Index: /trunk/src/bldprogs/VBoxDef2LazyLoad.cpp =================================================================== --- /trunk/src/bldprogs/VBoxDef2LazyLoad.cpp (revision 87281) +++ /trunk/src/bldprogs/VBoxDef2LazyLoad.cpp (revision 87282) @@ -27,4 +27,5 @@ #include #include +#include /* For RTLDRARCH. */ @@ -46,5 +47,4 @@ - /********************************************************************************************************************************* * Global Variables * @@ -59,4 +59,13 @@ static bool g_fWithExplictLoadFunction = false; static bool g_fSystemLibrary = false; +#if defined(RT_ARCH_AMD64) +static RTLDRARCH g_enmTarget = RTLDRARCH_AMD64; +#elif defined(RT_ARCH_X86) +static RTLDRARCH g_enmTarget = RTLDRARCH_X86; +#elif defined(RT_ARCH_ARM64) +static RTLDRARCH g_enmTarget = RTLDRARCH_ARM64; +#else +# error "Port me!" +#endif /** @} */ @@ -321,5 +330,6 @@ /** - * Generates the assembly source code, writing it to @a pOutput. + * Generates the assembly source code for AMD64 and x86, writing it + * to @a pOutput. * * @returns RTEXITCODE_SUCCESS or RTEXITCODE_FAILURE, in the latter case full @@ -328,5 +338,5 @@ * when closing). */ -static RTEXITCODE generateOutputInner(FILE *pOutput) +static RTEXITCODE generateOutputInnerX86AndAMD64(FILE *pOutput) { fprintf(pOutput, ";;\n"); @@ -983,4 +993,460 @@ /** + * Generates the assembly source code for ARM64, writing it + * to @a pOutput. + * + * @returns RTEXITCODE_SUCCESS or RTEXITCODE_FAILURE, in the latter case full + * details has been displayed. + * @param pOutput The output stream (caller checks it for errors + * when closing). + */ +static RTEXITCODE generateOutputInnerArm64(FILE *pOutput) +{ +// bool fMachO = true; +// bool fDarwin = true; + const char *pszNmPfx = "_"; + + fprintf(pOutput, ";;\n"); + for (unsigned i = 0; i < g_cInputs; i++) + fprintf(pOutput, ";; Autogenerated from '%s'.\n", g_apszInputs[i]); + + fprintf(pOutput, + ";; DO NOT EDIT!\n" + ";;\n" + "\n" + "\n" + /*"%%include \"iprt/asmdefs.mac\"\n"*/ + "\n" + "\n"); + + /* + * Put the thunks first for alignment and other reasons. It's the hot part of the code. + */ + fprintf(pOutput, + ";\n" + "; Thunks.\n" + ";\n" + ".section __TEXT,__text,regular,pure_instructions\n" + ".p2align 2\n"); + for (PMYEXPORT pExp = g_pExpHead; pExp; pExp = pExp->pNext) + fprintf(pOutput, + ".globl %s%s\n" + "%s%s:\n" + " ldr x9, =g_pfn%s\n" + " blr x9\n", + pszNmPfx, pExp->szName, pszNmPfx, pExp->szName, pExp->szName); + fprintf(pOutput, + "\n" + "\n"); + + /* + * Import pointers + */ + fprintf(pOutput, + ";\n" + "; Import pointers. Initialized to point a lazy loading stubs.