kerf-multikernel 0.4.0__tar.gz → 0.5.0__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/PKG-INFO +1 -1
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/pyproject.toml +1 -1
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/__init__.py +1 -1
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/cli.py +1 -1
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/create/main.py +11 -10
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/daxfs/__init__.py +2 -2
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/daxfs/mkdaxfs.py +30 -16
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/delete/main.py +1 -1
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/devices.py +34 -19
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/dtc/extractor.py +3 -12
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/dtc/overlay.py +4 -4
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/dtc/parser.py +10 -37
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/dtc/validator.py +11 -11
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/exec/main.py +6 -1
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/init/main.py +36 -56
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/load/main.py +86 -41
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/metadata.py +13 -11
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/models.py +4 -3
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/pool_diff.py +3 -3
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/resources.py +8 -4
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/show/main.py +46 -31
- kerf_multikernel-0.5.0/src/kerf/topology.py +253 -0
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/unload/main.py +17 -12
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/update/main.py +2 -2
- kerf_multikernel-0.4.0/src/kerf/topology.py +0 -139
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/LICENSE +0 -0
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/Makefile +0 -0
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/README.md +0 -0
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/build.py +0 -0
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/init/Makefile +0 -0
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/init/init.c +0 -0
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/init/proto.h +0 -0
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/init/session.c +0 -0
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/init/session.h +0 -0
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/init/session_test.c +0 -0
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/baseline.py +0 -0
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/console/__init__.py +0 -0
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/console/main.py +0 -0
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/create/__init__.py +0 -0
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/data/__init__.py +0 -0
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/delete/__init__.py +0 -0
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/docker/__init__.py +0 -0
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/docker/image.py +0 -0
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/dtc/__init__.py +0 -0
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/dtc/cells.py +0 -0
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/dtc/reporter.py +0 -0
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/dump/__init__.py +0 -0
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/dump/main.py +0 -0
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/exceptions.py +0 -0
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/exec/__init__.py +0 -0
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/exec/client.py +0 -0
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/exec/protocol.py +0 -0
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/exec/user.py +0 -0
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/init/__init__.py +0 -0
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/kill/__init__.py +0 -0
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/kill/main.py +0 -0
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/load/__init__.py +0 -0
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/runtime.py +0 -0
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/show/__init__.py +0 -0
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/sign/__init__.py +0 -0
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/sign/main.py +0 -0
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/signature.py +0 -0
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/start/__init__.py +0 -0
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/start/main.py +0 -0
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/unload/__init__.py +0 -0
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/update/__init__.py +0 -0
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/utils.py +0 -0
- {kerf_multikernel-0.4.0 → kerf_multikernel-0.5.0}/src/kerf/vmlinuz.py +0 -0
|
@@ -19,7 +19,7 @@ A comprehensive multikernel management system designed to orchestrate
|
|
|
19
19
|
and manage multiple kernel instances on a single host.
|
|
20
20
|
"""
|
|
21
21
|
|
|
22
|
-
__version__ = "0.
|
|
22
|
+
__version__ = "0.5.0"
|
|
23
23
|
__author__ = "Cong Wang"
|
|
24
24
|
|
|
25
25
|
# Export main runtime components for easy access
|
|
@@ -33,7 +33,7 @@ from .sign.main import sign
|
|
|
33
33
|
|
|
34
34
|
|
|
35
35
|
@click.group()
|
|
36
|
-
@click.version_option(version="0.
