1. 21 3月, 2018 1 次提交
    • D
      kbuild: disable clang's default use of -fmerge-all-constants · 87e0d4f0
      Daniel Borkmann 提交于
      Prasad reported that he has seen crashes in BPF subsystem with netd
      on Android with arm64 in the form of (note, the taint is unrelated):
      
        [ 4134.721483] Unable to handle kernel paging request at virtual address 800000001
        [ 4134.820925] Mem abort info:
        [ 4134.901283]   Exception class = DABT (current EL), IL = 32 bits
        [ 4135.016736]   SET = 0, FnV = 0
        [ 4135.119820]   EA = 0, S1PTW = 0
        [ 4135.201431] Data abort info:
        [ 4135.301388]   ISV = 0, ISS = 0x00000021
        [ 4135.359599]   CM = 0, WnR = 0
        [ 4135.470873] user pgtable: 4k pages, 39-bit VAs, pgd = ffffffe39b946000
        [ 4135.499757] [0000000800000001] *pgd=0000000000000000, *pud=0000000000000000
        [ 4135.660725] Internal error: Oops: 96000021 [#1] PREEMPT SMP
        [ 4135.674610] Modules linked in:
        [ 4135.682883] CPU: 5 PID: 1260 Comm: netd Tainted: G S      W       4.14.19+ #1
        [ 4135.716188] task: ffffffe39f4aa380 task.stack: ffffff801d4e0000
        [ 4135.731599] PC is at bpf_prog_add+0x20/0x68
        [ 4135.741746] LR is at bpf_prog_inc+0x20/0x2c
        [ 4135.751788] pc : [<ffffff94ab7ad584>] lr : [<ffffff94ab7ad638>] pstate: 60400145
        [ 4135.769062] sp : ffffff801d4e3ce0
        [...]
        [ 4136.258315] Process netd (pid: 1260, stack limit = 0xffffff801d4e0000)
        [ 4136.273746] Call trace:
        [...]
        [ 4136.442494] 3ca0: ffffff94ab7ad584 0000000060400145 ffffffe3a01bf8f8 0000000000000006
        [ 4136.460936] 3cc0: 0000008000000000 ffffff94ab844204 ffffff801d4e3cf0 ffffff94ab7ad584
        [ 4136.479241] [<ffffff94ab7ad584>] bpf_prog_add+0x20/0x68
        [ 4136.491767] [<ffffff94ab7ad638>] bpf_prog_inc+0x20/0x2c
        [ 4136.504536] [<ffffff94ab7b5d08>] bpf_obj_get_user+0x204/0x22c
        [ 4136.518746] [<ffffff94ab7ade68>] SyS_bpf+0x5a8/0x1a88
      
      Android's netd was basically pinning the uid cookie BPF map in BPF
      fs (/sys/fs/bpf/traffic_cookie_uid_map) and later on retrieving it
      again resulting in above panic. Issue is that the map was wrongly
      identified as a prog! Above kernel was compiled with clang 4.0,
      and it turns out that clang decided to merge the bpf_prog_iops and
      bpf_map_iops into a single memory location, such that the two i_ops
      could then not be distinguished anymore.
      
      Reason for this miscompilation is that clang has the more aggressive
      -fmerge-all-constants enabled by default. In fact, clang source code
      has a comment about it in lib/AST/ExprConstant.cpp on why it is okay
      to do so:
      
        Pointers with different bases cannot represent the same object.
        (Note that clang defaults to -fmerge-all-constants, which can
        lead to inconsistent results for comparisons involving the address
        of a constant; this generally doesn't matter in practice.)
      
      The issue never appeared with gcc however, since gcc does not enable
      -fmerge-all-constants by default and even *explicitly* states in
      it's option description that using this flag results in non-conforming
      behavior, quote from man gcc:
      
        Languages like C or C++ require each variable, including multiple
        instances of the same variable in recursive calls, to have distinct
        locations, so using this option results in non-conforming behavior.
      
