  bne  L4\r
-  tst  r1, #3\r
+  bne   memcopy_default\r
+  tst  r1, #0xF\r
   movne  r3, #0\r
   moveq  r3, #1\r
   cmp  r2, #31\r
   movls  r0, #0\r
   andhi  r0, r3, #1\r
-L4:\r
-  cmp  r11, r1\r
-  bcc  L26\r
-  bls  L7\r
-  rsb  r3, r1, r11\r
-  cmp  ip, r3\r
-  bcc  L26\r
-  cmp  ip, #0\r
-  beq  L7\r
-  add  r10, r11, ip\r
-  add  lr, ip, r1\r
-  b  L16\r
-L29:\r
-  sub  ip, ip, #8\r
-  cmp  ip, #7\r
-  ldrd  r2, [lr, #-8]!\r
+  b     memcopy_default\r
+    \r
+memcopy_check_optim_overlap:\r
+  // r10 = dest_end, r14 = source_end\r
+  add  r10, r11, r12\r
+  add  r14, r12, r1\r
+\r
+  // Are we in the optimized case ((length >= 32) && dest_end word-aligned && source_end word-aligned)\r
+  cmp  r2, #31\r
   movls  r0, #0\r
-  cmp  ip, #0\r
-  strd  r2, [r10, #-8]!\r
-  beq  L7\r
-L16:\r
-  cmp  r0, #0\r
-  bne  L29\r
-  sub  r3, lr, #1\r
-  sub  ip, ip, #1\r
+  movhi  r0, #1\r
+  tst  r10, #0xF\r
+  movne  r0, #0\r
+  tst  r14, #0xF\r
+  movne  r0, #0\r
+  b  memcopy_overlapped\r
+  \r
+memcopy_overlapped_non_optim:\r
+  // We read 1 byte from the end of the source buffer\r
+  sub  r3, r14, #1\r
+  sub  r12, r12, #1\r
   ldrb  r3, [r3, #0]  \r
   sub  r2, r10, #1\r
-  cmp  ip, #0\r
+  cmp  r12, #0\r
+  // We write 1 byte at the end of the dest buffer\r
   sub  r10, r10, #1\r
-  sub  lr, lr, #1\r
+  sub  r14, r14, #1\r
   strb  r3, [r2, #0]\r
-  bne  L16\r
-  b   L7\r
-L11:\r
-  ldrb  r3, [lr], #1  \r
-  sub  ip, ip, #1\r
+  bne  memcopy_overlapped_non_optim\r
+  b   memcopy_end\r
+\r
+// r10 = dest_end, r14 = source_end\r
+memcopy_overlapped:\r
+  // Are we in the optimized case ?\r
+  cmp  r0, #0\r
+  beq  memcopy_overlapped_non_optim\r
+  \r
+  // Optimized Overlapped - Read 32 bytes\r
+  sub  r14, r14, #32\r
+  sub  r12, r12, #32\r
+  cmp  r12, #31\r
+  ldmia  r14, {r2-r9}\r
+  \r
+  // If length is less than 32 then disable optim\r
+  movls  r0, #0\r
+  \r
+  cmp  r12, #0\r
+  \r
+  // Optimized Overlapped - Write 32 bytes  \r
+  sub  r10, r10, #32\r
+  stmia  r10, {r2-r9}\r
+  \r
+  // while (length != 0)\r
+  bne  memcopy_overlapped\r
+  b   memcopy_end\r
+  \r
+memcopy_default_non_optim:\r
+  // Byte copy\r
+  ldrb  r3, [r14], #1  \r
+  sub  r12, r12, #1\r
   strb  r3, [r10], #1\r
-L26:\r
-  cmp  ip, #0\r
-  beq  L7\r
-L30:\r
+  \r
+memcopy_default:\r
+  cmp  r12, #0\r
+  beq  memcopy_end\r
+  \r
+// r10 = dest, r14 = source\r
+memcopy_default_loop:\r
   cmp  r0, #0\r
-  beq  L11\r
-  sub  ip, ip, #32\r
-  cmp  ip, #31\r
-  ldmia  lr!, {r2-r9}\r
+  beq  memcopy_default_non_optim\r
+  \r
+  // Optimized memcopy - Read 32 Bytes\r
+  sub  r12, r12, #32\r
+  cmp  r12, #31\r
+  ldmia  r14!, {r2-r9}\r
+  \r
+  // If length is less than 32 then disable optim\r
   movls  r0, #0\r
-  cmp  ip, #0\r
+  \r
+  cmp  r12, #0\r
+  \r
+  // Optimized memcopy - Write 32 Bytes\r
   stmia  r10!, {r2-r9}\r
-  bne  L30\r
-L7:\r
+\r
+  // while (length != 0)\r
+  bne  memcopy_default_loop\r
+  \r
+memcopy_end:\r
   mov  r0, r11\r
   ldmfd  sp!, {r4-r11, pc}\r
   \r

diff --git a/ArmPkg/Library/BaseMemoryLibStm/Arm/CopyMem.asm b/ArmPkg/Library/BaseMemoryLibStm/Arm/CopyMem.asm
index 1a5e18e120b4b7f8a0bd6581edb580e376479e79..ca8d06a5502109cb0693ecfbeb741d14b4d532b6 100755 (executable)
--- a/ArmPkg/Library/BaseMemoryLibStm/Arm/CopyMem.asm
+++ b/ArmPkg/Library/BaseMemoryLibStm/Arm/CopyMem.asm
@@ -43,70 +43,129 @@ InternalMemCopyMem (
 \r
 InternalMemCopyMem\r
   stmfd  sp!, {r4-r11, lr}
-  tst  r0, #3
+  // Save the input parameters in extra registers (r11 = destination, r14 = source, r12 = length)
   mov  r11, r0
