BAW

I write about opinions in the consumer electronics industry and software , the stories in my experience.

[BAW @beandrewang]

I thought memory allocation was simple: Need 280KB? Just combine twenty-eight 10KB blocks from the pool, right?

Qualcomm’s ⁠pmalloc⁠ on the chip: No. Panic time. 🤡

Got hit with ⁠PANIC_CHIP_PRIVATE_MEMORY_EXHAUSTION⁠ today while working on embedded firmware.

At first, I thought it was basic memory fragmentation. So I went into the ⁠app_pools⁠ config and added smaller blocks like ⁠{10240, 3}⁠ (and more), assuming the allocator would just stitch them together to fulfill my 28145-byte (~27KB) request.

Still crashed. Hard.

Here are the low-level embedded hardware realities I was forcefully reminded of today:

1️⃣ ⁠malloc()⁠ demands strictly continuous physical address space
On bare-metal or RTOS chips without a dynamic MMU to map virtual pages, the memory manager cannot magically chain scattered blocks together. One ⁠malloc(N)⁠ call requires a single, physically contiguous block where size ≥ N.

2️⃣ ⁠pmalloc⁠ uses fixed-size Best-Fit allocation
It matches your request to the smallest available block size ≥ N. Allocating 28,123 bytes from a ⁠{32000, 1}⁠ pool works fine (wasting the leftover byte padding). But allocating 20,000 bytes from a pool of ten 10,000-byte blocks? Instant death.

3️⃣ Physical SRAM limits are unforgiving
Think you can just configure a ⁠{281280, 1}⁠ super-block in ⁠app_pools⁠? The Linker will immediately hit you with ⁠SRAM Overflow⁠. On a BLE/DSP App core, overall SRAM is often just a few hundred KB—one massive block eats the whole house.

🛠️ The Fix:
When you genuinely need massive buffers on constrained microcontrollers:

1. Application-layer Chunking: Maintain a pointer array of twenty-eight 10KB blocks in C, and map global offsets logically in your app code.

2. Offload to Dedicated RAM: Move large buffers out of the general OS ⁠pmalloc⁠ heap and map them directly into PSRAM, Audio, or DSP RAM regions.

High-level languages with garbage collection and virtual memory spoil us. The physical layer always brings you back to reality.

[BAW @beandrewang]

I would say to all the software student and junior software engineers,

DO NOT give up grow your skills of architecture design, failure analysis, performance optimizations.

AI is just your keyboard and Google search, not your enemy.

Integrate pieces to a system is your value.

[BAW @beandrewang]

@yashmp2004 AI is efficient in app levels, low level code is relative to huge modules and hardware, affecting the performance very much.

[BAW @beandrewang]

@jahirsheikh8 Can I say AI just replaces your keyboard and Google search?

[BAW @beandrewang]

@GohilHardy Are you saying software is coding? Big misunderstanding

[BAW @beandrewang]

@felix98813 Why not integrate deepseek to codex. Check it out in my post

[BAW @beandrewang]

Amazing, I made a wire with swift code only. deposit successfully by two working days.

I selected SHA mode, Standard Chartered charged $6.3, Charles Schwab didn’t charge.

Best brokerage firm ever!!!

[BAW @beandrewang]

I figured it out, don’t input ABA routing number though your bank support it, just input swift code if you are making an international wire.

[BAW @beandrewang]

@X X Money works in Japan?

[BAW @beandrewang]

@vikrambuilds If you treat AI as a tool rather than a competitor, you will know how to protect yourself and grow your skills and career.

[BAW @beandrewang]

IC design bugs
SDK bugs

Have you been a Lab mouse when you are working on a brand new SoC?

[BAW @beandrewang]

@Isaac_Dev0 AI is just a tool, it cannot replace software engineers. Like automation doesn’t replace farmers.

[BAW @beandrewang]

@eronbrutii @X Hi

[BAW @beandrewang]

- Will you buy the iPhone duo?

I don’t.

- What do you like?

iPhone 17 standard and iPad mini.