|
|
36
|
+
@click.version_option(version="0.5.0", prog_name="kerf")
|
|
37
37
|
@click.option("--debug", is_flag=True, help="Enable debug mode")
|
|
38
38
|
@click.pass_context
|
|
39
39
|
def main(ctx, debug):
|
|
@@ -35,6 +35,7 @@ from ..resources import (
|
|
|
35
35
|
get_available_cpus,
|
|
36
36
|
)
|
|
37
37
|
from ..exceptions import ValidationError, KernelInterfaceError, ResourceError, ParseError
|
|
38
|
+
from ..topology import cpu_id_name
|
|
38
39
|
|
|
39
40
|
|
|
40
41
|
def parse_cpu_spec(cpu_spec: str) -> List[int]:
|
|
@@ -137,7 +138,7 @@ def _allocate_spread(
|
|
|
137
138
|
numa_cpu_lists[numa_node_id] = sorted(numa_cpus)
|
|
138
139
|
|
|
139
140
|
if not numa_cpu_lists:
|
|
140
|
-
raise ResourceError(f"No available
|
|
141
|
+
raise ResourceError(f"No available {cpu_id_name()}s in specified NUMA nodes: {numa_nodes}")
|
|
141
142
|
|
|
142
143
|
allocated = []
|
|
143
144
|
numa_indices = {node_id: 0 for node_id in numa_cpu_lists}
|
|
@@ -187,7 +188,7 @@ def _allocate_local(
|
|
|
187
188
|
if len(numa_cpus) >= count:
|
|
188
189
|
return sorted(numa_cpus[:count])
|
|
189
190
|
raise ResourceError(
|
|
190
|
-
f"Not enough
|
|
191
|
+
f"Not enough {cpu_id_name()}s in NUMA node {numa_node_id}: "
|
|
191
192
|
f"requested {count}, but only {len(numa_cpus)} available"
|
|
192
193
|
)
|
|
193
194
|
|
|
@@ -200,7 +201,7 @@ def _allocate_local(
|
|
|
200
201
|
if len(numa_cpus) >= count:
|
|
201
202
|
return sorted(numa_cpus[:count])
|
|
202
203
|
|
|
203
|
-
raise ResourceError(f"No single NUMA node has {count} available
|
|
204
|
+
raise ResourceError(f"No single NUMA node has {count} available {cpu_id_name()}s for 'local' affinity")
|
|
204
205
|
|
|
205
206
|
|
|
206
207
|
def allocate_cpus_from_pool(
|
|
@@ -236,11 +237,11 @@ def allocate_cpus_from_pool(
|
|
|
236
237
|
if len(available) < count:
|
|
237
238
|
if numa_nodes:
|
|
238
239
|
raise ResourceError(
|
|
239
|
-
f"Not enough
|
|
240
|
+
f"Not enough {cpu_id_name()}s available in NUMA nodes {numa_nodes}: "
|
|
240
241
|
f"requested {count}, but only {len(available)} available"
|
|
241
242
|
)
|
|
242
243
|
raise ResourceError(
|
|
243
|
-
f"Not enough
|
|
244
|
+
f"Not enough {cpu_id_name()}s available: requested {count}, "
|
|
244
245
|
f"but only {len(available)} available in pool"
|
|
245
246
|
)
|
|
246
247
|
|
|
@@ -306,7 +307,7 @@ def parse_device_list(device_spec: Optional[str]) -> List[str]:
|
|
|
306
307
|
"""
|
|
307
308
|
Parse device specification string into list of device references.
|
|
308
309
|
|
|
309
|
-
A reference is
|
|
310
|
+
A reference is a name ("nvme0", "enP1s1f1"), a PCI address ("0000:4f:01.0") or a
|
|
310
311
|
device node name ("pci_0000_4f_01_0") from the baseline, comma-separated.
|
|
311
312
|
|
|
312
313
|
Args:
|
|
@@ -364,8 +365,8 @@ def dump_overlay_for_debug(
|
|
|
364
365
|
@click.option(
|
|
365
366
|
"--cpus",
|
|
366
367
|
"-c",
|
|
367
|
-
help='Explicit
|
|
368
|
-
'"128,130,132" for list)
|
|
368
|
+
help='Explicit CPU allocation by physical ID, the APIC ID on x86 or the MPIDR on arm64 '
|
|
369
|
+
'(e.g., "128", "128-134" for range, "128,130,132" for list), not logical CPU numbers. '
|
|
369
370
|
'Mutually exclusive with --cpu-count',
|
|
370
371
|
)
|
|
371
372
|
@click.option(
|
|
@@ -398,7 +399,7 @@ def dump_overlay_for_debug(
|
|
|
398
399
|
@click.option(
|
|
399
400
|
"--devices",
|
|
400
401
|
"-d",
|
|
401
|
-
help='Pool devices to assign, by
|
|
402
|
+
help='Pool devices to assign, by name, PCI address or node name '
|
|
402
403
|
'(comma-separated, e.g. "nvme0,0000:09:00.0"), as listed by kerf show.',
|
|
403
404
|
)
|
|
404
405
|
@click.option(
|
|
@@ -561,7 +562,7 @@ def create( # pylint: disable=too-many-arguments,too-many-positional-arguments
|
|
|
561
562
|
try:
|
|
562
563
|
cpu_spec_value = parse_cpu_spec(cpus)
|
|
563
564
|
except ValueError as e:
|
|
564
|
-
click.echo(f"Error: Invalid
|
|
565
|
+
click.echo(f"Error: Invalid {cpu_id_name()} specification '{cpus}': {e}", err=True)
|
|
565
566
|
sys.exit(2)
|
|
566
567
|
|
|
567
568
|
# Parse memory specification
|
|
@@ -19,7 +19,7 @@ from .mkdaxfs import (
|
|
|
19
19
|
unmount_daxfs,
|
|
20
20
|
DaxfsError,
|
|
21
21
|
DaxfsImage,
|
|
22
|
-
|
|
22
|
+
KERF_INIT_PATH,
|
|
23
23
|
)
|
|
24
24
|
|
|
25
25
|
__all__ = [
|
|
@@ -27,5 +27,5 @@ __all__ = [
|
|
|
27
27
|
"unmount_daxfs",
|
|
28
28
|
"DaxfsError",
|
|
29
29
|
"DaxfsImage",
|
|
30
|
-
"
|
|
30
|
+
"KERF_INIT_PATH",
|
|
31
31
|
]
|
|
@@ -92,14 +92,18 @@ class FileEntry:
|
|
|
92
92
|
data_offset: int = 0
|
|
93
93
|
child_count: int = 0
|
|
94
94
|
is_hardlink: bool = False
|
|
95
|
+
source: Optional[Path] = None
|
|
95
96
|
|
|
96
97
|
|
|
97
98
|
class DaxfsBuilder:
|
|
98
99
|
"""Builds a daxfs v8 filesystem image: base snapshot plus write overlay."""
|
|
99
100
|
|
|
100
101
|
def __init__(self, src_dir: str,
|
|
101
|
-
overlay_pool_size: int = DAXFS_DEFAULT_OVERLAY_POOL
|
|
102
|
+
overlay_pool_size: int = DAXFS_DEFAULT_OVERLAY_POOL,
|
|
103
|
+
overrides: Optional[Dict[str, Path]] = None):
|
|
102
104
|
self.src_dir = Path(src_dir)
|
|
105
|
+
# Image path -> host file placed there instead of the rootfs entry
|
|
106
|
+
self.overrides = overrides or {}
|
|
103
107
|
self.files: List[FileEntry] = []
|
|
104
108
|
self.files_by_path: Dict[str, FileEntry] = {}
|
|
105
109
|
# First entry seen for each hardlinked (st_dev, st_ino)
|
|
@@ -126,11 +130,13 @@ class DaxfsBuilder:
|
|
|
126
130
|
buckets <<= 1
|
|
127
131
|
return max(buckets, DAXFS_DEFAULT_BUCKET_COUNT)
|
|
128
132
|
|
|
129
|
-
def _add_file(self, relpath: str, stat_result: os.stat_result
|
|
133
|
+
def _add_file(self, relpath: str, stat_result: os.stat_result,
|
|
134
|
+
source: Optional[Path] = None) -> FileEntry:
|
|
130
135
|
entry = FileEntry(
|
|
131
136
|
path=relpath,
|
|
132
137
|
name=os.path.basename(relpath)[:DAXFS_NAME_MAX] if relpath else "",
|
|
133
138
|
stat=stat_result,
|
|
139
|
+
source=source,
|
|
134
140
|
)
|
|
135
141
|
|
|
136
142
|
# Hardlinked regular files share one inode and one data extent
|
|
@@ -162,6 +168,8 @@ class DaxfsBuilder:
|
|
|
162
168
|
|
|
163
169
|
for entry in sorted(entries, key=lambda e: e.name):
|
|
164
170
|
newrel = f"{relpath}/{entry.name}" if relpath else entry.name
|
|
171
|
+
if newrel in self.overrides:
|
|
172
|
+
continue
|
|
165
173
|
|
|
166
174
|
try:
|
|
167
175
|
file_stat = entry.lstat()
|
|
@@ -178,6 +186,16 @@ class DaxfsBuilder:
|
|
|
178
186
|
root_stat = self.src_dir.lstat()
|
|
179
187
|
self._add_file("", root_stat)
|
|
180
188
|
self._scan_directory_recursive("")
|
|
189
|
+
for relpath, source in self.overrides.items():
|
|
190
|
+
self._add_file(relpath, self._override_stat(source), source)
|
|
191
|
+
|
|
192
|
+
@staticmethod
|
|
193
|
+
def _override_stat(source: Path) -> os.stat_result:
|
|
194
|
+
st = os.stat(source)
|
|
195
|
+
return os.stat_result((
|
|
196
|
+
stat_mod.S_IFREG | 0o755, st.st_ino, st.st_dev, 1, 0, 0,
|
|
197
|
+
st.st_size, int(st.st_atime), int(st.st_mtime), int(st.st_ctime),
|
|
198
|
+
))
|
|
181
199
|
|
|
182
200
|
def build_tree(self) -> None:
|
|
183
201
|
"""Assign parent inodes and count directory children."""
|
|
@@ -324,7 +342,7 @@ class DaxfsBuilder:
|
|
|
324
342
|
def _write_file_data(self, mem: mmap.mmap, offset: int,
|
|
325
343
|
entry: FileEntry) -> None:
|
|
326
344
|
try:
|
|
327
|
-
with open(self.src_dir / entry.path, 'rb') as f:
|
|
345
|
+
with open(entry.source or self.src_dir / entry.path, 'rb') as f:
|
|
328
346
|
mem.seek(offset)
|
|
329
347
|
shutil.copyfileobj(f, mem)
|
|
330
348
|
except (PermissionError, FileNotFoundError, IsADirectoryError):
|
|
@@ -442,25 +460,18 @@ MOVE_MOUNT_F_EMPTY_PATH = 0x00000004
|
|
|
442
460
|
|
|
443
461
|
AT_FDCWD = -100
|
|
444
462
|
|
|
463
|
+
# Off /init so that an image's own /init stays usable as an entrypoint
|
|
464
|
+
KERF_INIT_PATH = "/.kerf-init"
|
|
445
465
|
|
|
446
|
-
def inject_kerf_init(rootfs_path: str) -> None:
|
|
447
|
-
"""Inject /init binary into rootfs.
|
|
448
466
|
|
|
449
|
-
|
|
450
|
-
"""
|
|
451
|
-
rootfs = Path(rootfs_path)
|
|
452
|
-
|
|
453
|
-
# Copy the pre-built init binary
|
|
467
|
+
def _kerf_init_binary() -> Path:
|
|
454
468
|
init_binary = get_init_binary_path()
|
|
455
469
|
if not init_binary.exists():
|
|
456
470
|
raise DaxfsError(
|
|
457
471
|
f"Init binary not found at {init_binary}. "
|
|
458
472
|
"Run 'make' to build it first."
|
|
459
473
|
)
|
|
460
|
-
|
|
461
|
-
init_path = rootfs / "init"
|
|
462
|
-
shutil.copy2(init_binary, init_path)
|
|
463
|
-
os.chmod(init_path, 0o755)
|
|
474
|
+
return init_binary
|
|
464
475
|
|
|
465
476
|
|
|
466
477
|
def _mount_daxfs(instance_name: str, dmabuf_fd: int) -> None:
|
|
@@ -521,7 +532,9 @@ def create_daxfs_image(
|
|
|
521
532
|
size: Optional[int] = None,
|
|
522
533
|
) -> DaxfsImage:
|
|
523
534
|
"""
|
|
524
|
-
Create a daxfs filesystem image from a directory
|
|
535
|
+
Create a daxfs filesystem image from a directory, with kerf-init at
|
|
536
|
+
KERF_INIT_PATH. The entrypoint is passed via kernel cmdline as
|
|
537
|
+
kerf.entrypoint=<path>.
|
|
525
538
|
|
|
526
539
|
Args:
|
|
527
540
|
rootfs_path: Path to the root filesystem directory
|
|
@@ -538,7 +551,8 @@ def create_daxfs_image(
|
|
|
538
551
|
if not os.path.isdir(rootfs_path):
|
|
539
552
|
raise DaxfsError(f"Rootfs directory '{rootfs_path}' does not exist")
|
|
540
553
|
|
|
541
|
-
builder = DaxfsBuilder(rootfs_path
|
|
554
|
+
builder = DaxfsBuilder(rootfs_path,
|
|
555
|
+
overrides={KERF_INIT_PATH[1:]: _kerf_init_binary()})
|
|
542
556
|
builder.scan()
|
|
543
557
|
builder.build_tree()
|
|
544
558
|
builder.calculate_offsets()
|
|
@@ -18,10 +18,9 @@ Naming of pool PCI devices.
|
|
|
18
18
|
A pool device is identified by its PCI address and nothing else: host
|
|
19
19
|
names such as nvme0n1 or enp9s0 are probe-order or udev artifacts that
|
|
20
20
|
differ between kernels, while the address is fixed by the fabric. The
|
|
21
|
-
device node is named after the address, and
|
|
22
|
-
|
|
23
|
-
|
|
24
|
-
stays nvme0 and its namespaces nvme0n1, "enp9s0" stays enp9s0).
|
|
21
|
+
device node is named after the address, and its label carries the name
|
|
22
|
+
the device has in the kernel that pooled it, verbatim, which every
|
|
23
|
+
spawned kernel then applies ("enP1s1f1" stays enP1s1f1).
|
|
25
24
|
"""
|
|
26
25
|
|
|
27
26
|
import re
|
|
@@ -29,8 +28,8 @@ from typing import Optional, Tuple
|
|
|
29
28
|
|
|
30
29
|
PCI_ID_RE = re.compile(r'^([0-9a-fA-F]{1,4}):([0-9a-fA-F]{1,2}):([0-9a-fA-F]{1,2})\.([0-9a-fA-F])$')
|
|
31
30
|
PCI_NODE_RE = re.compile(r'^pci_([0-9a-fA-F]{1,4})_([0-9a-fA-F]{1,2})_([0-9a-fA-F]{1,2})_([0-9a-fA-F])$')
|
|
32
|
-
ALIAS_RE = re.compile(r'^[a-z0-9-]{1,31}$')
|
|
33
31
|
NVME_NAME_RE = re.compile(r'^(nvme\d+)(?:n\d+(p\d+)?)?$')
|
|
32
|
+
LABEL_MAX = 31
|
|
34
33
|
IFNAMSIZ = 16
|
|
35
34
|
|
|
36
35
|
|
|
@@ -51,17 +50,33 @@ def pci_node_name(pci_id: str) -> str:
|
|
|
51
50
|
return "pci_" + normalized.replace(":", "_").replace(".", "_")
|
|
52
51
|
|
|
53
52
|
|
|
54
|
-
def
|
|
55
|
-
"""
|
|
56
|
-
name
|
|
57
|
-
|
|
58
|
-
|
|
59
|
-
|
|
53
|
+
def check_label(label: str, compatible: Optional[str] = None) -> None:
|
|
54
|
+
"""
|
|
55
|
+
Reject a name the kernels cannot carry or, for a network device, apply.
|
|
56
|
+
|
|
57
|
+
The rules are the kernel's own for an interface name, so any name the
|
|
58
|
+
host gave a device passes.
|
|
59
|
+
"""
|
|
60
|
+
if (not label or len(label) > LABEL_MAX or label in (".", "..")
|
|
61
|
+
or any(c in "/:,=" or c.isspace() for c in label)):
|
|
60
62
|
raise ValueError(
|
|
61
|
-
f"Invalid
|
|
62
|
-
"
|
|
63
|
+
f"Invalid name '{label}': use 1 to {LABEL_MAX} characters "
|
|
64
|
+
"without whitespace, '/', ':', ',' or '='"
|
|
63
65
|
)
|
|
64
|
-
|
|
66
|
+
if compatible == "pci-network" and len(label) >= IFNAMSIZ:
|
|
67
|
+
raise ValueError(
|
|
68
|
+
f"Invalid name '{label}' for a network device: "
|
|
69
|
+
f"an interface name is at most {IFNAMSIZ - 1} characters"
|
|
70
|
+
)
|
|
71
|
+
|
|
72
|
+
|
|
73
|
+
def split_name(entry: str) -> Tuple[str, Optional[str]]:
|
|
74
|
+
"""Split a "device" or "device=name" request into its parts."""
|
|
75
|
+
device, sep, name = entry.partition("=")
|
|
76
|
+
if not sep:
|
|
77
|
+
return device, None
|
|
78
|
+
check_label(name)
|
|
79
|
+
return device, name
|
|
65
80
|
|
|
66
81
|
|
|
67
82
|
def is_partition(host_name: str) -> bool:
|
|
@@ -69,19 +84,19 @@ def is_partition(host_name: str) -> bool:
|
|
|
69
84
|
return bool(match and match.group(2))
|
|
70
85
|
|
|
71
86
|
|
|
72
|
-
def
|
|
87
|
+
def default_label(host_name: str, compatible: Optional[str]) -> Optional[str]:
|
|
73
88
|
"""
|
|
74
|
-
The
|
|
89
|
+
The name a host device name implies, None when it implies nothing.
|
|
75
90
|
|
|
76
91
|
An NVMe name carries its controller name, which the namespaces follow.
|
|
77
|
-
A network interface name is the
|
|
78
|
-
|
|
92
|
+
A network interface name is the name itself. A raw PCI address, or a
|
|
93
|
+
name of something other than the PCI function, implies no name at all.
