Procfs enhancements #1066
Procfs enhancements #1066
Conversation
nimelehin
force-pushed
the
procfs
branch
4 times, most recently
from
d45c092
to
92abce3
Compare
|
Hello @tbodt! Asking you to review these changes, they make The patches can collect statistics on which tasks are blocked. This is done using a special wrapper in zones which can block our code (e.g calling real's os syscall). The wrapper is implemented with Also updated procfs, so |
|
Seems it closes #273 |
The func returns a count of vcpus
|
Thanks for the PR! The one thing I'm not sure about is the blocking regions, as that's a pretty large code change and it's easy to miss a spot. I guess nothing really bad happens if you do, though. |
There was a problem hiding this comment.
A slight problem here is some of these stats are for the whole system including any iOS daemons that are running and some of these stats are only for processes running inside iSH. This could get a bit confusing in htop if you can see CPU usage for everything but only see some processes in the breakdown. It's probably fine though.
kernel/signal.c
Outdated
| @@ -636,8 +636,10 @@ int_t sys_rt_sigsuspend(addr_t mask_addr, uint_t size) { | |||
|
|
|||
| int_t sys_pause() { | |||
| lock(¤t->sighand->lock); | |||
| while (wait_for(¤t->pause, ¤t->sighand->lock, NULL) != _EINTR) | |||
| continue; | |||
| TASK_BLOCKING_AREA { | |||
There was a problem hiding this comment.
Instead of doing this, it would make more sense to put the blocking area inside of wait_for.
There was a problem hiding this comment.
Unfortunately this might bring a deadlock. I just want this blocking defines as simple as possible, so I just put them on the most of blocking syscalls in Linux (list of blocking sys calls). From this list only fcntl and close are not wrapped with TASK_MAY_BLOCK, since their implementation in our case is not blocking.
There was a problem hiding this comment.
This list seems outdated? It doesn't include futex or sigsuspend, for example.
nimelehin
force-pushed
the
procfs
branch
2 times, most recently
from
18e7f68
to
856e13a
Compare
|
As an FYI, this mostly fixes the procps top command as well. The memory number looks extremely bogus.
|
nimelehin
force-pushed
the
procfs
branch
3 times, most recently
from
31c00cd
to
12167ba
Compare
Add apis to collect statisctics on blocked tasks. A scpecial wrapper should be used in areas which may block a task.
|
@tbodt, hello. I made these changes:
|
Updated /proc/stat to be compatible with Linux's implementation.
Added mainline PR ish-app#1066, procfs enhancements
Incorporated upstream PR ish-app#1066, procps enhancements
|
Can we get this merged? htop is broken till it is.. |
|
The reliance on locking/unlocking pids_lock in the TASK_MAY_BLOCK macro is a bit problematic from a performance/reliability perspective as so many other parts of iSH use it as well. |
What's added:
Closes #273