Unaligned access on M0 cores in FAT file system

[LMESTM]

Description

While running SPI SD card tests on STM32 L0 / M0 targets (e.g. NUCLEO_F091RC), the target would freeze and test end up in timeout:

[1498578756.92][CONN][RXD] >>> Running case #2: 'SPI - SD card exists'...
[1498578756.97][CONN][INF] found KV pair in stream: {{__testcase_start;SPI - SD card exists}}, queued...
[1498578796.53][HTST][INF] test suite run finished after 40.16 sec...
[...]
[1498578796.58][HTST][INF] {{result;timeout}}

Command used for test:
mbed test -m NUCLEO_F091RC -t GCC_ARM -n tests-api-spi

###Analysis
Investigations have shown that the platform aborts after a tentative to write 32bits to an unaligned address:

0x08007eb2 in disk_ioctl (pdrv=0 '\000', cmd=2 '\002',
    buff=0x200029a2 <_main_stack+3514>)
    at C:\Data\Workspace\ci-test-shield\mbed-os\features\filesystem\fat\FATFileSystem.cpp:220
220                     *((DWORD*)buff) = size;

(gdb) bt
#0  0x08007eb2 in disk_ioctl (pdrv=0 '\000', cmd=2 '\002',  buff=0x200029a2 <_main_stack+3514>)
    at C:\Data\Workspace\ci-test-shield\mbed-os\features\filesystem\fat\FATFileSystem.cpp:220
#1  0x08005b6e in find_volume (rfs=0x2000292c <_main_stack+3396>,  path=0x20002928 <_main_stack+3392>, wmode=0 '\000')
    at C:\Data\Workspace\ci-test-shield\mbed-os\features\filesystem\fat\ChaN\ff.cpp:2250
#2  0x08006230 in f_mount (fs=0x20002998 <_main_stack+3504>,    path=0x200029d2 <_main_stack+3562> "", opt=1 '\001')
    at C:\Data\Workspace\ci-test-shield\mbed-os\features\filesystem\fat\ChaN\ff.cpp:2451
#3  0x08008074 in FATFileSystem::mount (this=0x20002984 <_main_stack+3484>,    bd=0x200029d8 <_main_stack+3568>, mount=true)
    at C:\Data\Workspace\ci-test-shield\mbed-os\features\filesystem\fat\FATFileSystem.cpp:268
#4  0x08007fb8 in FATFileSystem::mount (this=0x20002984 <_main_stack+3484>,    bd=0x200029d8 <_main_stack+3568>)
    at C:\Data\Workspace\ci-test-shield\mbed-os\features\filesystem\fat\FATFileSystem.cpp:250
#5  0x0800031c in test_card_present () at .\TESTS\API\SPI\SPI.cpp:67
#6  0x0800a910 in utest::v1::Harness::run_next_case ()
    at C:\Data\Workspace\ci-test-shield\mbed-os\features\frameworks\utest\source\utest_harness.cpp:338
#7  0x0800ac42 in utest_us_ticker_run ()
    at C:\Data\Workspace\ci-test-shield\mbed-os\features\frameworks\utest\source\utest_shim.cpp:123
#8  0x0800a1bc in utest::v1::Harness::run (specification=...)
    at C:\Data\Workspace\ci-test-shield\mbed-os\features\frameworks\utest\source\utest_harness.cpp:155
#9  0x080005f2 in main () at .\TESTS\API\SPI\SPI.cpp:165

Proposed way forward

Need to align the struct overall or at least the ssize member ...

$ git diff features/filesystem/fat/ChaN/ff.h
diff --git a/features/filesystem/fat/ChaN/ff.h b/features/filesystem/fat/ChaN/ff.h
index 3653a80..b9cc04c 100644
--- a/features/filesystem/fat/ChaN/ff.h
+++ b/features/filesystem/fat/ChaN/ff.h
@@ -85,7 +85,7 @@ typedef struct {
        WORD    id;                             /* File system mount ID */
        WORD    n_rootdir;              /* Number of root directory entries (FAT12/16) */
 #if _MAX_SS != _MIN_SS
-       WORD    ssize;                  /* Bytes per sector (512, 1024, 2048 or 4096) */
+       WORD    ssize __attribute__ ((aligned (32)));                   /* Bytes per sector (512, 1024, 2048 or 4096) */
 #endif
 #if _FS_REENTRANT
        _SYNC_t sobj;                   /* Identifier of sync object */

[0xc0170]

cc @geky @simonqhughes

[geky]

Ah! Good catch! It looks like the issue here is actually an incorrect cast in the fatfs's disk_ioctl function (the porting layer from BlockDevice -> ChanFS).

The disk_ioctl function uses a void* for different argument types, and is incorrectly being interpreted as a 32-bit DWORD variable when ChanFS is expecting a 16-bit WORD.

Pull request up here:
#4652

Let us know if this doesn't fix the issue.

[geky]

I think this is fixed now, feel free to reopen if I'm wrong : )

[LMESTM]

I confirm this has been fixed indeed ! Thx