Re: kernel after 5.0-rc2 may not boot using BBL

From: Palmer Dabbelt <palmer@sifive.com>
To: atish.patra@wdc.com, anup@brainfault.org
Cc: Christoph Hellwig <hch@infradead.org>,
	atish.patra@wdc.com, linux-riscv@lists.infradead.org,
	Anup.Patel@wdc.com, schwab@suse.de
Subject: Re: kernel after 5.0-rc2 may not boot using BBL
Date: Fri, 08 Feb 2019 09:28:59 -0800 (PST)	[thread overview]
Message-ID: <mhng-487587c1-cd66-40f3-86fc-aa2f034f5028@palmer-si-x1c4> (raw)
In-Reply-To: <CAAhSdy3im4i0PJvKqDAfq2iDFvTA2O3ZsxjM6wSgdnFUU1eVcw@mail.gmail.com>

On Fri, 08 Feb 2019 01:17:00 PST (-0800), anup@brainfault.org wrote:
> On Fri, Feb 8, 2019 at 2:43 PM Christoph Hellwig <hch@infradead.org> wrote:
>>
>> On Fri, Feb 08, 2019 at 02:32:51PM +0530, Anup Patel wrote:
>> > It is a bug in BBL because it is not keeping sufficient gap between
>> > Linux kernel end and DTB. The change pointed by Atish actually
>> > reduced flat kernel binary size by almost 1MB.
>>
>> And that is all great.  But it breaks peoples few existing RISC-V
>> setups, which is the best way to frustate every single of the few
>> users out there.  And that is best way to lose any interested in this
>> already hard to use mess.
>
> There is no issue if you are building Linux RISC-V defconfig.
>
> The issue starts manifesting only if you more drivers enabled.
> Again, with growing kernel not everyone will see this issue because
> it will only show-up only when gap between payload_end and next
> 2MB aligned address is not sufficient to cover BSS section.

I agree with Christoph here: we should revert the commit.  There are various 
versions of BBL floating around that are a big headache to update, and since 
all the patch in question does is reduce the flat image size it's not worth 
breaking existing setups.

I also don't like the proposed fix, as that just statically sizes a 16MiB BSS.  
While that's fairly large, it is in essence just kicking the can down the road.  
The kernel on my laptop has a 7MiB BSS, and it's just an arbitrary smattering 
of the drivers I happen to use.  I'm not sure what the right long-term solution 
is here.  I kind of want to say "just use an ELF", as someone has already sat 
down and figured out how to put everything into a file that's required to make 
a binary run.  I feel like doing anything else here is really just going to 
involve re-inventing the wheel.

That's a bigger discussion, I've sent out the revert and will include it next 
week unless someone has an objection.

_______________________________________________
linux-riscv mailing list
linux-riscv@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-riscv