      There are also various clang bug reports open on that matter [1],
      where clang developers acknowledge the non-conforming behavior,
      and refer to disabling it with -fno-merge-all-constants. But even
      if this gets fixed in clang today, there are already users out there
      that triggered this. Thus, fix this issue by explicitly adding
      -fno-merge-all-constants to the kernel's Makefile to generically
      disable this optimization, since potentially other places in the
      kernel could subtly break as well.
      
      Note, there is also a flag called -fmerge-constants (not supported
      by clang), which is more conservative and only applies to strings
      and it's enabled in gcc's -O/-O2/-O3/-Os optimization levels. In
      gcc's code, the two flags -fmerge-{all-,}constants share the same
      variable internally, so when disabling it via -fno-merge-all-constants,
      then we really don't merge any const data (e.g. strings), and text
      size increases with gcc (14,927,214 -> 14,942,646 for vmlinux.o).
      
        $ gcc -fverbose-asm -O2 foo.c -S -o foo.S
          -> foo.S lists -fmerge-constants under options enabled
        $ gcc -fverbose-asm -O2 -fno-merge-all-constants foo.c -S -o foo.S
          -> foo.S doesn't list -fmerge-constants under options enabled
        $ gcc -fverbose-asm -O2 -fno-merge-all-constants -fmerge-constants foo.c -S -o foo.S
          -> foo.S lists -fmerge-constants under options enabled
      
      Thus, as a workaround we need to set both -fno-merge-all-constants
      *and* -fmerge-constants in the Makefile in order for text size to
      stay as is.
      
        [1] https://bugs.llvm.org/show_bug.cgi?id=18538Reported-by: NPrasad Sodagudi <psodagud@codeaurora.org>
      Signed-off-by: NDaniel Borkmann <daniel@iogearbox.net>
      Cc: Linus Torvalds <torvalds@linux-foundation.org>
      Cc: Chenbo Feng <fengc@google.com>
      Cc: Richard Smith <richard-llvm@metafoo.co.uk>
      Cc: Chandler Carruth <chandlerc@gmail.com>
      Cc: linux-kernel@vger.kernel.org
      Tested-by: NPrasad Sodagudi <psodagud@codeaurora.org>
      Acked-by: NAlexei Starovoitov <ast@kernel.org>
      Signed-off-by: NAlexei Starovoitov <ast@kernel.org>
      87e0d4f0
  2. 05 3月, 2018 1 次提交
  3. 02 3月, 2018 1 次提交
  4. 01 3月, 2018 2 次提交
    • L
      kbuild: disable sparse warnings about unknown attributes · 6c49f359
      Luc Van Oostenryck 提交于
      Currently, sparse issues warnings on code using an attribute
      it doesn't know about.
      
      One of the problem with this is that these warnings have no
      value for the developer, it's just noise for him. At best these
      warnings tell something about some deficiencies of sparse itself
      but not about a potential problem with code analyzed.
      
      A second problem with this is that sparse release are, alas,
      less frequent than new attributes are added to GCC.
      
      So, avoid the noise by asking sparse to not warn about
      attributes it doesn't know about.
      
      Reference: https://marc.info/?l=linux-sparse&m=151871600016790
      Reference: https://marc.info/?l=linux-sparse&m=151871725417322Signed-off-by: NLuc Van Oostenryck <luc.vanoostenryck@gmail.com>
      Acked-by: NRandy Dunlap <rdunlap@infradead.org>
      Tested-by: NRandy Dunlap <rdunlap@infradead.org>
      Signed-off-by: NMasahiro Yamada <yamada.masahiro@socionext.com>
      6c49f359
    • U
      Makefile: Fix lying comment re. silentoldconfig · 61277981
      Ulf Magnusson 提交于
      The comment above the silentoldconfig invocation is outdated.
      'make oldconfig' updates just .config and doesn't touch the
      include/config/ tree.
      
