lzma PackLinuxElf64amd works!

src/stub/src/amd64-linux.elf-entry.asm

@@ -73,7 +73,6 @@ ret_main:
 /* Returns 0 on success; non-zero on failure. */
 decompress:  // (uchar const *src, size_t lsrc, uchar *dst, u32 &ldst, uint method)
 
-  section NRV_COMMON
 /* Arguments according to calling convention */
 #define src  %arg1
 #define lsrc %arg2
@@ -82,6 +81,13 @@ decompress:  // (uchar const *src, size_t lsrc, uchar *dst, u32 &ldst, uint meth
 #define meth %arg5l
 #define methb %arg5b
 
+        push %rbp; push %rbx  // C callable
+        push ldst
+        push dst
+        addq src,lsrc; push lsrc  // &input_eof
+
+  section NRV_COMMON
+
 /* Working registers */
 #define off  %eax  /* XXX: 2GB */
 #define len  %ecx  /* XXX: 2GB */
@@ -89,11 +95,6 @@ decompress:  // (uchar const *src, size_t lsrc, uchar *dst, u32 &ldst, uint meth
 #define bits %ebx
 #define disp %rbp
 
-        push %rbp; push %rbx  // C callable
-        push ldst
-        push dst
-        addq src,lsrc; push lsrc  // &input_eof
-
         movq src,%rsi  // hardware src for movsb, lodsb
         movq dst,%rdi  // hardware dst for movsb
         xor bits,bits  // empty; force refill
@@ -272,7 +273,7 @@ unfold:
         ret
 
 main:
-        int3  # uncomment for debugging
+////    int3  # uncomment for debugging
         pop %rbp  // &decompress
         movl -4-(ret_main - _start)(%rbp),%r15d  // length which precedes stub
         subl $ sz_Ehdr + 2*sz_Phdr + sz_l_info + sz_p_info,%r15d

src/stub/src/arch/amd64/lzma_d.S

@@ -39,53 +39,56 @@ section LZMA_ELF00
 #define meth %arg5l /* %r8  */
 #define methb %arg5b
 
+// ELFMAINX has already done this for us:
+//      pushq %rbp; push %rbx  // C callable
+//      pushq ldst
+//      pushq dst
+//      addq src,lsrc; push lsrc  // &input_eof
+
 #define M_LZMA          14
         cmpb $ M_LZMA,methb; jne not_lzma
-                // C calling convention
+
-        pushq %rbp; movq %rsp,%rbp
+        pushq %rbp; movq %rsp,%rbp  // we use alloca
-        pushq %rbx
 
 //LzmaDecode(  // from lzmaSDK/C/7zip/Compress/LZMA_C/LzmaDecode.h
 //      %arg1= &CLzmaDecoderState,
 //      %arg2= in,  %arg3l= inSize, %arg4= &inSizeProcessed,
-//      %arg5= out, %arg6l= outSize, 8+%esp/ &outSizeProcessed
+//      %arg5= out, %arg6l= outSize, arg7@ 8+%esp/ &outSizeProcessed
 //)
+        movl (ldst),%arg6l  // &outSize  XXX: 4GB
+        movq dst,%arg5  // outp
+        movq lsrc,%arg3  // inSize
+        leaq 2(src),%arg2; pushq %arg2  // in; save @-8(%rbp) for size calc at eof
+
+        movb (src),%al; decl %arg3l  // first byte, replaces LzmaDecodeProperties()
+        movb %al,%cl  // cl= ((lit_context_bits + lit_pos_bits)<<3) | pos_bits
+        andb $7,%al  // al= pos_bits
+        shrb $3,%cl  // cl= lit_context_bits + lit_pos_bits
 
