add dependency to Send for the implementation of spinlock
#97
Conversation
stlankes
commented
Sep 1, 2020
stlankes
commented
- first draft to solve issue Spinlock is Unsound #96
- using of the core collection library to handle memory and tasks - the vector of PCI adpaters aren't longer protected by a lock because it is initialized at boot time - afterwards, we have only read-access
- avoids collisions in BtreeSet
…read - required to protect it by spinlock
|
@tlambertz I revise the interface to the filesystem. Does it meet your requirements? |
There was a problem hiding this comment.
I have only taken a cursory glance and found virtio-net vulnerable to deadlocks. Though I think similar issues would have caused a panic before. My own rewrite of virtio-fs uses a different interrupt logic, that might be worth adopting. I'll merge it next week.
| } | ||
| }); | ||
| if let Some(driver) = crate::arch::kernel::pci::get_network_driver() { | ||
| driver.lock().set_polling_mode(value); |
There was a problem hiding this comment.
This might deadlock. virtio_irqhandler also gets the network driver and calls
driver.lock().handle_interrupt()The interrupt might happen while the driver is locked here.
There was a problem hiding this comment.
That is the reason for SpinlockIrqSave. If a task hold this lock, all interrupt are blocked. => No interrupt during the critical section
There was a problem hiding this comment.
You find more details at https://github.com/hermitcore/libhermit-rs/blob/master/src/synch/spinlock.rs#L212-L221
…s#97) * add requirement to use only types T for spin lock, where the behavior Send is specified * remove obsolete implementation of a irqsave refcell * replace DoubleLinkedList by LinkedList of the alloc library * using of the core collection library to handle memory and tasks * the vector of PCI adapters aren't longer protected by a lock because these adapter are initialized at boot time. Afterwards, we have only read-access. * Freelist: add check if a reunion with the previous slot is possible
