On Dec 15, 2012, at 12:54 PM, David Laight wrote: > On Sat, Dec 15, 2012 at 11:24:09AM -0800, Matt Thomas wrote: >>> There are several problems though: >>> 1) gdb (etc) will find the wrong registers for the lwp. >> >> Not really since you can't return to user land before surrendering the PCU. > > An lwp could easily sleep with a PCU assigned (especially if the FP > regs are being used for copyin/out). > > I'm fairly sure there is an expectation of being able to read an lwp's > user-space registers at any time. While they might be stale (eg when the > process is in userspace), I doubt they are expected to be completely wrong. > > I'm not sure what effect delayed saving of FP registers has on this. > Possibly the call to retrieve the registers ensures the FP state > is flushed to memory.
It does. And the user-space register are flushed to memory before the kernel can steal it PCU.
