Hi Neale, Neale Ferguson wrote: >> How quickly do you want an answer to this ? >> > Last week! ;-) > > It's not a write fault that's the problem here. It's when a read is done the > page is marked dirty and fsflush will attempt to write it back out. Actually > it's any read of the page (whether caused by a read fault or a write fault). > The 1st time the page is brought in shouldn't result in the kernel seeing the > page as dirty. > > What I want to happen is > - trap > - as_fault -> segvn_fault -> getpage > - page is created, data read in (causes mod bit to be on), getpage resets > modified bit to 0 > - trap return > > Any dirtying of the page is now "legitimate". So if it was a write fault then > the reissued instruction will result in the page being dirtied. If it was a > read operation then the reference bit will be set but fsflush will not > identify the page as dirty until and unless something writes into that page. > > So, there is no way to turn off the "storage key" before faulting in the page, and then turning it back on after the input has completed? Also, I'm curious, why does the hardware care when something besides the CPU is modifying the page? By the way, your approach sounds fine. What happens when you try it?
max > Neale > > > This message posted from opensolaris.org > _______________________________________________ > opensolaris-code mailing list > [email protected] > http://mail.opensolaris.org/mailman/listinfo/opensolaris-code > > _______________________________________________ opensolaris-code mailing list [email protected] http://mail.opensolaris.org/mailman/listinfo/opensolaris-code
