Re: [PATCH v3 1/5] mtd: spi-nor: Fix gap in SR block protection locking

[<Tudor.Ambarus@xxxxxxxxxxxxx>]

On Monday, March 23, 2020 11:14:05 PM EET Michael Walle wrote:
> EXTERNAL EMAIL: Do not click links or open attachments unless you know the
> content is safe
> Am 2020-03-23 21:26, schrieb Tudor.Ambarus@xxxxxxxxxxxxx:
> > On Monday, March 23, 2020 9:54:38 PM EET Michael Walle wrote:
> >> EXTERNAL EMAIL: Do not click links or open attachments unless you know
> >> the
> >> content is safe
> >> 
> >> Am 2020-03-23 20:20, schrieb Tudor.Ambarus@xxxxxxxxxxxxx:
> >> > On Monday, March 23, 2020 8:27:13 PM EET Michael Walle wrote:
> >> >> EXTERNAL EMAIL: Do not click links or open attachments unless you know
> >> >> the
> >> >> content is safe
> >> >> 
> >> >> Hi,
> >> >> 
> >> >> Am 2020-03-23 10:24, schrieb Tudor.Ambarus@xxxxxxxxxxxxx:
> >> >> > From: Tudor Ambarus <tudor.ambarus@xxxxxxxxxxxxx>
> >> >> > 
> >> >> > Fix the gap for the SR block protection, the BP bits were set with
> >> >> > a +1 value than actually needed. This patch does not change the
> >> >> > behavior of the locking operations, just fixes the protected areas.
> >> >> 
> >> >> So instead of rounding up, it does round down now?
> >> > 
> >> > No. Why do you say that it rounds up? The behavior is not changed, the
> >> > patch
> >> > merely fix the protected area, which was wrong before. The round down
> >> > is
> >> > present before this patch.
> >> 
> >> TBH I don't understand what this patch should do. Could you give an
> >> example?
> > 
> > sure, let me try to be more explicit.
> > 
> >> >> > On a 16Mbit flash with 64KByte erase sector, the following changed
> >> 
> >> >> > for the lock operation:
> >> 16MBit is a bad example, because it is broken anyway, isn't it? We use
> >> a
> > 
> > it's not.
> 
> If I'm not mistaken this falls into the same category like the new 4bits
> BP
> flashes, because there are more slots free than needed. Ie. the last one
> "protect all" is either 110b or 111b and thus don't work with the old
> formula. This was actually my reason why there is no new formula for the
> 4 bits BP flashes; but the current one is not working with flashes
> <32Mbit.
> See the old long thread.
> 
> >> 32Mbit flash where 2MB are locked and the second 2MB are unlocked. Eg.
> >> a
> >> 50/50 split. I haven't seen any issued. Shouldn't it be then
> >> completely
> >> locked according this the following example?
> > 
> > I don't follow.
> 
> We've successfully used the "flash_lock 0 0x200" (with 4k sectors) on
> our
> boards to lock the first half of our 4MiB flash.
> 
> > The table from below was generated for the S25FL116K 16 Mbit flash.
> > BTW, one
> > has to disable CONFIG_MTD_SPI_NOR_USE_4K_SECTORS in order to test the
> > locking.
> > When you have a 4k sector erase, the locking is simply wrong, but this
> > is
> > another topic.
> 
> it should work with that too if you convert the number to the smaller
> sectors,
> ie multiply by 16; But yeah the cli tool has a broken interface. It
> should
> accept both offset and length in bytes; not one one in bytes and one in
> sectors,
> where the latter also changes with CONFIG_MTD_SPI_NOR_USE_4K_SECTORS.
> 
> >> >> > Number of blocks | BP2:0 before | BP2:0 now |
> >> >> > 
> >> >> >                1 | 010b         | 001b      |
> > 
> > - number of blocks is how many blocks you want to lock. One would do
> > for one
> > 
> > block:
> >     flash_lock /dev/mtd 0 1
> > 
> > i.e. lock a single erase block starting from offset 0.
> > 
> > - "BP0:2 before" is the result of the operation "flash_lock /dev/mtd 0
> > 1"
> > before this patch
> 
> Without your patch applied it works like expected:
> 
> [    1.914329] spi-nor spi0.0: w25q32dw (4096 Kbytes)
> # flash_lock -l /dev/mtd1 0 1
> # cat
> /sys/devices/platform/soc/20c0000.spi/spi_master/spi0/spi0.0/status_reg
> a4
> 
> A4 is 1010_0100, ie BP[2:0] = 001b and TB=1
> 

what happens if you request flash_lock -l /dev/mtd1 0 3?

> # flash_lock -u /dev/mtd1 0 64
> # flash_lock -l /dev/mtd1 0 32
> # cat
> /sys/devices/platform/soc/20c0000.spi/spi_master/spi0/spi0.0/status_reg
> b8
> 
> 
> With this patch applied:
> 
> # flash_lock -u /dev/mtd1 0 64
> # cat
> /sys/devices/platform/soc/20c0000.spi/spi_master/spi0/spi0.0/status_reg
> 00
> # flash_lock -l /dev/mtd1 0 1
> flash_lock: error!: could not lock device: /dev/mtd1
> 
>              error 22 (Invalid argument)

I'm wondering what was the reason for the -EINVAL.

> # flash_lock -l /dev/mtd1 0 2
> # cat
> /sys/devices/platform/soc/20c0000.spi/spi_master/spi0/spi0.0/status_reg
> a4
> 
> which is wrong, isn't it?
> 
Looks so. You should have obtained, 0xa8, right? Will recheck tomorrow 
morning.

Thanks for testing this! I don't have a 32Mbit flash ...

Cheers,
ta
> 
> > - "BP0:2 now" is the result of the operation "flash_lock /dev/mtd 0 1"
> > using
> > this patch
> > 
> > So before this patch, the lock operation was bad, because it locked 2
> > blocks
> > instead of one.
> > 
> >> >> >                2 | 110b         | 010b      |
> > 
> > - lock 2 erase blocks starting from offset 0. Results before this
> > patch, and
> > after this patch. Continue the logic on the following lines.
> > 
> > oops there's a typo in column 2, sorry. The value in column 2 should
> > have been
> > 011b.
> > 
> > So before this patch, when one requested to lock 2 block starting from
> > offset
> > 0, we would obtain 4 blocks locked, and he should have obtained just 2.
> > 
> > The scope of this patch is to first fix the locking ops, so that we can
> > introduce a more generic formula that gives the same results as before
> > introducing it. Without this patch, the new formula will silently fix
> > the bug
> > that is described here.
> > 
> >> >> >                3 | 110b         | 010b      |
> >               
> >               ^ typo s/110b/011b
> > 
> > rest of the examples are good.
> > 
> > Cheers,
> > ta
> > 
> >> >> >                4 | 100b         | 011b      |
> >> >> >                5 | 100b         | 011b      |
> >> >> >                6 | 100b         | 011b      |
> >> >> >                7 | 100b         | 011b      |
> >> >> >                8 | 101b         | 100b      |
> >> >> >                9 | 101b         | 100b      |
> >> >> >              
> >> >> >              ... | ...          | ...       |
> >> >> > 
> >> >> > For the lock operation, if one requests to lock an area that is not
> >> >> > matching the upper boundary of a BP protected area, we round down
> >> >> > the total length and lock less than the user requested, in order to
> >> >> > not lock more than the user actually requested.




______________________________________________________
Linux MTD discussion mailing list
http://lists.infradead.org/mailman/listinfo/linux-mtd/