Date: Mon, 24 Nov 2014 23:21:26 +0200 From: Andriy Gapon <avg@FreeBSD.org> To: Mark Johnston <markj@FreeBSD.org>, Rui Paulo <rpaulo@me.com> Cc: Konstantin Belousov <kostikbel@gmail.com>, freebsd-hackers@FreeBSD.org, freebsd-dtrace@FreeBSD.org, John Baldwin <jhb@FreeBSD.org>, Shrikanth Kamath <shrikanth07@gmail.com> Subject: Re: DTrace: stack() does not print kernel module functions for i386 Message-ID: <5473A156.6070502@FreeBSD.org> In-Reply-To: <20141123021856.GA54708@raichu> References: <CAEOAkMXnwqC42gZKc0f80cppff077pYGjs5PUPht0DBcyEi8Jw@mail.gmail.com> <20141109093632.GV53947@kib.kiev.ua> <9011F920-3092-4E61-9CDC-68FD9092BB7D@me.com> <201411131336.12334.jhb@freebsd.org> <AE2A9617-8E59-4F44-9C1F-8344EF5B05C7@me.com> <20141123021856.GA54708@raichu>
next in thread | previous in thread | raw e-mail | index | archive | help
On 23/11/2014 04:18, Mark Johnston wrote:
> On Thu, Nov 13, 2014 at 07:49:27PM -0800, Rui Paulo wrote:
>> On Nov 13, 2014, at 10:36, John Baldwin <jhb@freebsd.org> wrote:
>>> Why have the #ifdef? In theory other platforms besides amd64 could use
>>> sys/kern/link_elf_obj.c. It doesn't hurt to just let the code always accept
>>> both ET_DYN and ET_REL does it?
>>
>> No, it doesn't hurt.
>
> The suggested patch doesn't seem quite right; there are other functions
> in dt_module.c with the same assignment (i.e.
> "is_elf_obj = ehdr.e_type == ET_REL"), but the same modification is not
> correct in all cases - fixing it everywhere breaks stack() again - and
> "is_elf_obj" seems like the wrong name if DSOs are counted as well.
>
> The root of the problem is that dmp->dm_*_va offsets don't have the kld
> load address taken into account on i386, since they're currently set based
> only on the ELF section addresses. This is handled by
> dmp->dm_reloc_offset for symbols, but that's a separate case.
>
> When is_elf_obj is true, we include the load address when setting the
> dmp->dm_*_va fields. I suggest we do that unconditionally, and only set
> elements of dmp->dm_sec_offsets if is_elf_obj is true. This fixes the
> bug for me on i386. Any opinions?
This totally makes sense to me.
Thank you!
> diff --git a/cddl/contrib/opensolaris/lib/libdtrace/common/dt_module.c b/cddl/contrib/opensolaris/lib/libdtrace/common/dt_module.c
> index e3905c1..9dd52b5 100644
> --- a/cddl/contrib/opensolaris/lib/libdtrace/common/dt_module.c
> +++ b/cddl/contrib/opensolaris/lib/libdtrace/common/dt_module.c
> @@ -1211,13 +1211,13 @@ dt_module_update(dtrace_hdl_t *dtp, struct kld_file_stat *k_stat)
> #if defined(__FreeBSD__)
> if (sh.sh_size == 0)
> continue;
> - if (is_elf_obj && (sh.sh_type == SHT_PROGBITS ||
> - sh.sh_type == SHT_NOBITS)) {
> + if (sh.sh_type == SHT_PROGBITS || sh.sh_type == SHT_NOBITS) {
> alignmask = sh.sh_addralign - 1;
> mapbase += alignmask;
> mapbase &= ~alignmask;
> sh.sh_addr = mapbase;
> - dmp->dm_sec_offsets[elf_ndxscn(sp)] = sh.sh_addr;
> + if (is_elf_obj)
> + dmp->dm_sec_offsets[elf_ndxscn(sp)] = sh.sh_addr;
> mapbase += sh.sh_size;
> }
> #endif
>
--
Andriy Gapon
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?5473A156.6070502>