      This came up in https://lkml.org/lkml/2018/2/12/415.
      
      While fixing the comment, make it more informative by explaining the
      purpose of the unfortunately named silentoldconfig.
      
      I can't make sense of the comment re. auto.conf.cmd and a cleaned tree.
      include/config/auto.conf and include/config/auto.conf.cmd are both
      created simultaneously by silentoldconfig (in
      scripts/kconfig/confdata.c, by conf_write_autoconf()), and nothing seems
      to remove auto.conf.cmd that wouldn't remove auto.conf. Remove that part
      of the comment rather than blindly copying it. It might be a leftover
      from an older way of doing things.
      
      The include/config/auto.conf.cmd prerequisite might be there to ensure
      that silentoldconfig gets rerun if conf_write_autoconf() fails between
      writing out auto.conf.cmd and auto.conf (a comment in the function
      indicates that auto.conf is deliberately written out last to mark
      completion of the operation). It seems the Makefile dependency between
      include/config/auto.conf and .config would already take care of that
      though, since include/config/auto.conf would still be out of date re.
      .config if the operation fails.
      
      Cop out and leave the prerequisite in for now.
      Signed-off-by: NUlf Magnusson <ulfalizer@gmail.com>
      Signed-off-by: NMasahiro Yamada <yamada.masahiro@socionext.com>
      61277981
  5. 26 2月, 2018 1 次提交
  6. 21 2月, 2018 1 次提交
  7. 19 2月, 2018 1 次提交
  8. 12 2月, 2018 1 次提交
  9. 07 2月, 2018 5 次提交
  10. 29 1月, 2018 1 次提交
  11. 22 1月, 2018 1 次提交
  12. 15 1月, 2018 1 次提交
  13. 08 1月, 2018 1 次提交
  14. 01 1月, 2018 1 次提交
  15. 31 12月, 2017 1 次提交
    • L
      kbuild: add '-fno-stack-check' to kernel build options · 3ce120b1
      Linus Torvalds 提交于
      It appears that hardened gentoo enables "-fstack-check" by default for
      gcc.
      
      That doesn't work _at_all_ for the kernel, because the kernel stack
      doesn't act like a user stack at all: it's much smaller, and it doesn't
      auto-expand on use.  So the extra "probe one page below the stack" code
      generated by -fstack-check just breaks the kernel in horrible ways,
      causing infinite double faults etc.
      
      [ I have to say, that the particular code gcc generates looks very
        stupid even for user space where it works, but that's a separate
        issue.  ]
      Reported-and-tested-by: NAlexander Tsoy <alexander@tsoy.me>
      Reported-and-tested-by: NToralf Förster <toralf.foerster@gmx.de>
      Cc: stable@kernel.org
      Cc: Dave Hansen <dave.hansen@intel.com>
      Cc: Jiri Kosina <jikos@kernel.org>
      Cc: Andy Lutomirski <luto@amacapital.net>
      Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>
      3ce120b1
  16. 24 12月, 2017 1 次提交
  17. 18 12月, 2017 1 次提交
  18. 16 12月, 2017 1 次提交
  19. 11 12月, 2017 1 次提交
  20. 06 12月, 2017 1 次提交
  21. 04 12月, 2017 1 次提交
  22. 27 11月, 2017 1 次提交
  23. 23 11月, 2017 3 次提交
  24. 18 11月, 2017 3 次提交
    • M
      kbuild: create built-in.o automatically if parent directory wants it · f7adc312
      Masahiro Yamada 提交于
      "obj-y += foo/" syntax requires Kbuild to visit the "foo" subdirectory
      and link built-in.o from that directory.  This means foo/Makefile is
      responsible for creating built-in.o even if there is no object to
      link (in this case, built-in.o is an empty archive).
      
