Skip to contentSkip to conversation

MEMSTICK: Add driver for Ricoh R5C592 Card reader.

PATCH Alex Dubov 2 participants 35 messages Aug 5, 2010 – Aug 12, 2010

lore
Share & tools
lore

Alex Dubov

author
#0 lore raw
Quoted 13 lines
Quoted 4 lines · Maxim Levitsky
Quoted 3 lines · Alex Dubov

I see two immediate problems with this patch:

1. On cosmetic level, custom debug macros should

not be employed. Device

Quoted 1 line
Quoted 1 line

core already have this functionality (dynamic

debug levels and such). Please,

Quoted 5 lines
Quoted 1 line

use dev_dbg and friends for print-outs.

This allows much easier control for debug. Single module parameter is enough to adjust it. This helps me help users.
(Eg, kernel compilation is out of question)

I doubt it will be that useful, but it's not my call to make anyway.

Quoted 8 lines
Quoted 4 lines
Quoted 4 lines

2. On a structural level, I'd rather prefer host

drivers to not start their

Quoted 1 line
Quoted 1 line

own threads. If you look at both current host

implementations, they operate

Quoted 1 line
Quoted 1 line

in callback fashion. Apart from saving some

Quoted 11 lines

resources, this reduces the

Quoted 1 line
Quoted 1 line

amount of problems encountered during

suspend/resume and shutdown.

Quoted 2 lines

This isn't possible.
Hardware doesn't support interrupts on memstick bus

changes, it only

Quoted 1 line

supports DMA done from/to internal FIFO, and DMA it

only possible for

Quoted 2 lines

512 byte TPCs.

How depressing.

Quoted 10 lines

Another question.

I see that current code ignores MEMSTICK_CAP_AUTO_GET_INT Instread mspro_blk.c enables this capability for parallel mode, assuming
that hw supports it. Its true in my case, but might not be true in other
cases.
I think I should fix that, right?

This is mandated by the spec. INT should be available automatically in parallel mode, and some hardware does it in serial as well.

Quoted 8 lines

Also I see that you bath TPC_READ_LONG_DATA/TPC_READ_LONG_DATA
Does that mean that every HW sector is larger that 512? If so, you are doing copy on write, right?
I have small caching in my sm_ftl of last sector. It helps performance a
lot.

That's how its called in the spec.
Sectors can be larger than 512b on Pro-HG sticks, and there's additional TPC_READ/WRITE_QUAD_DATA which operates on larger quantities.

Quoted 10 lines

Also I want to clarify that the only kind of interrupts supported by hw
(besides usual card detection interrupt), is DMA done interrupt.
Thats why I have to use thread.
Doing polling in r592_submit_req (which runs in atomic context is just
cruel).

Yes, I see you have a timed wait there.

Quoted 9 lines

Besides, under moderate IO load, the IO thread doesn't sleep, thus there
is no overhead of wake/sleep.

Best regards,
Maxim Levitsky

#1 lore raw
Quoted 63 lines · Alex Dubov
Quoted 13 lines
Quoted 4 lines · Maxim Levitsky
Quoted 3 lines · Alex Dubov

I see two immediate problems with this patch:

1. On cosmetic level, custom debug macros should

not be employed. Device

Quoted 1 line
Quoted 1 line

core already have this functionality (dynamic

debug levels and such). Please,

Quoted 5 lines
Quoted 1 line

use dev_dbg and friends for print-outs.

This allows much easier control for debug. Single module parameter is enough to adjust it. This helps me help users.
(Eg, kernel compilation is out of question)

I doubt it will be that useful, but it's not my call to make anyway.

Quoted 8 lines
Quoted 4 lines
Quoted 4 lines

2. On a structural level, I'd rather prefer host

drivers to not start their

Quoted 1 line
Quoted 1 line

own threads. If you look at both current host

implementations, they operate

Quoted 1 line
Quoted 1 line

in callback fashion. Apart from saving some

Quoted 11 lines

resources, this reduces the

Quoted 1 line
Quoted 1 line

amount of problems encountered during

suspend/resume and shutdown.

Quoted 2 lines

This isn't possible.
Hardware doesn't support interrupts on memstick bus

changes, it only

Quoted 1 line

supports DMA done from/to internal FIFO, and DMA it

only possible for

Quoted 2 lines

512 byte TPCs.

