kernel/dma.c

/* $Id: dma.c,v 1.5 1992/11/18 02:49:05 root Exp root $
 * linux/kernel/dma.c: A DMA channel allocator. Inspired by linux/kernel/irq.c.
 * Written by Hennus Bergman, 1992.
 */

#include <linux/kernel.h>
#include <linux/errno.h>
#include <asm/dma.h>

/* A note on resource allocation:
 *
 * All drivers needing DMA channels, should allocate and release them
 * through the public routines `request_dma()' and `free_dma()'.
 *
 * In order to avoid problems, all processes should allocate resources in
 * the same sequence and release them in the reverse order.
 *
 * So, when allocating DMAs and IRQs, first allocate the IRQ, then the DMA.
 * When releasing them, first release the DMA, then release the IRQ.
 * If you don't, you may cause allocation requests to fail unnecessarily.
 * This doesn't really matter now, but it will once we get real semaphores
 * in the kernel.
 */

/* Channel n is busy iff dma_chan_busy[n] != 0.
 * DMA0 is reserved for DRAM refresh, I think.
 * DMA4 is reserved for cascading (?).
 */
 /* dma_chan_busy[n] != 0 表示该通道不可用
 * DMA0 用作DRAM的刷新.
 * DMA4 用作级连.
 */
static volatile unsigned int dma_chan_busy[MAX_DMA_CHANNELS] = {
    1, 0, 0, 0, 1, 0, 0, 0
};

/* Atomically swap memory location [32 bits] with `newval'.
 * This avoid the cli()/sti() junk and related problems.
 * [And it's faster too :-)]
 * Maybe this should be in include/asm/mutex.h and be used for
 * implementing kernel-semaphores as well.
 */
 // 将*p指向的数据替换为newval.由关键字xchgl看，这个值是32位的值。
 // xchgl是原子操作，可以省去cli和sti.
static __inline__ unsigned int mutex_atomic_swap(volatile unsigned int * p, unsigned int newval)
{
    unsigned int semval = newval;

/* If one of the operands for the XCHG instructions is a memory ref,
     * it makes the swap an uninterruptible RMW cycle.
     *
     * One operand must be in memory, the other in a register, otherwise
     * the swap may not be atomic.
     */

asm __volatile__ ("xchgl %2, %0\n"
            : /* outputs: semval   */ "=r" (semval)
            : /* inputs: newval, p */ "0" (semval), "m" (*p)
            );    /* p is a var, containing an address */
    return semval;
} /* mutex_atomic_swap */

//请求DMA，核心由上个函数mutex_atomic_swap完成
int request_dma(unsigned int dmanr)
{
    if (dmanr >= MAX_DMA_CHANNELS)
        return -EINVAL;

if (mutex_atomic_swap(&dma_chan_busy[dmanr], 1) != 0)
        return -EBUSY;
    else
        /* old flag was 0, now contains 1 to indicate busy */
        return 0;
} /* request_dma */

//释放DMA，核心由函数mutex_atomic_swap完成
void free_dma(unsigned int dmanr)
{
    if (dmanr >= MAX_DMA_CHANNELS) {
        printk("Trying to free DMA%d\n", dmanr);
        return;
    }

if (mutex_atomic_swap(&dma_chan_busy[dmanr], 0) == 0)
        printk("Trying to free free DMA%d\n", dmanr);
} /* free_dma */