|
|
79
94
|
"""
|
|
80
95
|
if normalize_pci_id(host_name):
|
|
81
96
|
return None
|
|
82
97
|
match = NVME_NAME_RE.match(host_name)
|
|
83
98
|
if match:
|
|
84
99
|
return match.group(1)
|
|
85
|
-
if compatible == "pci-network"
|
|
100
|
+
if compatible == "pci-network":
|
|
86
101
|
return host_name
|
|
87
102
|
return None
|
|
@@ -109,8 +109,6 @@ class InstanceExtractor:
|
|
|
109
109
|
|
|
110
110
|
fdt_sw.end_node() # End resources
|
|
111
111
|
|
|
112
|
-
self._add_aliases_sw(fdt_sw, tree.hardware.devices)
|
|
113
|
-
|
|
114
112
|
if tree.instances:
|
|
115
113
|
self._add_instances_section_sw(fdt_sw, tree.instances)
|
|
116
114
|
|
|
@@ -156,6 +154,9 @@ class InstanceExtractor:
|
|
|
156
154
|
if device_info.pci_id:
|
|
157
155
|
fdt_sw.property_string("pci-id", device_info.pci_id)
|
|
158
156
|
|
|
157
|
+
if device_info.label:
|
|
158
|
+
fdt_sw.property_string("label", device_info.label)
|
|
159
|
+
|
|
159
160
|
if device_info.vendor_id is not None:
|
|
160
161
|
fdt_sw.property_u32("vendor-id", device_info.vendor_id)
|
|
161
162
|
|
|
@@ -194,16 +195,6 @@ class InstanceExtractor:
|
|
|
194
195
|
|
|
195
196
|
fdt_sw.end_node()
|
|
196
197
|
|
|
197
|
-
def _add_aliases_sw(self, fdt_sw, devices):
|
|
198
|
-
aliases = {info.alias: name for name, info in (devices or {}).items() if info.alias}
|
|
199
|
-
if not aliases:
|
|
200
|
-
return
|
|
201
|
-
|
|
202
|
-
fdt_sw.begin_node("aliases")
|
|
203
|
-
for alias, name in aliases.items():
|
|
204
|
-
fdt_sw.property_string(alias, f"/resources/devices/{name}")
|
|
205
|
-
fdt_sw.end_node()
|
|
206
|
-
|
|
207
198
|
def _add_instances_section_sw(self, fdt_sw, instances):
|
|
208
199
|
"""Add instances section using FdtSw."""
|
|
209
200
|
fdt_sw.begin_node("instances")
|
|
@@ -236,7 +236,7 @@ class OverlayGenerator:
|
|
|
236
236
|
self._cpu_op(fdt_sw, "cpu-remove", diff.cpus_to_host)
|
|
237
237
|
self._cpu_op(fdt_sw, "cpu-add", diff.cpus_to_pool)
|
|
238
238
|
self._device_op(fdt_sw, "device-remove", diff.devices_to_host)
|
|
239
|
-
self._device_op(fdt_sw, "device-add", diff.devices_to_pool, diff.
|
|
239
|
+
self._device_op(fdt_sw, "device-add", diff.devices_to_pool, diff.device_labels)
|
|
240
240
|
|
|
241
241
|
fdt_sw.end_node() # End __overlay__
|
|
242
242
|
fdt_sw.end_node() # End fragment@0
|
|
@@ -269,7 +269,7 @@ class OverlayGenerator:
|
|
|
269
269
|
fdt_sw.end_node()
|
|
270
270
|
fdt_sw.end_node()
|
|
271
271
|
|
|
272
|
-
def _device_op(self, fdt_sw, operation, pci_ids,
|
|
272
|
+
def _device_op(self, fdt_sw, operation, pci_ids, labels=None):
|
|
273
273
|
"""Write a device operation node, or nothing when there are no devices."""
