HACKS revision 1.57
1# $NetBSD: HACKS,v 1.57 2006/05/10 21:54:58 mrg Exp $ 2# 3# This file is intended to document workarounds for currently unsolved 4# (mostly) compiler bugs. 5# 6# Format: 7# hack title 8# cdate creation date 9# mdate mod date 10# who responsible developer 11# port ... 12# affected ports, space separated, if not "all" 13# file affected file : revision : line from : line to 14# affected files and revision and line numbers describing hack 15# multiple lines if necessary. 16# pr NNNN ... 17# problem reports this hack works around, if known. Space 18# separated. 19# regress src/regress/directory/where/test/found 20# regression test directories, if available. 21# descr 22# insert short informal description (multi-line). (Longer ones 23# should be in the PR database. More formal descriptions might 24# be in the regress tree. See above). 25# kcah 26# closing bracket. 27# 28# this is a comment. 29 30hack gcc4 wrong uninitialised variable 31mdate 10 May 2006 32who mrg 33file bin/ksh/eval.c : 1.6 34file bin/sh/histedit.c : 1.39 35file bin/sh/parser.c : 1.60 36file bin/systrace/systrace.c : 1.33 37file crypto/dist/heimdal/kdc/524.c : 1.10 38file crypto/dist/ssh/sftp.c : 1.20 39file crypto/dist/ssh/ssh-keysign.c : 1.11 40file dist/ipf/lib/hostname.c : 1.2 41file dist/ipf/tools/ipmon.c : 1.8 42file dist/ntp/ntpd/refclock_shm.c : 1.4 43file dist/ntp/sntp/timing.c : 1.3 44file dist/pppd/pppstats/pppstats.c : 1.3 45file dist/smbfs/lib/smb/rap.c : 1.6 46file dist/tcpdump/print-zephyr.c : 1.5 47file gnu/libexec/uucp/uucico/uucico.c : 1.6 48file lib/libc/citrus/citrus_csmapper.c : 1.6 49file lib/libc/citrus/citrus_pivot_factory.c : 1.5 50file lib/libc/inet/inet_cidr_ntop.c : 1.3 51file lib/libc/inet/inet_ntop.c : 1.3 52file lib/libc/stdio/vfwprintf.c : 1.8 53file libexec/ld.elf_so/arch/sparc64/mdreloc.c : 1.39 54file sbin/fsck_ext2fs/dir.c : 1.19 55file sbin/routed/rtquery/rtquery.c : 1.18 56file sys/crypto/cast128/cast128.c : 1.9 57file sys/ddb/db_command.c : 1.86 58file sys/dev/fss.c : 1.25 59file sys/dev/ic/midway.c : 1.71 60file sys/dev/pci/machfb.c : 1.45 61file sys/dist/ipf/netinet/ip_nat.c : 1.10 62file sys/dist/ipf/netinet/ip_rpcb_pxy.c : 1.8 63file sys/kern/kern_sig.c : 1.219 64file sys/kern/tty.c : 1.181 65file sys/net/bpf.c : 1.116 66file sys/net/zlib.c : 1.26 67file sys/netinet/in.c : 1.107 68file sys/netsmb/smb_smb.c : 1.27 69file sys/netsmb/smb_trantcp.c : 1.22 70file sys/nfs/nfs_serv.c : 1.108 71file sys/nfs/nfs_socket.c : 1.129 72file sys/nfs/nfs_syscalls.c : 1.91 73file sys/ufs/lfs/lfs_vfsops.c : 1.207 [also (char *)] 74file usr.bin/ftp/ftp.c : 1.140 75file usr.bin/find/function.c : 1.54 76file usr.bin/mail/tty.c : 1.20 77file usr.bin/nl/nl.c : 1.7 78file usr.bin/systat/keyboard.c : 1.23 79file usr.bin/usbhidctl/usbhid.c : 1.29 80file usr.bin/vi/cl/cl_read.c : 1.5 81file usr.bin/vi/ex/ex_cscope.c : 1.12 82file usr.bin/vi/ex/ex_tag.c : 1.19 83file usr.bin/vi/vi/v_txt.c : 1.15 84file usr.sbin/altq/altqstat/qdisc_rio.c : 1.4 85file usr.sbin/cron/do_command.c : 1.19 86file usr.sbin/timed/timed/slave.c : 1.15 87descr 88 GCC 4.1 gets many uninitialised variable warnings wrong. We should 89 really audit all the old hacks like this when older compilers are 90 removed from the tree, as many are probably no longer required. 91kcah 92 93hack netstat ieee1394 address printing. 