\n" + ";\n" + ".section __DATA,__data\n" + ".p2align 2\n" + "g_apfnImports:\n"); + for (PMYEXPORT pExp = g_pExpHead; pExp; pExp = pExp->pNext) + fprintf(pOutput, + ".globl __imp_%s\n" + "__imp_%s:\n" + ".globl %sg_pfn%s\n" + "%sg_pfn%s:\n" + " .quad ___LazyLoad___%s\n" + "\n", + pExp->szName, pExp->szName, + pszNmPfx, pExp->szName, pszNmPfx, pExp->szName, + pExp->pszExportedNm); + fprintf(pOutput, + " .quad 0 ; Terminator entry for traversal.\n" + "\n" + "\n"); + + /* + * Now for the less important stuff, starting with the names. + * + * We keep the names separate so we can traverse them in parallel to + * g_apfnImports in the load-everything routine further down. + */ + fprintf(pOutput, + ";\n" + "; Imported names.\n" + ";\n" + ".section __TEXT,__cstring,cstring_literals\n" + "g_szLibrary:\n" + " .asciz \"%s\"\n" + "\n" + "g_szzNames:\n", + g_pszLibrary); + for (PMYEXPORT pExp = g_pExpHead; pExp; pExp = pExp->pNext) + if (!pExp->fNoName) + fprintf(pOutput, " g_sz%s:\n .asciz \"%s\"\n", pExp->pszExportedNm, pExp->pszExportedNm); + else + fprintf(pOutput, " g_sz%s:\n .asciz \"#%u\"\n", pExp->pszExportedNm, pExp->uOrdinal); + fprintf(pOutput, + "g_EndOfNames: .byte 0\n" + "\n" + "g_szFailLoadFmt: .asciz \"Lazy loader failed to load \\\"%%s\\\": %%Rrc\\n\"\n" + "g_szFailResolveFmt: .asciz \"Lazy loader failed to resolve symbol \\\"%%s\\\" in \\\"%%s\\\": %%Rrc\\n\"\n" + "\n" + "\n"); + + /* + * The per import lazy load code. + */ + fprintf(pOutput, + ";\n" + "; Lazy load+resolve stubs.\n" + ";\n" + ".section __TEXT,__text,regular,pure_instructions\n" + ".p2align 2\n"); + for (PMYEXPORT pExp = g_pExpHead; pExp; pExp = pExp->pNext) + { + if (!pExp->fNoName) + fprintf(pOutput, + "___LazyLoad___%s:\n" + " ldr x9, =g_sz%s\n" + " ldr x10, =g_pfn%s\n" + " bl LazyLoadResolver\n" + , pExp->pszExportedNm, pExp->pszExportedNm, pExp->pszExportedNm); + else + fprintf(pOutput, + "___LazyLoad___%s:\n" + " movk w9, #%u\n" + " ldr x10, =g_pfn%s\n" + " bl LazyLoadResolver\n" + , pExp->pszExportedNm, pExp->uOrdinal, pExp->pszExportedNm); + fprintf(pOutput, " b %s%s\n", pszNmPfx, pExp->szName); + fprintf(pOutput, "\n"); + } + fprintf(pOutput, + "\n" + "\n" + "\n"); + + /* + * The code that does the loading and resolving. + */ + fprintf(pOutput, + ";\n" + "; The module handle.\n" + ";\n" + ".section __DATA,__data\n" + "g_hMod:\n" + " .quad 0\n" + "\n" + "\n" + "\n"); + + /* + * Common lazy loader and resolved. + */ + fprintf(pOutput, + ";\n" + "; The resolver code.\n" + ";\n" + ".section __TEXT,__text,regular,pure_instructions\n" + ".p2align 2\n" + "LazyLoadResolver:\n" + " .cfi_startproc\n" + " ; Create frame.\n" + " sub sp, sp, #(16 + 192)\n" + " stp x29, x30, [sp, #192]\n" + " add x29, sp, #192\n" + " .cfi_def_cfa x29, 16\n" + " .cfi_offset x30, -8\n" + " .cfi_offset x29, -16\n" + " ; Save all argument registers and a handful of preserved ones.