Skip to contentSkip to conversation

Problem with cfi_probe.c and Intel chip

DISCUSSION Alice Hennessy 4 participants 8 messages Jan 10, 2002 – Jan 11, 2002 Memory technology devices (MTD)

lore
Share & tools
lore
#0 lore raw

Hello,

I just tried the latest code and discovered that the command used before the query command and also to
return to read mode in cfi_probe.c has been changed to 0xF0 - it used to be 0xFF . The board I'm testing
on has an Intel strata chip and is not responding to the query. If I use 0xFF, then everything is fine. AMD chips
seems to be happy with either command (I think because the AMD automatically returns to read mode after
a command anyway). It looks like this was changed with the jedec-probe

additions so maybe this is necessary
for older jedec chips (which I haven't tested).

Alice

#1 lore raw
Quoted 6 lines · [email protected] said:

I just tried the latest code and discovered that the command used before the query command and also to return to read mode in cfi_probe.c has been changed to 0xF0 - it used to be 0xFF . The board I'm testing on has an Intel strata chip and is not responding to the query. If I use 0xFF, then everything is fine. AMD chips seems to be happy with either command (

I seem to recall someone complaining before, and having to change it. But if so, I don't see why it only came in with the jedec-probe stuff.

Three options:
1. Change it to 0xFF and see if anyone screams.
2. Make it send both 0xFF and 0xF0.
3. Make it work out what type of chip it's talking to and send the _right_ command.

--
dwmw2

#7 lore raw
In reply to Alice Hennessy
Quoted 12 lines · Alice Hennessy

Hello,

I just tried the latest code and discovered that the command used before the query command and also to
return to read mode in cfi_probe.c has been changed to 0xF0 - it used to be 0xFF . The board I'm testing
on has an Intel strata chip and is not responding to the query. If I use 0xFF, then everything is fine. AMD chips
seems to be happy with either command (I think because the AMD automatically returns to read mode after
a command anyway). It looks like this was changed with the jedec-probe

Quoted 2 lines

additions so maybe this is necessary
for older jedec chips (which I haven't tested).

In this case it works for all of intel chips I have, so I thought the extra 0xFF in the reset, was put in by someone being extra careful, and not someone having problems...

I'll reenable it in just a minute.

Eric

#2 lore raw
In reply to David Woodhouse
Quoted 10 lines · David Woodhouse
Quoted 6 lines · [email protected] said:

I just tried the latest code and discovered that the command used before the query command and also to return to read mode in cfi_probe.c has been changed to 0xF0 - it used to be 0xFF . The board I'm testing on has an Intel strata chip and is not responding to the query. If I use 0xFF, then everything is fine. AMD chips seems to be happy with either command (

I seem to recall someone complaining before, and having to change it. But if so, I don't see why it only came in with the jedec-probe stuff.

O.k. I just did a skim through the JEDEC 21c section 3.5 assuming most of the time flash designers would follow a variant of the standard. And at various points both 0xF0 and 0xFF are mentioned as a reset, with perhaps a little preference towards 0xFF.

The probe algorithm is pretty much reset the chip, and then read it's id. So there isn't a lot that can be simplified there.

We are pretty much probing for the jedec 1 byte command set (intel) or the jedec multibyte command set (amd). With respect to the single byte command set, we are already sending the nonsense byte multibyte prefix and expecting it to ignore it. So expecting the multibyte command set to ignore the reset for the single byte command set is reasonable.

If we run into further problems we can detangle the single byte and the multibyte command set probes.

Quoted 2 lines

Three options:
1. Change it to 0xFF and see if anyone screams.

Well I just did remove 0xFF and see if anyone screams, I should have looked closer but I thought it was an intel specific then, and nothing I have required it. Though I admit 0xFF was not there originally.

Quoted 1 line

2. Make it send both 0xFF and 0xF0.

So far I haven't heard of any breakage with this combination. And now that I have double checked and discovered in some weird fashion that they are both equally valid, it is probably safe to leave them both in. At least until we hear of further breakage.

Quoted 2 lines

3. Make it work out what type of chip it's talking to and send the _right_ command.

This suffers from extreme catch22.

Eric

#3 lore raw
Quoted 47 lines · Eric W. Biederman
Quoted 10 lines · David Woodhouse
Quoted 6 lines · [email protected] said:

I just tried the latest code and discovered that the command used before the query command and also to return to read mode in cfi_probe.c has been changed to 0xF0 - it used to be 0xFF . The board I'm testing on has an Intel strata chip and is not responding to the query. If I use 0xFF, then everything is fine. AMD chips seems to be happy with either command (

I seem to recall someone complaining before, and having to change it. But if so, I don't see why it only came in with the jedec-probe stuff.

