公开(公告)号：US07770169B2

公开(公告)日：2010-08-03

申请号：US10847779

申请日：2004-05-17

申请人： Nedim Fresko ,  Christopher J. Plummer ,  Dean R. Long

发明人： Nedim Fresko ,  Christopher J. Plummer ,  Dean R. Long

IPC分类号： G06F9/46 ,  G06F12/00

CPC分类号： G06F9/3009 ,  G06F8/458 ,  G06F9/30043 ,  G06F9/3851 ,  G06F9/52 ,  G06F9/522 ,  G06F12/0253

摘要： Techniques for assuring thread rendezvous for a plurality of threads executing in a computing system are disclosed. Techniques can be used to assure thread rendezvous for read-only code in a manner that is more efficient than polling techniques. A Light-weight, Yet Trappable On Demand (LYTOD) instruction can be generated for code that is executed by one or more threads. Typically, a LYTOD instruction is generated at critical points of the code in order to assure safe-point thread rendezvous. The LYTOD is a lightweight instruction that can change its behavior from a lightweight instruction to an instruction that causes a trap when executed. The LYTOD can, for example, be implemented as a read-from-memory instruction that operates to load a register with a content of a valid memory location.

摘要翻译： 公开了用于确保在计算系统中执行的多个线程的线程会合的技术。 技术可以用于以比轮询技术更有效的方式来确保只读代码的线程会合。 可以为由一个或多个线程执行的代码生成轻量级，可追踪按需（LYTOD）指令。 通常，在代码的关键点生成LYTOD指令，以确保安全点线程会合。 LYTOD是一个轻量级的指令，可以将其行为从轻量级指令更改为在执行时引发陷阱的指令。 例如，LYTOD可以实现为从存储器读取指令，其操作用于加载具有有效存储器位置的内容的寄存器。

公开(公告)号：US08769250B2

公开(公告)日：2014-07-01

申请号：US12554595

申请日：2009-09-04

申请人： Nedim Fresko ,  Dean R. Long ,  Jiangli Zhou

发明人： Nedim Fresko ,  Dean R. Long ,  Jiangli Zhou

IPC分类号： G06F9/00

CPC分类号： G06F12/145 ,  G06F11/3636 ,  G06F12/109 ,  G06F2212/1012 ,  G06F2212/656

摘要： In general, the invention relates to a method. The method includes receiving notification, which includes context information, of a trap. The method further includes accessing, based at least partially upon the context information, a particular instruction that caused the trap, determining, based at least partially upon the context information, a particular address that is to be accessed by the particular instruction, updating a set of log information to indicate accessing of the particular address, causing subsequent accesses of the particular address to not give rise to a trap, after causing subsequent accesses of the particular address to not give rise to a trap, accessing the particular address, after accessing the particular address, causing subsequent accesses of the particular address to give rise to a trap, and causing the particular instruction to not be executed.

摘要翻译： 通常，本发明涉及一种方法。 该方法包括接收陷阱的通知，其包括上下文信息。 该方法还包括至少部分地基于上下文信息访问引起陷阱的特定指令，至少部分地基于上下文信息确定特定指令要访问的特定地址，更新集合 的日志信息以指示特定地址的访问，导致特定地址的后续访问不引起陷阱，在使特定地址的后续访问不引起陷阱，访问特定地址之后，在访问特定地址之后 特定地址，引起特定地址的后续访问以产生陷阱，并使特定指令不被执行。

公开(公告)号：US07711937B1

公开(公告)日：2010-05-04

申请号：US11206492

申请日：2005-08-17

申请人： Nedim Fresko ,  Dean R. Long ,  Jiangli Zhou

发明人： Nedim Fresko ,  Dean R. Long ,  Jiangli Zhou

IPC分类号： G06F9/00

CPC分类号： G06F12/145 ,  G06F11/3636 ,  G06F12/109 ,  G06F2212/1012 ,  G06F2212/656

摘要： A trap-based mechanism is provided for gaining greater visibility into the memory usage of a process. To detect and record the memory accesses of a process, a virtual address range (or a plurality of address ranges) of the process is set to a protected status. This address range represents the range of virtual addresses that are to be monitored for access. By setting the address range to a protected status, whenever a memory access (in one implementation, whenever a memory write) is made to a virtual address within that address range, a trap arises. When the trap arises, a trap handler is invoked. When invoked, the trap handler records the virtual address that was accessed. In this manner, the access of the virtual address is detected and recorded without having to add extensive instrumentation code to the process.

摘要翻译： 提供了基于陷阱的机制，可以更好地了解进程的内存使用情况。 为了检测和记录进程的存储器访问，将处理的虚拟地址范围（或多个地址范围）设置为受保护状态。 该地址范围表示要进行监视的虚拟地址的范围。 通过将地址范围设置为受保护的状态，每当存储器访问（在一个实现中，每当存储器写入时）对该地址范围内的虚拟地址进行，则会产生陷阱。 当陷阱出现时，调用陷阱处理程序。 调用时，陷阱处理程序记录访问的虚拟地址。 以这种方式，检测和记录虚拟地址的访问，而无需向该过程添加广泛的检测码。

公开(公告)号：US20090327659A1

公开(公告)日：2009-12-31

申请号：US12554595

申请日：2009-09-04

申请人： Nedim Fresko ,  Dean R. Long ,  Jiangli Zhou

发明人： Nedim Fresko ,  Dean R. Long ,  Jiangli Zhou

IPC分类号： G06F9/30 ,  G06F12/00

CPC分类号： G06F12/145 ,  G06F11/3636 ,  G06F12/109 ,  G06F2212/1012 ,  G06F2212/656

摘要： In general, the invention relates to a method. The method includes receiving notification, which includes context information, of a trap. The method further includes accessing, based at least partially upon the context information, a particular instruction that caused the trap, determining, based at least partially upon the context information, a particular address that is to be accessed by the particular instruction, updating a set of log information to indicate accessing of the particular address, causing subsequent accesses of the particular address to not give rise to a trap, after causing subsequent accesses of the particular address to not give rise to a trap, accessing the particular address, after accessing the particular address, causing subsequent accesses of the particular address to give rise to a trap, and causing the particular instruction to not be executed.

摘要翻译： 通常，本发明涉及一种方法。 该方法包括接收陷阱的通知，其包括上下文信息。 该方法还包括至少部分地基于上下文信息访问引起陷阱的特定指令，至少部分地基于上下文信息确定特定指令要访问的特定地址，更新集合 的日志信息以指示特定地址的访问，导致特定地址的后续访问不引起陷阱，在使特定地址的后续访问不引起陷阱，访问特定地址之后，在访问特定地址之后 特定地址，引起特定地址的后续访问以产生陷阱，并使特定指令不被执行。