Stand-alone tracing for ARM stub; "#define DEBUG 1" then re-compile.

https://sourceforge.net/tracker/?func=detail&atid=102331&aid=3105926&group_id=2331

src/stub/src/arm-linux.elf-entry.S

@@ -61,6 +61,10 @@ __NR_mmap64 = 0xc0 + __NR_SYSCALL_BASE
 __ARM_NR_BASE  = 0xf0000 + __NR_SYSCALL_BASE
 __ARM_NR_cacheflush =  2 + __ARM_NR_BASE
 
+#ifndef DEBUG  /*{*/
+#define DEBUG 0
+#endif  /*}*/
+
         //.long sz_pack2  // placed there by ::pack3()
   section ELFMAINX
 start_params:
@@ -74,6 +78,16 @@ _start: .globl _start
    the brk() of the _un_compressed program.  The address is pre-calculated
    calculated by PackLinuxElf32arm::addLinkerSymbols().
 */
+
+#if DEBUG  /*{*/
+#define TRACE_REGS r0-r12,r14,r15
+// sp (r13) is not included because the write-back might cause UNDEFINED behavior
+// if the write-back register is not first or last.  The actual value of sp
+// usually does not matter.  Just remember that lr (r14) and pc (r15) are stored
+// one word closer to the stack pointer because r13 has been omitted.
+
+        stmdb sp!,{TRACE_REGS}; bl trace1
+#endif  /*}*/
         adr r12,start_params -4  // &sz_pack2
         ldmia r12,{r1,r2, r10,r11}  // r1= sz_pack2; r2= ADRM; r10= LENF; r11= CPR0;
         add r10,r10,r12  // end_decompress
@@ -96,6 +110,9 @@ D_sz_unc=2*4  // stack displacement to sz_unc
         swi 0
 #else  /*}{*/
         swi __NR_mmap64
+#endif  /*}*/
+#if DEBUG  /*{*/
+        stmdb sp!,{TRACE_REGS}; bl trace2
 #endif  /*}*/
         cmn r0,#4096
         bcs msg_SELinux
@@ -133,9 +150,73 @@ D_stm1=2*4
         mov  r2,lr  // dst
         str  r3,[sp,#D_stm1 + D_sz_unc]  // sz_unc; lzma needs for EOF
         add  r3, sp,#D_stm1 + D_sz_unc   // &sz_unc
+#if DEBUG  /*{*/
+        stmdb sp!,{TRACE_REGS}; bl trace3
+#endif  /*}*/
         mov lr,pc; mov pc,r5  // decompress folded code [opcode 'bx' not in v4a]
         ldmia sp!,{r1, pc}  // discard 5th param; goto unfolded
 
+#if DEBUG  /*{*/
+trace3:
+        add r0,r0,#1
+trace2:
+        add r0,r0,#1
+trace1:
+        add r0,r0,#1
+trace0:
+
+TRACE_BUFLEN=512
+trace_print:
+        str lr,[sp,#(-1+ 15)*4]  @ return pc; [remember: sp is not stored]
+        mov r4,sp  @ &saved_r0
+        ldr r1,[sp]  @ saved r0
+        sub sp,sp,#TRACE_BUFLEN
+        mov r2,sp  @ output string
+
+        sub r0,r0,r1; mov r1,#'\n'; bl trace_hex  @ which trace
+        mov r1,#'>';  strb r1,[r2],#1
+
+        mov r5,#3  @ rows to print
+L600:  @ each row
+        sub r0,r4,#TRACE_BUFLEN
+        sub r0,r0,sp
+        mov r0,r0,lsr #2; mov r1,#'\n'; bl trace_hex  @ which block of 8
+
+        mov r6,#8  @ words per row
+L610:  @ each word
+        ldr r0,[r4],#4; mov r1,#' '; bl trace_hex  @ next word
+        subs r6,r6,#1; bgt L610
+
+        subs r5,r5,#1; bgt L600
+
+        mov r0,#'\n'; strb r0,[r2],#1
+        sub r2,r2,sp  @ count
+        mov r1,sp  @ buf
+        mov r0,#2  @ FD_STDERR
+#if defined(ARMEL_EABI4)  /*{*/
+        mov r7,#__NR_write
+        swi 0
+#else  /*}{*/
+        swi __NR_write
+#endif  /*}*/
+        add sp,sp,#TRACE_BUFLEN
+        ldmia sp!,{TRACE_REGS}
+
+trace_hex:  // In: r0=val, r1=punctuation before, r2=ptr; Uses: r3, ip
+        strb r1,[r2],#1  @ punctuation
+        mov r3,#4*(8 -1)  @ shift count
+        adr ip,hex
+L620:
+        mov r1,r0,lsr r3
+        and r1,r1,#0xf
+        ldrb r1,[ip, r1]
+        strb r1,[r2],#1
+        subs r3,r3,#4; bge L620
+        ret
+hex:
+        .ascii "0123456789abcdef"
+#endif  /*}*/
+
 f_decompress:
 #define LINUX_ARM_CACHEFLUSH 1
 

src/stub/src/arm-linux.elf-fold.S

@@ -119,7 +119,7 @@ top_unf:
 tst_unf:
         cmp len,#0
         bne top_unf
-        mov pc,lr
+        ret
 
