This patch does three things:
- Convert the global thread list from a HashMap to an IntrusiveList
- Combine the thread list and its lock into a SpinLockProtectedValue
- Customize Thread::unref() so it locks the list while unreffing
This closes the same race window for Thread as @sin-ack's recent changes
did for Process.
Note that the HashMap->IntrusiveList conversion means that we lose O(1)
lookups, but the majority of clients of this list are doing traversal,
not lookup. Once we have an intrusive hashing solution, we should port
this to use that, but for now, this gets rid of heap allocations during
a sensitive time.
We are not using this for anything and it's just been sitting there
gathering dust for well over a year, so let's stop carrying all this
complexity around for no good reason.
This patch replaces the remaining users of this API with the new
try_copy_kstring_from_user() instead. Note that we still convert to a
String for continued processing, and I've added FIXME about continuing
work on using KString all the way.
This makes the following scenario impossible with an SMP setup:
1) CPU A enters unref() and decrements the link count to 0.
2) CPU B sees the process in the process list and ref()s it.
3) CPU A removes the process from the list and continues destructing.
4) CPU B is now holding a destructed Process object.
By holding the process list lock before doing anything with it, we
ensure that other CPUs can't find this process in the middle of it being
destructed.
This allows us to 1) let go of the Process when an inode is ref'ing for
ProcFSExposedComponent related reasons, and 2) change our ref/unref
implementation.
A hub can technically have up to 255 ports, given that bNbrPorts is a
u8 and the DeviceRemovable field is a VLA to support up to 255 ports.
Source: USB 2.0 Specification Section 11.23.2.1
That means this enum is not going to scale well in terms of size.
Replacing it with a raw u8 allows me to remove the two port assumption
and a cast.
Previously it would create a contiguous AVMO manually and pass it to
MM. This uses supervisor pages that quickly run out as they never get
returned and crash the system.
Instead, use allocate_kernel_region as we're only allocating a page so
it will be contiguous and will be returned when destroyed.
A potentially better solution would be to use a pool of transfers to
avoid all the allocations. This just prevents the system from crashing
within ~5 seconds from the continuous hub polling.
