Backport35 doc fixes: PR#68 and PR#124 #125 by vstinner · Pull Request #126 · python/cpython
* Travis CI: run rstlint.py in the docs job (#68) Currently, http://buildbot.python.org/all/buildslaves/ware-docs buildbot is only run as post-commit. For example, bpo-29521 (PR#41) introduced two warnings, unnotified by the Travis CI docs job. Modify the docs job to run toosl/rstlint.py. Fix also the two minor warnings which causes the buildbot slave to fail. (cherry picked from commit 2b50186) * Doc/Makefile: set PYTHON to python3 (#124) rstlint.py run by "make check" doesn't support Python 2. "make venv" runs "$(PYTHON) -m venv", whereas Python 2 doens't provide the venv module: it's a module of Python 3 standard library. (cherry picked from commit 91b0e7d) (cherry picked from commit b300c66)
Backport35 doc
Backport35 doc fixes: PR#68 and PR#124 #125
akruis referenced this pull request in stackless-dev/stackless
akruis pushed a commit to akruis/cpython that referenced this pull request
akruis pushed a commit to akruis/cpython that referenced this pull request
akruis pushed a commit to akruis/cpython that referenced this pull request
This was referenced
SonicField added a commit to SonicField/cpython that referenced this pull request
Phoenix-introduced regression in the phoenix-asm x86_64 encoder.
phx_fs_ptr (TLS access for PyThreadState) returned PhxMem with
m.base default-initialized to {0} = RAX (id=0). encode_modrm_mem
treated this as `[RAX + disp]` addressing and emitted
64 4c 8b 58 c8 mov %r11, %fs:-0x38(%rax)
instead of the intended
64 4c 8b 1c 25 c8 ff ff ff mov %r11, %fs:[-0x38]
Under release-mode RAX happened to be 0 at the load_tstate call
site so fs:[disp+rax] resolved correctly. Under pydebug RAX was
non-zero (testkeeper 16:46:13Z lldb register dump showed rax=0x4
at the crash) so fs:[rax-0x38] read the wrong TLS slot, returned
NULL/garbage tstate, and crashed at the subsequent
mov rax, [r11+0x38] (read tstate->cframe).
Fix: phx_fs_ptr sets is_abs_addr=1 + abs_addr=sign-extend(offset)
so encode_modrm_mem's existing absolute-address branch
(x86_64.c:182-188) emits the SIB pure-disp32 form (ModR/M=04,
SIB=25, disp32). The FS segment prefix is independently emitted
by emit_segment_prefix; the byte sequence required for FS-segment
+ pure-displacement is identical to the absolute-address form.
Why valgrind didn't catch (testkeeper 15:46:07Z Phoenix valgrind
clean): valgrind tracked memory addresses correctly; the wrong
tstate was logically wrong but didn't trip valgrind's
instrumentation. The crash deferred to the later
[r11+0x38] read.
Why ARM64 unaffected (testkeeper 16:58:33Z 4/4 PASS without fix):
AArch64 has no FS segment; load_tstate reads TLS via
mrs(TPIDR_EL0) + ldr — different code path entirely.
Verification (testkeeper 16:52:24Z + 16:58:33Z):
x86_64 pydebug + fix: simple-add / closure / generator /
exception 4/4 PASS (was SIGSEGV pre-fix on 3 of 4)
ARM64 pydebug (no fix needed): 4/4 PASS
Dual-arch confirms (β) hypothesis: bug is x86_64-only
Structural debt: is_abs_addr is reused here for the SIB-disp32
shape, NOT for its conventional 'absolute 64-bit address'
semantics. DSecondary-2 (theologian 16:49:17Z, supervisor
16:55:58Z) tracks the proper refactor: add explicit has_base
flag to PhxMem; refactor phx_fs_ptr to use it instead of
overloading is_abs_addr. Scheduled immediately after this push
to pay structural debt while surface is fresh.
Auth chain: testkeeper root cause 16:46:13Z (lldb full prologue
disasm) + theologian patch-shape APPROVE 16:49:17Z (with
DSecondary-2 logged) + testkeeper 4/4 fixture verify x86_64
16:52:24Z + ARM64 (β) confirm 16:58:33Z; supervisor authorization
16:55:58Z queue re-sequence. Closes audit Group C C2 W-C2
workstream.
Closes pythia python#126 single-fixture concern (4-fixture pre-fix
verification per supervisor 16:15:41Z).
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters