It's funny to see what originally started as a simple list of rbyd attrs
slowly morph into a full isa. But it makes sense. What we really want is
an abstract description of operations that can be played and replayed as
necessary to atomically update the mtree.
Using a fixed lfs3_rattr_t struct to represent this in C is easy, and
avoids strict-aliasing issues, but ultimately limited when it comes to
the wide-range of data we want to attach to attributes.
Unlike a computer's isa, we want to be able to include full 12-24 byte
branch pointers directly in the instruction!
---
So here's a full variable-length isa organized by words (max(uintptr_t,
uint32_t)).
The first 32-bit word extends the 16-bit tag with an extra 16-bits of
control information:
wwll llff ffcc cccc tttt tttt tttt tttt
^'-.-''-.-''--.--' : :
'--|----|-----|----:-----------------:-- compressed weight
:: '----|-----|----:-----------------:-- total len
:: '-----|----:-----------------:-- from encoder
:: '----:-----------------:-- optional count
:: rgmm kkkk -kkk kkkk
11 => w=-1 ^^ ^ '-.' '---.---'
00 => w=0 '|-|---|------|------ rm bit
01 => w=+1 '-|---|------|------ grow bit
10 => w=attached '---|------|------ mask bits
'------|------ tag suptype
'------ tag subtype
The 4-bit length field always encodes the full length of the
instruction, including the instruction itself and optional weight. The
4-bit from + 6-bit count fields operate independently and tell
lfs3_rbyd_appendrattr_ how to actually encode the data related to the
instruction.
To work around strict-aliasing issues, complex structs are expected to
be broken down into words and reconstructed in lfs3_rbyd_appendrattr_.
Most of our structs are organized into words anyways. For example:
// new child
*r++ = LFS3_RATTR(5, LFS3_TAG_BRANCH, -2, LFS3_FROM_BRANCH);
*r++ = LFS3_RATTR_WEIGHT(+child_->weight);
*r++ = LFS3_RATTR_ARG(child_->blocks[0]);
*r++ = LFS3_RATTR_ARG(child_->trunk);
*r++ = LFS3_RATTR_ARG(child_->cksum);
This also changes rattr-lists to be null-terminated, which makes a bit
more sense in a variable-length isa:
*r++ = LFS3_RATTR_NULL; // all zeros, including length
One concern with null-terminated rattr-lists is how easy it is to
forget the null-terminator, but an assert that all non-null rattrs have
non-zero length seemed to catch the many many mistakes during adoption.
Alternatively, separate LFS3_FROM_NULL/LFS3_FROM_NIL from fields could
be used if encoding space gets tight.
I'm also quite happy with the 2-bit weight feild, which allows omitting
the optional weight word for -1,0,+1 weights. These should cover at
least all mdir operations.
Note the exact encoding of the rattr fields is less of a concern than
the tag fields, as it doesn't reside on-disk can be changed on whim.
---
Saves a nice chunk of code and stack:
code stack ctx
before: 35920 2280 660
after: 35324 (-1.7%) 2176 (-4.6%) 660 (+0.0%)
code stack ctx
gbmap before: 38812 2296 772
gbmap after: 38156 (-1.7%) 2192 (-4.5%) 772 (+0.0%)
The stack savings are obvious, but the code savings a bit less so. A
variable length isa _is_ more complicated, but by limiting most encoding
decisions to compile-time (2-bit weights vs 32-bit weights for example),
the savings from fewer word manipulations on the stack wins.
This more closely matches behavior of functions like mkdir and remove,
even though mkgbmap/rmgbmap operate on a special object and not files.
Besides, returning an error is more useful as users are always free to
ignore said error.
Adds what appears to be one literal to mkgbmap (curiously not rmgbmap?
snuck into alignment?):
code stack ctx
before: 35968 2280 660
after: 35968 (+0.0%) 2280 (+0.0%) 660 (+0.0%)
code stack ctx
gbmap before: 38824 2296 772
gbmap after: 38828 (+0.0%) 2296 (+0.0%) 772 (+0.0%)
- lfs3_gbmap_set* -> lfs3_gbmap_mark*
- lfs3_alloc_markfree -> lfs3_alloc_adopt
- lfs3_alloc_mark* -> lfs3_alloc_markinuse*
Mainly for consistency, since the gbmap and lookahead buffer are more or
less the same algorithm, ignoring nuances (lookahead only ors inuse
bits, gbmap rebuilding can result in multiple snapshots, etc).
The rename lfs3_gbmap_set* -> lfs3_gbmap_mark* also makes space for
lfs3_gbmap_set* to be used for range assignments with a payload, which
may be useful for erased ranges (gbmap tracked ecksums?)
These two functions allow changing whether or not the gbmap is in use
after format:
// Enable the global on-disk block-map
//
// Returns a negative error code on failure. Does nothing if a gbmap
// already exists.
int lfs3_fs_mkgbmap(lfs3_t *lfs3);
// Disable the global on-disk block-map
//
// Returns a negative error code on failure. Does nothing if no gbmap
// is found.
int lfs3_fs_rmgbmap(lfs3_t *lfs3);
rmgbmap was easy enough, but implementing mkgbmap turned out to be
surprisingly tricky due to how gstate permeates the system:
- Even if we zero gstate when we removing the gbmap, mounting the
image on a driver that doesn't understand the gbmap results in garbage
gstate over time as mdir compacts drop unknown gdeltas.
I think this sort of implicit gdelta cleanup is a good thing, but the
possibility of garbage gstate is a bit annoying.
Example A: the dbg scripts are currently printing a bunch of warnings
for corrupt gstate that can be safely ignored.
To support recovering from garbage gstate in mkgbmap, I changed
lfs3_fs_commitgdelta to _always_ track p state even when disabled. We
already needed to do this in lfs3_fs_flush/consumegdelta anyways,
since we don't know if the gbmap is used until parsing wcompat flags.
- The commit that enables the gbmap is tricky. We need the gbmap enabled
to calculate the new gdelta, but we also need it disabled so we don't
traverse the existing gbmap_p (which may be garbage).
As a workaround I added gbmap.b_p, which is in theory redundant with
gbmap_p, but (1) avoids needing to decode gbmap_p during traversals,
and (2) allows the two to temporarily fall out-of-sync in mkgbmap.
This means we potentially have 5 (!) snaphots flying around when
rebuilding the gbmap, which is starting to get a bit silly. But this
was also motivated by gbmap_p decoding adding roughly the same amount
of RAM to lfs3_mtree_traverse_, so the total RAM usage should in
theory be roughly the same.
There might be a better solution, but this at least gets mkgbmap
working. The gbmap builds are not our most RAM senstive configurations
anyways.
---
Also added a couple more tests in test_gbmap to test these:
- test_gbmap_files
- test_gbmap_rmgbmap
- test_gbmap_mkgbmap
- test_gbmap_rmmkgbmap
- test_gbmap_mkrmgbmap
And an explicit wraparound test to test_alloc. This was loosely implied
by the nospc tests, but it's probably better to have an explicit test.
The only downside is this implementation is limited to files:
- test_alloc_wraparound_files
---
Note we are currently dealing with three different configurations:
no-gbmap (the default), yes-gbmap (LFS3_YES_GBMAP), and maybe-gbmap
(LFS3_GBMAP + LFS3_F_GBMAP at runtime).
It only makes sense to include these in maybe-gbmap mode, so this is the
only mode with a notable code increase. However these functions are
relatively cheap. The stack/ctx changes also affect yes-gbmap, but
should mostly cancel out, see above:
code stack ctx
no-gbmap before: 37168 2352 684
no-gbmap after: 37168 (+0.0%) 2352 (+0.0%) 684 (+0.0%)
code stack ctx
maybe-gbmap before: 39292 2456 800
maybe-gbmap after: 39688 (+1.0%) 2392 (-2.6%) 852 (+6.5%)
code stack ctx
yes-gbmap before: 39116 2456 800
yes-gbmap after: 39156 (+0.1%) 2392 (-2.6%) 852 (+6.5%)
Having gbmap/bmap used in different places for the same thing was
confusing. Preferring gbmap as it is consistent with other gstate (grm
queue, gcksums), even if it is a bit noisy.
It's interesting to note what didn't change:
- The BM* range tags: LFS3_TAG_BMFREE, etc. These already differs from
the GBMAP* prefix enough, and adopting GBM* would risk confusion for
actual gstate.
- The gbmap revdbg string: "bb~r". We don't have enough characters for
anything else!
- dbgbmap.py/dbgbmapsvg.py. These aren't actually related to the gbmap,
so the name difference is a good thing.
Christopher Haster