   mov  r10, r0
-  mov  ip, r2
-  mov  lr, r1
+  mov  r12, r2
+  mov  r14, r1
+  \r
+memcopy_check_overlapped\r
+  cmp  r11, r1\r
+  // If (dest < source)\r
+  bcc  memcopy_check_optim_default\r
+  // If (dest <= source). But with the previous condition ->  If (dest == source)\r
+  bls  memcopy_end\r
+\r
+  // If (source + length < dest)\r
+  rsb  r3, r1, r11\r
+  cmp  r12, r3\r
+  bcc  memcopy_check_optim_default\r
+\r
+  // If (length == 0)  \r
+  cmp  r12, #0\r
+  beq  memcopy_end\r
+  \r
+  b     memcopy_check_optim_overlap\r
+\r
+memcopy_check_optim_default\r
+  // Check if we can use an optimized path ((length >= 32) && destination word-aligned && source word-aligned) for the memcopy (optimized path if r0 == 1)\r
+  tst  r0, #0xF\r
   movne  r0, #0
-  bne  L4
-  tst  r1, #3
+  bne   memcopy_default
+  tst  r1, #0xF
   movne  r3, #0
   moveq  r3, #1
   cmp  r2, #31
   movls  r0, #0
   andhi  r0, r3, #1
-L4
-  cmp  r11, r1
-  bcc  L26
-  bls  L7
-  rsb  r3, r1, r11
-  cmp  ip, r3
-  bcc  L26
-  cmp  ip, #0
-  beq  L7
-  add  r10, r11, ip
-  add  lr, ip, r1
-  b  L16
-L29
-  sub  ip, ip, #8
-  cmp  ip, #7
-  ldrd  r2, [lr, #-8]!
+  b     memcopy_default
+    
+memcopy_check_optim_overlap
+  // r10 = dest_end, r14 = source_end
+  add  r10, r11, r12
+  add  r14, r12, r1
+
+  // Are we in the optimized case ((length >= 32) && dest_end word-aligned && source_end word-aligned)
+  cmp  r2, #31
   movls  r0, #0
-  cmp  ip, #0
-  strd  r2, [r10, #-8]!
-  beq  L7
-L16
-  cmp  r0, #0
-  bne  L29
-  sub  r3, lr, #1
-  sub  ip, ip, #1
+  movhi  r0, #1
+  tst  r10, #0xF
+  movne  r0, #0
+  tst  r14, #0xF
+  movne  r0, #0
+  b  memcopy_overlapped\r
+  
+memcopy_overlapped_non_optim\r
+  // We read 1 byte from the end of the source buffer\r
+  sub  r3, r14, #1
+  sub  r12, r12, #1
   ldrb  r3, [r3, #0]  
   sub  r2, r10, #1
-  cmp  ip, #0
+  cmp  r12, #0
+  // We write 1 byte at the end of the dest buffer\r
   sub  r10, r10, #1
-  sub  lr, lr, #1
+  sub  r14, r14, #1
   strb  r3, [r2, #0]
-  bne  L16
-  b   L7
-L11
-  ldrb  r3, [lr], #1  
-  sub  ip, ip, #1
+  bne  memcopy_overlapped_non_optim\r
+  b   memcopy_end\r
+\r
+// r10 = dest_end, r14 = source_end\r
+memcopy_overlapped\r
+  // Are we in the optimized case ?\r
+  cmp  r0, #0\r
+  beq  memcopy_overlapped_non_optim\r
+  
+  // Optimized Overlapped - Read 32 bytes\r
+  sub  r14, r14, #32\r
+  sub  r12, r12, #32\r
+  cmp  r12, #31\r
+  ldmia  r14, {r2-r9}\r
+  
+  // If length is less than 32 then disable optim\r
+  movls  r0, #0\r
+  
+  cmp  r12, #0\r
+  \r
+  // Optimized Overlapped - Write 32 bytes  \r
+  sub  r10, r10, #32\r
+  stmia  r10, {r2-r9}\r
+  \r
+  // while (length != 0)\r
+  bne  memcopy_overlapped\r
+  b   memcopy_end\r
+  \r
+memcopy_default_non_optim\r
+  // Byte copy\r
+  ldrb  r3, [r14], #1  
+  sub  r12, r12, #1
   strb  r3, [r10], #1
-L26
-  cmp  ip, #0
-  beq  L7
-L30
+  
+memcopy_default\r
+  cmp  r12, #0
+  beq  memcopy_end
+  
+// r10 = dest, r14 = source\r
+memcopy_default_loop\r
   cmp  r0, #0
-  beq  L11
-  sub  ip, ip, #32
-  cmp  ip, #31
-  ldmia  lr!, {r2-r9}
+  beq  memcopy_default_non_optim\r
+  
+  // Optimized memcopy - Read 32 Bytes\r
+  sub  r12, r12, #32
+  cmp  r12, #31
+  ldmia  r14!, {r2-r9}
+  \r
+  // If length is less than 32 then disable optim\r
   movls  r0, #0
-  cmp  ip, #0
+  \r
+  cmp  r12, #0
+  \r
+  // Optimized memcopy - Write 32 Bytes\r
   stmia  r10!, {r2-r9}
-  bne  L30
-L7
+
+  // while (length != 0)\r
+  bne  memcopy_default_loop\r
+  
+memcopy_end\r
   mov  r0, r11
   ldmfd  sp!, {r4-r11, pc}\r
   \r