kernel: pipe: simple test fails for pipe write / read of 3 bytes

[cfriedt]

Describe the bug
There is a page fault when a simple test is run using k_pipe_put() and k_pipe_get(). No combination of static or __aligned(8) on the buffers or k_pipe resolves the issue.

If it were a stack smash, then making the pipe or the buffers static would presumably the issue.

Note: if BOARD=qemu_cortex_m3, then it works fine.

Also worrisome - test still passes with BOARD=qemu_x86_64 in spite of the page fault 🤨

void test_pipe_simple_write_read(void)
{
	int res;
	size_t bytes_written = 0;
	size_t bytes_read = 0;

	u8_t buf[16];
	u8_t rxbuf[16];
	u8_t txbuf[16];
	struct k_pipe p;

	txbuf[0] = 'H';
	txbuf[1] = 'i';
	txbuf[2] = '!';
	txbuf[3] = '\0';

	k_pipe_init(&p, buf, sizeof(buf));

	res = k_pipe_put(&p, txbuf, 3, &bytes_written,
			 3 /* min_xfer */, K_FOREVER);
	zassert_equal(res, 0, "did not write entire message");
	zassert_equal(bytes_written, 3, "did not write entire message");

	res = k_pipe_get(&p, rxbuf, 3, &bytes_read,
			 3 /* min_xfer */, K_FOREVER);
	zassert_equal(res, 0, "did not read at least one byte");
	zassert_equal(bytes_read, 3, "did not read all bytes available");

	zassert_equal(rxbuf[0], 'H',
		"rxbuf[0] expected: 'H' actual: 0x%02x", rxbuf[0]);
	zassert_equal(rxbuf[1], 'i',
		"rxbuf[1] expected: 'i' actual: 0x%02x", rxbuf[1]);
	zassert_equal(rxbuf[2], '!',
		"rxbuf[2] expected: '!' actual: 0x%02x", rxbuf[2]);
}

To Reproduce
Steps to reproduce the behavior:

1. west build -b qemu_x86_64 -p always tests/kernel/pipe/pipe
2. ninja -C build run
3. See error

Expected behavior
Expect tests to pass

Impact
blocking #24486
blocking implementation of #24366

Screenshots or console output

starting test - test_pipe_simple_write_read
E: Page fault at address 0x15cf2f (error code 0x4)
E: Linear address not present in page tables
E: PML4E: 0x0000000000134827 Writable, User, Execute Enabled
E: PDPTE: 0x0000000000135827 Writable, User, Execute Enabled
E:   PDE: 0x0000000000136827 Writable, User, Execute Enabled
E:   PTE: Non-present
E: RAX: 0x00000000000000cf RBX: 0x000000000013cf58 RCX: 0x0000000000000000 RDX: 0x0000000000000010
E: RSI: 0x000000000013cf28 RDI: 0x000000000013cf58 RBP: 0x0000000000000000 RSP: 0x000000000013cef0
E:  R8: 0x000000000013cff0  R9: 0x000000000013b000 R10: 0x0000000000000000 R11: 0x0000000000000202
E: R12: 0x0000000000000000 R13: 0x0000000000000000 R14: 0x0000000000000000 R15: 0x0000000000000000
E: RSP: 0x000000000013cef0 RFLAGS: 0x0000000000000206 CS: 0x003b CR3: 0x0000000000116000
E: RIP: 0x000000000010f1c3
E: >>> ZEPHYR FATAL ERROR 0: CPU exception on CPU 0
E: Current thread: 0x00000000001283c0 (unknown)
Caught system error -- reason 0
PASS - test_pipe_simple_write_read

Environment (please complete the following information):

• OS: Linux (Debian 10.2)
• Toolchain: cross-compile (aarch64-linux-gnu -> x86_64-linux-gnu)
• Toolchain: cross-compile (x86_64-linux-gnu -> x86_64-linux-gnu)
• Commit SHA or Version used: 8d984b3

Additional context
detected in #24486

[cfriedt]

Aha!
So it seems that k_pipes and other objects for sure cannot be stack-allocated. I looked through some more of the documentation and saw k_object_alloc() and friends.

The example test should have been in tests/kernel/pipe/pipe_api rather than tests/kernel/pipe/pipe.

I'll need to adjust the code in my implementation of #24366 to use k_object_alloc() before I call k_pipe_init(). The test in #24486 can simply use the statically-defined / initialized test_pipe.

[cfriedt]

@carlescufi - feel free to assign to me and close, or I can close it later.