Re: zero filling pages

• Subject: Re: zero filling pages
• From: Nat! <email@hidden>
• Date: Fri, 25 Jul 2003 19:58:41 +0200

On Thursday, July 24, 2003, at 02:34 PM, Nat! wrote:

Actually the time expended because of zero filling is more than Shark will tell you, because the zero fill routine (as I wrote before) not only wastes the 4K of data cache but also 4K of instruction cache.

The semantics of the icbi instruction are roughly "The data at the effective address has changed, so if there is a copy in icache, remove it". The "if" means that the second loop in pmap_zero_page should only zap icache if the zeroed page formerly contained executable code that had been used recently enough to still be in icache.

• Prev by Date: Darwin / OSX 10.2.6 equivalent of ld -E on Linux?
• Next by Date: Re: ni_setpassword() and ni_setuser() authentication
• Previous by thread: Re: zero filling pages
• Next by thread: Re: zero filling pages
• Index(es):
  • Date
  • Thread