[U-Boot] [PATCH] efi_loader: Rename sections to allow for implicit data
Simon Glass
sjg at chromium.org
Thu Aug 23 16:31:57 UTC 2018
Hi Tom.
On 23 August 2018 at 07:58, Tom Rini <trini at konsulko.com> wrote:
> On Thu, Aug 23, 2018 at 04:45:29AM -0600, Simon Glass wrote:
>> Hi Tom,
>>
>> On 21 August 2018 at 17:11, Tom Rini <trini at konsulko.com> wrote:
>> > On Tue, Aug 21, 2018 at 04:29:49PM -0600, Simon Glass wrote:
>> >> Hi Alex,
>> >>
>> >> On 21 August 2018 at 13:26, Alexander Graf <agraf at suse.de> wrote:
>> >> >
>> >> >
>> >> > On 21.08.18 19:30, Simon Glass wrote:
>> >> >> Hi Alex,
>> >> >>
>> >> >> On 20 August 2018 at 06:23, Alexander Graf <agraf at suse.de> wrote:
>> >> >>>
>> >> >>> On 08/17/2018 02:49 PM, Simon Glass wrote:
>> >> >>>>
>> >> >>>> Hi,
>> >> >>>>
>> >> >>>> On 9 August 2018 at 23:45, Bin Meng <bmeng.cn at gmail.com> wrote:
>> >> >>>>>
>> >> >>>>> Hi Alex,
>> >> >>>>>
>> >> >>>>> On Wed, Aug 8, 2018 at 1:16 AM, Alexander Graf <agraf at suse.de> wrote:
>> >> >>>>>>
>> >> >>>>>>
>> >> >>>>>>> Am 07.08.2018 um 18:12 schrieb Simon Glass <sjg at chromium.org>:
>> >> >>>>>>>
>> >> >>>>>>> Hi Alex,
>> >> >>>>>>>
>> >> >>>>>>>> On 11 June 2018 at 23:48, Alexander Graf <agraf at suse.de> wrote:
>> >> >>>>>>>> Some times gcc may generate data that is then used within code that may
>> >> >>>>>>>> be part of an efi runtime section. That data could be jump tables,
>> >> >>>>>>>> constants or strings.
>> >> >>>>>>>>
>> >> >>>>>>>> In order to make sure we catch these, we need to ensure that gcc emits
>> >> >>>>>>>> them into a section that we can relocate together with all the other
>> >> >>>>>>>> efi runtime bits. This only works if the -ffunction-sections and
>> >> >>>>>>>> -fdata-sections flags are passed and the efi runtime functions are
>> >> >>>>>>>> in a section that starts with ".text".
>> >> >>>>>>>>
>> >> >>>>>>>> Up to now we had all efi runtime bits in sections that did not
>> >> >>>>>>>> interfere with the normal section naming scheme, but this forces
>> >> >>>>>>>> us to do so. Hence we need to move the efi_loader text/data/rodata
>> >> >>>>>>>> sections before the global *(.text*) catch-all section.
>> >> >>>>>>>>
>> >> >>>>>>>> With this patch in place, we should hopefully have an easier time
>> >> >>>>>>>> to extend the efi runtime functionality in the future.
>> >> >>>>>>>>
>> >> >>>>>>>> Signed-off-by: Alexander Graf <agraf at suse.de>
>> >> >>>>>>>> ---
>> >> >>>>>>>> arch/arm/config.mk | 4 ++--
>> >> >>>>>>>> arch/arm/cpu/armv8/u-boot.lds | 24 +++++++++++++--------
>> >> >>>>>>>> arch/arm/cpu/u-boot.lds | 36 ++++++++++++++++++-------------
>> >> >>>>>>>> arch/arm/mach-zynq/u-boot.lds | 36 ++++++++++++++++++-------------
>> >> >>>>>>>> arch/riscv/cpu/ax25/u-boot.lds | 26 +++++++++++++---------
>> >> >>>>>>>> arch/sandbox/config.mk | 3 +++
>> >> >>>>>>>> arch/sandbox/cpu/u-boot.lds | 9 ++++----
>> >> >>>>>>>> arch/x86/config.mk | 2 +-
>> >> >>>>>>>> arch/x86/cpu/u-boot.lds | 32 ++++++++++++++-------------
>> >> >>>>>>>> board/qualcomm/dragonboard410c/u-boot.lds | 17 +++++++++++++--
>> >> >>>>>>>> board/qualcomm/dragonboard820c/u-boot.lds | 24 +++++++++++++--------
>> >> >>>>>>>> board/ti/am335x/u-boot.lds | 36 ++++++++++++++++++-------------
>> >> >>>>>>>> include/efi_loader.h | 4 ++--
>> >> >>>>>>>> 13 files changed, 154 insertions(+), 99 deletions(-)
>> >> >>>>>>>>
>> >> >>>>>>> I missed this at the time, probably thinking the subject made it sound
>> >> >>>>>>> innocuous. There is no 'sandbox:' tag.
