#15818 closed bug (fixed)
NVMe data loss
| Reported by: | KapiX | Owned by: | waddlesplash |
|---|---|---|---|
| Priority: | normal | Milestone: | R1/beta2 |
| Component: | Drivers/Disk/NVMe | Version: | R1/Development |
| Keywords: | Cc: | ||
| Blocked By: | Blocking: | ||
| Platform: | All |
Description
Symptoms:
- Invalid opcode when running executables from it, ninja crashes trying to rebuild (WebKit). After rebooting had to rebuild entire project.
- Now my WebKit git repo is corrupted (unknown index entry format 0x74650000).
Nothing suspicious in syslog, I'm running Linux from the same drive without issues (3 months now).
Drive: Intel 760p 512GB.
Change History (8)
comment:1 by , 5 years ago
comment:3 by , 5 years ago
| Component: | - General → Drivers/Disk/NVMe |
|---|---|
| Owner: | changed from to |
comment:4 by , 5 years ago
There's also #15123 which is a series of very strange KDLs under VMware, which I couldn't reproduce on QEMU, so I guess I should try again there.
X512 mentioned that a bug may be (and the stacktraces there seem to indicate it) in the kernel disk cache code, which may not be able to tolerate NVMe returning stuff in parallel. However, the "completion event for unknown cmd" is much more suspicious, I still don't know what to make of that.
comment:6 by , 5 years ago
Actually the "corruption" may have just been on the underlying disk and compounded by some BFS driver issues when corruption is present. See comment in #15123; I'd wait to retest until Diver confirms that is gone however.
comment:7 by , 5 years ago
| Resolution: | → fixed |
|---|---|
| Status: | new → closed |
Fixed in hrev54102, but be a bit cautious at first.
comment:8 by , 5 years ago
| Milestone: | Unscheduled → R1/beta2 |
|---|
Assign tickets with status=closed and resolution=fixed within the R1/beta2 development window to the R1/beta2 Milestone
hrev53992 64-bit, checkfs doesn't complain.