On Tue, Jun 01, 2021 at 04:59:52PM +0200, Patrick DELAUNAY wrote: > Hi Marek, > > On 5/20/21 1:23 PM, Marek Behún wrote: > > Hello, > > > > this is version 4 of patches adding support for LTO to U-Boot. > > > > This series is being tested by Github/Azure CI at > > https://github.com/u-boot/u-boot/pull/57 > > > > There is a problem with sandbox_clang test scenario, which I was > > unable to resolve yet, or even determine correctly whether the > > problem is with my patches or was there before. (One of the > > problems is with testing stack protector test_stackprotector.py, > > but for some reason I could not make this work even with GCC on > > my local machine.) > > Nevertheless I am sending these patches now so that at least you > > can review them. In the meantime I will try to resolve the issue > > with sandbox_clang test scenario. > > > > Changes since v3: > > - for some reason the mvneta driver does not work correctly when > > U-Boot is compiled with LTO. I have not debugged this issue yet, so > > for now I have removed Turris Omnia and Turris MOX from devices with > > enabled LTO > > - linker list entry symbols are now forced to emit by using the > > __ADDRESSABLE macro in a generated C file (keep-syms-lto.c). > > Previously this was done in the ll_entry_declare() macro, but this > > was not flexible, since, for example, trying to declare an entry as > > extern could not work that way. This came to attention now after > > Simon's patches for dtoc were merged, and ll_entry_declare() is > > being used with the extern keyword in include/generated/dt-decl.h > > - when compiling LTO with Clang, use llvm-ar and llvm-nm for AR and NM > > - several patches rebased > > - removed patch "api: fix a potential serious bug caused by undef > > CONFIG_SYS_64BIT_LBA" > > - updated patch converting __attribute__((section(...))) to > > __section(...) > > > > Changes sinve v2: > > - now linking with --build-id=none in order to avoid link failures with > > some toolchains (thanks Herald Seiler) > > - we don't use -flto=jobserver anymore, since it causes build errors for > > some people. Instead we link with -flto=NPROC > > - removed LTO exception for arch/arm/mach-omap2/omap3/clock.o, Adam Ford > > says it is not needed > > - added some Reviewed-by tags > > > > Changes since v1: > > - remove patches applied into u-boot-marvell > > - added Reviewed-by tags > > - addressed some issues discovered by Bin Meng, Marek Vasut, > > Heinrich Schuchardt > > - added more ARM boards (thanks to Adam Ford, Tim Harvey and Bin Meng) > > - removed --gc-sections for ARM if internal libgcc is used > > - remove -fwhole-program in final LTO LDFLAGS > > - declared all 4 functions (memcpy, memset, memcmp, memmove) __used, > > (these are mentioned in GCC man page for option -nodefaultlibs that > > the compiler may generate; this seems to be a bug in GCC that linking > > fails with LTO even if these functions are present, because the > > symbols can be renamed on some targets by optimization) > > > > Marek > > > > Marek Behún (36): > > regmap: fix a serious pointer casting bug > > checkpatch: require quotes around section name in the __section() > > macro > > treewide: Convert macro and uses of __section(foo) to __section("foo") > > compiler.h: align the __ADDRESSABLE macro with Linux' version > > test/py: improve regular expression for ut subtest symbol matcher > > string: make memcpy(), memset(), memcmp() and memmove() visible for > > LTO > > efi_loader: fix warning when linking with LTO > > efi_loader: add Sphinx doc for __efi_runtime and __efi_runtime_data > > efi_loader: add macro for const EFI runtime data > > efi_selftest: compiler flags for efi_selftest_miniapp_exception.o > > lib: crc32: put the crc_table variable into efi_runtime_rodata section > > Makefile, Makefile.spl: cosmetic change > > build: use thin archives instead of incremental linking > > build: support building with Link Time Optimizations > > build: link with --build-id=none > > sandbox: errno: avoid conflict with libc's errno > > sandbox: use sections instead of symbols for getopt array boundaries > > sandbox: make LTO available > > sandbox: enable LTO by default > > ARM: global_data: make set_gd() work for armv5 and armv6 > > ARM: make gd a function call for LTO and set via set_gd() > > ARM: fix LTO build for some thumb-interwork cases > > ARM: fix LTO for imx28_xea > > ARM: fix LTO for apf27 > > ARM: fix LTO for keystone > > ARM: kona: fix clk_bsc_enable() type mismatch for LTO > > ARM: imx8m: fix imx_eqos_txclk_set_rate() type mismatch for LTO > > ARM: fix LTO for seaboard > > ARM: fix LTO for rockchip and samsung > > ARM: omap3: fix LTO for DM3730 (and possibly other omap3 boards) > > armv8: SPL: discard relocation information > > ata: ahci: fix ahci_link_up() type mismatch for LTO > > ARM: make LTO available > > ARM: don't use -ffunction-sections/-fdata-sections with LTO build > > ARM: don't use --gc-sections with LTO when using private libgcc > > ARM: enable LTO for some boards > > > > Kbuild | 2 + > > Kconfig | 24 +++++++ > > Makefile | 81 ++++++++++++++++++++-- > > arch/Kconfig | 3 + > > arch/arm/config.mk | 10 ++- > > arch/arm/cpu/arm926ejs/Makefile | 2 + > > arch/arm/cpu/arm926ejs/mxs/mxs.c | 2 +- > > arch/arm/cpu/arm926ejs/mxs/spl_boot.c | 4 +- > > arch/arm/cpu/arm926ejs/spear/spl.c | 2 +- > > arch/arm/cpu/armv7/kona-common/clk-stubs.c | 2 +- > > arch/arm/cpu/armv7/ls102xa/ls102xa_psci.c | 2 +- > > arch/arm/cpu/armv8/spl_data.c | 4 +- > > arch/arm/cpu/armv8/u-boot-spl.lds | 1 + > > arch/arm/include/asm/global_data.h | 6 +- > > arch/arm/include/asm/secure.h | 6 +- > > arch/arm/include/asm/setup.h | 2 +- > > arch/arm/lib/Makefile | 3 + > > arch/arm/lib/sections.c | 31 +++++---- > > arch/arm/lib/spl.c | 2 +- > > arch/arm/mach-at91/spl.c | 2 +- > > arch/arm/mach-exynos/spl_boot.c | 2 +- > > arch/arm/mach-imx/imx8m/clock_imx8mm.c | 2 +- > > arch/arm/mach-imx/imx8m/soc.c | 2 +- > > arch/arm/mach-imx/spl_imx_romapi.c | 16 ++--- > > arch/arm/mach-k3/am642_init.c | 2 +- > > arch/arm/mach-k3/am6_init.c | 2 +- > > arch/arm/mach-k3/j721e_init.c | 4 +- > > arch/arm/mach-keystone/Makefile | 1 + > > arch/arm/mach-mvebu/mbus.c | 4 +- > > arch/arm/mach-mvebu/timer.c | 2 +- > > arch/arm/mach-nexell/clock.c | 6 +- > > arch/arm/mach-nexell/timer.c | 6 +- > > arch/arm/mach-omap2/omap3/Makefile | 1 + > > arch/arm/mach-rockchip/board.c | 2 +- > > arch/arm/mach-socfpga/spl_a10.c | 2 +- > > arch/arm/mach-sunxi/board.c | 2 +- > > arch/arm/mach-tegra/board.c | 2 +- > > arch/arm/mach-tegra/cboot.c | 8 +-- > > arch/arm/mach-tegra/tegra20/Makefile | 1 + > > arch/mips/mach-jz47xx/jz4780/jz4780.c | 2 +- > > arch/nds32/include/asm/setup.h | 6 +- > > arch/powerpc/include/asm/cache.h | 4 +- > > arch/riscv/cpu/cpu.c | 4 +- > > arch/sandbox/config.mk | 14 +++- > > arch/sandbox/cpu/os.c | 3 +- > > arch/sandbox/cpu/start.c | 3 +- > > arch/sandbox/cpu/u-boot-spl.lds | 8 ++- > > arch/sandbox/cpu/u-boot.lds | 8 ++- > > arch/sandbox/include/asm/getopt.h | 2 +- > > arch/sandbox/include/asm/sections.h | 21 +++++- > > arch/sandbox/lib/sections.c | 9 +-- > > arch/x86/cpu/coreboot/timestamp.c | 2 +- > > arch/x86/lib/coreboot/cb_sysinfo.c | 2 +- > > arch/x86/lib/sections.c | 9 +-- > > arch/xtensa/cpu/cpu.c | 2 +- > > board/bosch/shc/board.c | 2 +- > > board/broadcom/bcmstb/bcmstb.c | 2 +- > > board/samsung/arndale/arndale_spl.c | 2 +- > > board/samsung/common/exynos5-dt.c | 2 +- > > board/samsung/smdk5250/smdk5250_spl.c | 2 +- > > board/samsung/smdk5420/smdk5420_spl.c | 2 +- > > board/siemens/draco/board.c | 2 +- > > board/xilinx/common/fru_ops.c | 2 +- > > configs/am3517_evm_defconfig | 1 + > > configs/da850evm_defconfig | 1 + > > configs/da850evm_direct_nor_defconfig | 1 + > > configs/da850evm_nand_defconfig | 1 + > > configs/imx6q_logic_defconfig | 1 + > > configs/imx8mm_beacon_defconfig | 1 + > > configs/imx8mm_venice_defconfig | 1 + > > configs/imx8mn_beacon_2g_defconfig | 1 + > > configs/imx8mn_beacon_defconfig | 1 + > > configs/nokia_rx51_defconfig | 1 + > > configs/omap3_logic_defconfig | 1 + > > configs/r8a774a1_beacon_defconfig | 1 + > > configs/r8a774b1_beacon_defconfig | 1 + > > configs/r8a774e1_beacon_defconfig | 1 + > > drivers/ata/ahci.c | 2 +- > > drivers/bios_emulator/biosemu.c | 4 +- > > drivers/clk/kendryte/clk.c | 4 +- > > drivers/core/regmap.c | 59 +++++++++++++++- > > drivers/firmware/psci.c | 2 +- > > drivers/mtd/nand/raw/mxc_nand_spl.c | 2 +- > > drivers/pinctrl/nxp/pinctrl-imx5.c | 2 +- > > drivers/pinctrl/nxp/pinctrl-imx7.c | 2 +- > > drivers/pinctrl/nxp/pinctrl-imx8m.c | 2 +- > > drivers/power/pmic/pmic_tps62362.c | 2 +- > > drivers/power/pmic/pmic_tps65217.c | 2 +- > > drivers/power/pmic/pmic_tps65218.c | 2 +- > > drivers/power/pmic/pmic_tps65910.c | 2 +- > > drivers/serial/serial_pl01x.c | 4 +- > > include/efi_loader.h | 54 +++++++++++++-- > > include/errno.h | 8 ++- > > include/linker_lists.h | 24 +++---- > > include/linux/compiler.h | 10 +-- > > include/linux/compiler_attributes.h | 2 +- > > lib/crc32.c | 3 +- > > lib/efi_selftest/Makefile | 2 + > > lib/errno.c | 4 +- > > lib/string.c | 9 +-- > > lib/trace.c | 4 +- > > scripts/Makefile.build | 16 ++--- > > scripts/Makefile.lib | 3 + > > scripts/Makefile.spl | 63 +++++++++++++++-- > > scripts/checkpatch.pl | 2 +- > > scripts/gen_ll_addressable_symbols.sh | 12 ++++ > > test/py/conftest.py | 2 +- > > 107 files changed, 513 insertions(+), 183 deletions(-) > > create mode 100755 scripts/gen_ll_addressable_symbols.sh > > > > I am trying to activate CONFIG_LTO for STMicroelectronics stm32mp15 platform > > (stm32mp15_trusted_defconfig and stm32mp15_basic_defconfig). > > > The config stm32mp15_basic_defconfig, with SPL supported failed to boot > > with gcc-arm-9.2-2019.12. > > because a weak function is not correctly handle between > > 1/ ./arch/arm/lib/cache-cp15.c: 96 > >     __weak void dram_bank_mmu_setup(int bank); > > 2/ the stm32mp implementation in > >    arch/arm/mach-stm32mp/cpu.c:62 > >    void dram_bank_mmu_setup(int bank) > > => The used function in SPL is the cache library, not the stm32mp function. > >       and SPL boot it OK when I remove the function in the cache library... Good work figuring that out. It might help lead to people figuring out some other problems with LTO off/on. FWIW, I found using buildman's size display options can also be useful. > And the issue is also solved with I use a more recent version > > gcc-arm-10.2-2020.11 on my PC. > > > Do you see the same kind of issue for old version of gcc ? > > > Do you think CONFIG_LTO require a more recent version of gcc that > > the minimal version required by U-Boot  because I don't see this information > > in the serie? Note that we only test with gcc-9.2.x right now, moving up to gcc-11. But that had its own problems I didn't get feedback on (grub+riscv32 just doesn't build, and that's an upstream problem that's been ignored as of when I posted my patches at least). -- Tom