How depressing.

Quoted 10 lines

Another question.

I see that current code ignores MEMSTICK_CAP_AUTO_GET_INT Instread mspro_blk.c enables this capability for parallel mode, assuming
that hw supports it. Its true in my case, but might not be true in other
cases.
I think I should fix that, right?

This is mandated by the spec. INT should be available automatically in parallel mode, and some hardware does it in serial as well.

Quoted 8 lines

Also I see that you bath TPC_READ_LONG_DATA/TPC_READ_LONG_DATA
Does that mean that every HW sector is larger that 512? If so, you are doing copy on write, right?
I have small caching in my sm_ftl of last sector. It helps performance a
lot.

Quoted 3 lines

That's how its called in the spec.
Sectors can be larger than 512b on Pro-HG sticks, and there's additional TPC_READ/WRITE_QUAD_DATA which operates on larger quantities.

But not on ordinary PRO, right?

Small question, can I use Pro-HG stick in my reader that is designed for Standard/PRO only? Does your subsystem support it?

8-bit mode really isn't supported here, but it is optional I am sure.

Quoted 12 lines
Quoted 11 lines

Also I want to clarify that the only kind of interrupts supported by hw
(besides usual card detection interrupt), is DMA done interrupt.
Thats why I have to use thread.
Doing polling in r592_submit_req (which runs in atomic context is just
cruel).

Quoted 2 lines

Yes, I see you have a timed wait there.

Alex, how should I proceed in merge of my driver?
Do you have any objections against it now?

Best regards,
Maxim Levitsky

author
#2 lore raw
Quoted 7 lines
Quoted 3 lines

That's how its called in the spec.
Sectors can be larger than 512b on Pro-HG sticks, and

there's additional

Quoted 1 line

TPC_READ/WRITE_QUAD_DATA which operates on larger

quantities.
But not on ordinary PRO, right?

Pro sectors are normally 512 bytes.

Quoted 4 lines

Small question, can I use Pro-HG stick in my reader that is designed for
Standard/PRO only? Does your subsystem support it?

It should work. It works for me on TI, which has no 8b mode either.

Quoted 5 lines
Quoted 1 line

Alex, how should I proceed in merge of my driver?
Do you have any objections against it now?

I may have done a thing or two differently, but otherwise no objection. I suppose, you should ask Andrew Morton to put it in.

#3 lore raw
Quoted 8 lines · Alex Dubov
Quoted 7 lines
Quoted 3 lines

That's how its called in the spec.
Sectors can be larger than 512b on Pro-HG sticks, and

there's additional

Quoted 1 line

TPC_READ/WRITE_QUAD_DATA which operates on larger

quantities.
But not on ordinary PRO, right?

Quoted 8 lines

Pro sectors are normally 512 bytes.

Quoted 4 lines

Small question, can I use Pro-HG stick in my reader that is designed for
Standard/PRO only? Does your subsystem support it?

It should work. It works for me on TI, which has no 8b mode either.

In that case, is there an upper limit on sector size?

Quoted 9 lines
Quoted 6 lines
Quoted 2 lines

Alex, how should I proceed in merge of my driver?
Do you have any objections against it now?

I may have done a thing or two differently, but otherwise no objection. I suppose, you should ask Andrew Morton to put it in.

Ok, will do.

Best regards,
Maxim Levitsky

#4 lore raw
Quoted 12 lines · Maxim Levitsky
Quoted 11 lines · Alex Dubov
Quoted 7 lines
Quoted 3 lines

That's how its called in the spec.
Sectors can be larger than 512b on Pro-HG sticks, and

there's additional

Quoted 1 line

TPC_READ/WRITE_QUAD_DATA which operates on larger

quantities.
But not on ordinary PRO, right?

Pro sectors are normally 512 bytes.

Quoted 1 line
Quoted 6 lines
Quoted 5 lines
Quoted 3 lines

Small question, can I use Pro-HG stick in my reader that is designed for
Standard/PRO only? Does your subsystem support it?

It should work. It works for me on TI, which has no 8b mode either.

In that case, is there an upper limit on sector size?

Also, you don't use the MS_TPC_READ_QUAD_DATA at all. So mspro_blk.c won't work with >512 bytes/sector disk? Or there is some compatibility?

