这篇文章主要说明俩个问题：
1. 在APC_LEVEL上，Thread为何不能被suspend。
2. 在 APC_LEVEL上，可以使用分页内存的原因。
关于线程如何响应APC，要看是何种APC，请参考MSDN文档。我在看微软提供的资料的时候，发现俩个比较难懂的问题，把它们单独拿出来讨论。
首先看中断请求级：IRQL（Interrupt Request Levels）
| IRQL | IRQL value | Description | ||
| x86 | IA64 | AMD64 | ||
| PASSIVE_LEVEL | 0 | 0 | 0 | User threads and most kernel-mode operations |
| APC_LEVEL | 1 | 1 | 1 | Asynchronous procedure calls and page faults |
| DISPATCH_LEVEL | 2 | 2 | 2 | Thread scheduler and deferred procedure calls (DPCs) |
| CMC_LEVEL | N/A | 3 | N/A | Correctable machine-check level (IA64 platforms only) |
| Device interrupt levels (DIRQL) | 3-26 | 4-11 | 3-11 | Device interrupts |
| PC_LEVEL | N/A | 12 | N/A | Performance counter (IA64 platforms only) |
| PROFILE_LEVEL | 27 | 15 | 15 | Profiling timer for releases earlier than Windows 2000 |
| SYNCH_LEVEL | 27 | 13 | 13 | Synchronization of code and instruction streams across processors |
| CLOCK_LEVEL | N/A | 13 | 13 | Clock timer |
| CLOCK2_LEVEL | 28 | N/A | N/A | Clock timer for x86 hardware |
| IPI_LEVEL | 29 | 14 | 14 | Interprocessor interrupt for enforcing cache consistency |
| POWER_LEVEL | 30 | 15 | 14 | Power failure |
| HIGH_LEVEL | 31 | 15 | 15 | Machine checks and catastrophic errors; profiling timer for Windows XP and later releases |
微软说：
When a processor is running at a given IRQL, interrupts at that IRQL and lower are masked off (blocked) on the processor.
但其实这句话只适合在DISPATCH_LEVEL到HIGH_LEVEL之间。在APC_LEVEL和PASSIVE_LEVEL级别要特殊对待。而且这两个级别可以被调度器调度，就显得更加复杂。
所以微软又说：
IRQL分为： Processor-specific and Thread-specific IRQLs
-
Processor-specific IRQLS：
- DISPATCH_LEVEL
- DIRQL
- HIGHEST_LEVEL
规则：When a processor is running at a given IRQL, interrupts at that IRQL and lower are masked off (blocked) on the processor.
-
Thread-specific IRQLS:
- PASSIVE_LEVEL
- IRQL PASSIVE_LEVEL in a critical region. Intermediate level（KeEnterCriticalRegion, KeLeaveCriticalRegion）
- APC_LEVEL
在这三个级别上运行的线程，都能被调度器调度（调度器运行在DISPATCH_LEVEL），调度器考虑的只是优先级，优先级高的就能抢占优先级低的线程。所一个运行在APC_LEVEL低优先级的线程，可以被一个运行在PASSIVE_LEVEL优先级高的线程给抢占。所有微软说：
The thread scheduler considers only thread priority, and not IRQL, when preempting a thread. If a thread running at IRQL=APC_LEVEL blocks, the scheduler might select a new thread for the processor that was previously running at PASSIVE_LEVEL.
线程相关的IRQL，可以将线程想象为一个伪CPU，此CPU只有三个LEVEL：PASSIVE_LEVEL，Intermediate level and APC_LEVEL.
IRQL PASSIVE_LEVEL in a critical region
Code that is running at PASSIVE_LEVEL in a critical region is effectively running at an intermediate level between PASSIVE_LEVEL and APC_LEVEL. Calls to KeGetCurrentIrql return PASSIVE_LEVEL. Driver code can determine whether it is operating in a critical region by calling the function KeAreApcsDisabled (available in Windows XP and later releases).
Driver code that is running above PASSIVE_LEVEL (either at PASSIVE_LEVEL in a critical region or at APC_LEVEL or higher) cannot be suspended. Almost every operation that a driver can perform at PASSIVE_LEVEL can also be performed in a critical region. Two notable exceptions are raising hard errors and opening a file on storage media.
IRQL APC_LEVEL
APC_LEVEL is a thread-specific IRQL that is most commonly associated with paging I/O. Applications cannot suspend code that is running at IRQL=APC_LEVEL. The system implements fast mutexes (a type of synchronization mechanism) at APC_LEVEL. The KeAcquireFastMutex routine raises the IRQL to APC_LEVEL, and KeReleaseFastMutex returns the IRQL to its original value.
The only difference between a thread that is running at PASSIVE_LEVEL with APCs disabled and a thread that is running at APC_LEVEL is that while running at APC_LEVEL, the thread cannot be interrupted to deliver a special kernel-mode APC.