94mdate 14 Nov 2000 95who matt 96file lib/libc/net/getnameinfo.c : 1.32 : 497 : 503 97descr 98 Because the current implementation of IP over IEEE1394, the 99 fw device address contains more than just the IEEE1394 EUI-64. 100 So when printing out IEEE1394 addresses, ignore the extra stuff. 101kcah 102 103hack xterm vs. libterm 104mdate 01 Aug 2000 105who jdc 106file xsrc/xc/programs/xterm/main.c : 1.2 : 3609 : 3614 107pr 10383 108descr 109 In order to extend the termcap string over 1023 bytes, a ZZ entry was 110 introduced to point to a memory location containing the full entry. 111 Without this hack, xterm will export a termcap containing the ZZ 112 entry, which will then be ignored by libterm. As xterm modifies the 113 exported termcap, this would cause those modifications to be ignored. 114kcah 115 116hack wi-at-big-endian-bus 117cdate 15 Mar 2002 118who martin 119file dev/ic/wireg.h 120descr Add an option to access the underlying bus in big endian byte order 121 to work around deficiencies in bus_space_{read,write}_* macros. 122 Those don't allow the implementation of a proper pcmcia bus space 123 tag. 124kcah 125 126hack specific knowledge of colours in curses code 127cdate Sun Apr 6 11:05:24 BST 2003 128who jdc 129file lib/libcurses/color.c : r1.24 130descr 131 Swap red/blue and yellow/cyan colours for COLOR_OTHER. 132 Fix is to enhance libtermcap to understand terminfo-style % sequences. 133 See also: 134 http://mail-index.NetBSD.org/tech-userlevel/2003/04/06/0000.html 135kcah 136 137hack Compensation for differing types of LINUX_USRSTACK and USRSTACK 138cdate 21 Aug 2003 139who he 140file sys/miscfs/procfs/procfs_linux.c : 1.14 141descr 142 Not all ports have LINUX_USRSTACK and/or USRSTACK defined as 143 literals/constants, but refer to variables of a type which is 144 not "unsigned long", causing compilation of procfs_linux.c to 145 fail with "makes integer from pointer without a cast". This 146 is observed on e.g. the sun3 port. Ideally the "types" for 147 symbols should be consistent across all ports. 148kcah 149 150hack GCC2 isn't C99 151cdate 5 December 2003 152who mrg 153file gnu/dist/binutils/ld/ldlex.c: r1.2 154descr 155 ldlex.c is generated from ldlex.l and we get macro redefinition 156 errors from it if __STDC_VERSION__ is not C99 compliant. This 157 hack forces NetBSD 1.6 and later systems to also include <stdint.h> 158 to get the relevant definitions. This hack should be backed out 159 when all platforms have switched away from GCC2. 160kcah 161 162hack gcc-strict-aliasing 163cdate 6 August 2004 164who junyoung 165pr 26516 166file src/usr.sbin/racoon/racoon/Makefile : 1.21 167descr 168 GCC issues a warning "dereferencing type-punned pointer will break 169 strict-aliasing rules" when compiling 170 crypto/dist/kame/racoon/crypto/openssl.c with -O[23s]. Note that this 171 is due to use of type casting in a non-standard way in the code and 172 not a compiler bug. Once a new release of KAME with a proper fix 173 applied is imported this should be backed out. 174kcah 175 176 177port vax 178 179 hack gcc 2.95/vax doesn't like abort being used as function pointer 180 cdate Tue Dec 13 05:54:50 GMT 2005 181 who jmc 182 file lib/libc/rpc/svc_vc.c 1.15 183 descr 184 Provide a local definition for abort which doesn't include 185 the noreturn attribute which trips up gcc 2.95 on vax. 186 187 hack gcc 2.95/vax