Best regards,
Maxim Levitsky

author
#5 lore raw
Quoted 17 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines

That's how its called in the spec.
Sectors can be larger than 512b on

Pro-HG sticks, and

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

there's additional

Quoted 1 line

TPC_READ/WRITE_QUAD_DATA which operates

on larger

Quoted 7 lines
Quoted 7 lines
Quoted 2 lines

quantities.
But not on ordinary PRO, right?

Pro sectors are normally 512 bytes.

Quoted 2 lines

Small question, can I use Pro-HG stick in my

reader that is

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

designed for
Standard/PRO only? Does your subsystem

Quoted 9 lines

support it?

Quoted 2 lines
Quoted 2 lines

It should work. It works for me on TI, which has

no 8b mode either.

Quoted 1 line

In that case, is there an upper limit on sector size?

Also, you don't use the MS_TPC_READ_QUAD_DATA at all. So mspro_blk.c won't work with >512 bytes/sector disk? Or there is some compatibility?

The code can work with arbitrarily sized pages, 512b
value is not hard coded. All that is needed is to slightly alter h_mspro_block_transfer_data function, given, of course, that adapters support longer transfer sizes.

#6 lore raw
Quoted 27 lines · Alex Dubov
Quoted 17 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines

That's how its called in the spec.
Sectors can be larger than 512b on

Pro-HG sticks, and

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

there's additional

Quoted 1 line

TPC_READ/WRITE_QUAD_DATA which operates

on larger

Quoted 7 lines
Quoted 7 lines
Quoted 2 lines

quantities.
But not on ordinary PRO, right?

Pro sectors are normally 512 bytes.

Quoted 2 lines

Small question, can I use Pro-HG stick in my

reader that is

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

designed for
Standard/PRO only? Does your subsystem

Quoted 9 lines

support it?

Quoted 2 lines
Quoted 2 lines

It should work. It works for me on TI, which has

no 8b mode either.

Quoted 1 line

In that case, is there an upper limit on sector size?

Also, you don't use the MS_TPC_READ_QUAD_DATA at all. So mspro_blk.c won't work with >512 bytes/sector disk? Or there is some compatibility?

Quoted 4 lines

The code can work with arbitrarily sized pages, 512b
value is not hard coded. All that is needed is to slightly alter h_mspro_block_transfer_data function, given, of course, that adapters support longer transfer

No, I mean if I go and buy memstick PRO HG, that has > 512 bytes/sector, will it work with current code?

Btw, there is currently no way of telling core about maximum transfer length.
Here absolute maximum is 1024 (number of bit that _might_ hold TPC length), and FIFO size is 512 bytes (maybe its possible to use fifo twice)

Best regards,
Maxim Levitsky

author
#7 lore raw
Quoted 47 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

That's how its called in the

spec.

Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

Sectors can be larger than

512b on

Quoted 3 lines
Quoted 3 lines

Pro-HG sticks, and

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

there's additional

Quoted 1 line

TPC_READ/WRITE_QUAD_DATA

which operates

Quoted 8 lines
Quoted 8 lines

on larger

Quoted 7 lines
Quoted 7 lines
Quoted 2 lines

quantities.
But not on ordinary PRO, right?

Pro sectors are normally 512 bytes.

Quoted 2 lines

Small question, can I use Pro-HG

stick in my

Quoted 3 lines
Quoted 3 lines

reader that is

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

designed for
Standard/PRO only? Does your

subsystem

Quoted 3 lines
Quoted 3 lines

support it?

Quoted 2 lines
Quoted 2 lines

It should work. It works for me on TI,

which has

Quoted 2 lines
Quoted 2 lines

no 8b mode either.

Quoted 1 line

In that case, is there an upper limit on

sector size?

Quoted 1 line
Quoted 1 line

Also, you don't use the MS_TPC_READ_QUAD_DATA at

all.

Quoted 1 line
Quoted 1 line

So mspro_blk.c won't work with >512

bytes/sector disk?

Quoted 5 lines
Quoted 2 lines

Or there is some compatibility?

The code can work with arbitrarily sized pages, 512b value is not hard coded. All that is needed is to

slightly alter

Quoted 1 line

h_mspro_block_transfer_data function, given, of

course, that adapters

Quoted 1 line

support longer transfer

No, I mean if I go and buy memstick PRO HG, that has > 512 bytes/sector,
will it work with current code?

