Re: BTF without CONFIG_DEBUG_INFO_BTF=y

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

 



Arnaldo Carvalho de Melo <acme@xxxxxxxxxx> writes:

> Em Thu, Oct 01, 2020 at 09:50:29AM -0300, Arnaldo Carvalho de Melo escreveu:
>> Em Thu, Oct 01, 2020 at 12:33:18PM +0200, Toke Høiland-Jørgensen escreveu:
>> > Kevin Sheldrake <Kevin.Sheldrake@xxxxxxxxxxxxx> writes:
>> > > I've seen mention a few times that BTF information can be made
>> > > available from a kernel that wasn't configured with
>> > > CONFIG_DEBUG_INFO_BTF. Please can someone tell me if this is true and,
>> > > if so, how I could go about accessing and using it in kernels 4.15 to
>> > > 5.8?
>
>> > > I have built the dwarves package from the github latest and run pahole
>> > > with '-J' against my kernel image to no avail - it actually seg
>> > > faults:
>
>> > > ~/dwarves/build $ sudo ./pahole /boot/vmlinuz-5.3.0-1022-azure
>> > > btf_elf__new: cannot get elf header.
>> > > ctf__new: cannot get elf header.
>> > > ~/dwarves/build $ sudo ./pahole -J /boot/vmlinuz-5.3.0-1022-azure
>> > > btf_elf__new: cannot get elf header.
>> > > ctf__new: cannot get elf header.
>> > > Segmentation fault
>> > > ~/dwarves/build $ sudo ./pahole --version
>> > > v1.17
>
>> > > Judging by the output, I'm guessing that my kernel image isn't the
>> > > right kind of file. Can someone point me in the right direction?
>
>> > vmlinuz is a compressed image. There's a script in the kernel source
>> > tree (scripts/extract-vmlinux), however the kernel image in /boot/
>> > probably also has debug information stripped from it, so that likely
>> > won't help you. You'll need to get hold of a kernel image with debug
>> > information still intact somehow...
>
>> > (Either way, pahole shouldn't be segfaulting, so hopefully someone can
>> > take a look at that).
>
>> Reproduced:
>
>> [acme@five pahole]$ cp /boot/vmlinuz-5.9.0-rc6+ .
>> [acme@five pahole]$ pahole -J vmlinuz-5.9.0-rc6+
>> btf_elf__new: cannot get elf header.
>> ctf__new: cannot get elf header.
>> tag__check_id_drift: subroutine_type id drift, core_id: 1145, btf_type_id: 1143, type_id_off: 0
>> pahole: type 'vmlinuz-5.9.0-rc6+' not found
>> libbpf: Unsupported BTF_KIND:0
>> btf_elf__encode: btf__new failed!
>> free(): double free detected in tcache 2
>> Aborted (core dumped)
>> [acme@five pahole]$
>  
>> Working on a fix. Thanks for the report!
>
> commit 4e55425d9eaac78689fbd296283e1557bb6ca725
> Author: Arnaldo Carvalho de Melo <acme@xxxxxxxxxx>
> Date:   Thu Oct 1 10:10:50 2020 -0300
>
>     pahole: Only try using a single file name as a type name if not encoding BTF or CTF
>     
>     Otherwise we end up trying to encode without any debug info and this
>     causes a segfault:
>     
>     Before:
>     
>       $ pahole -J vmlinuz-5.9.0-rc6+
>       tag__check_id_drift: subroutine_type id drift, core_id: 1145, btf_type_id: 1143, type_id_off: 0
>       pahole: type 'vmlinuz-5.9.0-rc6+' not found
>       libbpf: Unsupported BTF_KIND:0
>       btf_elf__encode: btf__new failed!
>       free(): double free detected in tcache 2
>       Aborted (core dumped)
>       $
>     
>     The vmlinuz file doesn't contain any debugging info, fixing it we get:
>     
>       $ pahole -J vmlinuz-5.9.0-rc6+
>       pahole: vmlinuz-5.9.0-rc6+: No debugging information found
>       $
>     
>     If debugging info is available, it all works as before:
>     
>     Using /sys/kernel/btf/vmlinux
>     
>     $ ls -la /sys/kernel/btf/vmlinux
>     -r--r--r--. 1 root root 3393761 Oct  1 09:50 /sys/kernel/btf/vmlinux
>     
>       $ pahole -E fw_cache_entry
>       struct fw_cache_entry {
>             struct list_head {
>                     struct list_head * next;          /*     0     8 */
>                     struct list_head * prev;          /*     8     8 */
>             } list; /*     0    16 */
>             const char  *              name;          /*    16     8 */
>     
>             /* size: 24, cachelines: 1, members: 2 */
>             /* last cacheline: 24 bytes */
>       };
>       $
>     
>     Or explicitely asking for DWARF, where it will find the appropriate
>     vmlinux according to its buildid in /sys/kernel/notes:
>     
>       $ pahole -F dwarf pm_clock_entry
>       struct pm_clock_entry {
>             struct list_head           node;          /*     0    16 */
>             char *                     con_id;        /*    16     8 */
>             struct clk *               clk;           /*    24     8 */
>             enum pce_status            status;        /*    32     4 */
>     
>             /* size: 40, cachelines: 1, members: 4 */
>             /* padding: 4 */
>             /* last cacheline: 40 bytes */
>       };
>       $ pahole -F dwarf --expand_types pm_clock_entry
>       struct pm_clock_entry {
>             struct list_head {
>                     struct list_head * next;          /*     0     8 */
>                     struct list_head * prev;          /*     8     8 */
>             } node; /*     0    16 */
>             char *                     con_id;        /*    16     8 */
>             struct clk *               clk;           /*    24     8 */
>             enum pce_status            status;        /*    32     4 */
>     
>             /* size: 40, cachelines: 1, members: 4 */
>             /* padding: 4 */
>             /* last cacheline: 40 bytes */
>       };
>       $
>     
>     Reported-by: Kevin Sheldrake <Kevin.Sheldrake@xxxxxxxxxxxxx>
>     Cc: Toke Høiland-Jørgensen <toke@xxxxxxxxxx>
>     Signed-off-by: Arnaldo Carvalho de Melo <acme@xxxxxxxxxx>

Yeah, that's much better!

Acked-by: Toke Høiland-Jørgensen <toke@xxxxxxxxxx>





[Index of Archives]     [Linux Samsung SoC]     [Linux Rockchip SoC]     [Linux Actions SoC]     [Linux for Synopsys ARC Processors]     [Linux NFS]     [Linux NILFS]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]


  Powered by Linux