read array of 4 bytes atomically using read of 32-bit

Question

I am managing an array of 4 flags of uint8_t each.

uint8_t flags[4].

these flags can change by different threads at any moment.

i want to be able to get a snapshot of these flags by using one atomic operation. lets say i do the following :

uint32_t temp_val = read_atomic32((uint32_t *) &flags[0])

now i cast back to an array of flags:

uint8_t *my_flags = &temp_val;
printf("flag zero is %d\n, my_flags[0]);

i am suspecting few problems here:

maybe flags[4] is not aligned for uint32_t reads and might give a non-aligned crash or prehaps the atomic functionality will not be garunteed because of nonaligned access.
what about endians? should a problem occur? even though i am using uint32_t i am later on casting it back to an array of uint8_t and i don't think the memory layout will be changed because of this action - still, maybe i am missing something here. i am assuming here that if i read a memory location for 4 bytes, these pattern will be the same until cast back no matter if your machine is little endian or big endian.
is there a better way to manage 4 independent flags of uint8_t but still be able to read them as whole using one action in a portable safe way?

Jens · Accepted Answer · 2013-07-09 20:24:45Z

2

Place your flags in a union:

union {
  uint8_t  c[4];
  uint32_t t;
} flags;

flags.c[0] = ...;
flags.t = 0xffffffffU;

This takes care of the alignment issue and you can use the type punning to access the flags via t without a cast. Not quite kosher according to strict ISO C rules, but usually does what you want.

Endianness becomes an issue the moment you assign literal values (as in my example) or expect certain values when accessing flags.t. Just reading a value and then writing one just read (unmodified--no bit twiddling!) should be okay.

answered Jul 9, 2013 at 20:24

Jens

73.3k16 gold badges134 silver badges186 bronze badges

Sign up to request clarification or add additional context in comments.

7 Comments

Itay Marom Over a year ago

thanks. so using a union will force the compiler to put this structure on an aligned address as the code can access t.

Jens Over a year ago

Exactly. The alignment requirement of a union is the maximum alignment requirement of all its members. Since all members must be at offset zero this guarantees that c is aligned just like t.

Zan Lynx Over a year ago

Type punning with unions works in most compilers for practical reasons. However I believe that by the standard a compiler is allowed to store flags.c and flags.t in different registers which would result in disconnected values.

Jens Over a year ago

@ZanLynx I don't believe so because the Standard guarantees that any object can be accessed as an array of unsigned char. And all three of flags, c and t must have the same address.

Itay Marom Over a year ago

@Jens i didn't quite understand your comment about endianness. having flags.t used only for reading the array pattern and then simply assinging it to another flags2.t to be able to read flags2.c should be ok on all systems, do you agree ?

|

Collectives™ on Stack Overflow

read array of 4 bytes atomically using read of 32-bit

1 Answer 1

7 Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

7 Comments

Your Answer

Sign up or log in

Post as a guest

Related