Page fault: bad address: Code=6 (Instruction TLB miss fault) at addr 37dd3fa10c0

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Boot on parisc has been broken for some time:

[...]
Run /init as init process
process '/usr/bin/sh' started with executable stack
Loading, please wait...
Starting systemd-udevd version 254.1-3
Backtrace:
 [<0000000040433d0c>] kmalloc_trace+0x3c/0x50
 [<00000000402fceb4>] do_init_module+0xa4/0x4d0
 [<00000000402ff130>] load_module+0xe48/0x1028
 [<00000000402ff6ec>] __do_sys_finit_module+0x104/0x168
 [<00000000402ff7bc>] sys_finit_module+0x2c/0x40
 [<0000000040203e5c>] syscall_exit+0x0/0x10


Page fault: bad address: Code=6 (Instruction TLB miss fault) at addr 37dd3fa10c0
CPU: 1 PID: 825 Comm: (udev-worker) Not tainted 6.4.0-rc2+ #14
Hardware name: 9000/785/C8000

     YZrvWESTHLNXBCVMcbcbcbcbOGFRQPDI
PSW: 00001000000001001101111100001111 Not tainted
r00-03  000000ff0804df0f 37dd3fa10c6712d0 000000000816704c 0000000053390520
r04-07  0000000008161000 00000000081611c0 00000000520915a0 00000000411d8000
r08-11  0000000000000000 000000004115ad88 0000000053390240 0000000000000000
r12-15  00000000081631b8 0000000053390370 00000000410b9e40 0000000008165174
r16-19  0000000000000000 0000000000000000 0000000053390288 0000000008169000
r20-23  000000005381b0b0 000000000eaaa000 00000000533904c8 0000000000000001
r24-27  0000000000000cc0 0000000000000000 0000000008165290 081a024770650030
r28-31  0000000000000000 0000000053390570 00000000533905a0 0000000008167000
sr00-03  00000000000c3800 0000000000000000 0000000000000000 00000000000c6400
sr04-07  0000000000000000 0000000000000000 0000000000000000 0000000000000000

IASQ: 0000000037dd3fa1 0000000037dd3fa1 IAOQ: 37dd3fa10c6712d0 37dd3fa10c6712d4
 IIR: 43ffff80    ISR: 0000000040ca43a0  IOR: 0000000040c8c040
 CPU:        1   CR30: 00000000520915a0 CR31: ffffffffffffffff
 ORIG_R28: 000000000800000e
 IAOQ[0]: 0x37dd3fa10c6712d0
 IAOQ[1]: 0x37dd3fa10c6712d4
 RP(r2): usb_common_init+0x2c/0xfe0 [usb_common]
Backtrace:
 [<0000000040433d0c>] kmalloc_trace+0x3c/0x50
 [<00000000402fceb4>] do_init_module+0xa4/0x4d0
 [<00000000402ff130>] load_module+0xe48/0x1028
 [<00000000402ff6ec>] __do_sys_finit_module+0x104/0x168
 [<00000000402ff7bc>] sys_finit_module+0x2c/0x40
 [<0000000040203e5c>] syscall_exit+0x0/0x10

Kernel panic - not syncing: Page fault: bad address

I bisected this to the following commit:

dave@atlas:~/linux/linux$ git bisect good
ddb5cdbafaaad6b99d7007ae1740403124502d03 is the first bad commit
commit ddb5cdbafaaad6b99d7007ae1740403124502d03
Author: Masahiro Yamada <masahiroy@xxxxxxxxxx>
Date:   Mon Jun 12 00:50:52 2023 +0900

    kbuild: generate KSYMTAB entries by modpost

    Commit 7b4537199a4a ("kbuild: link symbol CRCs at final link, removing
    CONFIG_MODULE_REL_CRCS") made modpost output CRCs in the same way
    whether the EXPORT_SYMBOL() is placed in *.c or *.S.

    For further cleanups, this commit applies a similar approach to the
    entire data structure of EXPORT_SYMBOL().

    The EXPORT_SYMBOL() compilation is split into two stages.

    When a source file is compiled, EXPORT_SYMBOL() will be converted into
    a dummy symbol in the .export_symbol section.

    For example,

        EXPORT_SYMBOL(foo);
        EXPORT_SYMBOL_NS_GPL(bar, BAR_NAMESPACE);

    will be encoded into the following assembly code:

        .section ".export_symbol","a"
        __export_symbol_foo:
                .asciz ""                      /* license */
                .asciz ""                      /* name space */
                .balign 8
                .quad foo                      /* symbol reference */
        .previous

        .section ".export_symbol","a"
        __export_symbol_bar:
                .asciz "GPL"                   /* license */
                .asciz "BAR_NAMESPACE"         /* name space */
                .balign 8
                .quad bar                      /* symbol reference */
        .previous

    They are mere markers to tell modpost the name, license, and namespace
    of the symbols. They will be dropped from the final vmlinux and modules
    because the *(.export_symbol) will go into /DISCARD/ in the linker script.