doesn't have stdbool.h and gettext needs it 188 cdate Mon May 9 12:16:22 CDT 2005 189 who jmc 190 file gnu/usr.bin/gettext/Makefile.inc.prog 1.5 191 gnu/usr.bin/gettext/libgrep/Makefile 1.3 192 gnu/usr.bin/gettext/libnlspr/Makefile 1.5 193 gnu/usr.bin/gettext/libnlsut/Makefile 1.3 194 descr 195 The latest gettext assumes a C99 environment or at least 196 configure checks to work around this. As we don't use configure 197 provide a stdbool.h for vax builds locally in directories from 198 the template provided by the gettext code. 199 200 hack gcc 2.95/vax can't handle __used__ in mcount.c 201 cdate Fri May 6 14:30:53 CDT 2005 202 who jmc 203 file src/lib/libc/gmon/mcount.c 1.17 204 descr 205 Recent changes to mark this as __used__ to prevent optimization 206 away on sparc64 don't work on 2.95 vax. Ifdef'd back in the 207 old __unused__ if on vax. 208 209 hack gcc 2.95/vax cannot handle __builtin_ffs() 210 cdate Fri Aug 23 21:31:15 CEST 2002 211 who ragge 212 file sys/lib/libkern/libkern.h : 1.42 213 descr 214 __builtin_ffs() is ifdef'd away if __vax__. Matt Thomas has 215 added this feature to GCC 3 so it can be removed when switching. 216 217 218 hack gcc-vax-libbz2 219 mdate 27 Jun 2002 220 who thorpej 221 file lib/libbz2/Makefile 222 descr 223 libbz2 is mis-compiled with optimization with GCC 2.95.3 224 on VAX. -O0 works around this problem. 225 kcah 226 227 hack gawk 228 mdate 8 Nov 2003 229 who ragge 230 file gnu/usr.bin/gawk/Makefile 231 descr 232 gawk is mis-compiled with optimization with GCC 2.95.3 233 on VAX. -O0 works around this problem. 234 kcah 235 236 hack vax Toolchain bug 237 cdate 28 Mar 2003 238 who he 239 file sys/arch/vax/vax/intvec.S : 1.5 240 descr 241 Workaround for PR toolchain/20924. The assembler apparently 242 tries to range-check byte offsets when it doesn't have 243 sufficient information to make that decision. 244 Workaround uses `brw' instead of `brb' instruction. 245 kcah 246 247 hack vax gcc 2.95.3 -Wuninitialized workarounds 248 cdate 6 Nov 2003 249 who he 250 file sys/dev/bi/if_ni.c : 1.22 251 file sys/arch/vax/vax/ctu.c : 1.20 252 descr 253 Workaround for gcc 2.95.3 failing to detect that certain 254 variables will be initialized; so that this code compiles 255 with -Wuninitialized. Should be reviewed and possibly 256 reverted when gcc 3.3.2 is ready for vax. 257 kcah 258 259 hack vax gcc 2.95.3 structure initialization 260 cdat 13 May 2004 261 who cl 262 file sys/miscfs/kernfs/kernfs_vnops.c : 1.104 263 descr 264 Workaround for gcc 2.95.3 failing to initialize structures 265 and/or unions inside structures using nested designators. 266 Should be reverted when gcc >=3.3.3 is ready for vax. 267 kcah 268 269 hack vax gcc 2.95.3 needs -I. to build nslexer.c 270 cdat 6 Jun 2004 271 who mhitch 272 file lib/libc/arch/vax/Makefile.inc : 1.6 273 descr 274 It appears necessary to add -I. to CPPFLAGS when building 275 nslesxer.c. This may be caused by gcc 2.95.3 being used. 276 Should be reverted when a newer gcc is ready for vax. 277 kcah 278 279 hack vax gcc 2.95.3 doesn't understand "-std=c99" 280 cdat 5 April 2005 281 who tron 282 file src/share/mk/bsd.sys.mk 283 descr 284 We cannot use "-std=c99" for compiler warning level 4 and above 285 because gcc 2.95.3 doesn't support that option. 