Linux 4.7 kernel commit 9b94a8fba501f38368aef6ac1b30e7335252a220,
which changed the ring_buffer_per_cpu.nr_pages member from an int
to a long. Without the patch, the trace.so extension module fails
to load on big-endian machines, indicating "extend: Num of pages
is less than 0".
(feij.fnst@cn.fujitsu.com)
CONFIG_RANDOMIZE_BASE. Without the patch, the command fails with
the message "set: gdb cannot find text block for address: <symbol>".
This also affects extension modules that call gdb_set_crash_scope()
when running with KASLR kernels.
(anderson@redhat.com)
of a "symbol+offset" argument where the duplicates are contiguous
in the symbol list. In addition, reject "symbol+offset" arguments
if the resultant address goes beyond the end of the function.
(anderson@redhat.com)
not passed in the VMCOREINFO data of ELF vmcores. Without the patch,
it is possible that the base address of the vmalloc region is unknown
and initialized to an incorrect default address during the very early
stages of initialization, which causes the parsing of the PT_LOAD
segments for the START_KERNEL_map region to fail.
(anderson@redhat.com)
encountered that contains an invalid NULL pointer. Without the
patch, the "list -[hH]" options would complete/continue as if the
NULL were a legitimate end-of-list indicator, and no error would be
reported.
(rabin.vincent@axis.com)
on the host build machine. Without the patch, the build fails with
the error message "amd64-linux-nat.c:496:1: error: conflicting types
for 'ps_get_thread_area'".
(anderson@redhat.com)
to signficantly reduce the amount of time to complete the buddy
allocator free-list scan for the target address. On very large
memory systems, the patch may reduce the time spent by several orders
of magnitude.
(anderson@redhat.com)
in Linux 4.6 and later kernels if an ELF format dumpfile does not
contain its value in a VMCOREINFO note, or when running against
live systems using /dev/mem, /proc/kcore, or an older version of
/dev/crash.
(liyueyi@live.com)
in x86_64 kernels that contain, or have backports of, kernel commit
7c1da8d0d046174a4188b5729d7579abf3d29427, titled "crypto: sha - SHA1
transform x86_64 AVX2", which introduced an "_end" text symbol.
Without the patch, if a base kernel symbol address that is larger
than the "_end" text symbol is passed to "kmem <address>", its
symbol/filename information will not be displayed. Also, when the
"search" command scans the __START_KERNEL_map region that contains
kernel text and static data, the search will be truncated to stop at
the "_end" text symbol address.
(anderson@redhat.com)
of one or more symbol values of a kernel module. Without the patch,
when loading a kernel module, a message may indicate "mod: <module>:
last symbol: <symbol> is not _MODULE_END_<module>?" may be displayed,
and one or more symbols may be reassigned an incorrect symbol value.
If none of the erroneous symbol value reassignments are beyond the
end of the module's address space, then there will be no message.
(anderson@redhat.com)
when building with "make warn". The warning types are "[-Wnonnull]"
in filesys.c, and "[-Wformat-overflow=]" in kernel.c and cmdline.c.
(anderson@redhat.com)
in the situation where the task was running on its soft IRQ stack,
took a hard IRQ, and then the system crashed while it was running on
its hard IRQ stack.
(hirofumi@mail.parknet.co.jp)
kernel commit 198d208df4371734ac4728f69cb585c284d20a15, titled "x86:
Keep thread_info on thread stack in x86_32". The patch fixes the
stack transition symbol from "handle_IRQ" to "handle_irq" for usage
by the "bt" command.
(anderson@redhat.com)
commit 198d208df4371734ac4728f69cb585c284d20a15, titled "x86: Keep
thread_info on thread stack in x86_32". Without the patch, incorrect
addresses of each per-cpu hardirq_stack and softirq_stack were saved
for usage by the "bt" command.
(hirofumi@mail.parknet.co.jp, anderson@redhat.com)
the "bt" command on Linux 4.2 and later 32-bit X86 kernels. Without
the patch, the first "bt" command that is executed will be preceded
by "bt: WARNING: "system_call" symbol does not exist", followed by
"bt: WARNING: neither "ret_from_sys_call" nor "syscall_badsys"
symbols exist".
(anderson@redhat.com)
Without the patch, the backtrace displays the "cannot resolve stack
trace" warning, dumps the backtrace, and then the text symbols:
crash> bt
PID: 0 TASK: f0962180 CPU: 6 COMMAND: "swapper/6"
bt: cannot resolve stack trace:
#0 [f095ff1c] __schedule at c0b6ef8d
#1 [f095ff58] schedule at c0b6f4a9
#2 [f095ff64] schedule_preempt_disabled at c0b6f728
#3 [f095ff6c] cpu_startup_entry at c04b0310
#4 [f095ff94] start_secondary at c04468c0
bt: text symbols on stack:
[f095ff1c] __schedule at c0b6ef8d
[f095ff58] schedule at c0b6f4ae
[f095ff64] schedule_preempt_disabled at c0b6f72d
[f095ff6c] cpu_startup_entry at c04b0315
[f095ff94] start_secondary at c04468c5
crash>
The backtrace shown is actually correct.
(anderson@redhat.com)
containing kernel commit 0100301bfdf56a2a370c7157b5ab0fbf9313e1cd,
subject "sched/x86: Rewrite the switch_to() code". Without the
patch, backtraces for inactive (sleeping) tasks fail with the message
"bt: invalid structure member offset: task_struct_thread_eip".
