Fix vmlinux/ppc32 at compress time. Runtime needs work.

src/p_vmlinx.cpp

@@ -986,11 +986,14 @@ void PackVmlinuxPPC32::buildLoader(const Filter *ft)
         addFilter32(ft->id);
     }
     addLoader("LINUX030", NULL);
-         if (ph.method == M_NRV2E_LE32) addLoader("NRV2E", NULL);
+         if (ph.method == M_NRV2E_LE32) addLoader("NRV2E,NRV_TAIL", NULL);
-    else if (ph.method == M_NRV2B_LE32) addLoader("NRV2B", NULL);
+    else if (ph.method == M_NRV2B_LE32) addLoader("NRV2B,NRV_TAIL", NULL);
-    else if (ph.method == M_NRV2D_LE32) addLoader("NRV2D", NULL);
+    else if (ph.method == M_NRV2D_LE32) addLoader("NRV2D,NRV_TAIL", NULL);
     else if (M_IS_LZMA(ph.method))   addLoader("LZMA_ELF00,LZMA_DEC10,LZMA_DEC30", NULL);
     else throwBadLoader();
+    if (hasLoaderSection("CFLUSH"))
+        addLoader("CFLUSH");
+
     addLoader("IDENTSTR,UPX1HEAD", NULL);
 }
 

src/stub/src/powerpc-linux.kernel.vmlinux.S

@@ -90,6 +90,42 @@ section NRV2D
 section NRV2E
 #include "arch/powerpc/32/nrv2e_d.S"
 
+  section NRV_TAIL
+eof_nrv:
+#define dst0 a4
+#define tmp a1
+        lwz dst0,0(ldst)  // original dst
+        mtlr t3  // return address
+        subf a0,lsrc,src
+        subf tmp,dst0,dst  // -1+ dst length
+        addi a0,a0,1  // return 0: good; else: bad  [+1: correct for lbzu]
+        addi tmp,tmp,1  // dst length
+        stw  tmp,0(ldst)
+#undef tmp
+
+// CACHELINE=32 is the observed minimum line size of any cache.
+// Some caches may have larger lines, but it is cumbersome to lookup
+// {AT_DCACHEBSIZE, AT_ICACHEBSIZE, AT_UCACHEBSIZE: /usr/include/elf.h},
+// then save the correct size in a variable {where to put it?}, or to modify
+// the two instructions here.  If a cache has larger lines, then we expect
+// that the second dcbst (or icbi) on a the same line will be fast.
+// If not, then too bad.
+
+  section CFLUSH  // In: a2=dst= &highest stored byte; a4=dst0= &lowest stored byte
+CACHELINE=32
+        ori dst0,dst0,-1+ CACHELINE  // highest addr on cache line
+cfl_nrv:
+        dcbst  0,dst0  // initiate store (modified) cacheline to memory
+        cmpl cr0,dst0,dst  // did we cover the highest-addressed byte?
+        icbi   0,dst0  // discard instructions from cacheline
+        addi     dst0,dst0,CACHELINE  // highest addr on next line
+        blt  cr0,cfl_nrv  // not done yet
+#undef dst0
+        sync   // wait for all memory operations to finish
+        isync  // discard prefetched instructions (if any)
+cfl_ret:
+        ret
+
 #undef hibit
 
 #undef src

src/stub/tmp/powerpc-linux.kernel.vmlinux.bin.dump

@@ -9,21 +9,25 @@ Idx Name          Size      VMA       LMA       File off  Algn  Flags
   4 NRV2B         000000f0  00000000  00000000  000000a8  2**0  CONTENTS, RELOC, READONLY
   5 NRV2D         0000012c  00000000  00000000  00000198  2**0  CONTENTS, RELOC, READONLY
   6 NRV2E         00000148  00000000  00000000  000002c4  2**0  CONTENTS, RELOC, READONLY
-  7 LZMA          00000000  00000000  00000000  0000040c  2**0  CONTENTS, READONLY
+  7 NRV_TAIL      0000001c  00000000  00000000  0000040c  2**0  CONTENTS, READONLY
-  8 LZMA_ELF00    0000008c  00000000  00000000  0000040c  2**0  CONTENTS, RELOC, READONLY
+  8 CFLUSH        00000024  00000000  00000000  00000428  2**0  CONTENTS, READONLY
-  9 LZMA_DEC10    0000099c  00000000  00000000  00000498  2**0  CONTENTS, READONLY
+  9 LZMA          00000000  00000000  00000000  0000044c  2**0  CONTENTS, READONLY
- 10 LZMA_DEC20    0000099c  00000000  00000000  00000e34  2**0  CONTENTS, READONLY
+ 10 LZMA_ELF00    0000008c  00000000  00000000  0000044c  2**0  CONTENTS, RELOC, READONLY
- 11 LZMA_DEC30    00000020  00000000  00000000  000017d0  2**0  CONTENTS, READONLY
+ 11 LZMA_DEC10    0000099c  00000000  00000000  000004d8  2**0  CONTENTS, READONLY
- 12 UPX1HEAD      00000020  00000000  00000000  000017f0  2**0  CONTENTS, READONLY
+ 12 LZMA_DEC20    0000099c  00000000  00000000  00000e74  2**0  CONTENTS, READONLY
+ 13 LZMA_DEC30    00000020  00000000  00000000  00001810  2**0  CONTENTS, READONLY
+ 14 UPX1HEAD      00000020  00000000  00000000  00001830  2**0  CONTENTS, READONLY
 SYMBOL TABLE:
 00000000 l    d  LINUX020	00000000 LINUX020
 00000000 l    d  LINUX030	00000000 LINUX030
+00000000 l    d  NRV_TAIL	00000000 NRV_TAIL
 00000000 l    d  LZMA_DEC30	00000000 LZMA_DEC30
 00000000 l    d  LINUX000	00000000 LINUX000
 00000000 l    d  LINUX010	00000000 LINUX010
 00000000 l    d  NRV2B	00000000 NRV2B
 00000000 l    d  NRV2D	00000000 NRV2D
 00000000 l    d  NRV2E	00000000 NRV2E
+00000000 l    d  CFLUSH	00000000 CFLUSH
 00000000 l    d  LZMA	00000000 LZMA
 00000000 l    d  LZMA_ELF00	00000000 LZMA_ELF00
 00000000 l    d  LZMA_DEC10	00000000 LZMA_DEC10
@@ -31,7 +35,6 @@ SYMBOL TABLE:
 00000000 l    d  UPX1HEAD	00000000 UPX1HEAD
 00000000         *UND*	00000000 filter_length
 00000000         *UND*	00000000 filter_cto
-00000000         *UND*	00000000 eof_nrv
 
 RELOCATION RECORDS FOR [LINUX000]:
 OFFSET   TYPE              VALUE
@@ -45,15 +48,15 @@ OFFSET   TYPE              VALUE
 
 RELOCATION RECORDS FOR [NRV2B]:
 OFFSET   TYPE              VALUE
-00000090 R_PPC_REL14       eof_nrv
+00000090 R_PPC_REL14       NRV_TAIL
 
 RELOCATION RECORDS FOR [NRV2D]:
 OFFSET   TYPE              VALUE
-000000b4 R_PPC_REL14       eof_nrv
+000000b4 R_PPC_REL14       NRV_TAIL
 
 RELOCATION RECORDS FOR [NRV2E]:
 OFFSET   TYPE              VALUE
-000000b4 R_PPC_REL14       eof_nrv
+000000b4 R_PPC_REL14       NRV_TAIL
 
 RELOCATION RECORDS FOR [LZMA_ELF00]:
 OFFSET   TYPE              VALUE