Re: [Xen-devel] [PATCH v3 26/28] xen/arm64: head: Use a page mapping for the 1:1 mapping in create_page_tables()

From: Julien Grall <julien.grall@arm.com>
To: Stefano Stabellini <sstabellini@kernel.org>
Cc: xen-devel@lists.xenproject.org,
	Volodymyr Babchuk <Volodymyr_Babchuk@epam.com>
Subject: Re: [Xen-devel] [PATCH v3 26/28] xen/arm64: head: Use a page mapping for the 1:1 mapping in create_page_tables()
Date: Tue, 17 Sep 2019 18:53:42 +0100	[thread overview]
Message-ID: <b0633e83-cb2f-12fd-3c9e-73a6bceb9ef9@arm.com> (raw)
In-Reply-To: <alpine.DEB.2.21.1908231754060.26226@sstabellini-ThinkPad-T480s>

Hi Stefano,

On 8/24/19 2:16 AM, Stefano Stabellini wrote:
> On Mon, 12 Aug 2019, Julien Grall wrote:
>>           lsr   x2, x19, #THIRD_SHIFT  /* Base address for 4K mapping */
>>           lsl   x2, x2, #THIRD_SHIFT
>> @@ -674,21 +591,80 @@ create_page_tables:
>>           cmp   x1, #(LPAE_ENTRIES<<3) /* 512 entries per page */
>>           b.lt  1b
> 
> Why can't we use create_mapping_entry here?

If we re-use create_mapping_entry, then we will compute everything at 
each loop.

At the moment, the loop execute 3 instructions (excluding the branch and 
comparison). We would end up to have an extra 5 instructions for arm64 
(9 instruction for arm32).

At the moment, we iterate 512 times the loop, so this adds 2560 
instructions (resp. 4608) in the boot code. Long-term plan, we will want 
to increase the Xen mapping to 4MB (so you double up the number iteration).

This is a less than ideal solution. I haven't come up with a clever 
solution so far, hence why it is left alone.

> 
> 
>> -        /* Defer fixmap and dtb mapping until after paging enabled, to
>> -         * avoid them clashing with the 1:1 mapping. */
>> +        /*
>> +         * If Xen is loaded at exactly XEN_VIRT_START then we don't
>> +         * need an additional 1:1 mapping, the virtual mapping will
>> +         * suffice.
>> +         */
>> +        cmp   x19, #XEN_VIRT_START
>> +        bne   1f
>> +        ret
>> +1:
>> +        /*
>> +         * Only the first page of Xen will be part of the 1:1 mapping.
>> +         * All the boot_*_id tables are linked together even if they may
>> +         * not be all used. They will then be linked to the boot page
>> +         * tables at the correct level.
>> +         */
>> +        create_table_entry boot_first_id, boot_second_id, x19, FIRST_SHIFT, x0, x1, x2
>> +        create_table_entry boot_second_id, boot_third_id, x19, SECOND_SHIFT, x0, x1, x2
> 
> If I understood the code right, it is not actually required to link
> boot_first_id -> boot_second_id and/or boot_second_id -> boot_third_id:
> it depends on whether x19 clashes with XEN_FIRST_SLOT, XEN_SECOND_SLOT,
> etc. Do you think it would be possible without making the code complex
> to only call create_table_entry boot_first_id, boot_second_id and
> create_table_entry boot_second_id, boot_third_id when required?  So
> moving the calls below after the relevant checks? It looks like it could
> be done by adding those calls before "ret". I wouldn't mind if they are
> duplicated but we could avoid it by adding appropriate labels and having
> a single return path:
> 
> out1:
>    create_table_entry boot_first_id, boot_second_id, x19, FIRST_SHIFT, x0, x1, x2
> out2:
>    create_table_entry boot_second_id, boot_third_id, x19, SECOND_SHIFT, x0, x1, x2
> out3:
>    ret

I have implemented something similar. I will send it as part of the next 
version in a bit.

Cheers,

-- 
Julien Grall

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xenproject.org
https://lists.xenproject.org/mailman/listinfo/xen-devel