Thread 进入APC_LEVEL方式：
- Call KeAcquireFastMutex
- Delivery a APC
- KeRaiseIrql （一般不用）等。
使用Fast Mutex进入APC_LEVEL后，对于其它线程，若要获取此Mutex，则会被设置为等待状态。对于线程自己而言，微软说：
Code paths that are protected by a fast mutex run at IRQL=APC_LEVEL, thus disabling delivery of all APCs and preventing the thread from suspension.
即：阻止响应任何APC，而且线程不能被挂起（suspend），为什么不能被挂起？因为操作系统实现线程挂起的方式，就是Delivery APC，在APC的回调函数里面等待一个信号量（这个是我查阅WRK中找到的答案）。由于运行在APC_LEVEL上，会disabling delivery of all APCs。
如果将有一个线程理解为一个伪CPU，此CPU只有三个LEVEL：PASSIVE_LEVEL，Intermediate level and APC_LEVEL. 然后将Delivery APC当做一个中断来处理，使用微软的中断规则来处理：When a processor is running at a given IRQL, interrupts at that IRQL and lower are masked off (blocked) on the processor. 就可以理解为CPU在APC_LEVEL上，屏蔽了所有等于和小于它的中断。
APC有三种：kernel normal apc, special kernel normal apc, and user mode apc。Delivery a APC，此APC并不是会马上运行，要看情况而定，这个情况很复杂，不在本文说明，可以去微软MSDN寻找答案。
另外一个问题：
在 APC_LEVEL上，可以使用分页内存。但我查找资料发现，当在分页内存中发生了Page fault，系统会delivery APC，那岂不是此APC也不会得到执行了么？那不就形成死锁，和在DIPATCH_LEVEL一样，形成蓝屏了么？后来，我找到了一个关键点。在APC_LEVEL上，一旦发生Page fault，系统首先是将此线程挂起，此线程将等待一个内核同步对象，虽然线程运行在APC_LEVEL上但是线程被挂起，此时调度器会调度别的线程。当Page fault处理完成后，系统会给此线程delivery a APC，此线程又临时变为可调度状态，调度器就去调用这个APC，这个APC会激活线程等待的那个内核同步对象，但线程等待的那个点并不会马上得到执行，得等到这个APC退出后，因为这个APC是在这个线程中执行的。等到Page fault的Delivery APC退出，线程等待的对象处于激活状态，此线程变为可调度，调度器会在合适的时间，继续调度此线程，由于内存错误得到了解决，所以访问内存没有任何问题。
这段解释是不对的，一旦线程在APC_LEVEL上，而且线程处于等待状态，给它排队APC，这个APC是不会得到执行的。排队APC等于给线程发送一个软中断，是不会得到响应的。在DISPATCH_LEVEL不能使用分页的主要原因在于，一旦有分页错误，它就会等待分页完成，在DISPATCH_LEVEL上，是不能调用KeWaitXXX来等待的。而不是在于排队一个APC，而且一个同步的分页，是不会排队APC的。在IoCompleteRequest函数中有如下代码：
if (Irp->Flags & (IRP_PAGING_IO | IRP_CLOSE_OPERATION |IRP_SET_USER_EVENT)) {
if (Irp->Flags & (IRP_SYNCHRONOUS_PAGING_IO | IRP_CLOSE_OPERATION |IRP_SET_USER_EVENT)) {
ULONG flags;
flags = Irp->Flags & (IRP_SYNCHRONOUS_PAGING_IO|IRP_PAGING_IO);
*Irp->UserIosb = Irp->IoStatus;
(VOID) KeSetEvent( Irp->UserEvent, PriorityBoost, FALSE );
if (flags) {
if (IopIsReserveIrp(Irp)) {
IopFreeReserveIrp(PriorityBoost);
} else {
IoFreeIrp( Irp );
}
}
} else {
thread = Irp->Tail.Overlay.Thread;
KeInitializeApc( &Irp->Tail.Apc,
&thread->Tcb,
Irp->ApcEnvironment,
IopCompletePageWrite,
(PKRUNDOWN_ROUTINE) NULL,
(PKNORMAL_ROUTINE) NULL,
KernelMode,
(PVOID) NULL );
(VOID) KeInsertQueueApc( &Irp->Tail.Apc,
(PVOID) NULL,
(PVOID) NULL,
PriorityBoost );
}
return;
}
如果线程运行在DISPATCH_LEVEL，显而易见，单核CPU中调度器无法执行（多核CPU中可以执行在另外一个核中）。但主要的原因是发送的APC相当于一个APC_LEVEL的中断，它是得不到执行的。所以在DISPATCH_LEVEL上，会崩溃。而且在DISPATCH_LEVEL上，不能等待一个timer不为0的内核对象。
扫一扫
2595
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
