[PATCH] Do not sanitize kernel area on 32-bit targets

[Yury Gribov]

================
Comment at: lib/asan/asan_rtl.cc:485
@@ -484,1 +484,3 @@
     ProtectGap(kShadowGapBeg, kShadowGapEnd - kShadowGapBeg + 1);
+    // Allow accesses to unprotected part of address space
+    if (kShadowGapEnd != kHighShadowBeg - 1)
----------------
Kostya Serebryany wrote:
> this is weird... This should not be required... hm.. Let me see your comments first. 
With this patch kShadowGapEnd is now computed as (MEM_TO_SHADOW(kHighShadowEnd + 1) - 1) and kHighShadowEnd = MEM_TO_SHADOW(kHighMemEnd). kHighMemEnd can now be 3G instead of 4G so there is a 16M gap (== 128M / 8) between shadow gap and start of high shadow (see http://code.google.com/p/address-sanitizer/issues/detail?id=278 for exact memory maps). I think we should reserve this memory for ASan to avoid user code messing with ASan internals. The question is whether to mmap (this will allow access to kernel area) or protect it (this will force users to compile code that accesses kernel area without sanitization).


http://llvm-reviews.chandlerc.com/D3119