README.gcc12
1$NetBSD: README.gcc12,v 1.29 2024/09/23 10:28:54 rin Exp $
2
3
4new stuff:
5 libmpx
6 liboffloadmic
7 libvtv
8 libcet
9 libphobos
10 d
11 gfortran
12 libbacktrace: +dwarf5_CFLAGS = $(libbacktrace_TEST_CFLAGS) -gdwarf-5
13 c++tools
14
15todo:
16 - audit all uses of new warning-disable variables and
17 convert some from -Wno- to -Wno-error=
18
19arch/feature list.
20
21tools: does build.sh tools work?
22kernels: does a kernel run? y (yes), b (builds/ready), p (partially), ? (not ready)
23libgcc: does libgcc build?
24native-gcc: does a full mknative-gcc complete?
25sani: has sanitizer been ported to this CPU again for GCC 12? b (build) y (works)
26make release: does build.sh release complete? (before sanitizer - bs)
27runs: does the system boot with a full world?
28atf: does atf run / compare well
29switched: has port switched? y (yes), n (no/ready), ? (not ready)
30
31
32architecture tools kernels libgcc native-gcc/sani make release runs atf switched
33------------ ----- ------- ------ --------------- ------------ ---- --- --------
34aarch64 y y y y y[17] y y y y
35aarch64eb y y y y y[17] y y y y
36alpha y y y y b y y y y
37earmv4 y b ? y ? ? ? ? y
38earmv4eb y b ? y ? ? ? ? y
39earmv5 y y y y b y y y y
40earmv5eb y y y y b y y y y
41earmv5hf y b ? y ? ? ? ? y
42earmv5hfeb y b ? y ? ? ? ? y
43earmv6 y b ? y ? ? ? ? y
44earmv6eb y b ? y ? ? ? ? y
45earmv6hf y y y y b y y y y
46earmv6hfeb y y y y b y y y y
47earmv7 y b y y ? ? ? ? y
48earmv7eb y b ? y ? ? ? ? y
49earmv7hf y y y y b y y y y
50earmv7hfeb y y y y b y y y y
51hppa y y y y ? y y y y
52i386 y y y y y y y y y
53ia64 y b y y y y y[3] n y
54m68000 y y y y b n y[14] ? y
55m68k y y y y b y y[14] y[14] y
56mipseb y b y y b y y y y
57mipsel y b y y b y y ? y
58mips64eb y b y y b y y y y
59mips64el y b y y b y y ? y
60powerpc y y y y y y y y y
61powerpc64 y ? ? y b b ? ? y
62sh3eb y ? y y b y ? ? y
63sh3el y y y y b y y y y
64sparc y y y y n[26] y y y y
65sparc64 y b y y n[16] y y ? y
66vax y y y y n[23] n n[24,25] ? ?
67x86_64 y y y y y[13] y y y y
68riscv32 y y y y b y n[2] ? y
69riscv64 y y y y b y y y y
70------------ ----- ------- ------ --------------- ------------ ---- ---
71architecture tools kernels libgcc native-gcc/sani make release runs atf
72
73[2]: riscv32 kernels are not really implemented yet, not GCC 12 issue.
74[3]: GENERIC.SKI boots just as well as GCC 10.
75[13]: on amd64, with very basic test, lsan, ubsan, and asan work, but tsan crashes at start up:
76 ThreadSanitizer: CHECK failed: tsan_rtl.cpp:149 "((reinterpret_cast<uptr>(this) % 64)) == ((0))" (0x10, 0x0) (tid=21499)
77 ThreadSanitizer:DEADLYSIGNAL
78 ==21499==ERROR: ThreadSanitizer: SEGV on unknown address 0x000000000017 (pc 0x7f7ff4e5e14e bp 0x7f7fffffdbd0 sp 0x7f7fffffdb40 T21499)
79 ==21499==The signal is caused by a READ memory access.
80 ==21499==Hint: address points to the zero page.
81 ThreadSanitizer:DEADLYSIGNAL
82 ThreadSanitizer: nested bug in the same thread, aborting.
83[14]: "dd count=1", "env LC_CTYPE=en_US.UTF-8 locale", and "tests/crypto/libcrypto/h_evp_test evpmd_blake.txt" abort.
84 adding -fno-stack-protector to strsuftoll.c, citrus_module.c, and blake2_prov.c, respectively, works around the problem.
85 -- workarounds commited.
