Re: [Minios-devel] [UNIKRAFT PATCHv5 20/46] plat/common: Add early debug console library for Arm64

[Julien Grall <julien.grall@xxxxxxx>]

Hi Wei,

On 09/12/2018 03:58 AM, Wei Chen (Arm Technology China) wrote:
> > -----Original Message-----
> > From: Julien Grall <julien.grall@xxxxxxx>
> > Sent: 2018年9月11日 18:37
> > To: Wei Chen (Arm Technology China) <Wei.Chen@xxxxxxx>; Julien Grall
> > <julien.grall@xxxxxxxxxx>; minios-devel@xxxxxxxxxxxxxxxxxxxx;
> > simon.kuenzer@xxxxxxxxx
> > Cc: Kaly Xin (Arm Technology China) <Kaly.Xin@xxxxxxx>; nd <nd@xxxxxxx>
> > Subject: Re: [Minios-devel] [UNIKRAFT PATCHv5 20/46] plat/common: Add early
> > debug console library for Arm64
> >
> > Hi Wei,
> >
> > On 11/09/18 06:35, Wei Chen (Arm Technology China) wrote:
> > > > > Oh, You beat me. Yes, PL011 start at IPA 0 is possible. But I don't know
> > > > > how to distinguish PL011 at IPA 0 or #ifndef
> > > > CONFIG_KVM_EARLY_DEBUG_PL011_UART.
> > > > > I had tried not to check (!pl011_uart_bas), it will generate an exception,
> > > > > and the exception entry will call PL011 to print message. It's an infinite
> > > > loop.
> > > >
> > > > If I understand correctly, KVM_EARLY_DEBUG_PL011_UART will exist if
> > > > KVM_DEBUG_SERIAL_CONSOLE is set. So one solution would be to introduce
> > > > an extra variable to check whether the UART has been initialized.
> > > >
> > > > This would be set to 1 at boot when KVM_DEBUG_SERIAL_CONSOLE is set.
> > > Ok, I understand now.
> > >
> > > Just a digression, if an IPA 0 is possible, so I think most of the NULL
> > > Check would be unreliable. For example, lots of fdt_get_property will
> > > return a pointer. You know, most of us will use the if(!pointer) to check
> > > the return value.
> > Well, that the drawback with building page-table with VA == PA. You have
> > somehow to ensure that VA 0 is never mapped.
> >
> > For now, you could just check whether the VA 0 has been mapped and print
> > a warning. This would need to be properly fixed sooner or later as it
> > looks like some hardware have PA 0 valid.
> Yes, at least I know that boot rom (flash) of QEMU/KVM is starting from IPA 0.
> And we have already mapped it, because we have mapped 0 - 0x3FFFFFFF as device
> MMIO. But I think only flash code need to care about the valid IPA 0, other code
> can use NULL as normal. For Unikraft, I don’t think we would implement the
> library for QEMU/KVM boot rom in recently.
NULL pointer is not only about check in the code. It is also here to 
generate a fault if a NULL pointer is dereferenced (i.e because of a 
missing check). If you have the VA 0 mapped, then you will read/write 
the content. This is going to be extremely difficult to debug such code 
if it is ever noticed.
> Above comment is about current code for QEMU/KVM. I know some SoC would use PA0
> for RAM, that would be more crazy : )
Have a look under arch/arm64/dts, you may find quite a few with RAM 
starting a 0 ;).
Cheers,

--
Julien Grall

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