[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Xen-ia64-devel] [PATCH] Remove FORCE_CRASH from alt_itlb_miss
Thanks for your detail information.
I thought that my patch causes discord between Xen and Linux-kernel.
When I confirmed itc with ITP, I confirmed that a FPSWA code was
inserted at size of 16 kbytes. As you thought, I think that it is
good to use bigger page size so that we reduce tlb miss.
I remake my patch and should send it again? Or, at first you apply
my patch to xen-ia64-unstable tree, and do I send a patch setting
page size later?
Xu, Anthony wrote:
>>From: Masaki Kanno
>>Sent: 2006?4?25? 23:27
>>Subject: Re: [Xen-ia64-devel] [PATCH] Remove FORCE_CRASH from alt_itlb_miss
>>Please teach me in detail.
>For supporting discontinuous memory, ps in region register is always 16K,
>There are two implicit parameters for itc instruction,
>The other is cr.ifa indicating fault virtual address.
>One is cr.itir, cr.itir.ps determine the TLB page size,
>When dtlb_miss happens,
>cr.itir.ps = rr.ps (now this is 16K)
>But in identity mapping, we can use bigger page size to reduce tlb miss faults,
>Following is pseudo code for this
> cr.itir.ps = IA64_GRANULE_SHIFT
> itc.d // insert the TLB entry
>>I applied this patch and FPSWA supporting patch to Xen and tested it.
>>Because I ran LTP on dom0, and a test about a floating point succeeded,
>>I sent this patch.
>Your patch is correct. :-)
>>Xu, Anthony wrote:
>>>Since the page size of region 7 is 16K now,
>>>This patch make identity mapping based on 16K.
>>>Can we align with linux kernel using 16M identity mapping?
>>>>[mailto:xen-ia64-devel-bounces@xxxxxxxxxxxxxxxxxxx] On Behalf Of Masaki
>>>>Sent: 2006?4?24? 19:39
>>>>Subject: [Xen-ia64-devel] [PATCH] Remove FORCE_CRASH from alt_itlb_miss
>>>>This patch removed FORCE_CRASH from alt_itlb_miss handler.
>>>>Signed-off-by: Masaki Kanno <kanno.masaki@xxxxxxxxxxxxxx>
>>Xen-ia64-devel mailing list
Xen-ia64-devel mailing list