         .unreq ptr
         .unreq len
@@ -179,58 +179,95 @@ mmap:
         ldr r5,[sp],#4
         ret
 
-#if 0  /*{ "gcc-3.4 -fPIC" botches string constants when code moves! */
+
+#ifndef DEBUG  /*{*/
+#define DEBUG 0
+#endif  /*}*/
+
+#if DEBUG  /*{ "gcc-3.4 -fPIC" botches string constants when code moves! */
+
+div10: .globl div10
+	mov ip,r0  @ extra copy used at end
+	sub r1,r1,r1  @ hi
+
+	mov r2,r0  @ copy lo
+	adds r0,r0,r0,lsl #3   @ 9*lo
+	adc  r1,r1,r1,lsl #3   @ 9*hi + C
+	add  r1,r1,r2,lsr #(32 - 3)  @ bits shifted from lo to hi
+
+	mov r2,r0  @ copy lo
+	adds r0,r0,r0,lsl #4
+	adc  r1,r1,r1,lsl #4
+	add  r1,r1,r2,lsr #(32 - 4)  @ * 0x99
+
+	mov r2,r0  @ copy lo
+	adds r0,r0,r0,lsl #8
+	adc  r1,r1,r1,lsl #8
+	add  r1,r1,r2,lsr #(32 - 8)  @ * 0x9999
+
+	mov r2,r0  @ copy lo
+	adds r0,r0,r0,lsl #16
+	adc  r1,r1,r1,lsl #16
+	add  r1,r1,r2,lsr #(32 - 16)  @ * 0x99999999
+
+	subs r0,r0,ip,lsl #(32 - 1)  @ - * 0x80000000
+	sbc  r1,r1,ip,lsr #1  @ * 0x19999999
+
+	adds r0,r0,r0
+	adc  r0,r1,#0  @ round and move to result
+	ret
+
         .globl STR_0x
 STR_0x:
-        mov r0,pc; mov pc,lr
+        mov r0,pc; ret
         .asciz "0x"
         .balign 4
 
         .globl STR_hex
 STR_hex:
-        mov r0,pc; mov pc,lr
+        mov r0,pc; ret
         .asciz "0123456789abcdef"
         .balign 4
 
         .globl STR_xread
 STR_xread:
-        mov r0,pc; mov pc,lr
+        mov r0,pc; ret
         .asciz "xread %p(%x %p) %p %x\n"
         .balign 4
 
         .globl STR_unpackExtent
 STR_unpackExtent:
-        mov r0,pc; mov pc,lr
+        mov r0,pc; ret
         .asciz "unpackExtent in=%p(%x %p)  out=%p(%x %p)  %p %p\n"
         .balign 4
 
         .globl STR_make_hatch_arm
 STR_make_hatch_arm:
-        mov r0,pc; mov pc,lr
+        mov r0,pc; ret
         .asciz "make_hatch_arm %p %x\n"
         .balign 4
 
         .globl STR_auxv_up
 STR_auxv_up:
-        mov r0,pc; mov pc,lr
+        mov r0,pc; ret
         .asciz "auxv_up  %p %x %x\n"
         .balign 4
 
         .globl STR_xfind_pages
 STR_xfind_pages:
-        mov r0,pc; mov pc,lr
+        mov r0,pc; ret
         .asciz "xfind_pages  %x  %p  %d  %p\n"
         .balign 4
 
         .globl STR_do_xmap
 STR_do_xmap:
-        mov r0,pc; mov pc,lr
+        mov r0,pc; ret
         .asciz "do_xmap  fdi=%x  ehdr=%p  xi=%p(%x %p)  av=%p  p_reloc=%p  f_unf=%p\n"
         .balign 4
 
         .globl STR_upx_main
 STR_upx_main:
-        mov r0,pc; mov pc,lr
+        mov r0,pc; ret
         .asciz "upx_main av=%p  szc=%x  f_dec=%p  f_unf=%p  xo=%p(%x %p)  xi=%p(%x %p)  dynbase=%x\n",
         .balign 4
 #endif  /*}*/