|
|
274
274
|
if not pci_ids:
|
|
275
275
|
return
|
|
@@ -278,8 +278,8 @@ class OverlayGenerator:
|
|
|
278
278
|
for idx, pci_id in enumerate(pci_ids):
|
|
279
279
|
fdt_sw.begin_node(f"pci@{idx}")
|
|
280
280
|
fdt_sw.property_string("pci-id", pci_id)
|
|
281
|
-
if
|
|
282
|
-
fdt_sw.property_string("
|
|
281
|
+
if labels and pci_id in labels:
|
|
282
|
+
fdt_sw.property_string("label", labels[pci_id])
|
|
283
283
|
fdt_sw.end_node()
|
|
284
284
|
fdt_sw.end_node()
|
|
285
285
|
|
|
@@ -108,14 +108,7 @@ class DeviceTreeParser:
|
|
|
108
108
|
self.fdt = libfdt.Fdt(dtb_data)
|
|
109
109
|
except Exception as e:
|
|
110
110
|
raise ParseError(f"Failed to parse instance DTB: {e}") from e
|
|
111
|
-
|
|
112
|
-
offsets = {}
|
|
113
|
-
for offset, (name, device_info) in self._parse_pci_tree().items():
|
|
114
|
-
devices[name] = device_info
|
|
115
|
-
offsets[offset] = name
|
|
116
|
-
for alias, name in self._parse_aliases(offsets).items():
|
|
117
|
-
devices[name].alias = alias
|
|
118
|
-
return devices
|
|
111
|
+
return dict(self._parse_pci_tree().values())
|
|
119
112
|
|
|
120
113
|
def _build_global_tree(self) -> GlobalDeviceTree:
|
|
121
114
|
"""Build GlobalDeviceTree from parsed FDT."""
|
|
@@ -241,14 +234,10 @@ class DeviceTreeParser:
|
|
|
241
234
|
if free_prop is not None:
|
|
242
235
|
available_free = unpack_cpu_ids(free_prop)
|
|
243
236
|
|
|
244
|
-
if available:
|
|
245
|
-
total = max(available) + 1
|
|
246
|
-
else:
|
|
247
|
-
total = 0
|
|
248
237
|
host_reserved = []
|
|
249
238
|
|
|
250
239
|
return CPUAllocation(
|
|
251
|
-
total=
|
|
240
|
+
total=len(available),
|
|
252
241
|
host_reserved=host_reserved,
|
|
253
242
|
available=available,
|
|
254
243
|
available_free=available_free
|
|
@@ -325,7 +314,6 @@ class DeviceTreeParser:
|
|
|
325
314
|
def _parse_devices(self, resources_node: int) -> Dict[str, DeviceInfo]:
|
|
326
315
|
"""Parse device information from resources node."""
|
|
327
316
|
devices = {}
|
|
328
|
-
offsets = {}
|
|
329
317
|
|
|
330
318
|
devices_node = self.fdt.subnode_offset(resources_node, 'devices',
|
|
331
319
|
libfdt.QUIET_NOTFOUND)
|
|
@@ -333,16 +321,11 @@ class DeviceTreeParser:
|
|
|
333
321
|
name = self.fdt.get_name(offset)
|
|
334
322
|
try:
|
|
335
323
|
devices[name] = self._parse_device_info(offset, name)
|
|
336
|
-
offsets[offset] = name
|
|
337
324
|
except ParseError:
|
|
338
325
|
pass
|
|
339
326
|
|
|
340
|
-
for
|
|
327
|
+
for name, device_info in self._parse_pci_tree().values():
|
|
341
328
|
devices[name] = device_info
|
|
342
|
-
offsets[offset] = name
|
|
343
|
-
|
|
344
|
-
for alias, name in self._parse_aliases(offsets).items():
|
|
345
|
-
devices[name].alias = alias
|
|
346
329
|
|
|
347
330
|
return devices
|
|
348
331
|
|
|
@@ -387,25 +370,11 @@ class DeviceTreeParser:
|
|
|
387
370
|
device_info = self._parse_device_info(node, name)
|
|
388
371
|
device_info.device_type = 'pci'
|
|
389
372
|
device_info.pci_id = pci_id
|
|
373
|
+
iommu = self.fdt.getprop(node, 'multikernel,iommu-base', libfdt.QUIET_NOTFOUND)
|
|
374
|
+
if iommu != -libfdt.NOTFOUND:
|
|
375
|
+
device_info.iommu_base = iommu.as_uint64()
|
|
390
376
|
found[node] = (name, device_info)
|
|
391
377
|
|
|
392
|
-
def _parse_aliases(self, offsets: Dict[int, str]) -> Dict[str, str]:
|
|
393
|
-
"""Map each /aliases entry to the device node it points at."""
|
|
394
|
-
aliases = {}
|
|
395
|
-
aliases_node = self.fdt.path_offset('/aliases', libfdt.QUIET_NOTFOUND)
|
|
396
|
-
if aliases_node < 0:
|
|
397
|
-
return aliases
|
|
398
|
-
|
|
399
|
-
prop_offset = self.fdt.first_property_offset(aliases_node, libfdt.QUIET_NOTFOUND)
|
|
400
|
-
while prop_offset >= 0:
|
|
401
|
-
prop = self.fdt.get_property_by_offset(prop_offset)
|
|
402
|
-
target = self.fdt.path_offset(prop.as_str(), libfdt.QUIET_NOTFOUND)
|
|
403
|
-
if target in offsets:
|
|
404
|
-
aliases[prop.name] = offsets[target]
|
|
405
|
-
prop_offset = self.fdt.next_property_offset(prop_offset, libfdt.QUIET_NOTFOUND)
|
|
406
|
-
|
|
407
|
-
return aliases
|
|
408
|
-
|
|
409
378
|
def _parse_device_info(self, node_offset: int, name: str) -> DeviceInfo:
|
|
410
379
|
"""Parse individual device information."""
|
|
411
380
|
compatible = ""
|
|
@@ -442,6 +411,9 @@ class DeviceTreeParser:
|
|
|
442
411
|
except libfdt.FdtException:
|
|
443
412
|
pass
|
|
444
413
|
|
|
414
|
+
label = self.fdt.getprop(node_offset, 'label', libfdt.QUIET_NOTFOUND)
|
|
415
|
+
label = None if label == -libfdt.NOTFOUND else label.as_str()
|
|
416
|
+
|
|
445
417
|
try:
|
|
446
418
|
vendor_id = self.fdt.getprop(node_offset, 'vendor-id').as_uint32()
|
|
447
419
|
except libfdt.FdtException:
|
|
@@ -488,6 +460,7 @@ class DeviceTreeParser:
|
|
|
488
460
|
device_type=device_type,
|
|
489
461
|
device_name=device_name,
|
|
490
462
|
pci_id=pci_id,
|
|
463
|
+
label=label,
|
|
491
464
|
vendor_id=vendor_id,
|
|
492
465
|
device_id=device_id,
|
|
493
466
|
sriov_vfs=sriov_vfs,
|
|
@@ -288,17 +288,17 @@ class MultikernelValidator:
|
|
|
288
288
|
cpus = tree.hardware.cpus
|
|
289
289
|
instance_cpus = set(instance.resources.cpus)
|
|
290
290
|
|
|
291
|
-
|
|
292
|
-
|
|
293
|
-
|
|
294
|
-
|
|
295
|
-
|
|
296
|
-
|
|
297
|
-
|
|
298
|
-
|
|
299
|
-
|
|
300
|
-
|
|
301
|
-
|
|
291
|
+
known = sorted(set(cpus.host_reserved) | set(cpus.available))
|
|
292
|
+
for cpu in sorted(instance_cpus - set(known)):
|
|
293
|
+
error_msg = self._format_error_with_context(
|
|
294
|
+
error_type="CPU allocation error",
|
|
295
|
+
instance_name=instance.name,
|
|
296
|
+
problem=f"CPU {cpu} does not exist in hardware inventory",
|
|
297
|
+
current_state=f"Hardware has CPUs {known}",
|
|
298
|
+
suggestion="Use a CPU from the hardware inventory",
|
|
299
|
+
alternative="Check hardware inventory configuration",
|
|
300
|
+
)
|
|
301
|
+
self.errors.append(error_msg)
|
|
302
302
|
|
|
303
303
|
host_reserved = set(cpus.host_reserved)
|
|
304
304
|
reserved_cpus = instance_cpus.intersection(host_reserved)
|
|
@@ -18,6 +18,7 @@ kerf exec: run a command inside a running instance.
|
|
|
18
18
|
|
|
19
19
|
import errno
|
|
20
20
|
import os
|
|
21
|
+
import random
|
|
21
22
|
import signal
|
|
22
23
|
import socket
|
|
23
24
|
import sys
|
|
@@ -76,7 +77,11 @@ def connect_agent(instance_id: int, name: str) -> socket.socket:
|
|
|
76
77
|
try:
|
|
77
78
|
sock.setsockopt(socket.AF_VSOCK, SO_VM_SOCKETS_TRANSPORT, VSOCK_TRANSPORT_MULTIKERNEL)
|
|
78
79
|
# kerf-init only trusts reserved ports, which only root can bind.
|
|
79
|
-
|
|
80
|
+
# Start at a random one: reusing the port of an exec that just gave
|
|
81
|
+
# up can pick up the late answer to that exec's connect.
|
|
82
|
+
ports = list(range(protocol.AGENT_PORT, 511, -1))
|
|
83
|
+
start = random.randrange(len(ports))
|
|
84
|
+
for port in ports[start:] + ports[:start]:
|
|
80
85
|
try:
|
|
81
86
|
sock.bind((socket.VMADDR_CID_ANY, port))
|
|
82
87
|
break
|