> WAY BACK WHEN I worked at a time-sharing service bureau running systems > based on CP67/CMS, we used still another mechanism. CP67 was modified to > use a BAL to a particular address in low storage, which went to a setup > routine, which in turn called the appropriate service. Reason: > interrupts took too long to process The actual service and its > parameters were determined by the contents of storage immediately > following the BAL instruction. All SVC's in the virtual machine were > processed by the CMS code in the user's virtual machine and interactions > between CMS and CP67 were handled via HVC (Hypervisor Call aka DIAGNOSE) > and its parm list. > > Interrupts are expensive; avoid them whenever possible; even the > PSW-SWAP is very slow!!!
In CP/67 or VM/370, an SVC under CMS in would give control to the hypervisor in real supervisor state, which would simulate the the SVC for the guest (CMS) by giving control to the guest SVC interrupt handler in virtual supervisor state (which is still real problem state). I would expect that the hypervisor simulation was a much more significant part of the cost than the PSW swap. A DIAGNOSE issued by CMS would cause a program check interrupt (privileged-operation exception) since CMS was running in real problem state, and that was the meachanism for getting into the hypervisor in real supervisor state. Jim Mulder z/OS System Test IBM Corp. Poughkeepsie, NY ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