\n" + " stp x0, x1, [sp, #(192 - 16)]\n" + " .cfi_offset x0, -32\n" + " .cfi_offset x1, -24\n" + " stp x2, x3, [sp, #(192 - 32)]\n" + " .cfi_offset x3, -40\n" + " .cfi_offset x2, -48\n" + " stp x4, x5, [sp, #(192 - 48)]\n" + " .cfi_offset x6, -56\n" + " .cfi_offset x5, -64\n" + " stp x6, x7, [sp, #(192 - 64)]\n" + " .cfi_offset x7, -72\n" + " .cfi_offset x6, -80\n" + " stp x16, x17, [sp, #(192 - 80)]\n" + " .cfi_offset x17, -88\n" + " .cfi_offset x16, -96\n" + " stp x18, x19, [sp, #(192 - 96)]\n" + " .cfi_offset x19, -104\n" + " .cfi_offset x18, -112\n" + " stp x20, x21, [sp, #(192 - 112)]\n" + " .cfi_offset x21, -120\n" + " .cfi_offset x20, -128\n" + " stp x22, x23, [sp, #(192 - 128)]\n" + " .cfi_offset x23, -136\n" + " .cfi_offset x22, -144\n" + " str x8, [sp, #(192 - 144)]\n" + "\n" + " ; Shift the symbol name to x19 and g_pfnXXXX pointer to x20 as these are preserved registers\n" + " ; (in case we need to call LazyLoadModule/RTLdrLoad)\n" + " mov x19, x9\n" + " mov x20, x10\n" + "\n" + " ; Get the module handle and call RTLdrGetSymbol(RTLDRMOD hLdrMod, const char *pszSymbol, void **ppvValue)\n" + " ldr x0, =g_hMod\n" + " ldr x0, [x0]\n" + " cmp x0, #0\n" + " b.eq LazyLoading\n" + " mov x1, x19\n" + " mov x2, x20\n" + " bl %sRTLdrGetSymbol\n" + "\n" + " cmp w0, #0\n" + " b.eq Lreturn\n" + "\n" + "Lbadsym: ; Call sRTAssertMsg2Weak. Variadic (...) arguments are passed on the stack it seems.\n" + " mov x3, x0\n" + " ldr x2, =g_szLibrary\n" + " mov x1, x19\n" + " ldr x0, =g_szFailLoadFmt\n" + " stp x1, x2, [sp]\n" + " str x3, [sp, #16]\n" + " bl %sRTAssertMsg2Weak\n" + "Lbadsymloop:\n" + " brk #0x1\n" + " b Lbadsymloop\n" + + "Lreturn:\n" + " ; Restore saved register\n" + " ldr x8, [sp, #(192 - 144)]\n" + " .cfi_restore x8\n" + " ldp x22, x23, [sp, #(192 - 128)]\n" + " .cfi_restore x23\n" + " .cfi_restore x22\n" + " ldp x20, x21, [sp, #(192 - 112)]\n" + " .cfi_restore x21\n" + " .cfi_restore x20\n" + " ldp x18, x19, [sp, #(192 - 96)]\n" + " .cfi_restore x19\n" + " .cfi_restore x18\n" + " ldp x16, x17, [sp, #(192 - 80)]\n" + " .cfi_restore x17\n" + " .cfi_restore x18\n" + " ldp x6, x7, [sp, #(192 - 64)]\n" + " .cfi_restore x7\n" + " .cfi_restore x6\n" + " ldp x4, x5, [sp, #(192 - 48)]\n" + " .cfi_restore x5\n" + " .cfi_restore x4\n" + " ldp x2, x3, [sp, #(192 - 32)]\n" + " .cfi_restore x3\n" + " .cfi_restore x2\n" + " ldp x0, x1, [sp, #(192 - 16)]\n" + " .cfi_restore x1\n" + " .cfi_restore x0\n" + "\n" + " ldp x29, x30, [sp, #192]\n" + " .cfi_restore x29\n" + " .cfi_restore x30\n" + " add sp, sp, #(16 + 192)\n" + " ret\n" + " .cfi_endproc\n" + "\n" + "\n" + , pszNmPfx, pszNmPfx); + + fprintf(pOutput, + ";\n" + "; Loads the module.