| commit | 2efe13d1a55e5c342a32b0da7cc15ec58ab5dce3 | |
| author | Douglas Katzman <dougk@google.com> | |
| Sun, 30 Nov 2014 00:28:06 +0000 (29 19:28 -0500) | ||
| committer | Douglas Katzman <dougk@google.com> | |
| Sun, 30 Nov 2014 00:28:06 +0000 (29 19:28 -0500) | ||
| tree | d269901d107b21df10f91d1c9629fca2f9fecff9 | treesnapshot (tar.gz zip) |
| parent | ab5bc60afbd1d90844db43f8fe90debab8d5d5dd | commitdiff |
Make disassembler correctly detect multiple code .ENTRY points
If pad bytes were inadvertently decoded as opcodes, they might bump
the PC beyond where the fun-header-widetag is, and so the .entry hook
would fail to trigger, and everything would be wrong until by chance
the decoder could sync to a valid opcode.
And the use of sb!kernel:code-header-ref in 'add-fun-header-hooks'
was insane. SEG-CODE could be NIL, but it "worked" to dereference NIL
as if a code-header and get back NIL by random coincidence.
If pad bytes were inadvertently decoded as opcodes, they might bump
the PC beyond where the fun-header-widetag is, and so the .entry hook
would fail to trigger, and everything would be wrong until by chance
the decoder could sync to a valid opcode.
And the use of sb!kernel:code-header-ref in 'add-fun-header-hooks'
was insane. SEG-CODE could be NIL, but it "worked" to dereference NIL
as if a code-header and get back NIL by random coincidence.
| src/compiler/target-disassem.lisp | diffblobblamehistory |