pktbuf: optimize size overhead in packet buffer

[miri64]

As long as PKTBUF_SIZE remains under a certain value, the bit-size of the size parameter of packets can be smaller and thus saving a few byte of loss in the packet buffer due to overhead.

On most platforms every packet now needs at least 2 byte less of pktbuf space, reducing the overhead from 9 byte to 7.

[Kijewski]

If you want to safe bits:

#if PKTBUF_SIZE < (1ull << 7)
typedef uint8_t _pktsize_t;
#elif PKTBUF_SIZE < (1ull << 15)
typedef uint16_t _pktsize_t;
#elif PKTBUF_SIZE < (1ull << 31)
typedef uint32_t _pktsize_t;
#elif PKTBUF_SIZE < (1ull << 63)
typedef uint64_t _pktsize_t;
#endif

typedef struct __attribute__((packed)) _packet {
     volatile struct _packet *next;
     _pktsize_t _size;
} _packet_t;

static inline int packet_get_processing(const _packet_t *p) {
     return p->_size & 1;
}

static inline _pktsize_t packet_get_size(const _packet_t *p) {
     return p->_size >> 1;
}

static inline void packet_set_size(_packet_t *p, _pktsize_t size) {
     p->_size = (size << 1) | (p->_size & 1);
}

static inline void packet_set_processing(_packet_t *p, int processing) {
     p->_size = (p->_size >> 1 << 1) | !!processing;
}

static inline void packet_set_size_and_processing(_packet_t *p, _pktsize_t size, int processing) {
     p->_size = (size << 1) | !!processing;
}

[miri64]

First of all: uint64_t for size just isn't reasonable, second of all: yes, but that would require also some heavy rewriting of the whole module, for what I currently do not have the nerves ;-)

[miri64]

Furthermore, I still don't know if it might be a good idea to make size public (but not processing), having the model in mind

[miri64]

Furthermore 2, your code implies, that processing is boolean, which it isn't. It's a [edit]semaphoric[/edit] counter.

[OlegHahm]

ACK - 2 bytes are better than nothing.