Quoted 9 lines

Btw, there is currently no way of telling core about maximum transfer
length.
Here absolute maximum is 1024 (number of bit that _might_ hold TPC
length), and FIFO size is 512 bytes (maybe its possible to use fifo
twice)

On TI adapters, FIFO can be reused and DMA restarted. Jmicron adapters are funny beasts, but their team is keen to support MSpro-HG, so at some point it will be fully possible, though, probably, not with every version of the chipset.

#8 lore raw
Quoted 57 lines · Alex Dubov
Quoted 47 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

That's how its called in the

spec.

Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

Sectors can be larger than

512b on

Quoted 3 lines
Quoted 3 lines

Pro-HG sticks, and

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

there's additional

Quoted 1 line

TPC_READ/WRITE_QUAD_DATA

which operates

Quoted 8 lines
Quoted 8 lines

on larger

Quoted 7 lines
Quoted 7 lines
Quoted 2 lines

quantities.
But not on ordinary PRO, right?

Pro sectors are normally 512 bytes.

Quoted 2 lines

Small question, can I use Pro-HG

stick in my

Quoted 3 lines
Quoted 3 lines

reader that is

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

designed for
Standard/PRO only? Does your

subsystem

Quoted 3 lines
Quoted 3 lines

support it?

Quoted 2 lines
Quoted 2 lines

It should work. It works for me on TI,

which has

Quoted 2 lines
Quoted 2 lines

no 8b mode either.

Quoted 1 line

In that case, is there an upper limit on

sector size?

Quoted 1 line
Quoted 1 line

Also, you don't use the MS_TPC_READ_QUAD_DATA at

all.

Quoted 1 line
Quoted 1 line

So mspro_blk.c won't work with >512

bytes/sector disk?

Quoted 5 lines
Quoted 2 lines

Or there is some compatibility?

The code can work with arbitrarily sized pages, 512b value is not hard coded. All that is needed is to

slightly alter

Quoted 1 line

h_mspro_block_transfer_data function, given, of

course, that adapters

Quoted 1 line

support longer transfer

No, I mean if I go and buy memstick PRO HG, that has > 512 bytes/sector,
will it work with current code?

Quoted 9 lines

Btw, there is currently no way of telling core about maximum transfer
length.
Here absolute maximum is 1024 (number of bit that _might_ hold TPC
length), and FIFO size is 512 bytes (maybe its possible to use fifo
twice)

Quoted 5 lines

On TI adapters, FIFO can be reused and DMA restarted. Jmicron adapters are funny beasts, but their team is keen to support MSpro-HG, so at some point it will be fully possible, though, probably, not with every version of the chipset.

I still need an answer for whether HG sticks need MS_TPC_READ_QUAD_DATA to be used.

But anyway, I buy a HG stick and see how it works.

Best regards,
Maxim Levitsky

author
#9 lore raw
Quoted 85 lines · Maxim Levitsky

From: Maxim Levitsky <[email protected]>
Subject: Re: [PATCH 2/2] MEMSTICK: Add driver for Ricoh R5C592 Card reader.
To: "Alex Dubov" <[email protected]>
Cc: "LKML" <[email protected]>
Received: Saturday, 7 August, 2010, 8:58 AM

Quoted 2 lines · Alex Dubov
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

That's how its

called in the

Quoted 2 lines
Quoted 2 lines

spec.

Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

Sectors can be

larger than

Quoted 4 lines
Quoted 4 lines

512b on

Quoted 3 lines
Quoted 3 lines

Pro-HG sticks, and

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

there's additional

Quoted 1 line

TPC_READ/WRITE_QUAD_DATA

Quoted 4 lines
Quoted 4 lines

which operates

Quoted 3 lines
Quoted 3 lines

on larger

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

quantities.
But not on ordinary PRO,

right?

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

Pro sectors are normally 512

bytes.

Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines

Small question, can I

use Pro-HG

Quoted 4 lines
Quoted 4 lines

stick in my

Quoted 3 lines
Quoted 3 lines

reader that is

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

designed for
Standard/PRO only? Does

your

Quoted 4 lines
Quoted 4 lines

subsystem

Quoted 3 lines
Quoted 3 lines

support it?

Quoted 2 lines
Quoted 2 lines

It should work. It works for

me on TI,

Quoted 3 lines
Quoted 3 lines

which has

Quoted 2 lines
Quoted 2 lines

no 8b mode either.

Quoted 1 line

In that case, is there an upper

limit on

Quoted 2 lines
Quoted 2 lines

sector size?

Quoted 1 line
Quoted 1 line

Also, you don't use the

MS_TPC_READ_QUAD_DATA at

Quoted 7 lines
Quoted 7 lines

all.

Quoted 1 line
Quoted 1 line

So mspro_blk.c won't work with >512

bytes/sector disk?

Quoted 4 lines
Quoted 2 lines

Or there is some compatibility?

The code can work with arbitrarily sized

pages, 512b

Quoted 1 line
Quoted 1 line
Quoted 1 line

value is not hard coded. All that is needed

is to

Quoted 2 lines
Quoted 2 lines

slightly alter

Quoted 1 line

h_mspro_block_transfer_data function, given,

of

Quoted 4 lines
Quoted 4 lines

course, that adapters

Quoted 1 line

support longer transfer

No, I mean if I go and buy memstick PRO HG, that

has >

Quoted 4 lines
Quoted 4 lines

512 bytes/sector,
will it work with current code?

Btw, there is currently no way of telling core

about

Quoted 3 lines
Quoted 3 lines

maximum transfer
length.
Here absolute maximum is 1024 (number of bit that

_might_

Quoted 2 lines
Quoted 2 lines

hold TPC
length), and FIFO size is 512 bytes (maybe its

possible to

Quoted 5 lines
Quoted 3 lines

use fifo
twice)

On TI adapters, FIFO can be reused and DMA restarted.

Jmicron adapters are

Quoted 1 line

funny beasts, but their team is keen to support

MSpro-HG, so at some

Quoted 8 lines
Quoted 1 line

point it will be fully possible, though, probably, not

with every version

Quoted 2 lines

of the chipset.

I still need an answer for whether HG sticks need
MS_TPC_READ_QUAD_DATA
to be used.

No, they work fine with LONG_DATA, at least the ones I tried.

Quoted 2 lines

But anyway, I buy a HG stick and see how it works.

This is always a right thing to do.

Quoted 5 lines

Best regards,
Maxim Levitsky

author
#10 lore raw
In reply to Maxim Levitsky
Quoted 12 lines
Quoted 3 lines
Quoted 3 lines

Small question, can I use Pro-HG stick in my

reader that is

Quoted 2 lines
Quoted 2 lines

designed for
Standard/PRO only? Does your subsystem support

it?

Quoted 2 lines

It should work. It works for me on TI, which has no 8b

mode either.
In that case, is there an upper limit on sector size?

Good code has no artificially induced limits. Less than machine page.

#11 lore raw
In reply to Alex Dubov
Quoted 50 lines · Alex Dubov
Quoted 13 lines
Quoted 4 lines · Maxim Levitsky
Quoted 3 lines · Alex Dubov

I see two immediate problems with this patch:

1. On cosmetic level, custom debug macros should

not be employed. Device

Quoted 1 line
Quoted 1 line

core already have this functionality (dynamic

debug levels and such). Please,

Quoted 5 lines
Quoted 1 line

use dev_dbg and friends for print-outs.

This allows much easier control for debug. Single module parameter is enough to adjust it. This helps me help users.
(Eg, kernel compilation is out of question)

I doubt it will be that useful, but it's not my call to make anyway.

Quoted 8 lines
Quoted 4 lines
Quoted 4 lines

2. On a structural level, I'd rather prefer host

drivers to not start their

Quoted 1 line
Quoted 1 line

own threads. If you look at both current host

implementations, they operate

Quoted 1 line
Quoted 1 line

in callback fashion. Apart from saving some

Quoted 11 lines

resources, this reduces the

Quoted 1 line
Quoted 1 line

amount of problems encountered during

suspend/resume and shutdown.

Quoted 2 lines

This isn't possible.
Hardware doesn't support interrupts on memstick bus

changes, it only

Quoted 1 line

supports DMA done from/to internal FIFO, and DMA it

only possible for

Quoted 2 lines

512 byte TPCs.

How depressing.

Quoted 10 lines

Another question.