      We have had several fixups like commit 4b024242 ("kbuild: Fix
      linking error built-in.o no such file or directory"), then ended up
      with a complex condition as follows:
      
        ifneq ($(strip $(obj-y) $(obj-m) $(obj-) $(subdir-m) $(lib-target)),)
        builtin-target := $(obj)/built-in.o
        endif
      
      We still have more cases not covered by the above, so we need to add
        obj- := dummy.o
      in several places just for creating empty built-in.o.
      
      A key point is, the parent Makefile knows whether built-in.o is needed
      or not.  If a subdirectory needs to create built-in.o, its parent can
      tell the fact when descending.
      
      If non-empty $(need-builtin) flag is passed from the parent, built-in.o
      should be created.  $(obj-y) should be still checked to support the
      single target "%/".  All of ugly tricks will go away.
      Signed-off-by: NMasahiro Yamada <yamada.masahiro@socionext.com>
      Reviewed-by: NSam Ravnborg <sam@ravnborg.org>
      f7adc312
    • B
      kbuild: /bin/pwd -> pwd · 16f8259c
      Bjørn Forsman 提交于
      Most places use pwd and rely on $PATH lookup. Moving the remaining
      absolute path /bin/pwd users over for consistency.
      
      Also, a reason for doing /bin/pwd -> pwd instead of the other way around
      is because I believe build systems should make little assumptions on
      host filesystem layout. Case in point, we do this kind of patching
      already in NixOS.
      
      Ref. commit 028568d8
      ("kbuild: revert $(realpath ...) to $(shell cd ... && /bin/pwd)").
      Signed-off-by: NBjørn Forsman <bjorn.forsman@gmail.com>
      Signed-off-by: NMasahiro Yamada <yamada.masahiro@socionext.com>
      16f8259c
    • V
      Makefile: support flag -fsanitizer-coverage=trace-cmp · d677a4d6
      Victor Chibotaru 提交于
      The flag enables Clang instrumentation of comparison operations
      (currently not supported by GCC).  This instrumentation is needed by the
      new KCOV device to collect comparison operands.
      
      Link: http://lkml.kernel.org/r/20171011095459.70721-2-glider@google.comSigned-off-by: NVictor Chibotaru <tchibo@google.com>
      Signed-off-by: NAlexander Potapenko <glider@google.com>
      Cc: Dmitry Vyukov <dvyukov@google.com>
      Cc: Andrey Konovalov <andreyknvl@google.com>
      Cc: Mark Rutland <mark.rutland@arm.com>
      Cc: Alexander Popov <alex.popov@linux.com>
      Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
      Cc: Kees Cook <keescook@chromium.org>
      Cc: Vegard Nossum <vegard.nossum@oracle.com>
      Cc: Quentin Casasnovas <quentin.casasnovas@oracle.com>
      Cc: <syzkaller@googlegroups.com>
      Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
      Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>
      d677a4d6
  25. 16 11月, 2017 2 次提交
  26. 13 11月, 2017 4 次提交
    • M
      kbuild: do not call cc-option before KBUILD_CFLAGS initialization · 433dc2eb
      Masahiro Yamada 提交于
      Some $(call cc-option,...) are invoked very early, even before
      KBUILD_CFLAGS, etc. are initialized.
      
      The returned string from $(call cc-option,...) depends on
      KBUILD_CPPFLAGS, KBUILD_CFLAGS, and GCC_PLUGINS_CFLAGS.
      
      Since they are exported, they are not empty when the top Makefile
      is recursively invoked.
      
      The recursion occurs in several places.  For example, the top
      Makefile invokes itself for silentoldconfig.  "make tinyconfig",
      "make rpm-pkg" are the cases, too.
      
      In those cases, the second call of cc-option from the same line
      runs a different shell command due to non-pristine KBUILD_CFLAGS.
      
      To get the same result all the time, KBUILD_* and GCC_PLUGINS_CFLAGS
      must be initialized before any call of cc-option.  This avoids
      garbage data in the .cache.mk file.
      
