Re: [Xen-devel] [PATCH] xen/arm: mm: optimize setup_pagetables

[Peng Fan <van.freenix@xxxxxxxxx>]

Hi Julien,

On Wed, May 11, 2016 at 10:31:49AM +0100, Julien Grall wrote:
>Hi Peng,
>
>I would rename the title: "xen/arm: mm: remove unnecessary tlb flush in
>setup_pagetables".

Thanks. Will fix in V2.

>
>On 11/05/2016 08:59, Peng Fan wrote:
>>Before relocating xen to high address, need to build
>>the mapping BOOT_RELOC_VIRT_START to xen_paddr into
>>boot_pgtable and xen_pgtable.
>>
>>In setup_pagetables, relocate_xen will switch the root
>>page table from boot_pgtable to xen_pgtable/cpu0_pgtable.
>>
>>Before relocate_xen:
>>If touching xen_pgtable, no need to flush_xen_data_tlb_range_va_local.
>>We only need to flush tlb when touch boot_pgtable.
>>And no need to use write_pte for xen_pgtable, we can simply use
>>"xen_second[second_table_offset(dest_va)] = pte;"
>>
>>Also move the piece code near DTB/Fixmap/XEN_VIRT_START to
>>be more easier to understand.
>
>The commit message is rather hard to understand. Can you please rework it
>say:
>  - CPU0 is using the boot pages table and xen_second is not part of them =>
>No need to flush the TLB
>  - Clean up the code (i.e the code movement)
>

Sorry for my bad english. Will refine the commit log in V2.

>However, I would prefer to see 2 patches, one for removing the tlb flush, the
>other to move the code.

Will fix in V2.

>
>[...]
>
>>diff --git a/xen/arch/arm/mm.c b/xen/arch/arm/mm.c
>>index 94ea054..bebd82f 100644
>>--- a/xen/arch/arm/mm.c
>>+++ b/xen/arch/arm/mm.c
>>@@ -443,12 +443,6 @@ void __init setup_pagetables(unsigned long 
>>boot_phys_offset, paddr_t xen_paddr)
>>      lpae_t pte, *p;
>>      int i;
>>
>>-    /* Map the destination in the boot misc area. */
>>-    dest_va = BOOT_RELOC_VIRT_START;
>>-    pte = mfn_to_xen_entry(xen_paddr >> PAGE_SHIFT, WRITEALLOC);
>>-    write_pte(xen_second + second_table_offset(dest_va), pte);
>>-    flush_xen_data_tlb_range_va_local(dest_va, SECOND_SIZE);
>>-
>>      /* Calculate virt-to-phys offset for the new location */
>>      phys_offset = xen_paddr - (unsigned long) _start;
>>
>>@@ -498,6 +492,11 @@ void __init setup_pagetables(unsigned long 
>>boot_phys_offset, paddr_t xen_paddr)
>>      /* Map the destination in the boot misc area. */
>>      dest_va = BOOT_RELOC_VIRT_START;
>>      pte = mfn_to_xen_entry(xen_paddr >> PAGE_SHIFT, WRITEALLOC);
>>+    xen_second[second_table_offset(dest_va)] = pte;
>>+
>>+    /* Map the destination in the boot misc area. */
>>+    dest_va = BOOT_RELOC_VIRT_START;
>>+    pte = mfn_to_xen_entry(xen_paddr >> PAGE_SHIFT, WRITEALLOC);
>
>Why do you need to recompute "pte" and "dest_va"? They will be the same for
>"boot_second" and "xen_second".

I just change "write_pte(xen_second + second_table_offset(dest_va), pte);" to
"xen_second[second_table_offset(dest_va)] = pte;".

The pte and dest_va are not changed.

>
>Also, I think the code is still confusing in the current state because of the
>comments "Map the destination in the boot misc area.". They need to be
>updated to differentiate xen_second vs boot_second.

Yeah. This confused me when I first read this piece code. I can update
the comments in V2.

Thanks,
Peng.

>
>>      write_pte(boot_second + second_table_offset(dest_va), pte);
>>      flush_xen_data_tlb_range_va_local(dest_va, SECOND_SIZE);
>>  #ifdef CONFIG_ARM_64
>>
>
>Regards,
>
>-- 
>Julien Grall

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel