stm32h7 tests: spi_loopback fails due to mem_region

[FRASTM]

Describe the bug
The tescase tests/drivers/spi/spi_loopback/ now fails when ruuning on a stm32h7 target

After the merge of #57786
There is a mem region definition missing to PASS the testcase

To Reproduce
Steps to reproduce the behavior:

1. west build -p auto -b nucleo_h753zi tests/drivers/spi/spi_loopback/
2. west flash
3. connect D11 <--> D12 on the nucleo 144 target board
4. See error

Expected behavior
test PASSED

PASS - test_spi_loopback in 0.049 seconds
===================================================================
TESTSUITE spi_loopback succeeded

------ TESTSUITE SUMMARY START ------

SUITE PASS - 100.00% [spi_loopback]: pass = 1, fail = 0, skip = 0, total = 1 dus
 - PASS - [spi_loopback.test_spi_loopback] duration = 0.049 seconds

------ TESTSUITE SUMMARY END -----

Impact
CI failure

Logs and console output

*** Booting Zephyr OS build zephyr-v3.4.0-1785-gdfeb2d902756 ***                
Running TESTSUITE spi_loopback                                                  
===================================================================             
START - test_spi_loopback                                                       
I: SPI test on buffers TX/RX 0x240001b5/0x240009ce                              
I: SPI test slow config                                                         
I: Start complete multiple                                                      
E: Code -14                                                                     

    Assertion failed at WEST_TOPDIR/zephyr/tests/drivers/spi/spi_loopback/src/spi.c:111: spi_complete_multiple: (ret is true)
SPI transceive failed                                                                                                                                                                              
 FAIL - test_spi_loopback in 0.024 seconds                                      
===================================================================             
TESTSUITE spi_loopback failed.  

Environment (please complete the following information):

• OS: Ubuntu
• Toolchain 0.16.1
• Commit SHA : dfeb2d9

[FRASTM]

@dgastonochoa what would be a right definition to prepare a spi buffer in no cache region ?

[dgastonochoa]

@dgastonochoa what would be a right definition to prepare a spi buffer in no cache region ?

Hi @FRASTM ,

First, enable a memory region that DMA can access and set it as NOCACHE, e.g.:

&sram2 {
	status = "okay";
	zephyr,memory-attr = "RAM_NOCACHE";
};

Then, declare a buffer in that region, e.g.:

#define NOCACHE_SECTION __attribute__((section("SRAM2")))
#define ALIGNED __attribute__((aligned(32)))
...
static uint32_t s_buf[32] NOCACHE_SECTION ALIGNED;

Finally, you can use it normally:

    s_buf[0] = 0xaaaaaaaa;
    s_buf[1] = 0x55555555;
    struct spi_buf tx_buf = {
        .buf = s_buf, .len = 8,
    };
    struct spi_buf_set tx_bufs = { .buffers = &tx_buf, .count = 1 };


    int rc = spi_write(spi1, &cfg, &tx_bufs);

You can see a complete example in #57786 (comment), see "How to test" section.

To execute the test, you can follow the instructions in the PR description at the top of the page. It's a normal Zephyr app. so it only needs to be copied to the right place and built and flashed with west on a Nucleo board.

Regards.

[dgastonochoa]

I have tried to reproduce the error as explained in the description but I get a build failure. However, I can build and run the same test for h743.

Also, I have tried to find references (grep) to h753 within tests/drivers/spi/spi_loopback (to see if I can understand why the build fails), but I didn't find any.

[FRASTM]

That does not seem sufficient,
the SPI test on buffers TX/RX 0x300200a0/0x30020080 --> SRAM2
but still fails even if the SRAM2 is defined in the soc/arm/st_stm32/stm32h7/mpu_regions.c

	MPU_REGION_ENTRY("SRAM2", 0x30020000,
					 REGION_RAM_NOCACHE_ATTR(REGION_128K)),

[dgastonochoa]

That does not seem sufficient, the SPI test on buffers TX/RX 0x24000260/0x24000240 --> not SRAM2 area

I think I see the problem. The buffers in this test are declared as __attribute__((__section__(".nocache")));. That puts them in a nocache memory area (sram0) but that is not declared as so in devicetree. Because of this, even them being in a nocache region, the STM32 SPI driver does not recognize it as so.