286 kcah 287 288 hack declare boolean_t in two IPF user-mode programs 289 cdate Tue Mar 7 19:19:20 CET 2006 290 who he 291 file dist/ipf/ipsend/iptests.c : 1.8 292 dist/ipf/ipsend/sock.c : 1.7 293 descr 294 The IPF user-mode programs ipsend and iptest first 295 include <sys/types.h> without _KERNEL defined, and 296 later include <sys/file.h> with _KERNEL defined. 297 This causes a build failure when building for vax, 298 since <sys/device.h> ends up being included without 299 bollean_t being defined by <sys/types.h>. 300 Build failure and further details documented in 301 PR#32907. 302 kcah 303 304port sparc64 305 306 hack binutil-2.11-sparc64-pltrela 307 mdate 14 Aug 2001 308 who eeh 309 file libexec/ld.elf_so/arch/sparc64/mdreloc.c 310 descr 311 The first four PLT entries are reserved. There is some 312 disagreement whether they should have associated relocation 313 entries. Both the SPARC 32-bit and 64-bit ELF specifications 314 say that they should have relocation entries, but the 32-bit 315 SPARC binutils do not generate them, and now the 64-bit SPARC 316 binutils have stopped generating them too. 317 318 To provide binary compatibility, we will check the first entry, 319 if it is reserved it should not be of the type JMP_SLOT. If it 320 is JMP_SLOT, then the 4 reserved entries were not generated and 321 our index is 4 entries too far, so we frob the rela pointer. 322 kcah 323 324 325port arm 326 327 hack gcc-unsigned-compare 328 cdate 09 Mar 2002 329 mdate 18 Mar 2002 330 who bjh21 331 file dist/bind/lib/nameser/ns_parse.c : 1.3 332 file dist/dhcp/minires/ns_parse.c : 1.3 333 file dist/dhcp/omapip/result.c : 1.2 334 file dist/dhcp/server/failover.c : 1.3 335 file gnu/dist/gawk/eval.c : 1.4 336 file gnu/dist/toolchain/bfd/bfd.c : 1.2 337 file gnu/dist/toolchain/bfd/format.c : 1.2 338 file gnu/dist/toolchain/gdb/target.c : 1.2 339 file sys/kern/vfs_subr.c : 1.172 340 descr When checking that a potentially-unsigned enum is >= 0, assign 341 it to an int first. This is necessary to avoid "comparison is 342 always true" warnings with -fshort-enums. Casting to an int 343 really should be enough, but turns out not to be. 344 kcah 345 346 hack gcc 3.3.2/arm sys/kern/uipc_socket.c 347 cdate Tue Oct 28 18:02:16 GMT 2003 348 who skrll 349 pr 23044 350 file sys/arch/arm/conf/Makefile.arm : 1.19 351 descr 352 Performing a build.sh tools on a CATS with NFS mounted 353 obj and src directories results in "panic: receive 1" 354 The hack is to compile sys/kern/uipc_socket.c with 355 -fno-strict-aliasing 356 kcah 357 358 359port pc532 360 361 hack egcs-pc532-ip6_mroute 362 cdate 09 Jul 2002 363 who simonb 364 file sys/arch/pc532/conf/Makefile.pc532 : 1.70 365 file sys/arch/pc532/conf/files.pc532 : 1.47 366 descr 367 egcs 1.1.2 gets an "internal error--insn does not satisfy its 368 constraints" error compiling ip6_mroute.c with -O2 or greater. 369 -O1 works around this problem. 370 kcah 371 372 hack gcc-pc532-duffs_device 373 cdate Mon Oct 27 07:23:05 UTC 2003 374 who simonb 375 file sys/arch/pc532/conf/Makefile.pc532 : 1.71 376 file sys/arch/pc532/conf/files.pc532 : 1.52 377 pr GCC PR optimization/5230 378 descr 379 gcc incorrectly gives an "unreachable code at beginning of 380 switch statement" for a Duff's device construct in 381 arch/pc532/dev/ncr.c. There is no way to disable just this 382 warning, so -Wno-error is turned on for this file. 383 kcah 384 385 hack gcc332-cppmacro-ice 386 cdate Sun Dec 7 12:48:33 UTC 2003 387 who simonb 388 file gnu/usr.bin/gcc3/libcpp/Makefile : 1.2 389 descr 390 gcc 3.3.2 gets an ICE when compiling cppmacro.c with -O2. 