Currently clear_screen() calls set_pixel() to set all pixels. However,
the actual order of pixels being set depends on the framebuffer
orientation. With NORMAL orientation, the framebuffer is accessed
sequentially; with LEFT_UP/RIGHT_UP orientation, it is accessed back and
forth, leading to performance drop (>1 second on bugzzy).
Therefore, ensure sequential access to the framebuffer, regardless of
the orientation.
BUG=b:194967458
TEST=emerge-cherry libpayload
BRANCH=dedede
Change-Id: Iecaff5b6abc24ba4b3859cbc44c0d61b2a90b2d9
Signed-off-by: Yu-Ping Wu <yupingso@chromium.org>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/57104
Reviewed-by: Hung-Te Lin <hungte@chromium.org>
Reviewed-by: Julius Werner <jwerner@chromium.org>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
The timer structure (in particular, the offset to memory addresses)
on recent MTK SoCs for example MT8195 has been changed.
BUG=b:195274787
Signed-off-by: Rex-BC Chen <rex-bc.chen@mediatek.com>
Change-Id: Ifd6ff65a825c4309c47f3b115b80a8ecd42fedac
Reviewed-on: https://review.coreboot.org/c/coreboot/+/56845
Reviewed-by: Hung-Te Lin <hungte@chromium.org>
Reviewed-by: Rex-BC Chen <rex-bc.chen@mediatek.corp-partner.google.com>
Reviewed-by: Yu-Ping Wu <yupingso@google.com>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Building nvramcui with i386-elf-gcc (coreboot toolchain
v2021-04-06_7014f8258e) 8.3.0 and Link Time Optimization (LTO) enabled
in libpayload (`CONFIG_LP_LTO=y`) fails with the error below.
LPGCC nvramcui.bin
curses/PDCurses/pdcurses/refresh.c: In function 'wrefresh':
curses/pdcurses-backend/pdcdisp.c:217:4: error: 'bg' may be used uninitialized in this function [-Werror=maybe-uninitialized]
curses/pdcurses-backend/pdcdisp.c:214:18: note: 'bg' was declared here
curses/pdcurses-backend/pdcdisp.c:217:4: error: 'fg' may be used uninitialized in this function [-Werror=maybe-uninitialized]
curses/pdcurses-backend/pdcdisp.c:214:14: note: 'fg' was declared here
lto1: all warnings being treated as errors
lto-wrapper: fatal error: i386-elf-gcc returned 1 exit status
compilation terminated.
/opt/xgcc/lib/gcc/i386-elf/8.3.0/../../../../i386-elf/bin/ld.bfd: error: lto-wrapper failed
collect2: error: ld returned 1 exit status
`pair_content()` returns in case `PAIR_NUMBER(attr)` is invalid, so
guard the usage of `serial_set_color()`.
if (pair < 0 || pair >= COLOR_PAIRS || !fg || !bg)
return ERR;
Note, building with x86_64-linux-gnu-gcc-10 (Debian 10.2.1-6) 10.2.1
20210110 does *not* fail.
Change-Id: Ic63e34f2b5bc9f826db37597bebc6b20542481d7
Signed-off-by: Paul Menzel <pmenzel@molgen.mpg.de>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/51914
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Angel Pons <th3fanbus@gmail.com>
This was originally several commits that had to be squashed into one
because the intermediate states weren't able to build coreboot:
- one to remove everything that wasn't our own code, leaving only
regex.[ch], toada.c, description.md and Makefile.inc.
- one to copy in Linux 5.13's scripts/kconfig and adapt Makefile.inc
to make the original Makefile work again.
- adapt abuild to use olddefconfig, simplifying matters.
- apply patches in util/kconfig/patches.
- Some more adaptations to the libpayload build system.
The patches are now in util/kconfig/patches/, reverse applying them
should lead to a util/kconfig/ tree that contains exactly the Linux
version + our own 5 files.
Change-Id: Ia0e8fe4e9022b278f34ab113a433ef4d45e5c355
Signed-off-by: Patrick Georgi <pgeorgi@google.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/37152
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Raul Rangel <rrangel@chromium.org>
CB:51638 separated Chrome OS NVS from global NVS by allocating it
separately in CBMEM. CNVS is used in depthcharge to fill firmware
information at boot time. Thus, location of CNVS needs to be shared in
coreboot tables for depthcharge to use.