86[16]: sanitizers crash early:
87 Program received signal SIGSEGV, Segmentation fault.
88 (gdb) bt
89 #0 0x000000004064265c in __sanitizer::CheckASLR () at /usr/src/external/gpl3/gcc/dist/libsanitizer/sanitizer_common/sanitizer_linux.cpp:2181
90 #1 0xffffffffffffc704 in ?? ()
91 Backtrace stopped: previous frame identical to this frame (corrupt stack?)
92 -- some wip patches make this less bad
93[17]: (NOT A REGRESSION) libtsan is newly introduced, which fails with:
94 FATAL: ThreadSanitizer: unexpected memory mapping 0x200100000-0x200101000
95[23]: sanitizers fail to build:
96 sanitizer_flag_parser.h:141:72: error: format '%zx' expects argument of type 'size_t', but argument 4 has type 'unsigned int' [-Werror=format=]
97[24]: (NOT A REGRESSION) c++ exception handling is broken. mechanically applying
98 http://www.execsw.org/netbsd/changeset.cgi?id=20140318T234341Z.c59ac4642e7d4b345f6dc5aa9887e9e1f9aa459a#src/gnu/dist/gcc4/gcc/config/vax/vax.c
99 does not help (with adjustments for gcc 10.5.0, which is broken, too):
100 (gdb) invalid-command
101 [1] Illegal instruction (core dumped) gdb
102[25]: (maybe not related to [24]) native gcc is broken:
103 simh# cc hello.c
104 during RTL pass: final
105 hello.c: In function 'main':
106 hello.c:9:1: internal compiler error: Segmentation fault
107[26]: divide by zero with ubsan hits CheckLocked assertion failure, and stalls at LockSlow indefinitely:
108 (gdb) bt
109 #0 __sanitizer::StaticSpinMutex::LockSlow at sanitizer_mutex.cpp:24
110 ...
111 #17 __ubsan::ScopedReport::CheckLocked ubsan_diag.h:255
112 #18 __ubsan::Diag::~Diag ubsan_diag.cpp:347
113 #19 handleDivremOverflowImpl ubsan_diag.h:210
114 #20 __ubsan::__ubsan_handle_divrem_overflow ubsan_handlers.cpp:320
115
116
117CPU vs platform test table (for CPUs with multiple ports). this is "make release" or just kernels.
118values: y (yes), k (kernels only), n (failed), r (running), ? (not attempted), nx (yes, no x11), nk (no kernels available)
119
120CPU platform list
121--- -------- ----
122 acorn32 cats epoc32 evbarm-eb evbarm-el hpcarm iyonix netwinder shark zaurus
123earmv4: k k k k k k k y
124earm: nx nx k k
125earmhf: k k
126earmv6: k k
127earmv6hf: nx nx
128earmv7: k k
129earmv7hf: nx nx
130
131 amiga atari cesfic hp300 luna68k mac68k mvme68k news68k next68k sun3 x68k
132m68k: nx k k k k nx k k k nx k
133
134 evbmips emips ews4800mips mipsco newsmips sgimips
135mipseb: k k k k k y
136
137 evbmips algor arc cobalt hpcmips pmax
138mipsel: k k k k k y
139
140 algor evbmips pmax
141mips64eb: y
142mips64el: k y y
143
144 amigappc bebox evbppc ibmnws macppc mvmeppc ofppc prep rs6000 sandpoint
145powerpc: k k k k y k k k k k
146
147 evbppc macppc ofppc
148powerpc64: nk nk nk
149
150 dreamcast evbsh3 hpcsh landisk mmeye
151sh3eb: y k
152sh3el: k y k k
153
README.warnings
1$NetBSD: README.warnings,v 1.5 2023/08/08 11:50:22 riastradh Exp $
2
3What to do about GCC warnings and NetBSD.
4
5
6New GCC releases always come with a host of new warnings and
7each new warning can find real bugs, find odd code, or simply
8be a pain of new useless warnings, or all three and more.
9
10As each warning has its own set of issues they each have their
11own section and it is expected that this document will be
12modified for updates to warnings and new warnings.
13
14
15<bsd.own.mk> provides several variables for use in Makefiles:
16 COPTS.foo.c += ${CC_WNO_FORMAT_TRUNCATION}
17 COPTS.foo.c += ${CC_WNO_FORMAT_OVERFLOW}
18 COPTS.foo.c += ${CC_WNO_STRINGOP_OVERFLOW}
19 COPTS.foo.c += ${CC_WNO_STRINGOP_TRUNCATION}
20 COPTS.foo.c += ${CC_WNO_CAST_FUNCTION_TYPE}
21 COPTS.foo.c += ${CC_WNO_IMPLICIT_FALLTHROUGH}
22 COPTS.foo.c += ${CC_WNO_ADDRESS_OF_PACKED_MEMBER}
23 COPTS.foo.c += ${CC_WNO_MAYBE_UNINITIALIZED}
24 COPTS.foo.c += ${CC_WNO_RETURN_LOCAL_ADDR}
25 COPTS.foo.c += ${CC_WNO_MISSING_TEMPLATE_KEYWORD}
26 COPTS.foo.c += ${CC_WNO_STRINGOP_OVERREAD}
27 COPTS.foo.c += ${CC_WNO_REGISTER}
28 COPTS.foo.c += ${CC_WNO_ARRAY_BOUNDS}
29
30
31new GCC 12 warnings:
32
33 -Wno-missing-template-keyword
34
35 This warning trips on older C++ code, and should only be applyed to 3rd
36 party code.
