JFFS2 mounting performance improvement

JFFS2 mounting performance improvement

[Davide CASCONE]

It seems that during the JFFS2 mounting, the file system uses the function cfi_intelext_point (in jffs2_scan_medium function) to point to the flash memory. This function refers to the flash memory using the virtual address (map->virt). But if the flash is cached and the burst mode is enabled, it should be better to refer to the flash with the address map->cached (in the same way of cfi_intelext_read function, that calls the map_copy_from function, in which there is a check to verify if the flash is cached). This improves in a considerable way the performance during the mounting phase, reducing the mounting time. 
What do you think about? Do you agree with me? Do you know if there are other points in JFFS2 code that uses the virtual address map->virt instead of map->cached.

Best Regards,
Davide

Re: JFFS2 mounting performance improvement

[Balaji]

Best Regards
====================================
 Balaji Kadloor
 Philips Semiconductors Division (HIC-B)
 Philips Innovation Campus
1 Murphy Rd, Ulsoor, Bangalore - 560008, INDIA
Tel: (91)-80-25579000 X2269,
Fax: (91)-80-25560580
====================================

"  Success doesn't go to the Faster or the Stonger Man.
   Sooner or Later,the man who wins is the man who thinks HE CAN "
----- Original Message ----- 
From: "Davide CASCONE" <davide.cascone@st.com>
To: <linux-mtd@lists.infradead.org>
Sent: Tuesday, June 20, 2006 1:55 PM
Subject: JFFS2 mounting performance improvement


>
> It seems that during the JFFS2 mounting, the file system uses the function
cfi_intelext_point (in jffs2_scan_medium function) to point to the flash
memory. This function refers to the flash memory using the virtual address
(map->virt). But if the flash is cached and the burst mode is enabled, it
should be better to refer to the flash with the address map->cached (in the
same way of cfi_intelext_read function, that calls the map_copy_from
function, in which there is a check to verify if the flash is cached). This
improves in a considerable way the performance during the mounting phase,
reducing the mounting time.
> What do you think about? Do you agree with me? Do you know if there are
other points in JFFS2 code that uses the virtual address map->virt instead
of map->cached.
>
> Best Regards,
> Davide
>
>
>
>
>
> ______________________________________________________
> Linux MTD discussion mailing list
> http://lists.infradead.org/mailman/listinfo/linux-mtd/

Re: JFFS2 mounting performance improvement

[David Woodhouse]

On Tue, 2006-06-20 at 10:25 +0200, Davide CASCONE wrote:
> 
> It seems that during the JFFS2 mounting, the file system uses the
> function cfi_intelext_point (in jffs2_scan_medium function) to point
> to the flash memory. This function refers to the flash memory using
> the virtual address (map->virt). But if the flash is cached and the
> burst mode is enabled, it should be better to refer to the flash with
> the address map->cached (in the same way of cfi_intelext_read
> function, that calls the map_copy_from function, in which there is a
> check to verify if the flash is cached). This improves in a
> considerable way the performance during the mounting phase, reducing
> the mounting time. 
> What do you think about? Do you agree with me? 

Yes. As long as you handle the cache invalidation where appropriate, you
should definitely use the cached address for the 'point' function.

> Do you know if there are other points in JFFS2 code that uses the
> virtual address map->virt instead of map->cached. 

JFFS2 wouldn't touch that directly. It's down to the chip driver. JFFS2
just uses the read() and point() methods (during scan), so I think the
answer would be 'no'.

-- 
dwmw2