nRF52840 BLE causing display glitch on GC9A01 using SPIM

[wyatt_tritium]

We have a product based on the nRF52840 which uses the GC9A01 display driven using SPI on battery power. The device also functions as a Bluetooth connectable so it is normally advertising data at a 1 second interval. If the display is left on for several hours while BLE is enabled, the display will randomly become glitched and stays this way until reset. It appears that the BLE is somehow interfering with SPI and causing corrupt data eventually. We are fairly confident that the issue is not hardware related, as we have eliminated noise using stable power supplies and impedance matching the SPI lines.

Is it possible that the nRF is low on resources and corrupting the SPI data or display buffer when using BLE?

The pin-out for the display is as follows.

• DC - P0.06
• CS - P0.08
• RESET - P0.11
• SDA - P0.12
• SCL - P1.09

We are using nRF Connect SDK v2.5.1.

[Naeem Maroof]

Hi,

There is no known issue like this. 

How much resources your program is using? If there is an overflow then the compilation would tell you about that.

If there is run time issue then you should debug and locate where the problem is occurring.

/br

Naeem

[wyatt_tritium]

Here is the build output. There is no overflow.

Memory region         Used Size  Region Size  %age Used
           FLASH:      953705 B         1 MB     90.95%
             RAM:      145660 B       256 KB     55.56%
        IDT_LIST:          0 GB         2 KB      0.00%

The issue occurs during runtime, but I do not see how to debug it because the issue occurs randomly after several hours to several days.

[wyatt_tritium]

Thanks for your insight! We're looking into this and will report back. If you have any additional information on how to separate RAM areas in nRF Connect, please let me know.

[hmolesworth]

Should be straightforward, but I see issues and have not bothered yet to try myself; here's a recent ticket where the user is trying to use a specific RAM address, pretty common requirement. noinit-section-at-a-specific-address

Edit: tried __attribute__((at(0x20001000))) but zephyr ignored me:

struct gpio_callback button_cb_data;
C:\ncs>type C:\ncs\v2.7.0\zephyr\samples\basic\button\build\button\zephyr\zephyr.map | find "button_cb_data"
 .bss.button_cb_data
                0x0000000020000288                button_cb_data

struct gpio_callback button_cb_data __attribute__((at(0x20001000)));
C:\ncs>type C:\ncs\v2.7.0\zephyr\samples\basic\button\build\button\zephyr\zephyr.map | find "button_cb_data"
 .bss.button_cb_data
                0x0000000020000288                button_cb_data

[wyatt_tritium]

We have set `NRFX_SPIM3_NRF52840_ANOMALY_198_WORKAROUND_ENABLED` in `nrfx_config_nrf52840.h` to 1 and running tests now. See this note.

https://infocenter.nordicsemi.com/index.jsp?topic=%2Ferrata_nRF52840_Rev2%2FERR%2FnRF52840%2FRev2%2Flatest%2Fanomaly_840_198.html

[hmolesworth]

Didn't know nRFConnect had this workaround; does the map file show that it works? I'm curious what the setting does to the build. I'll try searching for it ..

This is what I found so far in v2.7.0 with Nordic pre-release:

nrfx_config.h
#ifdef CONFIG_NRF52_ANOMALY_198_WORKAROUND
#define NRFX_SPIM3_NRF52840_ANOMALY_198_WORKAROUND_ENABLED 1

nrfx_config_nrf52840.h
 * @brief NRFX_SPIM3_NRF52840_ANOMALY_198_WORKAROUND_ENABLED
#ifndef NRFX_SPIM3_NRF52840_ANOMALY_198_WORKAROUND_ENABLED
#define NRFX_SPIM3_NRF52840_ANOMALY_198_WORKAROUND_ENABLED 0

[wyatt_tritium]

Enabling the workaround does not seem to do anything. The map files are identical before and after enabling this value. Unless I am not enabling the workaround correctly.

Edit: Is this value enabled by default? I looked in the Kconfig and it was already enabled.

CONFIG_NRF52_ANOMALY_198_WORKAROUND

[hmolesworth]

The 198 handler code is here, not sure how it is ensuring buffer placement:

v2.7.0\modules\hal\nordic\nrfx\drivers\src\nrfx_spim.c

No 198 handler here:

v2.7.0\zephyr\drivers\spi\spi_nrfx_spim.c

Wonder just what this code is doing - 0x40000E00 some secret power control register?

// Workaround for nRF52840 anomaly 198: SPIM3 transmit data might be corrupted.

static uint32_t m_anomaly_198_preserved_value;

static void anomaly_198_enable(uint8_t const * p_buffer, size_t buf_len)
{
    m_anomaly_198_preserved_value = *((volatile uint32_t *)0x40000E00);
    if (buf_len == 0)
    {
        return;
    }
    uint32_t buffer_end_addr = ((uint32_t)p_buffer) + buf_len;
    uint32_t block_addr      = ((uint32_t)p_buffer) & ~0x1FFF;
    uint32_t block_flag      = (1UL << ((block_addr >> 13) & 0xFFFF));
    uint32_t occupied_blocks = 0;
    if (block_addr >= 0x20010000)
    {
        occupied_blocks = (1UL << 8);
    }
    else
    {
        do {
            occupied_blocks |= block_flag;
            block_flag <<= 1;
            block_addr  += 0x2000;
        } while ((block_addr < buffer_end_addr) && (block_addr < 0x20012000));
    }
    *((volatile uint32_t *)0x40000E00) = occupied_blocks;
}

[wyatt_tritium]

Yes it isn't clear to us what the CONFIG_NRF52_ANOMALY_198_WORKAROUND does. It appears to be enabled by default in ncs v2.5.1 so it is clear this is not the fix we need.