>> >> >>>>>>>
>> >> >>>>>>> This seems to break sandbox in a pretty strange way:
>> >> >>>>>>>
>> >> >>>>>>> gdb --args /tmp/crosfw/sandbox/u-boot -D
>> >> >>>>>>> GNU gdb (Debian 7.12-6) 7.12.0.20161007-git
>> >> >>>>>>> Copyright (C) 2016 Free Software Foundation, Inc.
>> >> >>>>>>> License GPLv3+: GNU GPL version 3 or later <http://gnu.org/licenses/gpl.html>
>> >> >>>>>>> This is free software: you are free to change and redistribute it.
>> >> >>>>>>> There is NO WARRANTY, to the extent permitted by law. Type "show copying"
>> >> >>>>>>> and "show warranty" for details.
>> >> >>>>>>> This GDB was configured as "x86_64-linux-gnu".
>> >> >>>>>>> Type "show configuration" for configuration details.
>> >> >>>>>>> For bug reporting instructions, please see:
>> >> >>>>>>> <http://www.gnu.org/software/gdb/bugs/>.
>> >> >>>>>>> Find the GDB manual and other documentation resources online at:
>> >> >>>>>>> <http://www.gnu.org/software/gdb/documentation/>.
>> >> >>>>>>> For help, type "help".
>> >> >>>>>>> Type "apropos word" to search for commands related to "word"...
>> >> >>>>>>> Reading symbols from /tmp/crosfw/sandbox/u-boot...done.
>> >> >>>>>>> (gdb) r
>> >> >>>>>>> Starting program: /tmp/crosfw/sandbox/u-boot -D
>> >> >>>>>>> [Thread debugging using libthread_db enabled]
>> >> >>>>>>> Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".
>> >> >>>>>>>
>> >> >>>>>>> Program received signal SIGSEGV, Segmentation fault.
>> >> >>>>>>> 0x0000555555571520 in open at plt ()
>> >> >>>>>>> (gdb) up
>> >> >>>>>>> #1 0x0000555555571e9a in sandbox_read_fdt_from_file ()
>> >> >>>>>>> at /home/sjg/c/src/third_party/u-boot/files/arch/sandbox/cpu/cpu.c:264
>> >> >>>>>>> 264 fd = os_open(fname, OS_O_RDONLY);
>> >> >>>>>>> (gdb) print fname
>> >> >>>>>>> $1 = 0x7ffff7ff0000 "/tmp/crosfw/sandbox/u-boot.dtb"
>> >> >>>>>>> (gdb) q
>> >> >>>>>>>
>> >> >>>>>>>
>> >> >>>>>>> Also the commit message suggests that this patch changes sandbox to
>> >> >>>>>>> use --gc-sections, which is not obvious from the subject. I think that
>> >> >>>>>>> should be a separate commit and in fact it should really be separate
>> >> >>>>>>> commits for each arch, I think. That might help people notice it...
>> >> >>>>>>>
>> >> >>>>>>> I only noticed now since the EFI pull request has landed.
>> >> >>>>>>
>> >> >>>>>> Can you try my bss patch really quick? Maybe we're just overwriting gd.
>> >> >>>>>>
>> >> >>>>>> Alex
>> >> >>>>>>
>> >> >>>>> This patch breaks efi-x86_app_defconfig. The EFI application no longer
>> >> >>>>> boots. I was testing on top of u-boot/master.
>> >> >>>>>
>> >> >>>>> If I do:
>> >> >>>>>
>> >> >>>>> diff --git a/arch/x86/config.mk b/arch/x86/config.mk
>> >> >>>>> index 586e11a..fc119ec 100644
>> >> >>>>> --- a/arch/x86/config.mk
>> >> >>>>> +++ b/arch/x86/config.mk
>> >> >>>>> @@ -24,7 +24,6 @@ endif
>> >> >>>>> ifeq ($(IS_32BIT),y)
>> >> >>>>> PLATFORM_CPPFLAGS += -march=i386 -m32
>> >> >>>>> # TODO: These break on x86_64; need to debug further
>> >> >>>>> -PLATFORM_RELFLAGS += -fdata-sections
>> >> >>>>> else
>> >> >>>>> PLATFORM_CPPFLAGS += $(if $(CONFIG_SPL_BUILD),,-fpic) -fno-common -m64
>> >> >>>>> endif
>> >> >>>>>
>> >> >>>>> Then it boots again. Can you please take a look?