As a quick solution for the tests, I would set sram2 as nocache in the required devicetree files and then declare the buffers to be in this region, in other words, replace this

static __aligned(32) char buffer_tx[BUF_SIZE] __used __attribute__((__section__(".nocache")));

by this

#define NOCACHE_SECTION __attribute__((section("SRAM2")))
...
static __aligned(32) char buffer_tx[BUF_SIZE] __used NOCACHE_REGION

and so on for the rest of the buffers. I have tried this with h743 (because I cannot build h753) and it works.

Also, sram0 could be made "zephyr,memory-region" in stm32h743.dtsi, and then mark it as nocache in devicetree. Although I imagine if it is not "zephyr,memory-region" yet, it is for a reason.

As I see it, the problem is that sram0 is not marked as nocache in devicetree, but it ends up being so by a different mechanism; Or more in general, there are 2 different mechanisms to achieve a nocache area and they can clash. One solution would be to abandon the __attribute__((__section__(".nocache"))); mechanism in favor the devicetree mechanism, so there is no room for contradictions. However, I understand this breaks backwards compatibility. Then, the STM32 SPI driver should check if the region is nocache regardless of what devicetree says but I'm not sure how to do this. It could also check the linker symbols:

_nocache_ram_size
_nocache_ram_end
_nocache_ram_start

to verify the buffer is actually in the nocache region. Although these symbols could not exist if the non-devicetree mechanism has not been used. Their existence can be checked either by defining them as __attribute__((weak)) and then checking if they are NULL or checking if CONFIG_NOCACHE_MEMORY is defined.

[FRASTM]

I have a branch https://github.com/FRASTM/zephyr/tree/issue60977 for testing on the nucleo_h753zi
(should also add the following to the the prj.conf when building $ west build -p auto -b nucleo_h753zi tests/drivers/spi/spi_loopback/) :

CONFIG_SPI_STM32_DMA=y
CONFIG_SPI_STM32_INTERRUPT=n
CONFIG_SPI_ASYNC=n
CONFIG_NOCACHE_MEMORY=y

[dgastonochoa]

I have tried that branch and still don't see any reference to h753 in tests/spi/spi_loopback.

[FRASTM]

1. soc/arm/st_stm32/stm32h7/mpu_regions.c does not require the MPU_REGION_ENTRY for the SRAM2
2. Adding the

&sram2 {
	zephyr,memory-attr = "RAM_NOCACHE";
};

to the sram2 node (or through the overlay file for that board tests/drivers/spi/spi_loopback/boards/nucleo_h753zi.overlay)
will improve the test result:

START - test_spi_loopback
I: SPI test on buffers TX/RX 0x300200a0/0x30020080
I: SPI test slow config
I: Start complete multiple
I: Passed
I: Start complete loop
I: Passed
I: Start null tx
E: Code -14

    Assertion failed at WEST_TOPDIR/zephyr/tests/drivers/spi/spi_loopback/src/spi.c:230: spi_null_tx_buf: (ret is true)
SPI transceive failed

[FRASTM]

I have tried that branch and still don't see any reference to h753 in tests/spi/spi_loopback.

Use the existing https://github.com/zephyrproject-rtos/zephyr/tree/main/tests/drivers/spi/spi_loopback

[dgastonochoa]

Use the existing https://github.com/zephyrproject-rtos/zephyr/tree/main/tests/drivers/spi/spi_loopback

Not sure what do you mean, I have checked out your branch and executed the command you provided above and the still the tests don't build. I still cannot find references to h7 so it seems to me the tests don't exist in this branch.

With regards to the test that fails:

The test that keeps failing says:

/* According to documentation, when sending NULL tx buf -
		 *  NOP frames should be sent on MOSI line
		 */

To achieve this, the if ((tx_bufs != NULL && !spi_buf_set_in_nocache(tx_bufs)) statement needs to be modified to account for the fact that, if tx_bufs contains one buffer and it is NULL, it should not return error. So replace this

if ((tx_bufs != NULL && !spi_buf_set_in_nocache(tx_bufs))

by

static bool is_null_buffer(const struct spi_buf_set *tx_bufs)
{
   return (tx_bufs.count ==1 && tx_bufs.buffers[0].buf == NULL;
}

if ((tx_bufs != NULL && !is_null_buffer(tx_bufs) && !spi_buf_set_in_nocache(tx_bufs))

Or something similar.

[dgastonochoa]

Use the existing https://github.com/zephyrproject-rtos/zephyr/tree/main/tests/drivers/spi/spi_loopback

Nevermind, I needed to pull.