This change adds a new coreboot table tag
`CB_TAG_ACPI_CNVS`/`CB_TAG_ACPI_CNVS`(0x41) which provides the
location of CNVS in CBMEM to payload (depthcharge).
Additionally, CB:51639 refactored device nvs(DNVS) and moved it to the
end of GNVS instead of the fixed offset 0x1000. DNVS is used on older
Intel platforms like baytrail, braswell and broadwell and depthcharge
fills this at boot time as well. Since DNVS is no longer used on any
new platforms, this information is not passed in coreboot
tables. Instead depthcharge is being updated to use statically defined
offsets for DNVS.
BUG=b:191324611, b:191324611
TEST=Verified that `crossystem fwid` which reads fwid information from
CNVS is reported correctly on brya.
Signed-off-by: Furquan Shaikh <furquan@google.com>
Change-Id: I3815d5ecb5f0b534ead61836c2d275083e397ff0
Reviewed-on: https://review.coreboot.org/c/coreboot/+/55665
Reviewed-by: EricR Lai <ericr_lai@compal.corp-partner.google.com>
Reviewed-by: Ivy Jian <ivy_jian@compal.corp-partner.google.com>
Reviewed-by: Tim Wawrzynczak <twawrzynczak@chromium.org>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Add a Kconfig option to set the keyboard translation state on exit and
set the default to true. This restores the keyboard to the power-up
defaults for firmware that does not always run libpayload keyboard init
to have consistent state, and provides an option to disable translation
for keyboards that might need it.
Change-Id: I25dfe3f425a5bb57e97476564886672b707aa3bd
Signed-off-by: Duncan Laurie <dlaurie@google.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/52737
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Furquan Shaikh <furquan@google.com>
Reviewed-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-by: Felix Singer <felixsinger@posteo.net>
Reviewed-by: Raul Rangel <rrangel@chromium.org>
Reviewed-by: Nico Huber <nico.h@gmx.de>
If we select scancode set #1 and keep that, it can confuse Linux
with keyboards that don't return to set #2 when asked to load the
defaults. This happens for instance with various integrated Think-
Pad keyboards but was also seen with an external PS/2 one.
The chosen configuration, scancode set #2 without translation, seems
to be the default for many systems. So we can expect other payloads
and kernels to work with it.
Change-Id: I28d74590e9f04d32bb2bbd461b67f15014f927ec
Signed-off-by: Nico Huber <nico.h@gmx.de>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/47594
Reviewed-by: Angel Pons <th3fanbus@gmail.com>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Instead of ignoring keyboards indefinitely when they failed to
initialize, we wait 5s and then start over with the hotplug
detection. As we always assume a present keyboard at first,
we'd otherwise never have a chance to hot plug a device after
the initial 30s timer ran out.
Change-Id: I8dec4921b2e932442d52b5118cdcf27090633498
Signed-off-by: Nico Huber <nico.h@gmx.de>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/48774
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Angel Pons <th3fanbus@gmail.com>
While we assume a keyboard is attached, we send an echo command every
500ms. If there is no data coming from the keyboard within 200ms, we
assume it was detached.
Correspondingly, if we assume no keyboard is attached, we run an echo
command once per second.
Change-Id: I2c75182761729bf30711305f3d8b9d43eafad675
Signed-off-by: Nico Huber <nico.h@gmx.de>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/47593
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Angel Pons <th3fanbus@gmail.com>
This is already the case on x86 but not on the ARM platforms, and
{read,write}[bwl] are using volatile pointers, too, so follow suit.
Change-Id: I6819df62016990e12410eaa9c3c97b8b90944b51
Signed-off-by: Patrick Georgi <pgeorgi@google.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/50918
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Angel Pons <th3fanbus@gmail.com>
The new `Makefile.payload` can be included by the Makefiles of pay-
loads for in-tree builds. The basic idea is to use libpayload's
build results without the `make install` step, and to ensure that
incremental builds work. For instance, if libpayload's code changes,
a `make` for the payload would automatically update the libpayload
build and rebuild the payload. But if there are no code changes in
libpayload, only updated files of the payload will be re-built.
The configuration of libpayload is supposed to be automatically
generated from a `defconfig` file. If this `defconfig` changes,
libpayload and the payload will be re-built.
Change-Id: If5319f1bf0bcd09964416237c5cf7f8e59f487a2
Signed-off-by: Nico Huber <nico.h@gmx.de>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/47633
Reviewed-by: Angel Pons <th3fanbus@gmail.com>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Add read64 and write64 for consistency with x86.