 #define LZMA_BASE_SIZE 1846
 #define LZMA_LIT_SIZE   768
-#define szSizeT 8
+#define szSizeT 4
-
-        movq dst,%arg5  // outp
-        movq lsrc,%arg3  // inSize
-        movq  src,%arg2  // in
-        movq ldst,%r10
-
-        lodsb; decl %arg3l  // first byte, replaces LzmaDecodeProperties()
-        movb %al,%cl  // cl= ((lit_context_bits + lit_pos_bits)<<3) | pos_bits
-        andb $6,%al  // al= pos_bits
-        shrb $3,%cl  // cl= lit_context_bits + lit_pos_bits
 
         movq $-LZMA_LIT_SIZE,%rbx
-        shlq %cl,%rbx
+        shlq %cl,%rbx; movb %al,%cl  // %cl= pos_bits
-// inSizeProcessed, outSizeProcessed, *_bits, CLzmaDecoderState
+// alloca{inSizeProcessed, outSizeProcessed, *_bits, CLzmaDecoderState}
         leaq -(2*szSizeT +4) - 2*LZMA_BASE_SIZE(%rsp,%rbx,2), %rbx
         andq $~0<<6,%rbx  // 64-byte align
 1:
-        pushq $0
+        pushq $0  // clera CLzmaDecoderState on stack
         cmpq %rbx,%rsp
         jne 1b
 
-        pushq %rbx; addq $ szSizeT,%rbx  // &outSizeProcessed
-        movl (%r10),%arg6l  // &outSize  XXX: 4GB
-        movq %rbx,%arg4; addq $ szSizeT,%rbx  // &inSizeProcessed
 
-        movb %al,2(%rbx)  // store pos_bits
+        pushq %rbx  // &outSizeProcessed [arg7]
-        lodsb; decl %arg3l  // second byte, replaces LzmaDecodeProperties()
+        leaq 2*szSizeT(%rbx),%arg1  // &CLzmaDecoderState
-        movb %al,%cl  // cl= (lit_pos_bits<<4) | lit_context_bits
+        movb -1(%arg2),%cl; decl %arg3l  // second byte, replaces LzmaDecodeProperties()
-        andb $0xf,%al; movb %al, (%rbx)  // store lit_context_bits
+                       movb %al,2(%arg1)  // store pos_bits
-        shrb  $4, %cl; movb %cl,1(%rbx)  // store lit_pos_bits
+        movb  %cl,%al  // al= (lit_pos_bits<<4) | lit_context_bits
+        shrb   $4,%cl; movb %cl,1(%arg1)  // store lit_pos_bits
+        andb $0xf,%al; movb %al, (%arg1)  // store lit_context_bits
+        leaq -szSizeT(%arg1),%arg4  // &inSizeProcessed
 
-        movq %rbx,%arg1  // &CLzmaDecoderState
         pushq %rax  // return address slot (dummy CALL)
 
 section LZMA_DEC10
@@ -94,25 +97,12 @@ section LZMA_DEC10
 section LZMA_DEC20
 #include "lzma_d_cf.S"
 
-
-//
-// cleanup
 section LZMA_DEC30
-
+        movq -1*8(%rbp),%rsi  // src [after header]
-        movl -1*szSizeT(%rbx),%eax; addq %rax,%rsi  // inSizeProcessed
+        movq  2*8(%rbp),%rdi  // dst
-        movl -2*szSizeT(%rbx),%eax; addq %rax,%rdi  // outSizeProcessed
+        movl szSizeT(%rbx),%ecx; addq %rcx,%rsi  //  inSizeProcessed
-        xorq %rax,%rax
+        movl        (%rbx),%edx; addq %rdx,%rdi  // outSizeProcessed
-
+        leave  // movl %ebp,%rsp; popq %rbp
-        leaq -256(%rsp),%rcx
-        movq %rbp,%rsp  // restore stack
-1:
-        pushq %rax
-        cmpq %rcx,%rsp
-        jne 1b
-
-        movq -8(%rbp),%rbx
-        xorq %rcx,%rcx
-        leave
 not_lzma:
 
 // vi:ts=8:et