TeamForge : Post

Forum Topic - Intel flash parts not performing buffered writes -- bug in mtd-flash or fs-flash3?: (5 Items)

View: as

Josh Radel

06/30/2009 8:54 PM

post32898

Intel flash parts not performing buffered writes -- bug in mtd-flash or fs-flash3?

We have Intel StrataFlash parts and are using the stock devf-generic. The flash is correctly recognized as 16-bit and 
supporting buffered writes (f3s_iCFI_ident() finds dbase->buffer_size to be 64 and sets the F3S_WRITE_BUFFER on dbase->
flags).

However, I've stepped through the f3s_iCFI_v2write() code (at least that available from the MPC8641d BSP) when we're 
doing flash writes on the raw flash device, and it is not using buffered writes (the (flags & F3S_WRITE_BUFFER) check fails). The call tree for our writes looks like:
f3s_raw_write() -> f3s_flash_write() -> 
f3s_iCFI_v2write().

Grepping the fs-flash3 code, it looks like the "flags" parameter that is passed to f3s_iCFI_v2write() is the write_flags
 parameter passed to f3s_flash_write(), which doesn't at all care about dbase->flags (it is always explicitly set to 0, 
1, or F3S_VERIFY_WRITE when f3s_flash_write() is called) -- so the "buffered writes" path in f3s_iCFI_v2write() is never
 taken.

Is this a bug in the fs-flash3 code (not correctly passing the F3S_WRITE_BUFFER flag) or in f3s_iCFI_v2write() (should 
be checking (dbase->flags & F3S_WRITE_BUFFER) instead of (flags & F3S_WRITE_BUFFER))?

David Sarrazin

06/30/2009 10:10 PM

post32899

RE: Intel flash parts not performing buffered writes -- bug in mtd-flash or fs-flash3?

Josh,

Thank you for reporting this.  Yes, this is a bug in the MTD library.
The function iCFI_ident() was setting the capability flag incorrectly,
and f3s_iCFI_v2write() was checking the wrong flag.  This has been
fixed, and the fix is available on the Foundry27 "BSP and Drivers"
project.

http://community.qnx.com/integration/viewcvs/viewcvs.cgi/trunk/hardware/
flash/mtd-flash/intel/?root=bsp&system=exsy1001

That's the link to the trunk (ie. 6.4.1) branch.  The change is also
available on the 6.3.0 branch.

David 

> -----Original Message-----
> From: Josh Radel [mailto:community-noreply@qnx.com] 
> Sent: June 30, 2009 8:55 PM
> To: general-filesystems
> Subject: Intel flash parts not performing buffered writes -- 
> bug in mtd-flash or fs-flash3?
> 
> We have Intel StrataFlash parts and are using the stock 
> devf-generic. The flash is correctly recognized as 16-bit and 
> supporting buffered writes (f3s_iCFI_ident() finds 
> dbase->buffer_size to be 64 and sets the F3S_WRITE_BUFFER on 
> dbase->flags).
> 
> However, I've stepped through the f3s_iCFI_v2write() code (at 
> least that available from the MPC8641d BSP) when we're doing 
> flash writes on the raw flash device, and it is not using 
> buffered writes (the (flags & F3S_WRITE_BUFFER) check fails). 
> The call tree for our writes looks like:
> f3s_raw_write() -> f3s_flash_write() -> f3s_iCFI_v2write().
> 
> Grepping the fs-flash3 code, it looks like the "flags" 
> parameter that is passed to f3s_iCFI_v2write() is the 
> write_flags parameter passed to f3s_flash_write(), which 
> doesn't at all care about dbase->flags (it is always 
> explicitly set to 0, 1, or F3S_VERIFY_WRITE when 
> f3s_flash_write() is called) -- so the "buffered writes" path 
> in f3s_iCFI_v2write() is never taken.
> 
> Is this a bug in the fs-flash3 code (not correctly passing 
> the F3S_WRITE_BUFFER flag) or in f3s_iCFI_v2write() (should 
> be checking (dbase->flags & F3S_WRITE_BUFFER) instead of 
> (flags & F3S_WRITE_BUFFER))?
> 
> _______________________________________________
> General
> http://community.qnx.com/sf/go/post32898
> 
>

Josh Radel

08/03/2009 3:37 PM

post35148

Re: RE: Intel flash parts not performing buffered writes -- bug in mtd-flash or fs-flash3?

Hi David,

I might be missing something... but it looks to me like the bug is still there. The devf-generic shipped in 6.4.1 is 
still slow -- not taking advantaged of buffered writes -- and the code in http://community.qnx.com/integration/viewvc/
viewvc.cgi/trunk/hardware/flash/mtd-flash/intel/?root=bsp&system=exsy1001&pathrev=137 still appears to have the bug.

How I fixed it in my local BSP:

In iCFI_v2write.c, I changed the line:

if ((flags & F3S_WRITE_BUFFER) && (left >= (4 * flashcfg.bus_width)))

to:

if ((dbase->flags & F3S_WRITE_BUFFER) && (left >= (4 * flashcfg.bus_width)))

David Sarrazin

08/04/2009 2:39 AM

post35153

RE: RE: Intel flash parts not performing buffered writes -- bug in mtd-flash or fs-flash3?

Josh,

I think you're right.  That code used to check for an entirely different
(incorrect) flag.  That was fixed for 6.4.1, but you're right that it
should be the dbase->flags, and not the passed-in flags.  I'll open a PR
and make a change to the code.

Thanks,

David 

> -----Original Message-----
> From: Josh Radel [mailto:community-noreply@qnx.com] 
> Sent: August 3, 2009 3:38 PM
> To: general-filesystems
> Subject: Re: RE: Intel flash parts not performing buffered 
> writes -- bug in mtd-flash or fs-flash3?
> 
> Hi David,
> 
> I might be missing something... but it looks to me like the 
> bug is still there. The devf-generic shipped in 6.4.1 is 
> still slow -- not taking advantaged of buffered writes -- and 
> the code in 
> http://community.qnx.com/integration/viewvc/viewvc.cgi/trunk/h
> ardware/flash/mtd-flash/intel/?root=bsp&system=exsy1001&pathre
v=137 still appears to have the bug.
> 
> How I fixed it in my local BSP:
> 
> In iCFI_v2write.c, I changed the line:
> 
> if ((flags & F3S_WRITE_BUFFER) && (left >= (4 * flashcfg.bus_width)))
> 
> to:
> 
> if ((dbase->flags & F3S_WRITE_BUFFER) && (left >= (4 * 
> flashcfg.bus_width)))
> 
> 
> 
> 
> _______________________________________________
> 
> General
> http://community.qnx.com/sf/go/post35148
> 
>

Albrecht Uhlmann(deleted)

11/26/2011 7:05 AM

post90307

Re: RE: RE: Intel flash parts not performing buffered writes -- bug in mtd-flash or fs-flash3?

Hello, I just ran into the same problem. I use BSP for i.MX35 3DS, the MTD library shipped with this still contains that
 bug. I modified it accordingly and now I have write speed increased from 11KB/s (!!) to 70KB/s. Still not really 
satisfying, but might be that the flash part (RC28F256P33 from Numonyx) requires some special code.

I downloaded bsp-freescale-p1022-ds-6.5.0.zip, which is only Sept 2011 released, and the MTD library shipped with this 
still contains that bug.

Can you tell us the PR# for this issue and what the status is, please. We should much prefer to use an unmodified MTD 
library from QSS.

Can the MTD library accessed separately anywhere?

Thanks,
Albrecht

Return