BUG=b:178785769
Change-Id: I342e3a23201d0b804ea5ecfe47ee3e4bb516de4c
Signed-off-by: Patrick Georgi <pgeorgi@google.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/50115
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Julius Werner <jwerner@chromium.org>
It either doesn't exist (in-tree builds) or is the same as $_LIBDIR.
Change-Id: I9551cbfc3295d86c22a3785be7cdc0f65eeb08c4
Signed-off-by: Nico Huber <nico.h@gmx.de>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/47632
Reviewed-by: Angel Pons <th3fanbus@gmail.com>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
We only need `$_OBJ` in the include path for in-tree builds. Also,
curses only need special handling for those and PDCurses turned out
to need many more include paths.
Change-Id: Idd29ef33065033e26ba61b09d412d8ca3566d643
Signed-off-by: Nico Huber <nico.h@gmx.de>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/47631
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Angel Pons <th3fanbus@gmail.com>
Instead of checking for an already fully build `libpayload.a`, we check
for the `libpayload.config` which is the actual prerequisite to start
using `lpgcc`. This will allow compilation of payload sources before or
in parallel with the build of `libpayload.a`.
Change-Id: Ic0143fefe33560af8b013ae48bbbe231b3ad46f3
Signed-off-by: Nico Huber <nico.h@gmx.de>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/48938
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Raul Rangel <rrangel@chromium.org>
Introduce a `$_OBJ` variable, that points to the build directory for
in-tree usage of `lpgcc`. If unset, the default `../build` relative
to the location of `lpgcc` is used.
Change-Id: I35112d7533d69aa51252dd2bceec010a62522403
Signed-off-by: Nico Huber <nico.h@gmx.de>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/47629
Reviewed-by: Raul Rangel <rrangel@chromium.org>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
This should make it easier to find the correct config for in-tree
builds.
Change-Id: I08d396ae3cedc65f63c4b8865701ea123c7d56cb
Signed-off-by: Nico Huber <nico.h@gmx.de>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/47628
Reviewed-by: Raul Rangel <rrangel@chromium.org>
Reviewed-by: Angel Pons <th3fanbus@gmail.com>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Keep libpayload's xcompile in its build dir. While we are at it,
align things with the top-level version.
Having `.xcompile` in a central place led to race conditions when
multiple payloads try to build their own libpayloads in parallel.
Change-Id: I504e1862db79b368289867f7568c9169f27a1549
Signed-off-by: Nico Huber <nico.h@gmx.de>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/47651
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Raul Rangel <rrangel@chromium.org>
Reviewed-by: Angel Pons <th3fanbus@gmail.com>
The keyboard self-test is required for some devices. At least one
device (integrated keyboard in a ThinkPad X201) actually starts the
test automatically leading to spurious output and no response for
the first seconds.
We wait up to 5s for the self-test result. On failure or timeout,
the command will be repeated until the 30s init timer runs out. This
happens all in the background of the UI polling loop.
To not unnecessarily delay the boot process, we first try an oppor-
tunistic initialization which skips the self-test.
Change-Id: Ie07b31e74d06e116ac81e76309621eed39a19b49
Signed-off-by: Nico Huber <nico.h@gmx.de>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/47088
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Angel Pons <th3fanbus@gmail.com>
Will be used to time out in states that don't always advance.
Change-Id: I28235e7638d8157cedf81fd915a41d28a1fc070b
Signed-off-by: Nico Huber <nico.h@gmx.de>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/47087
Reviewed-by: Angel Pons <th3fanbus@gmail.com>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
We'll process the init sequence as part of the polling loop. This
should have several advantages:
* It eases error handling, i.e. we can return to an earlier state.
* We don't have to stall initialization when a keyboard takes a
little longer.
* Generally, these keyboards can be hot-plugged (albeit not by
design).
Change-Id: I9cf5cf31eb420b3994bec20e56a72d37f3d2996e
Signed-off-by: Nico Huber <nico.h@gmx.de>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/47086
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Angel Pons <th3fanbus@gmail.com>
Draining the keyboard's buffer is only possible when the keyboard
port is enabled. We should also disable input scanning before, as
the buffer could be filled again with new keystrokes otherwise.
Change-Id: Ibac9c0d04880ff4a3efda5ac53da2f9731f6602c
Signed-off-by: Nico Huber <nico.h@gmx.de>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/47085
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Angel Pons <th3fanbus@gmail.com>
Move the input-buffer draining into a function. It uses the low-level
i8042 API directly to avoid conflicts with changes in the high-level
keyboard API.