Ensure the RAM areas used by the BLE RADIO and the SPIM are in separate RAM AHB bus masters

How can we do this in nRF Connect SDK?

---

We are confident that our issue and solution are related to the SPI speed/channel and the supporting clock. We are running a test with spim1 at 8MHz and so far there are no issues. Unfortunately, we need the spi to run at 32 MHz for better rendering performance and this is only possible with spim3.

So far we have only been using the 32kHz clock. Now we are trying a test where the 32MHz clock is enabled to see if that has any effect on our issue.

Edit: using the 32MHz clock does not seem to have an improvement for this issue.

[hmolesworth]

Finally a simple solution after much time spent by both wyatt and I. Double buffers for SPIM3:

uint8_t Spim3BufferL[8192] __attribute__((section(".spim3_data"), aligned(0x2000), used)) = {0};
uint8_t Spim3BufferR[8192] __attribute__((section(".spim3_data"), aligned(0x2000), used)) = {0};

// zephyr_final_map C:\ncs\v2.7.0\zephyr\samples\basic\button\build\button\zephyr\zephyr_final.map
.spim3_data     0x0000000020002000     0x2000 load address 0x0000000000005dac
 .spim3_data    0x0000000020002000     0x2000 app/libapp.a(main.c.obj)
                0x0000000020002000                Spim3BufferR
                0x0000000020004000                Spim3BufferL

Just need to align to next (any) SRAM AHB section - RAM sections are 0x2000 apart on nRF52833/nRF52840 so could just bump by 0x2000 and stuff at least 0x1000 of data in that section to block other cpu/radio access to same SRAM block as SPIM3.

[hmolesworth]

Some updates:

First note that to search the map file as well as source files the .map file (in "Output Files") has to be open in the Visual Code IDE editor.

Second, if a specific section is not required but just bss or .noinit (.noinit is probably best option) then the placeholder in main.c or other early link file is not required thus:

volatile uint8_t Spim3BufferL[8192] __attribute__((section(".noinit"), aligned(0x2000), used));
volatile uint8_t Spim3BufferR[8192] __attribute__((section(".noinit"), aligned(0x2000), used));
// Map file: .\build\zephyr\zephyr.map
noinit          0x000000002000c000    0x13700
 *(SORT_BY_ALIGNMENT(.noinit))
 .noinit        0x000000002000c000     0x3000 zephyr/libzephyr.a(blahBlah.c.obj)
                0x000000002000c000                Spim3BufferR
                0x000000002000e000                Spim3BufferL

The RAM is broken up into 0x2000 block segments each with its own AHB bus (on the nRF52840) so the buffers can be anywhere in RAM provided they occupy the entire 0x2000 (8192) byte block to stop the linker putting other data into the same block.

I use two buffers, fill one while the other is sending to eliminate any access over the AHB bus the SPIM is using while the transmission is in progress. If transmission is in bursts then a single buffer is all that is required. If monitoring transmitted data by receiving in loopback mode (wise for testing) then you can confirm if any data gets corrupted by comparing Rx and Tx buffers. Rx and Tx buffers require separate AHB busses, so must reserve 8192 bytes each. Loopback by either linking pins or declare the Tx pin to be both input and output and set the Rx pin to the same pin. Even if only using (say) 256 byte SPI data the full 8192 bytes must be reserved. Keep to range RAM0 up to RAM7

// nRF52833/nRF52840 Bus Masters
// =============================
// Each bus master is connected to all the slave devices using an interconnection matrix. The bus masters are
// assigned priorities, which are used to resolve access when two (or more) bus masters request access to
// the same slave device. When that occurs, the following rules apply:
// - If two (or more) bus masters request access to the same slave device, the master with the highest
// priority is granted the access first.
// - Bus masters with lower priority are stalled until the higher priority master has completed its transaction.
// To avoid AHB bus contention when using multiple bus masters, follow these guidelines:
// - Avoid situations where more than one bus master is accessing the same slave.
// - If more than one bus master is accessing the same slave, make sure that the bus bandwidth is not exhausted.
// Bus Slaves (RAM):
//           Name  AdrSpace    StartAdr    EndAdr      AccType
// Memory =  RAM0  Memory      0x20000000  0x20001FFF  RW
// Memory =  RAM1  Memory      0x20002000  0x20003FFF  RW
// Memory =  RAM2  Memory      0x20004000  0x20005FFF  RW
// Memory =  RAM3  Memory      0x20006000  0x20007FFF  RW
// Memory =  RAM4  Memory      0x20008000  0x20009FFF  RW
// Memory =  RAM5  Memory      0x2000A000  0x2000BFFF  RW
// Memory =  RAM6  Memory      0x2000C000  0x2000DFFF  RW
// Memory =  RAM7  Memory      0x2000E000  0x2000FFFF  RW
// Memory =  RAM8  Memory      0x20010000  0x2001FFFF  RW