Re: [PATCH] mtd: rawnand: Do not check for bad block if bbt is unavailable

[Miquel Raynal <miquel.raynal@bootlin.com>]

Hi Manivannan,

Manivannan Sadhasivam <manivannan.sadhasivam@linaro.org> wrote on Tue,
2 Feb 2021 09:46:14 +0530:

> Hi,
> 
> On Mon, Feb 01, 2021 at 03:18:24PM +0100, Miquel Raynal wrote:
> > Hi Manivannan,
> > 
> > Manivannan Sadhasivam <manivannan.sadhasivam@linaro.org> wrote on Sat,
> > 30 Jan 2021 09:24:12 +0530:
> >   
> > > The bbt pointer will be unavailable when NAND_SKIP_BBTSCAN option is
> > > set for a NAND chip. The intention is to skip scanning for the bad
> > > blocks during boot time.  
> > 
> > I don't have the same understanding: this flag skips the bad block
> > table scan, not the bad block scan. We do want to scan all the devices
> > in order to construct a RAM based table.
> >   
> > > However, the MTD core will call
> > > _block_isreserved() and _block_isbad() callbacks unconditionally for
> > > the rawnand devices due to the callbacks always present while collecting
> > > the ecc stats.
> > > 
> > > The _block_isreserved() callback for rawnand will bail out if bbt
> > > pointer is not available. But _block_isbad() will continue without
> > > checking for it. So this contradicts with the NAND_SKIP_BBTSCAN option
> > > since the bad block check will happen anyways (ie., not much difference
> > > between scanning for bad blocks and checking each block for bad ones).
> > > 
> > > Hence, do not check for the bad block if bbt pointer is unavailable.  
> > 
> > Not checking for bad blocks at all feels insane. I don't really get the
> > scope and goal of such change?
> >   
> 
> The issue I encountered is, on the Telit FN980 device one of the
> partition seems to be protected. So trying to read the bad blocks in
> that partition makes the device to reboot during boot.

o_O

Reading a protected block makes the device to reboot?

What is the exact device? Can you share the datasheet? Is this behavior
expected? Because it seems really broken to me, a read should not
trigger *anything* that bad.

> There seems to be no flag passed by the parser for this partition. So
> the only way I could let the device to boot is to completely skip the
> bad block check.

We do have a "lock" property which informs the host to first unlock the
device, would this help? Is this locking reversible?

> AFAIK, MTD core only supports checking for the reserved blocks to be
> used for BBM and there is no way to check for a reserved partition like
> this.

It sounds like a chip specificity/bug, would it make sense to add a
specific vendor implementation for that?

> I agree that skipping bad block check is not a sane way but I don't know
> any other way to handle this problem.
> 
> Thanks,
> Mani
> 

Thanks,
Miquèl