>> >> >>>>>
>> >> >>>>> Regards,
>> >> >>>>> Bin
>> >> >>>>
>> >> >>>> Please can we revert the offending patch quickly for the release? I am
>> >> >>>> not comfortable with the sandbox changes either (data-sections, etc.).
>> >> >>>
>> >> >>> I can not reproduce the sandbox breakage (and travis doesn't seem to either, otherwise it would be broken for everyone, no?). Can you give me some guidelines on how to reproduce the failures for you and I'll just fix it?
>> >> >>
>> >> >> I would like to revert the sandbox changes at least. I don't want to
>> >> >> enable -ffunction-sections, for example.
>> >> >
>> >> > Could you please explain why? In general I always thought the sandbox
>> >> > target was meant as debugging aid which allows you to find and debug
>> >> > bugs more easily.
>> >> >
>> >> > I would assume that chances for breakage are higher with function and
>> >> > data sections, because the linker could remove code it considers dead?
>> >> > So for a debugging target, I would think it makes sense to have it
>> >> > enabled rather than disabled.
>> >>
>> >> Yes I think removing dead could could cause problems. But so could not
>> >> garbage-collecting sections, so it is not a great argument. Sandbox is
>> >> targeted at building as much code as possible. Ideally every piece of
>> >> non-arch-specific code should be built with sandbox.
>> >
>> > I feel like this is the argument we had about enabling gc on other
>> > platforms. gc'ing dead code cannot cause problems that aren't real
>> > bugs. In fact, if we build something and it results in dead code that
>> > we don't expect to be dead that is a problem. We have everything else
>> > doing this collection so I think we do want sandbox to match.
>>
>> Fair enough, this is the consistency argument that Alex made also. I
>> do understand that. Sandbox does potentially have a wider toolchain
>> target, since it uses whatever is installed on the machine, and
>> apparently runs on non-Linux devices. I suppose toolchains that people
>> use handle this reliably now.
>
> Yes, this is something that toolchains get right and have been getting
> right for a long while now. Honestly, I didn't realize that sandbox
> wasn't doing this already or I would have brought it up.
>
>> >> Maybe I am being conservative, but I see no reason to enable it for
>> >> sandbox. I'll try to think of some better reasons and reply if I can.
>> >> I also feel that it slipped in under the radar with no review.
>> >
>> > This is something I want to be sensitive to. If you really want this
>> > change to go in for v2018.11, OK, it's your arch. But I really think we
>> > should move sandbox in this direction for consistency with all the
>> > hardware platforms.
>>
>> At the least I would like a revert followed by a patch to change
>> sandbox over. The commit that changed this is titled:
>>
>> efi_loader: Rename sections to allow for implicit data
>>
>> It does not have a sandbox: tag, nor does it mention the compiler flag
>> changes. No one would guess that this major change is happening. It
>> broke x86 and I am not yet confident that sandbox still functions
>> correctly.
>>
>> I would be happier if the second patch went in for the next release,
>> but if we really are not seeing problems, then OK.
>
> OK, now I'm trying to see what we need to do next. Regardless of Alex's
> current EFI PR, we have 7e21fbca26d1 ("efi_loader: Rename sections to
> allow for implicit data") in master. Looking at the PR, I see that it
> does turn on -fdata-sections on x86, and that Bin reviewed/tested it.
> So I'm going to push the PR there shortly.
>
> If, for sandbox you want to backout --gc-sections now (the commit
> message does spell out that we need -ffunction/data-sections and what
> for) and re-introduce that for the next release please post the patches
> and I'll grab the backout one and apply it (or post it then PR it if
> you'd rather, that's fine too).
I'd like to back out all of the compilel/link flag changes:
Here is my patch:
http://patchwork.ozlabs.org/patch/954875/
Then I would like to get the sandbox EFI stuff reviewed by Alex, and
get that applied when the merge window opens. I can bring it in via
sandbox but I do want Alex to look at it. It has been waiting for two
releases and I want to avoid any more barriers to this work.
Herinrich's testing work is excellent, but having sandbox able to
actually run apps is a good step forward I think.
Regards,
Simon
More information about the U-Boot
mailing list