OS学习笔记：Lecture 6

Communicationsmodels

1. 需要kernel参与 => 性能更好
2. kernel只提供一个shared memory => 安全性更好，不直接暴露数据

 

IPC

Interprocess communication

Models: shared memory; message passing

In xv6: pipe

通过read write交互

 

Message passing

• 过程：
  • Communication link
  • Exchange message via send/receive
• Direct communication
  • Send(P, message); receive(Q, message)
  • Links are established automatically
  • One link between any two processes
  • P, Q connect directly
• Indirect communication
  • Mailboxes
  • Operations:
    • Create a new mailbox
    • Send/receive through mailbox
    • Destroy a mailbox
• Synchronization & asynchronous
  • Message passing => blocking/non-blocking
  • Blocking => synvhronous
    • Sender block until the message is received
    • Receiver block until a message is available
  • Non-blocking => asynchronous
    • Sender send message and continue
    • Receiver receive a valid message or null

 

Interrupt, trap, sys call

Sys call

User ask for an OS service

Exception

Refer to illegal program action

Interrupt

Refer to sigal generated by a hardware device

 

 

LRPC (lightweight RPC)

• Simple control transfer: Client's thread executes in server's domain
• Simple data transfer: Shared argument stack
• Simple stubs: highly optimized marshalling
• Better concurrency
• High overhead of previous efforts
  • Stubs copy lots of data
  • Message copied from kernel
  • Access validation => security
  • Message transfer => wait
  • Scheduling
  • Context switch => 2 times
  • Dispatch =>one after another
• Connection setup phase
  • PDs (procedure descriptors) registered with kernel => like FDs
  • Argument stacks (a-stack) => 也保存返回值
  • Linkage records => store return address
  • Kernel return a-stack list to client
• Calling sequence
  • Client get a-stack
  • Find linkage
  • Ensure no thread using that a-stack/linkage
  • Put caller's return address and stack pointer in linkage
  • Push linkage on to thread control block's stack
  • Find an execution stack (E-stack => server's stack, like kernel stack) in server's domain
  • Update thread's SP to run off e-stack
  • CR3 perform address space switch to server domain
  • Upcall server's stub at address given in PD
• Discussion
  • Main kernel houskeeping task is allocationg A-stacks and E-stacks
  • Shared A-stacks safe and quick
  • Address spack switch is most of the overhead (no TLB tags => don't need to clear up the TLB)