I see that current code ignores MEMSTICK_CAP_AUTO_GET_INT Instread mspro_blk.c enables this capability for parallel mode, assuming
that hw supports it. Its true in my case, but might not be true in other
cases.
I think I should fix that, right?

Quoted 2 lines

This is mandated by the spec. INT should be available automatically in parallel mode, and some hardware does it in serial as well.

Thinking again about that I agree with you.
Then I can remove 2 more lines from my driver.

Btw, I want to add a callback from driver to card driver to be able to reset card in case of error (windows driver does that in case of any error)

I would use most of the mspro_block_resume for the implementation for mspro.

Any objections, suggestions?

Quoted 26 lines
Quoted 9 lines

Also I see that you bath TPC_READ_LONG_DATA/TPC_READ_LONG_DATA
Does that mean that every HW sector is larger that 512? If so, you are doing copy on write, right?
I have small caching in my sm_ftl of last sector. It helps performance a
lot.

That's how its called in the spec.
Sectors can be larger than 512b on Pro-HG sticks, and there's additional TPC_READ/WRITE_QUAD_DATA which operates on larger quantities.

Quoted 10 lines

Also I want to clarify that the only kind of interrupts supported by hw
(besides usual card detection interrupt), is DMA done interrupt.
Thats why I have to use thread.
Doing polling in r592_submit_req (which runs in atomic context is just
cruel).

Quoted 11 lines

Yes, I see you have a timed wait there.

Quoted 8 lines

Besides, under moderate IO load, the IO thread doesn't sleep, thus there
is no overhead of wake/sleep.

Best regards,
Maxim Levitsky

Best regards,
Maxim Levitsky

author
#12 lore raw
Quoted 7 lines

Btw, I want to add a callback from driver to card driver to be able to
reset card in case of error (windows driver does that in case of any
error)

Quoted 6 lines

I would use most of the mspro_block_resume for the implementation for
mspro.

Any objections, suggestions?

Just toggle a power on it. Power off/power on will do the full reset of the controller and the media. You don't have to reinitialize it, as you are sure that it's the same stick.

#13 lore raw
Quoted 11 lines · Alex Dubov
Quoted 11 lines

Btw, I want to add a callback from driver to card driver to be able to
reset card in case of error (windows driver does that in case of any
error)

I would use most of the mspro_block_resume for the implementation for
mspro.

Quoted 6 lines
Quoted 2 lines

Any objections, suggestions?

Just toggle a power on it. Power off/power on will do the full reset of the controller and the media. You don't have to reinitialize it, as you are sure that it's the same stick.

Yea, but after such reboot, the device will be in serial mode. So, I will need to send switch TPC to device. My driver doesn't know how to do that...., so it would be nice to have callback to the core.

Best regards,
Maxim Levitsky

author
#14 lore raw
Quoted 21 lines
Quoted 2 lines
Quoted 2 lines

Btw, I want to add a callback from driver to card

driver to

Quoted 2 lines
Quoted 2 lines

be able to
reset card in case of error (windows driver does

that in

Quoted 4 lines
Quoted 4 lines

case of any
error)

I would use most of the mspro_block_resume for

the

Quoted 7 lines
Quoted 5 lines

implementation for
mspro.

Any objections, suggestions?

Just toggle a power on it. Power off/power on will do

the full reset of

Quoted 1 line

the controller and the media. You don't have to

reinitialize it, as you

Quoted 9 lines
Quoted 1 line

are sure that it's the same stick.

Yea, but after such reboot, the device will be in serial mode. So, I
will need to send switch TPC to device. My driver doesn't know how to do
that...., so it would be nice to have callback to the core.

Lets ask a different question: why do you think this particular functionality is needed at all? Have you encountered any problems which require it (I haven't, btw)?

Windows drivers do a lot of things, but it does not mean all of them must be copied literally.

#15 lore raw
Quoted 31 lines · Alex Dubov
Quoted 21 lines
Quoted 2 lines
Quoted 2 lines

Btw, I want to add a callback from driver to card

driver to

Quoted 2 lines
Quoted 2 lines

be able to
reset card in case of error (windows driver does

that in

Quoted 4 lines
Quoted 4 lines

case of any
error)

I would use most of the mspro_block_resume for

the

Quoted 7 lines
Quoted 5 lines

implementation for
mspro.

Any objections, suggestions?

