Thread: Kernel: Legal to kfree() a down()'ed semaphore

I have something like:

Code:

struct buffer {
        uint32_t head, tail;
        void **buf;
        struct semaphore sem;
}

that I'm using to control several input buffers in my device driver. As a particular part of the driver is released, I need to dispose of that buffer. What I want to do is something along these lines:

Code:

        struct buffer *buf;
        down(buf->sem);
        kfree(buf->buf);
        kfree(buf);
        // Uh oh, where'd the semaphore go?

I'm almost certain that this would be valid since the semaphore is controlled by me and ONLY by me. I believe that the kernel has no interest in the semaphore at any point, so I'm the one completely responsible for it.

Am I right in these assumptions?

Further research leads me to believe that my assumptions are incorrect. It appears that I'll have to lock the list when I need to free the buffer. Otherwise, there could be a memory leak as there is a linked list on with int the struct sem that keeps track of who is in line for the semaphore next. To that end, if I hit a race condition where I want to free it, but I then tap the buf->sem and want to add data to it, I'll forever wait -- which would be "bad".