Re: [Minios-devel] [UNIKRAFT PATCHv5 17/46] plat/kvm: Add linker script for Arm64

[Simon Kuenzer <simon.kuenzer@xxxxxxxxx>]

On 11.09.2018 07:26, Wei Chen (Arm Technology China) wrote:
> Hi Julien,
>
> > -----Original Message-----
> > From: Julien Grall <julien.grall@xxxxxxx>
> > Sent: 2018年9月10日 18:16
> > To: Wei Chen (Arm Technology China) <Wei.Chen@xxxxxxx>; minios-
> > devel@xxxxxxxxxxxxxxxxxxxx; simon.kuenzer@xxxxxxxxx
> > Cc: Kaly Xin (Arm Technology China) <Kaly.Xin@xxxxxxx>; nd <nd@xxxxxxx>
> > Subject: Re: [Minios-devel] [UNIKRAFT PATCHv5 17/46] plat/kvm: Add linker
> > script for Arm64
> >
> > Hi Wei,
> >
> > On 10/09/18 09:31, Wei Chen (Arm Technology China) wrote:
> > > > -----Original Message-----
> > > > From: Julien Grall <julien.grall@xxxxxxx>
> > > > Sent: 2018年9月7日 22:50
> > > > To: Wei Chen (Arm Technology China) <Wei.Chen@xxxxxxx>; minios-
> > > > devel@xxxxxxxxxxxxxxxxxxxx; simon.kuenzer@xxxxxxxxx
> > > > Cc: Kaly Xin (Arm Technology China) <Kaly.Xin@xxxxxxx>; nd <nd@xxxxxxx>
> > > > Subject: Re: [Minios-devel] [UNIKRAFT PATCHv5 17/46] plat/kvm: Add linker
> > > > script for Arm64
> > > >
> > > > Hi,
> > > >
> > > > On 08/10/2018 08:08 AM, Wei Chen wrote:
> > > > > +       /* Read-write data that is not initialized explicitly in code */
> > > > > +       .bss :
> > > > > +       {
> > > > > +               *(.bss)
> > > > > +               *(.bss.*)
> > > > > +               *(COMMON)
> > > > > +               . = ALIGN(__PAGE_SIZE);
> > > > If you define the stack and page-table here. You can then free once boot
> > > > is done.
> > > I had thought a similar method before, but I am afraid to increase the
> > > image size, so I didn't use this method. Maybe I was considering too much
> > > for it?
> > IHMO trying to reduce the image size in early stage of the arm64 support
> > is very aggressive. This should be looked once you are done with the
> > core support.
> >
> > But I am not entirely sure to understand the concern regarding the image
> > size. As this is part of BSS, the actual binary will not be bigger in
> > the disk.
> >
> > In the guest memory, you will indeed use more memory. You can introduce
> > a new variable to tell you where the "boot memory". So you can discard
> > it easily once you are done with it.
> Ok, I will try it : )
Hey,

it is fine to me, to do such optimizations with follow-up patches.
Some TODO comments in the code should indicate that there is room for 
improvements wherever we can foresee it already during development.
So, I would document for now that the boot stack could be claimed back 
later during boot. I actually think we should find a generic solution 
with a ukplat interface since claiming back the boot stack could be 
possible also on other platforms. For some Unikernels it even may make 
sense to keep using the boot stack for the whole lifetime (e.g., no need 
for a scheduler), for others it makes sense. So the decision of claiming 
back the memory should come from a higher level.
> > This is not a proper "init section" but a way to avoid rely on free
> > space after the image in memory.
> >
> > Cheers,
> >
> > --
> > Julien Grall
Cheers,

Simon

_______________________________________________
Minios-devel mailing list
Minios-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/minios-devel