Change-Id: I9427c5b8be4d59c2ee3da12d6168d34590043682
Signed-off-by: Nico Huber <nico.h@gmx.de>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/47084
Reviewed-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-by: Paul Menzel <paulepanter@users.sourceforge.net>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Even if we are careful, it's still possible that we read spurious
data from the keyboard, e.g. keystrokes. Namely, when we send the
reset/disable command, there is a race before the command is pro-
cessed.
So we should always process data from the keyboard in a loop. We
break it, when an ACK (0xfa) or a NAK (0xfe) is received, and warn
on unexpected data unless it might be due to the mentioned race.
This also gives us the opportunity to use command-specific timeouts
which we take from Linux: 1s for the keyboard self-test (as there
are keyboards that perform the test before acking the command) and
200ms for all other commands.
Change-Id: I60a2643a8ff4b9231c63bf970c8749c97c7d8926
Signed-off-by: Nico Huber <nico.h@gmx.de>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/47083
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Angel Pons <th3fanbus@gmail.com>
Some background first: The original XT keyboards used what we call
scancode set #1 today. The PC/AT keyboards introduced scancode set #2,
but for compatibility, its controller translated scancodes back to
set #1 by default. Newer keyboards (maybe all we have to deal with)
also support switching the scancode set.
This means the translation option in the controller and the scancode
set selection in the keyboard have to match. In libpayload, we only
support set #1 scancodes. So we either need the controller's trans-
lation on and set #2 selected in the keyboard, or the controller's
translation off and set #1 selected in the keyboard.
Valid configurations:
* SET #1 + XLATE off
* SET #2 + XLATE on
Both with and without the PC_KEYBOARD_AT_TRANSLATED option, we were
only configuring one of the two settings, leaving room for invalid
configurations. With this change, we try to select scancode set #2
first, which seems to be the most supported one, and configure the
controller's translation accordingly. We try to fall back to set #1
on failure.
We also keep translation disabled during configuration steps to
ensure that the controller doesn't accidentally translate confi-
guration data.
On the coreboot side, we leave the controller's translation at its
default setting, unless DRIVERS_PS2_KEYBOARD is enabled. The latter
enables the translation unconditionally. For QEMU this means that
the option effectively toggles the translation, as QEMU's controller
has it disabled by default. This probably made a lot of earlier
testing inconsistent.
Fixes: commit a95a6bf646 (libpayload/drivers/i8402/kbd: Fix qemu)
The reset introduced there effectively reverted the scancode
selection made before (because 2 is the default). It's unclear
if later changes to the code were only necessary to work
around it.
Change-Id: Iad85af516a7b9f9c0269ff9652ed15ee81700057
Signed-off-by: Nico Huber <nico.h@gmx.de>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/46724
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Angel Pons <th3fanbus@gmail.com>
This change adds details about the memory map windows to translate
addresses between SPI flash space and host address space to coreboot
tables. This is useful for payloads to setup the translation using the
decode windows already known to coreboot. Until now, there was a
single decode window at the top of 4G used by all x86
platforms. However, going forward, platforms might support more decode
windows and hence in order to avoid duplication in payloads this
information is filled in coreboot tables.
`lb_spi_flash()` is updated to fill in the details about these windows
by making a call to `spi_flash_get_mmap_windows()` which is
implemented by the driver providing the boot media mapping device.
BUG=b:171534504
Signed-off-by: Furquan Shaikh <furquan@google.com>
Change-Id: I00ae33d9b53fecd0a8eadd22531fdff8bde9ee94
Reviewed-on: https://review.coreboot.org/c/coreboot/+/48185
Reviewed-by: Duncan Laurie <dlaurie@chromium.org>
Reviewed-by: Tim Wawrzynczak <twawrzynczak@chromium.org>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Provide get_mmu_ranges() for ARM64 to let payloads could get
MMU ranges for all used memory regions.
BUG=b:171858277
TEST=Build in x86, arm, arm64.
emerge-zork libpayload depthcharge
emerge-nyan libpayload depthcharge
emerge-asurada libpayload depthcharge
Signed-off-by: Meng-Huan Yu <menghuan@google.com>
Change-Id: I39b24aefc9dbe530169b272e839d0e1e7c697742
Reviewed-on: https://review.coreboot.org/c/coreboot/+/48113
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Julius Werner <jwerner@chromium.org>
Reviewed-by: Yu-Ping Wu <yupingso@google.com>
This turns on the compiler's printf style format string checker.