src/stub/src/i386-linux.elf-fold.S

@@ -238,5 +238,56 @@ mmap: .globl mmap
         pop ebx
         ret
 
+#ifndef DEBUG  /*{*/
+#define DEBUG 0
+#endif  /*}*/
+
+#if DEBUG  /*{ truly-PIC string constants for debugging */
+        .globl STR_0x
+STR_0x:
+        call 0f; .asciz "0x"; 0:
+        pop eax; ret
+
+        .globl STR_hex
+STR_hex:
+        call 0f; .asciz "0123456789abcdef"; 0:
+        pop eax; ret
+
+        .globl STR_xread
+STR_xread:
+        call 0f; .asciz "xread %p(%x %p) %p %x\n"; 0:
+        pop eax; ret
+
+        .globl STR_unpackExtent
+STR_unpackExtent:
+        call 0f; .asciz "unpackExtent in=%p(%x %p)  out=%p(%x %p)  %p %p\n"; 0:
+        pop eax; ret
+
+        .globl STR_make_hatch_arm
+STR_make_hatch_arm:
+        call 0f; .asciz "make_hatch_arm %p %x\n"; 0:
+        pop eax; ret
+
+        .globl STR_auxv_up
+STR_auxv_up:
+        call 0f; .asciz "auxv_up  %p %x %x\n"; 0:
+        pop eax; ret
+
+        .globl STR_xfind_pages
+STR_xfind_pages:
+        call 0f; .asciz "xfind_pages  %x  %p  %d  %p\n"; 0:
+        pop eax; ret
+
+        .globl STR_do_xmap
+STR_do_xmap:
+        call 0f; .asciz "do_xmap  fdi=%x  ehdr=%p  xi=%p(%x %p)  av=%p  p_reloc=%p  f_unf=%p\n"; 0:
+        pop eax; ret
+
+        .globl STR_upx_main
+STR_upx_main:
+        call 0f; .asciz "upx_main av=%p  szc=%x  f_dec=%p  f_unf=%p  xo=%p(%x %p)  xi=%p(%x %p)  dynbase=%x\n"; 0:
+        pop eax; ret
+#endif  /*}*/
+
 // vi:ts=8:et:nowrap
 

src/stub/src/i386-linux.elf-main.c

@@ -43,9 +43,13 @@ ssize_t write(int, void const *, size_t);
 // it at an address different from it load address:  there must be no
 // static data, and no string constants.
 
-#if 1  /*{*/
+#ifndef DEBUG  /*{*/
+#define DEBUG 0
+#endif  /*}*/
+
+#if !DEBUG || defined(__mips__)  /*{*/
 #define DPRINTF(a) /* empty: no debug drivel */
-#else  /*}{*/
+#else  /*}{ DEBUG */
 #if 0
 #include "stdarg.h"
 #else
@@ -55,12 +59,23 @@ ssize_t write(int, void const *, size_t);
 #define va_start    __builtin_va_start
 #endif
 
+#ifdef __arm__  /*{*/
+extern unsigned div10(unsigned);
+#else  /*}{*/
+static unsigned
+div10(unsigned x)
+{
+    return x / 10u;
+}
+#endif  /*}*/
+
 static int
 unsimal(unsigned x, char *ptr, int n)
 {
     if (10<=x) {
-        n = unsimal(x/10, ptr, n);
+        unsigned const q = div10(x);
-        x %= 10;
+        n = unsimal(q, ptr, n);
+        x -= 10 * q;
     }
     ptr[n] = '0' + x;
     return 1+ n;

src/stub/src/mipsel.r3000-linux.elf-fold.S

@@ -207,4 +207,65 @@ mmap: .globl mmap
           addiu sp,sp, sp_frame
 #endif  /*}*/
 
+#if 0  /*{ truly-PIC string constants for debugging */
+/* UNFORTUNATELY THIS DOES NOT WORK because gcc uses 'jal' instead of 'bal' */
+
+        .globl STR_0x
+ret_at:
+        jr at
+          nop
+STR_0x:
+        move at,ra; bal ret_at; move v0,ra
+        .asciz "0x"
+        .balign 4
+
+        .globl STR_hex
+STR_hex:
+        move at,ra; bal ret_at; move v0,ra
+        .asciz "0123456789abcdef"
+        .balign 4
+
+        .globl STR_xread
+STR_xread:
+        move at,ra; bal ret_at; move v0,ra
+        .asciz "xread %p(%x %p) %p %x\n"
+        .balign 4
+
+        .globl STR_unpackExtent
+STR_unpackExtent:
+        move at,ra; bal ret_at; move v0,ra
+        .asciz "unpackExtent in=%p(%x %p)  out=%p(%x %p)  %p %p\n"
+        .balign 4
+
+        .globl STR_make_hatch_arm
+STR_make_hatch_arm:
+        move at,ra; bal ret_at; move v0,ra
+        .asciz "make_hatch_arm %p %x\n"
+        .balign 4
+
+        .globl STR_auxv_up
+STR_auxv_up:
+        move at,ra; bal ret_at; move v0,ra
+        .asciz "auxv_up  %p %x %x\n"
+        .balign 4
+
+        .globl STR_xfind_pages
+STR_xfind_pages:
+        move at,ra; bal ret_at; move v0,ra
+        .asciz "xfind_pages  %x  %p  %d  %p\n"
+        .balign 4
+
+        .globl STR_do_xmap
+STR_do_xmap:
+        move at,ra; bal ret_at; move v0,ra
+        .asciz "do_xmap  fdi=%x  ehdr=%p  xi=%p(%x %p)  av=%p  p_reloc=%p  f_unf=%p\n"
+        .balign 4
+
+        .globl STR_upx_main
+STR_upx_main:
+        move at,ra; bal ret_at; move v0,ra
+        .asciz "upx_main av=%p  szc=%x  f_dec=%p  f_unf=%p  xo=%p(%x %p)  xi=%p(%x %p)  dynbase=%x\n",
+        .balign 4
+#endif  /*}*/
+
 # vi:ts=8:et:nowrap