\n" + "; ASSUMES called from LazyLoadResolver where all relevant registers are already saved.\n" + ";\n" + "LazyLoading:\n" + " .cfi_startproc\n" + " ; Create frame.\n" + " sub sp, sp, #(16 + 48)\n" + " stp x29, x30, [sp, #48]\n" + " add x29, sp, #48\n" + " .cfi_def_cfa x29, 16\n" + " .cfi_offset x30, -8\n" + " .cfi_offset x29, -16\n" + "\n"); + + if (!g_fSystemLibrary) + fprintf(pOutput, + " ; Call SUPR3HardenedLdrLoadAppPriv(const char *pszFilename, PRTLDRMOD phLdrMod, uint32_t fFlags, PRTERRINFO pErrInfo);\n" + " mov x3, #0\n" + " mov x2, #0\n" + " ldr x1, =g_hMod\n" + " ldr x0, =g_szLibrary\n" + " bl %sSUPR3HardenedLdrLoadAppPriv\n" + , pszNmPfx); + else + fprintf(pOutput, + " ; Call RTLdrLoadSystem(const char *pszFilename, bool fNoUnload, PRTLDRMOD phLdrMod);\n" + " ldr x2, =g_hMod\n" + " mov x1, #1\n" + " ldr x0, =g_szLibrary\n" + " bl %sRTLdrLoadSystem\n" + , pszNmPfx); + + fprintf(pOutput, + " cmp w0, #0\n" + " b.eq Lload_return\n" + "\n" + "Lbadload: ; Call sRTAssertMsg2Weak. Variadic (...) arguments are passed on the stack it seems.\n" + " mov x2, x0\n" + " ldr x1, =g_szLibrary\n" + " ldr x0, =g_szFailResolveFmt\n" + " stp x1, x2, [sp]\n" + " bl %sRTAssertMsg2Weak\n" + "Lbadloadloop:\n" + " brk #0x1\n" + " b Lbadloadloop\n" + "Lload_return:\n" + " ldr x0, =g_hMod\n" + " ldr x0, [x0]\n" + " ldp x29, x30, [sp, #48]\n" + " .cfi_restore x29\n" + " .cfi_restore x30\n" + " add sp, sp, #(16 + 48)\n" + " ret\n" + " .cfi_endproc\n" + "\n" + "\n" + , pszNmPfx); + + /* + * C callable method for explicitly loading the library and optionally + * resolving all the imports. + */ + if (g_fWithExplictLoadFunction) + { + if (g_fSystemLibrary) /* Lazy bird. */ + { + fprintf(stderr, "error: cannot use --system with --explicit-load-function, sorry\n"); + return RTEXITCODE_FAILURE; + } + + int cchLibBaseName = (int)(strchr(g_pszLibrary, '.') ? strchr(g_pszLibrary, '.') - g_pszLibrary : strlen(g_pszLibrary)); + fprintf(pOutput, + ";;\n" + "; ExplicitlyLoad%.*s(bool fResolveAllImports, pErrInfo);\n" + ";\n" + ".section __TEXT,__text,regular,pure_instructions\n" + ".p2align 2\n" + ".globl ExplicitlyLoad%.*s\n" + "ExplicitlyLoad%.*s:\n" + " .cfi_startproc\n" + " ; Create frame.\n" + " sub sp, sp, #(16 + #96)\n" + " stp x29, x30, [sp, #96]\n" + " add x29, sp, #96\n" + " .cfi_def_cfa x29, 16\n" + " .cfi_offset x30, -8\n" + " .cfi_offset x29, -16\n" + "\n" + " stp x20, x21, [sp, #(96 - 16)]\n" + " .cfi_offset x21, -24\n" + " .cfi_offset x20, -32\n" + " stp x22, x23, [sp, #(96 - 32)]\n" + " .cfi_offset x23, -40\n" + " .cfi_offset x22, -48\n" + + " ; Save the input parameters.