391 The hack is to use -O1 for this file. Fixed in gcc 392 -current mainline, unable to work out where the fix is 393 right now. 394 kcah 395 396 397port sh3 398 399 hack gcc-sh3-sed 400 mdate 23 Apr 2002 401 who thorpej 402 file usr.bin/sed/Makefile : 1.9 : 9 : 13 403 descr 404 The in-tree GCC 2.95.3-based compiler ICEs when building 405 with optimization for SuperH. Hack is to build with -O0. 406 kcah 407 408 409port sh5 410 411 hack SuperH SH5 Toolchain Bugs 412 cdate 11 Jul 2002 413 who scw 414 file usr.sbin/ndbootd/ndbootd.c : 1.6 415 file usr.sbin/traceroute/traceroute.c : 1.48 416 descr 417 The SuperH SH5 toolchain (2.97-sh5-010522) gets an internal 418 compiler error when assigning a bit-wise inverted value 419 under some circumstances. 420 Work around it by splitting the statement into two. 421 kcah 422 423 hack SuperH SH5 Toolchain Bugs 424 cdate 17 May 2003 425 who scw 426 file lib/libpthread/pthread_lock.c : 1.7 427 descr 428 The SuperH SH5 toolchain generates incorrect PIC code when 429 faced with a symbol which is declared extern, but has local 430 scope due to being defined within an asm() statement (without 431 being declared .globl in that statement). Work around it by 432 adding the .globl. 433 kcah 434 435 hack gcc 3.3.2 on sh5 - ICE compiling reload1.c at -O2 436 cdate Sun Dec 7 19:44:05 GMT 2003 437 who scw 438 file gnu/usr.bin/gcc3/backend/Makefile : 1.13 439 descr 440 Gcc3 for sh5 gets an ICE compiling reload1.c at -O2. 441 Drop to -O1 for now when building for sh5. 442 kcah 443 444 hack gcc 3.3.2 on sh5 - ICE compiling elf32.c/elf64.c at -O2 445 cdate Sun Dec 7 21:10:46 GMT 2003 446 who scw 447 file gnu/lib/libbfd/Makefile : 1.39 448 descr 449 Gcc3 for sh5 gets an ICE compiling libbfd's elf32.c 450 and elf64.c at -O2. 451 Drop to -O1 for now when building for sh5. 452 kcah 453 454 455 hack gcc 3.3.2 on sh5 - fatal warning compiling function.c 456 cdate Sun Dec 7 19:44:05 GMT 2003 457 who scw 458 file gnu/usr.bin/gcc3/backend/Makefile : 1.13 459 descr 460 Gcc3 for sh5 issues a warning: 461 .../gnu/dist/gcc/gcc/function.c:4742: warning: `and' of 462 mutually exclusive equal-tests is always 0. 463 This causes the build to fail because of -Werror. 464 Work-around by defining NOGCCERROR when building for sh5. 465 kcah 466 467 hack gcc 3.3.3 on sh5 - ICE compiling rbtdb.c and rbtdb64.c 468 cdate Thu Nov 11 00:35:33 UTC 2004 469 who he 470 file usr.sbin/bind/libdns/Makefile/1.2 471 descr 472 Gcc3 for sh5 gets an ICE compiling these files at -O2. 473 Drop to -O1 for now when building for sh5. 474 kcah 475 476 hack gcc 3.3.3 nb3 on sh5 - ICE compiling bsd-comp.c 477 cdate Thu May 25 23:05:20 UTC 2005 478 who he 479 file sys/lkm/net/bsdcomp/Makefile : 1.3 480 descr 481 Gcc3 for sh5 gets an ICE compiling this at -O2, 482 "could not split insn". Drop to -O1 for sh5. 483 kcah 484 485 hack gcc 3.3.3 nb3 on sh5 - ICE compiling strtod.c 486 cdate Tue Jan 31 10:51:52 GMT 2006 487 who scw 488 file lib/libc/gdtoa/Makefile.inc : 1.2 489 descr 490 Gcc3 for sh5 gets an ICE compiling this at -O2, 491 "could not split insn". Drop to -O1 for sh5. 