On some architectures (e.g. x86_64) the vmlinux is ET_EXEC, while on others (e.g. ppc64) it is ET_DYN. In both cases the phdr p_vaddr will be non-zero. We want the image to be placed as if it was ET_DYN, so pass true for add_p_vaddr which will do the right thing (in combination with a zero base) in either case.
Signed-off-by: Mark Wielaard mjw@redhat.com --- libdwfl/ChangeLog | 5 +++++ libdwfl/linux-kernel-modules.c | 8 +++++++- 2 files changed, 12 insertions(+), 1 deletions(-)
diff --git a/libdwfl/ChangeLog b/libdwfl/ChangeLog index 88bdfc6..90ba686 100644 --- a/libdwfl/ChangeLog +++ b/libdwfl/ChangeLog @@ -1,3 +1,8 @@ +2013-08-25 Mark Wielaard mjw@redhat.com + + * linux-kernel-modules.c (report_kernel): Pass add_p_vaddr as true + to dwfl_report_elf. + 2013-07-25 Jan Kratochvil jan.kratochvil@redhat.com
* dwfl_segment_report_module.c (dwfl_segment_report_module): Check for diff --git a/libdwfl/linux-kernel-modules.c b/libdwfl/linux-kernel-modules.c index dec1a59..fe01028 100644 --- a/libdwfl/linux-kernel-modules.c +++ b/libdwfl/linux-kernel-modules.c @@ -216,8 +216,14 @@ report_kernel (Dwfl *dwfl, const char **release,
if (report) { + /* Note that on some architectures (e.g. x86_64) the vmlinux + is ET_EXEC, while on others (e.g. ppc64) it is ET_DYN. + In both cases the phdr p_vaddr load address will be non-zero. + We want the image to be placed as if it was ET_DYN, so + pass true for add_p_vaddr which will do the right thing + (in combination with a zero base) in either case. */ Dwfl_Module *mod = INTUSE(dwfl_report_elf) (dwfl, KERNEL_MODNAME, - fname, fd, 0, false); + fname, fd, 0, true); if (mod == NULL) result = -1; else
On Thu, 2013-08-22 at 13:45 +0200, Mark Wielaard wrote:
On some architectures (e.g. x86_64) the vmlinux is ET_EXEC, while on others (e.g. ppc64) it is ET_DYN. In both cases the phdr p_vaddr will be non-zero. We want the image to be placed as if it was ET_DYN, so pass true for add_p_vaddr which will do the right thing (in combination with a zero base) in either case.
I pushed this and will also backport it to the fedora elfutils package since it fixes an issue on ppc64 the systemtap hackers are seeing.
Cheers,
Mark
elfutils-devel@lists.fedorahosted.org