01:07 <Tartarus> Sigh, LTO causes qemu_arm efi_selftest failure

01:07 <Tartarus> test/py/tests/test_efi_selftest.py:19: in test_efi_selftest

01:07 <Tartarus> disabling, moving on

01:29 <xypron> Tartarus: I guess you refer to test_hii_database_new_package_list crashing. Please, keep the branch. This could also be a bug in the HII protocol implementation.

01:30 <xypron> I will have a look at it.

02:20 <xypron> Tartarus Compiling with lto results in warnings for examples:

02:20 <xypron> +arm-linux-gnueabi-ld.bfd: examples/standalone/hello_world.o: plugin needed to handle lto object

02:20 <xypron> +arm-linux-gnueabi-ld.bfd: examples/standalone/libstubs.o: plugin needed to handle lto object

02:22 <Tartarus> We may need some logic to exempt examples/ from LTO

14:37 <Tartarus> marex: LTO for all of ARM causes failure to build problems (that should be fixable) with gardena-smart-gateway-at91sam omapl138_lcdk pogo_e02 pogo_v4 iconnect openrd_base openrd_client openrd_ultimate sheevaplug dockstar goflexhome nsa310s ti816x_evm imx6q_bosch_acc chromebook_jerry chromebook_speedy gxp corstone1000 imx8mp_rsb3720a1_4G imx8mp_rsb3720a1_6G imx8mm-cl-iot-gate imx8mm-cl-iot-gate-optee librem5 mt7981_emmc_rfb mt7981_rfb

14:37 <Tartarus> mt7981_sd_rfb mt7986_rfb mt7986a_bpir3_emmc mt7986a_bpir3_sd

14:37 <Tartarus> https://source.denx.de/u-boot/u-boot/-/jobs/590013 isn't too bad to read for the 64bit ones

14:37 <Tartarus> https://source.denx.de/u-boot/u-boot/-/jobs/590012 is less helpful

15:24 <Tartarus> One 32bit problem wasn't bad

15:33 <Tartarus> marex: Remember way back when I asked about if you could backport that __gnu_thumb1_case_si like you did some other asm funcs from an older gcc? That's the next blocking bit for default LTO on arm

15:34 <Tartarus> Or at least for more widely

15:34 <Tartarus> doing a no-thumb dep for now

15:52 <Tartarus> Rest of aarch64 seems easy, lets see

16:25 <apalos> so, xypron, Tartarus that linker messup is "known"

16:25 <apalos> mess*

16:25 <apalos> And apparently there are hacks to work around it. But gdb/objdump etc all agree

16:25 <apalos> LTO just completely strips that function....

16:25 <apalos> I'll have a look into known fixes and send something

16:29 <Tartarus> apalos: Well, we could just rework the logic as I suggested :)

16:29 <Tartarus> It's really an arm32 thing

16:30 <apalos> Tartarus: ah i just saw the mail, yea that would work

16:43 <Blok> I have a sama5d35-ek like board that I fail to start U-boot on. Should be trivial, but I fail. I have loaded the at91bootstrap to 0x0 and u-boot to 0x8000 of SPI flash. At the end of the at91bootstrap boot I get "FLASH: copy 0xa0000 bytes from 0x10008000 to 0x26f00000, FLASH: Done to load image" then nothing. I use sama5d3xekdf_uboot as defconfig

16:43 <Blok> for both. Any ideas?

19:05 <Tartarus> marex: Do you have time / inclination to look in to some of the oddities of size growth with LTO?

19:32 <marex> Tartarus: not this and next week, after that, possibly

19:33 <Tartarus> OK

19:33 <Tartarus> I'll make some further comments in the commit and mark the last part as RFC for now

19:34 <Tartarus> There's a few clear issues that I'm resolving for CI to pass

19:34 <Tartarus> And then there's stuff like 0fcc1c76d1acaa68a0675f0baa0e5d9a25908bae doesn't work right with LTO, which may mean there's more issues with weak functions than I thought

19:34 <Tartarus> apalos? :)

19:35 <Tartarus> 915047048f0acd3dbfe8605b854f151815f9be96 is the other half that's not-right

19:35 <Tartarus> What we get is sha256_process_one is not discarded, so 8kb of useless function included

20:32 <rfs613> any tips to debugging why patman seemed to ignore the [bounces] in my ~/.patman file?

20:51 <apalos> Tartarus: ok i'll have a look, i guess that 'urgent' before 2023.04 ?

20:53 <Tartarus> Nope, this is all vs next

20:53 <Tartarus> and may or may not be a good idea for 2023.07

20:54 <apalos> ok, i'll have a relaxed look then :>

20:55 <apalos> do we need LTO for anything other than size reduction?

21:06 <Tartarus> It's fairly significant size reduction, I don't know if anyone has done any performance testing

21:06 <Tartarus> some platforms are 30kb smaller

21:06 <Tartarus> 5-6kb on the low end, 10-15 feels average