From 2f4488521db663520c703a9a836d5549d679266c Mon Sep 17 00:00:00 2001 From: athenian200 Date: Thu, 10 Oct 2019 15:38:27 -0500 Subject: MoonchildProductions#1251 - Part 23: Allow AMD64 build to work. https://developer.mozilla.org/en-US/docs/Mozilla/Developer_guide/Build_Instructions/Compiling_32-bit_Firefox_on_a_Linux_64-bit_OS Setting this up turned out to be easier than I thought it would be. All I had to do was apply these instructions in reverse and add the following to my .mozconfig file: CC="gcc -m64" CXX="g++ -m64" AS="gas --64" ac_add_options --target=x86_64-pc-solaris2.11 export PKG_CONFIG_PATH=/usr/lib/amd64/pkgconfig ac_add_options --libdir=/usr/lib/amd64 ac_add_options --x-libraries=/usr/lib/amd64 Most of these changes were fairly trivial, just requiring me to make a few of the changes I made earlier conditional on a 32-bit build. The biggest challenge was figuring out why the JavaScript engine triggered a segfault everytime it tried to allocate memory. But this patch fixes it: https://github.com/OpenIndiana/oi-userland/blob/oi/hipster/components/web/firefox/patches/patch-js_src_gc_Memory.cpp.patch Turns out that Solaris on AMD64 handles memory management in a fairly unusual way with a segmented memory model, but it's not that different from what we see on other 64-bit processors. In fact, I saw a SPARC crash for a similar reason, and noticed that it looked just like mine except the numbers in the first segment were reversed. Having played around with hex editors before, I had a feeling I might be dealing with a little-endian version of a big-endian problem, but I didn't expect that knowledge to actually yield an easy solution. https://bugzilla.mozilla.org/show_bug.cgi?id=577056 https://www.oracle.com/technetwork/server-storage/solaris10/solaris-memory-135224.html As far as I can tell, this was the last barrier to an AMD64 Solaris build of Pale Moon. --- js/src/gc/Memory.cpp | 11 +++++++++-- 1 file changed, 9 insertions(+), 2 deletions(-) (limited to 'js/src') diff --git a/js/src/gc/Memory.cpp b/js/src/gc/Memory.cpp index 5cf0cd2f7..2bc1b9f50 100644 --- a/js/src/gc/Memory.cpp +++ b/js/src/gc/Memory.cpp @@ -415,8 +415,15 @@ InitMemorySubsystem() static inline void* MapMemoryAt(void* desired, size_t length, int prot = PROT_READ | PROT_WRITE, int flags = MAP_PRIVATE | MAP_ANON, int fd = -1, off_t offset = 0) + +// Solaris manages 64-bit address space in a different manner from every other +// AMD64 operating system, but fortunately the fix is the same one +// required for every operating system on 64-bit SPARC, Itanium, and ARM. +// Most people's intuition failed them here and they thought this couldn't +// possibly be correct on AMD64, but for Solaris/illumos it is. + { -#if defined(__ia64__) || (defined(__sparc64__) && defined(__NetBSD__)) || defined(__aarch64__) +#if defined(__ia64__) || (defined(__sparc64__) && defined(__NetBSD__)) || defined(__aarch64__) || (defined(__sun) && defined(__x86_64__)) MOZ_ASSERT((0xffff800000000000ULL & (uintptr_t(desired) + length - 1)) == 0); #endif void* region = mmap(desired, length, prot, flags, fd, offset); @@ -466,7 +473,7 @@ MapMemory(size_t length, int prot = PROT_READ | PROT_WRITE, return nullptr; } return region; -#elif defined(__aarch64__) +#elif defined(__aarch64__) || (defined(__sun) && defined(__x86_64__)) /* * There might be similar virtual address issue on arm64 which depends on * hardware and kernel configurations. But the work around is slightly -- cgit v1.2.3