O.k. I just did a skim through the JEDEC 21c section 3.5 assuming most of the time flash designers would follow a variant of the standard. And at various points both 0xF0 and 0xFF are mentioned as a reset, with perhaps a little preference towards 0xFF.

The probe algorithm is pretty much reset the chip, and then read it's id. So there isn't a lot that can be simplified there.

We are pretty much probing for the jedec 1 byte command set (intel) or the jedec multibyte command set (amd). With respect to the single byte command set, we are already sending the nonsense byte multibyte prefix and expecting it to ignore it. So expecting the multibyte command set to ignore the reset for the single byte command set is reasonable.

If we run into further problems we can detangle the single byte and the multibyte command set probes.

Quoted 2 lines

Three options:
1. Change it to 0xFF and see if anyone screams.

Well I just did remove 0xFF and see if anyone screams, I should have looked closer but I thought it was an intel specific then, and nothing I have required it. Though I admit 0xFF was not there originally.

Quoted 1 line

2. Make it send both 0xFF and 0xF0.

So far I haven't heard of any breakage with this combination. And now that I have double checked and discovered in some weird fashion that they are both equally valid, it is probably safe to leave them both in. At least until we hear of further breakage.

Quoted 2 lines

3. Make it work out what type of chip it's talking to and send the _right_ command.

This suffers from extreme catch22.

Quoted 7 lines

Eric

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

From reading the Intel spec, I don't see any reason putting the chip into read

array
mode (0xff) would be necessary before the query command so I'm not sure why my board is having problems. Even if the bios has issued a command to the flash
between power up and cfi_probe, I don't think read array command is necessary between commands.
If my board is the only Intel with a problem, then I need to investigate it further.

Alice

#6 lore raw
In reply to David Woodhouse
Quoted 9 lines · David Woodhouse
Quoted 6 lines · [email protected] said:

I just tried the latest code and discovered that the command used before the query command and also to return to read mode in cfi_probe.c has been changed to 0xF0 - it used to be 0xFF . The board I'm testing on has an Intel strata chip and is not responding to the query. If I use 0xFF, then everything is fine. AMD chips seems to be happy with either command (

Quoted 5 lines

I seem to recall someone complaining before, and having to change it. But if so, I don't see why it only came in with the jedec-probe stuff.

Three options:
1. Change it to 0xFF and see if anyone screams.

I will! Toshiba 8MiB (TC58FVT641FT) won't leave autoselect mode if you change that. Then amd_flash_probe (and jedec_probe I suppose) will not be able to detect "mirrors" of previously found chips.

/Jonas

Quoted 14 lines

2. Make it send both 0xFF and 0xF0.
3. Make it work out what type of chip it's talking to and send the _right_ command.

--
dwmw2

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

#4 lore raw
In reply to Alice Hennessy
Quoted 8 lines · Alice Hennessy

array
mode (0xff) would be necessary before the query command so I'm not sure why my board is having problems. Even if the bios has issued a command to the flash
between power up and cfi_probe, I don't think read array command is necessary between commands.
If my board is the only Intel with a problem, then I need to investigate it further.

That would certainly be useful. My real suspicion is that this code hasn't been utilized all that much, and we are entering it's first real stabalization phase. Since all I did was comment out what looked like redundant code, I actually think you are the second person to notice
this. Personally I have deal with boot block type flash memory,
which may have different characteristics than other intel product lines.

Having a couple of people get a firm grasp on what things cause problems, is probably needed right now. So if you come up with anything interesting I'd be glad to hear it.

Eric

#5 lore raw
Quoted 18 lines · Eric W. Biederman
Quoted 8 lines · Alice Hennessy

array
mode (0xff) would be necessary before the query command so I'm not sure why my board is having problems. Even if the bios has issued a command to the flash
between power up and cfi_probe, I don't think read array command is necessary between commands.
If my board is the only Intel with a problem, then I need to investigate it further.

That would certainly be useful. My real suspicion is that this code hasn't been utilized all that much, and we are entering it's first real stabalization phase. Since all I did was comment out what looked like redundant code, I actually think you are the second person to notice
this. Personally I have deal with boot block type flash memory,
which may have different characteristics than other intel product lines.

Quoted 5 lines

Having a couple of people get a firm grasp on what things cause problems, is probably needed right now. So if you come up with anything interesting I'd be glad to hear it.

Eric

I will do that.

Alice