    Then, modpost extracts all the information about EXPORT_SYMBOL() from the
    .export_symbol section, and generates the final C code:

        KSYMTAB_FUNC(foo, "", "");
        KSYMTAB_FUNC(bar, "_gpl", "BAR_NAMESPACE");

    KSYMTAB_FUNC() (or KSYMTAB_DATA() if it is data) is expanded to struct
    kernel_symbol that will be linked to the vmlinux or a module.

    With this change, EXPORT_SYMBOL() works in the same way for *.c and *.S
    files, providing the following benefits.

    [1] Deprecate EXPORT_DATA_SYMBOL()

    In the old days, EXPORT_SYMBOL() was only available in C files. To export
    a symbol in *.S, EXPORT_SYMBOL() was placed in a separate *.c file.
    arch/arm/kernel/armksyms.c is one example written in the classic manner.

    Commit 22823ab419d8 ("EXPORT_SYMBOL() for asm") removed this limitation.
    Since then, EXPORT_SYMBOL() can be placed close to the symbol definition
    in *.S files. It was a nice improvement.

    However, as that commit mentioned, you need to use EXPORT_DATA_SYMBOL()
    for data objects on some architectures.

    In the new approach, modpost checks symbol's type (STT_FUNC or not),
    and outputs KSYMTAB_FUNC() or KSYMTAB_DATA() accordingly.

    There are only two users of EXPORT_DATA_SYMBOL:

      EXPORT_DATA_SYMBOL_GPL(empty_zero_page) (arch/ia64/kernel/head.S)
      EXPORT_DATA_SYMBOL(ia64_ivt) (arch/ia64/kernel/ivt.S)

    They are transformed as follows and output into .vmlinux.export.c

      KSYMTAB_DATA(empty_zero_page, "_gpl", "");
      KSYMTAB_DATA(ia64_ivt, "", "");

    The other EXPORT_SYMBOL users in ia64 assembly are output as
    KSYMTAB_FUNC().

    EXPORT_DATA_SYMBOL() is now deprecated.

    [2] merge <linux/export.h> and <asm-generic/export.h>

    There are two similar header implementations:

      include/linux/export.h        for .c files
      include/asm-generic/export.h  for .S files

    Ideally, the functionality should be consistent between them, but they
    tend to diverge.

    Commit 8651ec01daed ("module: add support for symbol namespaces.") did
    not support the namespace for *.S files.

    This commit shifts the essential implementation part to C, which supports
    EXPORT_SYMBOL_NS() for *.S files.

    <asm/export.h> and <asm-generic/export.h> will remain as a wrapper of
    <linux/export.h> for a while.

    They will be removed after #include <asm/export.h> directives are all
    replaced with #include <linux/export.h>.

    [3] Implement CONFIG_TRIM_UNUSED_KSYMS in one-pass algorithm (by a later commit)

    When CONFIG_TRIM_UNUSED_KSYMS is enabled, Kbuild recursively traverses
    the directory tree to determine which EXPORT_SYMBOL to trim. If an
    EXPORT_SYMBOL turns out to be unused by anyone, Kbuild begins the
    second traverse, where some source files are recompiled with their
    EXPORT_SYMBOL() tuned into a no-op.

    We can do this better now; modpost can selectively emit KSYMTAB entries
    that are really used by modules.

    Signed-off-by: Masahiro Yamada <masahiroy@xxxxxxxxxx>
    Reviewed-by: Nick Desaulniers <ndesaulniers@xxxxxxxxxx>

 arch/ia64/include/asm/Kbuild      |   1 +
 arch/ia64/include/asm/export.h    |   3 --
 include/asm-generic/export.h      |  84 ++----------------------------
 include/asm-generic/vmlinux.lds.h |   1 +
 include/linux/export-internal.h   |  49 ++++++++++++++++++
 include/linux/export.h            | 101 +++++++++++++++---------------------
 include/linux/pm.h                |   4 +-
 kernel/module/internal.h          |  12 +++++
 scripts/Makefile.build            |   8 ++-
 scripts/check-local-export        |   4 +-
 scripts/mod/modpost.c             | 106 ++++++++++++++++++++++++++------------
 scripts/mod/modpost.h             |   1 +
 12 files changed, 190 insertions(+), 184 deletions(-)
 delete mode 100644 arch/ia64/include/asm/export.h

Dave

--
John David Anglin  dave.anglin@xxxxxxxx




[Index of Archives]     [Linux SoC]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]

  Powered by Linux