1*4882a593SmuzhiyunChinese translated version of Documentation/driver-api/io_ordering.rst 2*4882a593Smuzhiyun 3*4882a593SmuzhiyunIf you have any comment or update to the content, please contact the 4*4882a593Smuzhiyunoriginal document maintainer directly. However, if you have a problem 5*4882a593Smuzhiyuncommunicating in English you can also ask the Chinese maintainer for 6*4882a593Smuzhiyunhelp. Contact the Chinese maintainer if this translation is outdated 7*4882a593Smuzhiyunor if there is a problem with the translation. 8*4882a593Smuzhiyun 9*4882a593SmuzhiyunChinese maintainer: Lin Yongting <linyongting@gmail.com> 10*4882a593Smuzhiyun--------------------------------------------------------------------- 11*4882a593SmuzhiyunDocumentation/driver-api/io_ordering.rst 的中文翻译 12*4882a593Smuzhiyun 13*4882a593Smuzhiyun如果想评论或更新本文的内容，请直接联系原文档的维护者。如果你使用英文 14*4882a593Smuzhiyun交流有困难的话，也可以向中文版维护者求助。如果本翻译更新不及时或者翻 15*4882a593Smuzhiyun译存在问题，请联系中文版维护者。 16*4882a593Smuzhiyun 17*4882a593Smuzhiyun中文版维护者： 林永听 Lin Yongting <linyongting@gmail.com> 18*4882a593Smuzhiyun中文版翻译者： 林永听 Lin Yongting <linyongting@gmail.com> 19*4882a593Smuzhiyun中文版校译者： 林永听 Lin Yongting <linyongting@gmail.com> 20*4882a593Smuzhiyun 21*4882a593Smuzhiyun 22*4882a593Smuzhiyun以下为正文 23*4882a593Smuzhiyun--------------------------------------------------------------------- 24*4882a593Smuzhiyun 25*4882a593Smuzhiyun在某些平台上，所谓的内存映射I/O是弱顺序。在这些平台上，驱动开发者有责任 26*4882a593Smuzhiyun保证I/O内存映射地址的写操作按程序图意的顺序达到设备。通常读取一个“安全” 27*4882a593Smuzhiyun设备寄存器或桥寄存器，触发IO芯片清刷未处理的写操作到达设备后才处理读操作， 28*4882a593Smuzhiyun而达到保证目的。驱动程序通常在spinlock保护的临界区退出之前使用这种技术。 29*4882a593Smuzhiyun这也可以保证后面的写操作只在前面的写操作之后到达设备（这非常类似于内存 30*4882a593Smuzhiyun屏障操作，mb()，不过仅适用于I/O）。 31*4882a593Smuzhiyun 32*4882a593Smuzhiyun假设一个设备驱动程的具体例子： 33*4882a593Smuzhiyun 34*4882a593Smuzhiyun ... 35*4882a593SmuzhiyunCPU A: spin_lock_irqsave(&dev_lock, flags) 36*4882a593SmuzhiyunCPU A: val = readl(my_status); 37*4882a593SmuzhiyunCPU A: ... 38*4882a593SmuzhiyunCPU A: writel(newval, ring_ptr); 39*4882a593SmuzhiyunCPU A: spin_unlock_irqrestore(&dev_lock, flags) 40*4882a593Smuzhiyun ... 41*4882a593SmuzhiyunCPU B: spin_lock_irqsave(&dev_lock, flags) 42*4882a593SmuzhiyunCPU B: val = readl(my_status); 43*4882a593SmuzhiyunCPU B: ... 44*4882a593SmuzhiyunCPU B: writel(newval2, ring_ptr); 45*4882a593SmuzhiyunCPU B: spin_unlock_irqrestore(&dev_lock, flags) 46*4882a593Smuzhiyun ... 47*4882a593Smuzhiyun 48*4882a593Smuzhiyun上述例子中，设备可能会先接收到newval2的值，然后接收到newval的值，问题就 49*4882a593Smuzhiyun发生了。不过很容易通过下面方法来修复： 50*4882a593Smuzhiyun 51*4882a593Smuzhiyun ... 52*4882a593SmuzhiyunCPU A: spin_lock_irqsave(&dev_lock, flags) 53*4882a593SmuzhiyunCPU A: val = readl(my_status); 54*4882a593SmuzhiyunCPU A: ... 55*4882a593SmuzhiyunCPU A: writel(newval, ring_ptr); 56*4882a593SmuzhiyunCPU A: (void)readl(safe_register); /* 配置寄存器？*/ 57*4882a593SmuzhiyunCPU A: spin_unlock_irqrestore(&dev_lock, flags) 58*4882a593Smuzhiyun ... 59*4882a593SmuzhiyunCPU B: spin_lock_irqsave(&dev_lock, flags) 60*4882a593SmuzhiyunCPU B: val = readl(my_status); 61*4882a593SmuzhiyunCPU B: ... 62*4882a593SmuzhiyunCPU B: writel(newval2, ring_ptr); 63*4882a593SmuzhiyunCPU B: (void)readl(safe_register); /* 配置寄存器？*/ 64*4882a593SmuzhiyunCPU B: spin_unlock_irqrestore(&dev_lock, flags) 65*4882a593Smuzhiyun 66*4882a593Smuzhiyun在解决方案中，读取safe_register寄存器，触发IO芯片清刷未处理的写操作， 67*4882a593Smuzhiyun再处理后面的读操作，防止引发数据不一致问题。 68