Re: [RFC PATCH 1/3] Unified trace buffer

[Martin Bligh]

> OK, then how about this?
>
> Each page will start with a time stamp (I'm still aligning everything by 8
> bytes, just because it simplifies things). Then we can have a 3 byte
> (24 bit) counter offset? Then we can have a header that looks like:
>
> struct {
>        unsigned char time[3];
>        unsigned char length;
>        unsigned char buff[];
> };
>
> This still allows me to have the 2048 byte size buffer.
>
> Or is 24 bits for time too small? The offest will be from the previous
> entry, and not the beginning of the page.
>
> If one defines a fixed size entry, we could just use the full 32 bits for
> the timestamp, since the length will be ignored in that case, and will
> become part of the buffer.
>
> Hence,
>
> struct {
>        unsigned int time;
>        unsigned char length;
>        unsigend char buff[];
> };

How about we just steal 5 bits from the timestamp to indicate event
lengths up to 32 bytes, and if it's 0, that means there's a length
field following? Also that'd mean you could use a longer length field
and get beyond 256 bytes to 4096, without impacting most events.

struct {
        u32 length:5, time_delta:27;
        u16 length;
        u8 buf[];
};

struct {
        u32 length:5, time_delta:27;        /* where length == 0 */
        u8 buf[];
};

Obviously we could less than 5 bits, even just 1 for a flag ...
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/