Date: Fri, 1 Apr 2011 19:53:17 +0800 From: fuzhli <fuzhli1007@gmail.com> To: Alan Cox <alc@rice.edu> Cc: alc@freebsd.org, questions@freebsd.org Subject: Re: about NKPT on amd64 Message-ID: <AANLkTikCHuMrKQ6weORdKSo7OUMveZcavZUUK-a-5L4R@mail.gmail.com> In-Reply-To: <4D935139.8080701@rice.edu> References: <AANLkTimdxMo3HU8aLuj9F9PCpYU9M5C5X0o5HkHCvqtC@mail.gmail.com> <4D935139.8080701@rice.edu>
next in thread | previous in thread | raw e-mail | index | archive | help
Hi, Alan Cox
Thanks for your reply. I want to know why the NKPT set to 32 in
revision187465,
is it possible that set NKPT less or lager than 32? What's the limitation i=
f
set NKPT to 32?
I also guess the KPTphys[] not used in kernel initialization on amd64 i=
n
revision187465. And
I notice that in revision189075, in function pmap_init() "the vm page array
entries for the kernel
pmap's page table pages" is initialized; but in revision187458, the
initialization is not performed.
I guess "the earlist part of the kernel's initialization" is the
pmap_bootstrap(firstaddr),
this will call function create_pagetables().
In function create_pagetables(), we first use allocpages(firstaddr,
NKPT) to
allocate NKPT pages, then initialize the KPTphys[] and KPDphys[]. But then
the KPDphys[] re-initialize the range (0, *firstaddr) in 2MB page:
/* Map from zero to end of allocations under 2M pages */
/* This replaces some of the KPTphys entries above */
for (i =3D 0; (i << PDRSHIFT) < *firstaddr; i++) {
((pd_entry_t *)KPDphys)[i] =3D i << PDRSHIFT;
((pd_entry_t *)KPDphys)[i] |=3D PG_RW | PG_V | PG_PS | PG_G;
}
This means that the KPDphys[] will be override completely if the range
(KERNBASE, virtual_avail)'s size larger than 64MB, so I guess that the
KPTphys[]
not used anymore if the range (KERNBASE, virtual_avail)'s size larger than
64MB.
=D4=DA 2011=C4=EA3=D4=C230=C8=D5 =CF=C2=CE=E711:50=A3=ACAlan Cox <alc@rice.=
edu>=D0=B4=B5=C0=A3=BA
> On 03/30/2011 01:47, fuzhli wrote:
> > Hi, Alan
> > I'm study the "Revision 187465" : "Prepare for a larger kernel virtual
> > address space". After read some relative source code, I have an
> > question about the macro NKPT on amd64: why 32 is enough for the
> > kernel page table pages? Do it means that the range (KERNBASE,
> > virtual_avail) should always less than 64MB( 32 * 2MB)?
>
> NKPT sets the size of the kernel page table during the earliest part of
>
the kernel's initialization. After that, the size of the page table
> grows dynamically according to usage.
>
> Regards,
> Alan
>
>
--=20
=B1=F0=D7=F6=C3=CE,=C4=E3=D2=D124=CB=EA=C1=CB
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?AANLkTikCHuMrKQ6weORdKSo7OUMveZcavZUUK-a-5L4R>