\n" + " mov x20, x0\n" + " mov x21, x1\n" + "\n" + " ;\n" + " ; Is the module already loaded?\n" + " ;\n" + " ldr x0, =g_hMod\n" + " ldr x0, [x0]\n" + " cmp x0, #0\n" + " b.ne Lexplicit_loaded_module\n" + "\n" + , + cchLibBaseName, g_pszLibrary, + cchLibBaseName, g_pszLibrary, + cchLibBaseName, g_pszLibrary); + fprintf(pOutput, + "Lexplicit_load_module:\n" + " ; Call SUPR3HardenedLdrLoadAppPriv(const char *pszFilename, PRTLDRMOD phLdrMod, uint32_t fFlags, PRTERRINFO pErrInfo);\n" + " mov x3, #0\n" + " mov x2, #0\n" + " ldr x1, =g_hMod\n" + " ldr x0, =g_szLibrary\n" + " bl %sSUPR3HardenedLdrLoadAppPriv\n" + " cmp x0, #0\n" + " b.ne Lexplicit_load_return\n" + "\n" + , pszNmPfx); + + fprintf(pOutput, + " ;\n" + " ; Resolve the imports too if requested to do so.\n" + " ;\n" + "Lexplicit_loaded_module:\n" + " cmp w20, #0\n" + " b.eq Lexplicit_load_return\n" + "\n" + " ldr x22, =g_szzNames\n" + " ldr x23, =g_apfnImports\n" + "Lexplicit_load_next_import:\n" + " ldr x0, [x23]\n" + " cmp x0, #0\n" + " b.eq Lexplicit_load_return\n" + "\n" + " ; Get the module handle and call RTLdrGetSymbol(RTLDRMOD hLdrMod, const char *pszSymbol, void **ppvValue)\n" + " ldr x0, =g_hMod\n" + " ldr x0, [x0]\n" + " mov x1, x22\n" + " mov x2, x23\n" + " bl %sRTLdrGetSymbol\n" + " cmp x0, #0\n" + " b.ne Lexplicit_load_symbol_error\n" + "\n" + " ; Advance.\n" + " add x23, x23, #8\n" + "Lexplict_load_advance_string:\n" + " ldrb w0, [x22]\n" + " add x22, x22, #1\n" + " cmp w0, #0\n" + " b.ne Lexplict_load_advance_string\n" + " b Lexplicit_load_next_import\n" + "\n" + " ;\n" + " ; Error loading a symbol. Call RTErrInfoSet(PRTERRINFO pErrInfo, int rc, const char *pszMsg) on pErrInfo (preserves x0).\n" + " ;\n" + "Lexplicit_load_symbol_error:\n" + " mov x2, x22\n" + " mov x1, x0\n" + " mov x0, x21\n" + " bl %sRTErrInfoSet\n" + " b Lexplicit_load_return" + " " + "\n" + "Lexplicit_load_return:\n" + " ldp x22, x23, [sp, #(96 - 32)]\n" + " .cfi_restore x23\n" + " .cfi_restore x22\n" + " ldp x20, x21, [sp, #(96 - 16)]\n" + " .cfi_restore x21\n" + " .cfi_restore x20\n" + "\n" + " ldp x29, x30, [sp, #96]\n" + " .cfi_restore x29\n" + " .cfi_restore x30\n" + " add sp, sp, #(16 + 96)\n" + " ret\n" + " .cfi_endproc\n" + "\n" + "\n" + , pszNmPfx, pszNmPfx); + } + + return RTEXITCODE_SUCCESS; +} + + +/** * Generates the assembly source code, writing it to g_pszOutput. * @@ -994,5 +1460,17 @@ if (pOutput) { - rcExit = generateOutputInner(pOutput); + switch (g_enmTarget) + { + case RTLDRARCH_AMD64: + case RTLDRARCH_X86_32: + rcExit = generateOutputInnerX86AndAMD64(pOutput); + break; + case RTLDRARCH_ARM64: + rcExit = generateOutputInnerArm64(pOutput); + break; + default: + rcExit = RTEXITCODE_FAILURE; + break; + } if (fclose(pOutput)) {