      Move all calls of cc-option below the config targets because target
      compiler flags are unnecessary for Kconfig.
      Signed-off-by: NMasahiro Yamada <yamada.masahiro@socionext.com>
      Reviewed-by: NDouglas Anderson <dianders@chromium.org>
      433dc2eb
    • D
      kbuild: Cache a few more calls to the compiler · 4e562071
      Douglas Anderson 提交于
      These are a few stragglers that I left out of the original patch to
      cache calls to the C compiler ("kbuild: Add a cache for generated
      variables") because they bleed out into the main Makefile and thus
      uglify things a little bit.  The idea is the same here, though.
      Signed-off-by: NDouglas Anderson <dianders@chromium.org>
      Tested-by: NIngo Molnar <mingo@kernel.org>
      Tested-by: NGuenter Roeck <linux@roeck-us.net>
      Signed-off-by: NMasahiro Yamada <yamada.masahiro@socionext.com>
      4e562071
    • D
      kbuild: Add a cache for generated variables · 3298b690
      Douglas Anderson 提交于
      While timing a "no-op" build of the kernel (incrementally building the
      kernel even though nothing changed) in the Chrome OS build system I
      found that it was much slower than I expected.
      
      Digging into things a bit, I found that quite a bit of the time was
      spent invoking the C compiler even though we weren't actually building
      anything.  Currently in the Chrome OS build system the C compiler is
      called through a number of wrappers (one of which is written in
      python!) and can take upwards of 100 ms to invoke even if we're not
      doing anything difficult, so these invocations of the compiler were
      taking a lot of time.  Worse the invocations couldn't seem to take
      advantage of the multiple cores on my system.
      
      Certainly it seems like we could make the compiler invocations in the
      Chrome OS build system faster, but only to a point.  Inherently
      invoking a program as big as a C compiler is a fairly heavy
      operation.  Thus even if we can speed the compiler calls it made sense
      to track down what was happening.
      
      It turned out that all the compiler invocations were coming from
      usages like this in the kernel's Makefile:
      
      KBUILD_CFLAGS += $(call cc-option,-fno-delete-null-pointer-checks,)
      
      Due to the way cc-option and similar statements work the above
      contains an implicit call to the C compiler.  ...and due to the fact
      that we're storing the result in KBUILD_CFLAGS, a simply expanded
      variable, the call will happen every time the Makefile is parsed, even
      if there are no users of KBUILD_CFLAGS.
      
      Rather than redoing this computation every time, it makes a lot of
      sense to cache the result of all of the Makefile's compiler calls just
      like we do when we compile a ".c" file to a ".o" file.  Conceptually
      this is quite a simple idea.  ...and since the calls to invoke the
      compiler and similar tools are centrally located in the Kbuild.include
      file this doesn't even need to be super invasive.
      
      Implementing the cache in a simple-to-use and efficient way is not
      quite as simple as it first sounds, though.  To get maximum speed we
      really want the cache in a format that make can natively understand
      and make doesn't really have an ability to load/parse files. ...but
      make _can_ import other Makefiles, so the solution is to store the
      cache in Makefile format.  This requires coming up with a valid/unique
      Makefile variable name for each value to be cached, but that's
      solvable with some cleverness.
      
      After this change, we'll automatically create a ".cache.mk" file that
      will contain our cached variables.  We'll load this on each invocation
      of make and will avoid recomputing anything that's already in our
      cache.  The cache is stored in a format that it shouldn't need any
      invalidation since anything that might change should affect the "key"
      and any old cached value won't be used.
      Signed-off-by: NDouglas Anderson <dianders@chromium.org>
      Tested-by: NIngo Molnar <mingo@kernel.org>
      Tested-by: NGuenter Roeck <linux@roeck-us.net>
      Signed-off-by: NMasahiro Yamada <yamada.masahiro@socionext.com>
      3298b690
    • L
      Linux 4.14 · bebc6082
      Linus Torvalds 提交于
      bebc6082
  27. 09 11月, 2017 1 次提交