Revert "ace: mm: tlb: Check tlb translation enabled before flushing cache"

[lyakh]

This reverts commit 311ddf9. That commit breaks firmware boot on both MTL and LNL.

[lyakh]

https://sof-ci.01.org/sofpr/PR9455/build8033/devicetest/index.html
https://sof-ci.01.org/sofpr/PR9455/build8034/devicetest/index.html

[jsarha]

The Zephyr main now indeed crashes at boot at least on MTL, and reverting 311ddf9 fixes it.

[softwarecki]

I have checked on our mtl fpga test platforms and the firmware boots successfully using sof/main + zephyr/main. Is there any overlay being used? In particular the debugging options like:
CONFIG_ZTEST, CONFIG_SOF_BOOT_TEST, CONFIG_DEBUG or CONFIG_ASSERT? I will check on rvp tomorrow.

[softwarecki]

ACK. Activating the CONFIG_ASSERT option actually break firmware loading. It seems that somewhere we trying to unmap an memory address that is not mapped. After my fix the function returns an appropriate error code. In my opinion we should look for the cause and not treat the effect by reverting my commit.

[lyakh]

testing with thesofproject/sof#9455

ACK. Activating the CONFIG_ASSERT option actually break firmware loading. It seems that somewhere we trying to unmap an memory address that is not mapped. After my fix the function returns an appropriate error code. In my opinion we should look for the cause and not treat the effect by reverting my commit.

@softwarecki sorry, which your fix? I don't see any open PRs from you for Zephyr ATM?

[softwarecki]

@lyakh The one you're trying to revert in this PR. It causes the sys_mm_drv_unmap_page_wflush function to return an error code when trying to unmap a memory page that is not mapped. This error is probably detected by unmap_locked or sys_mm_drv_unmap_region_initial and the firmware stops at assert. I think we should stop trying to unmap an unmapped memory instead of throwing the error checking code away.

[kv2019i]

@softwarecki Ack a fix is preferred, but time is of essence here as well. Let's go with standard procedure. If you can fix the MTL/LNL boot assert issue quickly, please submit ASAP, or if not, let's merge this revert PR and @softwarecki you submit the original patch again later with the updates to avoid panics with asserts. This is currently blocking all Zephyr updates to SOF mainline, so we need to address this quickly to not block other developers while this issue is addressed for Intel.

[softwarecki]

@kv2019i I will create a PR with the fix right away.

[lyakh]

I will block it as fix is proposed in #78612

@softwarecki have you also submitted a SOF PR to test your fix?

[softwarecki]

Approve to not block others