37
38 bsd.own.mk variable: ${CC_WNO_MISSING_TEMPLATE_KEYWORD}
39
40 -Wno-stringop-overread
41
42 This warning triggers when array bounds appear to be exceeded. There
43 maybe some bugs related to this warning in GCC 12.
44
45 bsd.own.mk variable: ${CC_WNO_STRINGOP_OVERREAD}
46
47 -Wno-register
48
49 This warning triggers in C++17 mode where 'register' has been removed,
50 and should only be applied to 3rd party code.
51
52 bsd.own.mk variable: ${CC_WNO_REGISTER}
53
54 -Wno-array-bounds
55
56 This warning triggers with a number of code issues that tend to be real
57 problem but require careful adjustments to fix properly, when there are
58 platform related accesses beyond the immediately size and address of
59 known variables (real bugs to fix), but also may trigger when passing
60 C arrays vs C pointers to functions, often incorrectly. See
61 https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110878
62
63 bsd.own.mk variable: ${CC_WNO_ARRAY_BOUNDS}
64
65
66new GCC 10 warnings:
67
68 GCC 10 switched the default from "-fcommon" to "-fno-common",
69 which can cause multiply defined symbol issues. Ideally we
70 fix all of these, but "-fcommon" can be used otherwise.
71
72 -Wno-maybe-uninitialized
73
74 This warning was introduced in an ancient GCC but was
75 significantly enhanced in GCC 10, unfortunately, many of
76 the new instances are incorrect.
77
78 bsd.own.mk variable: ${CC_WNO_MAYBE_UNINITIALIZED}
79
80 -Wno-return-local-addr
81
82 This warning was introduced in GCC 5 and was enhanced in GCC
83 10. Unfortunately, the new instances are failing to correctly
84 analyze code flow and miss that most of the are handled.
85
86 bsd.own.mk variable: ${CC_WNO_RETURN_LOCAL_ADDR}
87
88
89new GCC 9 warnings:
90
91 -Wno-address-of-packed-member
92
93 This warning was introduced in GCC 8.
94 This warning is similar to -Wformat-truncation, but for the
95 general family of string functions (str*(), etc.), and has
96 similar issues of false positives.
97
98 bsd.own.mk variable: ${CC_WNO_ADDRESS_OF_PACKED_MEMBER}
99
100
101new GCC 8 warnings:
102
103 -Wstringop-truncation
104
105 This warning was introduced in GCC 8.
106 This warning is similar to -Wformat-truncation, but for the
107 general family of string functions (str*(), etc.), and has
108 similar issues of false positives.
109
110 bsd.own.mk variable: ${CC_WNO_STRINGOP_TRUNCATION}
111
112
113 -Wcast-function-type
114
115 This warning was introduced in GCC 8.
116 This warning can find real problems. Most instances are
117 false positives, and hopefully this warning will become
118 more useful in the future. See __FPTRCAST().
119
120 bsd.own.mk variable: ${CC_WNO_CAST_FUNCTION_TYPE}
121
122
123new GCC 7 warnings:
124
125 -Wstringop-overflow
126
127 This warning was introduced in GCC 7.
128 This warning can find issues where source length is
129 passed as destination length (eg, strncpy() where the
130 length is strlen(src)) that are potential buffer overflow
131 cases and should always be inspected, but false positives
132 are also seen.
133
134 bsd.own.mk variable: ${CC_WNO_STRINGOP_OVERFLOW}
135
136
137 -Wformat-truncation
138
139 This warning was introduced in GCC 7.
140 This warning has many false positives where truncation is
141 either expected or unavoidable, but also finds several real
142 code bugs.
143
144 Code should always be manually inspected for this warning
145 as it does pick up real issues.
146
147 bsd.own.mk variable: ${CC_WNO_FORMAT_TRUNCATION}
148
149
150 -Wformat-overflow
151
152 This warning was introduced in GCC 7.
153 This warning typically identifies a real problem, but it may
154 fail to notice the code handles this case.
155
156 Code should always be manually inspected for this warning
157 as it does pick up real issues.
158
159 bsd.own.mk variable: ${CC_WNO_FORMAT_OVERFLOW}
160
161
162 -Wimplicit-fallthrough
163
164 This warning was introduced in GCC 7.
165 This warning has many false positives in GCC 7, and many are
166 fixed in newer GCC 10. Old uses should be checked occasionally.
167
168 Code should always be manually inspected for this warning
169 as it does pick up real issues.
170
171 bsd.own.mk variable: ${CC_WNO_IMPLICIT_FALLTHROUGH}
172