(anderson@redhat.com)
Without the patch, an invalid kernel PAGE_OFFSET value is calculated
and as a result the session fails during session initialization just
after the patching of the gdb minimal_symbol values message, showing
the warning message "WARNING: cannot read linux_banner string",
followed by "crash: /vmlinux and /dev/crash do not match!". This
patch also adds a new "--machdep page_offset=<value>" option that
can be used if the CONFIG_PAGE_OFFSET value is not the default
address of 0xc0000000.
(anderson@redhat.com)
Without the patch, if the target machine has been up for a long
enough time, an arithmetic overflow will occur and the time value
displayed will be incorrect.
(shane.seymour@hpe.com)
by the "kmem" command when SLAB_RED_ZONE has been enabled in Linux 4.6 and later
kernels. By default, CONFIG_SLUB object addresses that are displayed by the
"kmem" command will point to the SLAB_RED_ZONE padding inserted at the
beginning of the object. As an alternative, a new "redzone" environment
variable has been addedd that can be toggled on or off. If "set redzone off"
is entered, the object addresses will point to the address that gets returned
to the allocator.
(hirofumi@mail.parknet.co.jp, anderson@redhat.com)
by "max_tr_ring_buffer" is not available in all configurations of the
kernel so the unitialized max_tr_ring_buffer variable should not be
used. A similar check existed previously before the recent rework of
the trace extension module to support multiple buffers.
(rabinv@axix.com)
in Linux 4.6 and later kernels due to upstream changes in the radix
tree facility. Without the patch, the commands will fail with the
message "radix trees do not exist or have changed their format".
(hirofumi@mail.parknet.co.jp)
and earlier x86_64 kernels configured with CONFIG_RANDOMIZE_BASE.
A segmentation violation may occur during session initialization,
just after the patching of the gdb minimal_symbol values message,
depending upon the value of KERNEL_IMAGE_SIZE, which was variable
in the earlier KASLR kernels. This patch sets the KERNEL_IMAGE_SIZE
default value to 1GB for those earlier kernels, and also adds a
new "--machdep kernel_image_size=<value>" option that can be
used to override the default KERNEL_IMAGE_SIZE value if necessary.
(anderson@redhat.com)
FADUMP-generated dumpfiles (Firmware Assisted Dump facility).
Without the patch, backtraces of those tasks may be of the form
"#0 [c0000000700b3a90] (null) at c0000000700b3b50 (unreliable)".
This patch uses and displays the ptregs register set saved in the
dumpfile header for the non-panicking active tasks.
(hbathini@linux.vnet.ibm.com)
display from generating a fatal session-killing error if relevant
kernel symbol names or data structures change in the future (again).
(anderson@redhat.com)
member names to be changed to "c_true" and "c_false", which fixes
build problems when an out-of-tree module defines "true" or "false".
(anderson@redhat.com)
the change in size of the module.taints flag from an int to a long,
while allowing for a kernel backport that keeps it as an int.
(anderson@redhat.com)
and later x86_64 kernels configured with CONFIG_RANDOMIZE_BASE, which
randomizes the unity-mapping PAGE_OFFSET value. Without the patch,
the crash session fails during session initialization with the error
message "crash: seek error: kernel virtual address: <address>
type: page_offset_base".
(anderson@redhat.com)
"taint/module: Clean up global and module taint flags handling".
Without the patch, when running against Linux 4.10-rc1 and later
kernels, the crash utility fails during session initialization with
the message "crash: invalid structure size: tnt".
(panand@redhat.com)
are configured with CONFIG_THREAD_INFO_IN_TASK. Without the patch,
the "bt" command will fail for active tasks in dumpfiles that were
generated by the kdump facility.
(takahiro.akashi@linaro.org)
which it isn't now since the internal machdep->pagesize is unsigned.
Without this patch, module loading fails during initialization on a
kernel that has a page size of 16KB, with messages that indicate
"please wait... (gathering module symbol data)" followed by
"crash: invalid size request: 0 type: pgd page".
(rabinv@axis.com)
running against a 32-bit MIPS ELF kdump or compressed kdump if a
per-cpu NT_PRSTATUS note cannot be be gathered from the dumpfile
header. Without the the patch, a segmentation violation occurs after
the message "WARNING: cannot find NT_PRSTATUS note for cpu: <number>"
is displayed.
(rabinv@axis.com)
if the kernel crashed while running a module's initcall. Without
the patch, the crash session fails during initialation with a message
similar to "crash: store_module_symbols_v2: total: 7 mcnt: 8".
(rabinv@axis.com)
which introduced the new slab management type OBJFREELIST_SLAB.
In this mode, the freelist can be an object, and if the slab is full,
there is no freelist. On the next free, an object is recycled to be
used as the freelist but not cleaned-up. This patch will go through
only known freed objects, and will prevent "kmem -S" errors that
indicate "invalid/corrupt freelist entry" on kernels configured
with CONFIG_SLAB.
(thgarnie@google.com)
to version 1.3, which adds support for Linux 4.6 and later ARM64
kernels, kernels configured with CONFIG_HARDENED_USERCOPY, and
S390X kernels use xlate_dev_mem_ptr() and unxlate_dev_mem_ptr()
instead of kmap() and kunmap().
(anderson@redhat.com)
