build/tools/amebasmart/gnu_utility: fix stuck tash when key held on boot

[lhenry-realtek]

This fix does the following:

After KM0 core start, IER bit set to 0 to disable LOGUART After CA32 ipc_send_message in board_initialize(), set IER bit to 1 in KM0 IPC command handler (ROM LOGUART is disabled here) CA32 can use tizen RAM handler now, and tash shell is working properly

Root cause:

• Due to multicore environment, LOGUART control is passed from KM4->KM0->CA32 on boot, where TASH shell is executing on AIDual.
• Control is passed by disabling interrupt handling for LOGUART on each core, until only 1 core is using it
• When TizenRT is starting up, in board_initialize there is a short period of time before CA32 sends IPC to disable KM0's LOGUART interrupt handler. In this short period it is possible for CA32 LOGUART handler and KM0 LOGUART handler to fire.
• Due to handling of the Rx interrupt event enable (ERBFI) bit in each handler, it is possible for KM0 side to read and set an invalid value of the enable bit as CA32 side is also accessing it
• When the ERBFI bit is 0 after KM0 finishes IRQ and then KM0 LOGUART is disabled, CA32 now has access to the LOGUART peripheral but no events will fire, as the bit is 0, thus tash shell also hangs

[lhenry-realtek]

Please check rtk jira SDAIOT30E-291 for more info, there is another proposed fix for same issue

[sunghan-chang]

Please check rtk jira SDAIOT30E-291 for more info, there is another proposed fix for same issue

Did you leave this for RTK engineers?

[lhenry-realtek]

Hi @sunghan-chang,

We are discussing the way to implement this with Dane, I was unsure if you had access to our outward-facing jira as I am relatively new to the project. I will paste the triage and suggestion here

---

Cause due to LOGUART ERBFI interrupt bit not being re-enabled, causing the peripheral to not generate any data receive events from the Host (PC) input

This cause Tash shell to also lock up due to shared peripheral, so the bit is re-enabled in the IPC handler of KM0 when IPC_A2L_DISLOGUART is received by KM0

Currently we have found an alternative solution, so the fix can be introduced in 2 methods

---

The fix is a one liner: LOGUART_INTConfig(LOGUART_DEV, LOGUART_BIT_ERBI, ENABLE);

For Method 1, this line is placed inside KM0's IPC handler, which is fired after https://github.com/Samsung/TizenRT/blob/2f722e6c907268b892fb1a77719857f3ea05ddb6/os/board/rtl8730e/src/rtl8730e_boot.c#L513

For Method 2, instead of placing it in our KM0 bin, we set the bit after the line above in rtl8730e_boot.c:

ipc_send_message(IPC_AP_TO_LP, IPC_A2L_DISLOGUART, &ipc_msg_loguart);

/* enable event generation for rx data interrupt again */
LOGUART_INTConfig(LOGUART_DEV, LOGUART_BIT_ERBI, ENABLE);

[lhenry-realtek]

This fix does the following:

after KM0 core start, IER bit set to 0 after CA32 ipc_send_message, set IER bit to 1 in KM0 handler (ROM LOGUART is disabled here) CA32 can use tizen RAM handler now, and tash shell is working properly

Root cause will follow in a separate comment after I have confirm with Dane, but it is a concurrency issue due to LOGUART peripheral being shared by KM0, KM4 and CA32, during init certain operations overwrite the interrupt bit and cause the peripheral to lock up in an invalid state. Once locked up, it will not respond to any sort of key press, thus causing Tash shell to hang as well as it rely on LOGUART at low level

[sunghan-chang]

@lhenry-realtek Could you leave what you said in the commit description? @jylee9613 Could you verify this?

[lhenry-realtek]

Use loadable_apps_ext_ddr or ext_psram configuration for quickest reproduce

Negative example: Clone TizenRT master branch, use above config and disable USB so all the output is on UART only, then press and hold a key while reset board to boot. Shell will hang

Positive example: Clone TizenRT master branch and apply this patch, repeat above step

[lhenry-realtek]

• Move set bit operation out of KM0 handler and into CA32 board_initialize
• Apply suggested changes from Samsung such as critical section around setting LOGUART bit
• Move rootcause into commit message