On 12/14/2014 09:03 AM, Ivan Shapovalov wrote:
On Saturday 13 December 2014 at 23:38:02, Edward Shishkin wrote:
On 12/12/2014 10:10 PM, Ivan Shapovalov wrote:
So, this patchset uses your algorithm of padding/gluing extents to be discarded.
I've added exact allocation primitives (is it what you've been talking about?
So, we found out that the old bitmap primitives don't allow us to implement
precise discard, and I asked for new ones. Also we wanted to implement the
general case, when precise discard works for any pair (offset, unit).
Currently
it works only for ones determined by the function convert_lattice_params().
For some reasons (that I don't remember) I was sure, that the general case
should be obtained from this partial case by a small transformation of the
function discard_sorted_merged_extents().
Well, I'll try to dive into details and generalize things on the holidays.
If generalization is impossible, then we'll switch to non-precise
discard every
time when convert_lattice_params() fails.
I don't quite understand you here... Do you mean that some non-insane combinations
of discard unit, discard offset and blocksize are not handled (but worth
handling)?
Yes. As you can see, if (discard_offset % block_size != 0), then we
don't issue discard requests. Neither precise, nor non-precise.
It is hard to implement the precise discard support in the generic case
straight away.
TBH, convert_lattice_params() looks magic-ish to me...
OK, I'll try to provide some background (geometric minimum).
Edward.
--
To unsubscribe from this list: send the line "unsubscribe reiserfs-devel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