- Rich, I think you absolutely buy iPhone duo,

Yes, of course

[BAW @beandrewang]

@rezwana1926 Yes

[BAW @beandrewang]

@X_supporttop1 Ready

[BAW @beandrewang]

@Mi_Zu_Ki_20 too

[BAW @beandrewang]

@fuyu_kasumii Yes

[BAW @beandrewang]

@X_supporttop1 Hello

[BAW @beandrewang]

@DeveloperNehaa @X lets connect

[BAW @beandrewang]

@hiwise_aigc let’s contact

[BAW @beandrewang]

FUNCTION Initialize_MAX17262(sensor_id):

// Step 0: Read Status register (0x00) and check Power-On Reset (POR) flag
status_reg = Read_Register(MAX17262_STATUS_REG)

// If POR bit (Bit 1) is 0, configuration is intact. Skip full initialization.
IF (status_reg AND POR_MASK) == 0 THEN
LOG_INFO("No POR detected. Fuel gauge configuration is valid.")
Set_Sensor_Healthy(sensor_id, TRUE)
RETURN SUCCESS
END IF

LOG_INFO("POR detected. Starting full MAX17262 initialization.")

// Step 1: Wait until hardware finishes initial conversion (FSTAT.DNR == 0)
IF Wait_Until_Bit_Clear(FSTAT_REG, DNR_MASK, TIMEOUT_MS) == FAIL THEN
LOG_ERROR("Timeout waiting for FSTAT.DNR to clear before config")
RETURN FAIL
END IF

// Step 2: Backup Hibernate configuration and exit Hibernate mode
original_hibcfg = Read_Register(HIBCFG_REG)

IF Exit_Hibernate_Mode(sensor_id) == FAIL THEN
LOG_ERROR("Failed to exit hibernate mode")
RETURN FAIL
END IF

// Step 3: Write custom battery parameters and custom model
IF Load_Battery_Model(sensor_id) == FAIL THEN RETURN FAIL
IF Write_Custom_Parameters(sensor_id) == FAIL THEN RETURN FAIL
IF Load_Optional_Registers(sensor_id) == FAIL THEN RETURN FAIL

// Step 4: Refresh ModelCFG (0xDB) and wait for hardware auto-clear (Bit 15)
IF Initiate_Model_Loading(sensor_id) == FAIL THEN
LOG_ERROR("Failed to initiate model loading")
RETURN FAIL
END IF

IF Write_Update_QR_Table(sensor_id) == FAIL THEN RETURN FAIL

// Restore original Hibernate configuration
Write_Register(HIBCFG_REG, original_hibcfg)

// Step 5: Clear POR bit in Status register (Bitwise AND-NOT)
status_reg = Read_Register(MAX17262_STATUS_REG)
status_reg = status_reg AND (NOT POR_MASK)
Write_Register(MAX17262_STATUS_REG, status_reg)

// Verify POR bit is successfully cleared
verify_status = Read_Register(MAX17262_STATUS_REG)
IF (verify_status AND POR_MASK) != 0 THEN
LOG_ERROR("Failed to clear POR bit in Status register")
RETURN FAIL
END IF

// Step 6: Wait until FSTAT.DNR clears to guarantee SOC calculation is ready
IF Wait_Until_Bit_Clear(FSTAT_REG, DNR_MASK, TIMEOUT_MS) == FAIL THEN
LOG_ERROR("Timeout waiting for FSTAT.DNR to clear after config")
RETURN FAIL
END IF

Set_Sensor_Healthy(sensor_id, TRUE)
LOG_INFO("MAX17262 initialization complete.")
RETURN SUCCESS

END FUNCTION

[BAW @beandrewang]

1/3 🧵 Why is your ADI/Maxim MAX17262 fuel gauge always reading 1% SOC after initialization?
Many embedded developers see a fixed 1% battery state after an MCU reset. After hours spent blaming the hardware, it usually turns out to be flawed driver logic.
Here are 2 fatal code pitfalls causing the 1% SOC bug 👇

[BAW @beandrewang]

3/3 💡 Correct MAX17262 Init Flow:
Check POR: Only run full init if ⁠Status.POR == 1⁠. If ⁠0⁠, skip init.
Config: Exit Hibernate & load Custom Model parameters.
Wait Refresh: Poll until ⁠ModelCFG.Refresh⁠ (Bit 15) clears automatically.
Clear POR: Use ⁠&= ~POR_MASK⁠.
Wait Ready: Ensure ⁠FSTAT.DNR == 0⁠ before reading SOC.
Always distinguish MCU resets from fuel gauge power loss!