Fix LTO & -O2 offset out of range

[swift-tk]

The arch.shared_interrupt.lto.speed compile fail on apollo5 platform.
Remove ldr instruction and pass function pointer by input operand.

This PR could fix #90777.

[sonarqubecloud]

Quality Gate passed

Issues
2 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud

[swift-tk]

This wouldn’t work for platform that can take only two input operands… Any suggestion on making the two function pointer less PC relative so it doesn’t exceed ldr offset?

[wearyzen]

This wouldn’t work for platform that can take only two input operands… Any suggestion on making the two function pointer less PC relative so it doesn’t exceed ldr offset?

@swift-tk I haven't tried this but I think preloading the pointers in r4 r5 could work, something like :

register uintptr_t __unlock_reg  __asm__("r6");
__unlock_reg = (uintptr_t)arch_irq_unlock_outlined;

and then

__asm__ volatile(
...
"movs  r0, #0       \n"
"blx   r6           \n"
...
)

[swift-tk]

@wearyzen Thanks for the help, the compile seems to pass now.

[sonarqubecloud]

Quality Gate passed

Issues
2 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud

[swift-tk]

The modified version works on apollo5 platform, but for some reason, qemu_cortex_m0/nrf51822 still tries to use r5 and r6 as input operands when they are in the clobber list.. Thus emits impossible operand constraints... Deleting them from clobber seems to do the trick for compile but failed twister tests. I think those entry point in r5 and r6 eventually gets lost.

Using volatile keyword on register value wouldn't work as they gets optimized out....

[wearyzen]

Hi @swift-tk I tried to reproduce the error today with west build -p always -b apollo510_evb tests/kernel/interrupt -T arch.shared_interrupt.lto.speed but wasn't able to. Could you try to rebase your branch to main and try reproducing the issue without this fix? I think we don't need any change at all for this if rebased to main.

[swift-tk]

Hi @swift-tk I tried to reproduce the error today with west build -p always -b apollo510_evb tests/kernel/interrupt -T arch.shared_interrupt.lto.speed but wasn't able to. Could you try to rebase your branch to main and try reproducing the issue without this fix? I think we don't need any change at all for this if rebased to main.

There was a workaround merged to main to allow us merge our blocked PRs, but it is not ideal. I think it is better to try solving it fundamentally.
The commit is a54197b.
The issue might appear again even with this if there are other changes in tree.

Please also see #90777 (comment) for the root cause.

[stephanosio]

See #92569