BUG=b:167517417
TEST=enabled all USB controllers on volteer and fixed resulting
compiler errors when USB_DEBUG is enabled.
Change-Id: Ic94ebcbafdde8a5f79278b5635111b99af40f892
Signed-off-by: Caveh Jalali <caveh@chromium.org>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/45025
Reviewed-by: Tim Wawrzynczak <twawrzynczak@chromium.org>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
This fixes format string mismatch errors in the USB subsystem found by
the compiler's format string checker.
BUG=b:167517417
TEST=enabled all USB controllers on volteer and fixed resulting
compiler errors when USB_DEBUG is enabled.
Change-Id: I4dc70baefb3cd82fcc915cc2e7f68719cf6870cc
Signed-off-by: Caveh Jalali <caveh@chromium.org>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/45024
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Tim Wawrzynczak <twawrzynczak@chromium.org>
The current timeout of 500ms is too low. For instance self-test
of the KBC integrated into IT8516E took almost 1s in tests. We
already check for presence of the KBC before the self-test. So
the timeout should only trigger on a hardware defect and we can
leave some margin.
Change-Id: I95f01a4e605a9c7deb894a71e102c3a881759bb1
Signed-off-by: Nico Huber <nico.h@gmx.de>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/47588
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Angel Pons <th3fanbus@gmail.com>
Work on this mainboard was abandoned and never finished. It's not really
usable in its current state, so let's get rid of it.
Signed-off-by: Julius Werner <jwerner@chromium.org>
Change-Id: I4cd2e2cd0ee69d9846472653a942fa074e2b924d
Reviewed-on: https://review.coreboot.org/c/coreboot/+/47427
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Stefan Reinauer <stefan.reinauer@coreboot.org>
Reviewed-by: Angel Pons <th3fanbus@gmail.com>
The OHCI header file declares various enums as follows:
enum { ... } enum_name;
Since the name is at the end, this is actually declaring a variable
called enum_name and *not* a type, which is causing a multiple
definition error in GCC 10. Move the enum_name before the opening brace
to prevent this.
Signed-off-by: Jacob Garber <jgarber1@ualberta.ca>
Change-Id: I452c0a1b118990942aa53f1e7e77f5e8378e8975
Reviewed-on: https://review.coreboot.org/c/coreboot/+/47224
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-by: Nico Huber <nico.h@gmx.de>
Headers in libpayload define various structs like so:
struct struct_name { ... } __packed;
However, these header files do not include the compiler.h macro that
defines what __packed is, so they are actually defining a variable named
__packed and *not* declaring a packed struct. This leads to defining the
same variable multiple times, which was caught by GCC 10. Add compiler.h
to the compiler parameters so it is included in all files automatically.
Signed-off-by: Jacob Garber <jgarber1@ualberta.ca>
Change-Id: Ia67182520dc94149e06fe9e03a14b3fc2ee29973
Reviewed-on: https://review.coreboot.org/c/coreboot/+/47153
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Nico Huber <nico.h@gmx.de>
Link time optimization is a technique for whole-program optimization.
Instead of doing code generation during compilation, the compiler saves
its intermediate representation to the object files. During the final
linking step, it will then merge all the object files together and
perform optimizations on the entire program. This can often reduce the
final binary size, but also may increase the total compilation time.
This patch introduces a Kconfig option for enabling link time
optimization in libpayload. Since libpayload does no linking of its own,
its LTO archive files will contain only IR and no generated code.
Downstream projects will need to use LTO-aware tools when manipulating
the archives (eg. gcc-ar and gcc-nm), but otherwise do not need to use
LTO themselves -- the compiler will recognize which files are LTO and
which are not, so enabling this option should mostly be "drop in".
For example, when building coreinfo.elf using tinycurses libpayload:
binary size compilation time
default 114 KiB 11.49s
LTO 95 KiB 10.36s
In this case the total compilation time was actually shorter -- despite
the final linking step taking longer, this was offset by the shorter
compilation times for each individual file (since there is no code gen
until the very end).
Change-Id: I048f2ff6298ed0d891098942e1e8b29d35487b91
Signed-off-by: Jacob Garber <jgarber1@ualberta.ca>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/38291
Reviewed-by: Nico Huber <nico.h@gmx.de>
Reviewed-by: Martin Roth <martinroth@google.com>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Yu-Ping Wu