492 kcah 493 494port sun2 495 496 hack gcc 3.3.2/mdsetimage 497 cdate Tue Oct 28 18:43:05 EST 2003 498 who fredette 499 file gnu/usr.sbin/mdsetimage/Makefile : 1.15 500 descr 501 mdsetimage.c causes cc1 to barf for as-yet unknown reasons. 502 this hack uses -O0 to avoid the problem. 503 kcah 504 505 hack gcc 3.3.2/installboot/ffs.c 506 cdate Tue Oct 28 18:43:05 EST 2003 507 who fredette 508 file usr.sbin/installboot/Makefile : 1.26 509 descr 510 ffs.c causes cc1 to barf for as-yet unknown reasons. this 511 hack uses -O0 to avoid the problem. 512 kcah 513 514 515port m68000 516 517 hack gcc 3.3.2/gcc/ra-colorize.c 518 cdate Wed Nov 5 11:03:13 EST 2003 519 who mrg 520 file gnu/usr.bin/gcc3/backend/Makefile : 1.11 521 descr 522 ra-colorize.c causes a whole spew of unreferenced .L* local 523 symbols. using -O0 (or even -O2 reportedly) avoids the problem. 524 Note this has been fixed in 3.3.3 at least on m68k, 525 but it hasn't been confirmed on sun2/m68000. 526 (it's too slow even on emulator and unlikely to build 527 toolchains on sun2 with its maximum 4MB RAM) 528 kcah 529 530port sparc64 531 532 hack disable optimzations for uvm_bio.c on 32 bit kernels 533 cdate Sun Mar 21 14:14:04 MET 2004 534 who martin 535 file sys/arch/sparc64/conf/Makefile.sparc64 : 1.53-1.54 536 file sys/arch/sparc64/conf/files.sparc64 : 1.93 537 descr 538 The sparc compiler miscompiles uvm_bio.c when using 539 any optimization. This results in ubc_release panics. 540 kcah 541 542port macppc 543 544 hack remove inline from rotate_{left,right} 545 cdate Tue Feb 22 13:43:25 EST 2005 546 who christos 547 file src/dist/bind/lib/dns/rbt.c 548 descr 549 The macppc gcc-3.3 compiler miscompiles rbt.c when these 550 functions are inlined. The error is: 551 .* named[368]: rbt.c:1584: INSIST(child != (void *)0) >failed 552 .* named[368]: exiting (due to assertion failure) 553 This is reportedly fixed in gcc-3.4 554 kcah 555 556port hppa 557 558 hack hppa-gcc-3.3.3/grep/dfa.c 559 cdate Sun Aug 8 08:34:32 PDT 2004 560 who chs 561 file gnu/usr.bin/grep/grep/Makefile : 1.4 562 descr 563 gcc 3.3.3 generates bad code for dfa.c with -O2, 564 reduce optimization to -O1 to avoid the bug. 565 kcah 566 567 hack hppa-gcc-3.3.3 / make/buf.c 568 cdate Mon May 8 13:02:36 CEST 2006 569 who he 570 file usr.bin/make/Makefile : 1.34 571 descr 572 gcc 3.3.3 nb3 gets an "unrecognized insn" internal 573 compiler error when building this file with -O2, 574 reduce optimization to -O1 to avoid the bug. 575 kcah 576 577 hack hppa boot-from-disk memory corruption 578 cdate Wed May 18 06:56:07 PDT 2005 579 who chs 580 file src/sys/dev/scsipi/scsipiconf.h : 1.98 581 descr 582 There is some bug with DMA on hppa that corrupts scsipi_xfer 583 structures, but it only shows up when booting from disk. 584 For now, we'll add some padding to scsipi_xfer so that 585 the corrupted memory is not otherwise used. 586 kcah 587 588port mips 589 590 hack mips-shared-linker-load-address 591 cdate Fri Oct 7 08:33:10 UTC 2005 592 who simonb 593 file src/sys/kern/exec_elf32.c : 1.107 594 descr 595 With COMPAT_16 or previous enabled (which enables 596 ELF_INTERP_NON_RELOCATABLE) a recent ld.elf_so will 597 load and run at address 0. The check to fix this in 598 rev 1.107 only checks the first psection of the ELF 599 executable, which may not be loadable. A more correct 600 fix is to check the first loadable psection instead of 601 just the first psection. 602 kcah 603 604