Just toggle a power on it. Power off/power on will do

the full reset of

Quoted 1 line

the controller and the media. You don't have to

reinitialize it, as you

Quoted 9 lines
Quoted 1 line

are sure that it's the same stick.

Yea, but after such reboot, the device will be in serial mode. So, I
will need to send switch TPC to device. My driver doesn't know how to do
that...., so it would be nice to have callback to the core.

Quoted 3 lines

Lets ask a different question: why do you think this particular functionality is needed at all? Have you encountered any problems which require it (I haven't, btw)?

I would image using this to verify that stick is still in place on I/O error.

If I just power down/up it, it will loose state, for example IO mode, reg range, etc...

But anyway, this isn't very important.

Best regards,
Maxim Levitsky

author
#16 lore raw
Quoted 48 lines · Maxim Levitsky
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

Btw, I want to add a callback from

driver to card

Quoted 3 lines
Quoted 3 lines

driver to

Quoted 2 lines
Quoted 2 lines

be able to
reset card in case of error (windows

driver does

Quoted 5 lines
Quoted 5 lines

that in

Quoted 4 lines
Quoted 4 lines

case of any
error)

I would use most of the

mspro_block_resume for

Quoted 8 lines
Quoted 8 lines

the

Quoted 7 lines
Quoted 5 lines

implementation for
mspro.

Any objections, suggestions?

Just toggle a power on it. Power off/power

on will do

Quoted 2 lines
Quoted 2 lines

the full reset of

Quoted 1 line

the controller and the media. You don't have

to

Quoted 4 lines
Quoted 4 lines

reinitialize it, as you

Quoted 1 line

are sure that it's the same stick.

Yea, but after such reboot, the device will be in

serial

Quoted 2 lines
Quoted 2 lines

mode. So, I
will need to send switch TPC to device. My driver

doesn't

Quoted 2 lines
Quoted 2 lines

know how to do
that...., so it would be nice to have callback to

the

Quoted 5 lines
Quoted 2 lines

core.

Lets ask a different question: why do you think this particular functionality is needed at all? Have you

encountered any

Quoted 1 line

problems which require it (I haven't, btw)?

I would image using this to verify that stick is still in place on I/O
error.

Quoted 6 lines

If I just power  down/up it, it will loose state, for example IO mode,
reg range, etc...

But anyway, this isn't very important.

My point was, that problems are better solved as they come. Windows drivers do a lot of things (like SCSI device emulation), but it doesn't mean all of them are necessary. Adding code "just because" often means more work later on.

#17 lore raw
In reply to Alex Dubov

Hi,

I have few more questions about memsticks.

First of all, I need an explanation of overwrite flag. You already explained it to me once, but I still not sure about few things.

#define MEMSTICK_OVERWRITE_UDST 0x10
This one I understand, thinking about xD again, I think it is very handy.

My idea (from xD of course) is that copyonwrite is done this way:

1. read old sector
2. allocate new sector
2. write what was just read to new sector.
3. erase old sector.

Could you explain when I need to set and reset the MEMSTICK_OVERWRITE_UDST?

#define MEMSTICK_OVERWRITE_PGST1 0x20
#define MEMSTICK_OVERWRITE_PGST0 0x40
I suppose these indicate that page(sector) contains incorrect data, just like in xD there is page status?
Again, better explanation is welcome.
Also, should I touch that flag when I update sector?

#define MEMSTICK_OVERWRITE_BKST 0x80
This marks bad blocks?

Another question is about write of oob data.
When I write it, overwrite flag is updated, or I need to use MEMSTICK_CP_OVERWRITE to update it?
I think former is true.

When I write a sector, I just write 0 to management flag, right?

And last question,
If I use MEMSTICK_CP_BLOCK, can I start reading a block from non-zero page offset?

And surely last question, what is 'MS_CMD_BLOCK_END'

Thanks again for all help so far,

Best regards,
Maxim Levitsky

author
#18 lore raw
Quoted 8 lines

Hi,

I have few more questions about memsticks.

First of all, I need an explanation of overwrite flag. You already explained it to me once, but I still not sure about few
things.

Overwrite register can be accessed either as part of extra data access or separately (CP_OVERWRITE access mode).

Quoted 6 lines

#define MEMSTICK_OVERWRITE_UDST  0x10
This one I understand, thinking about xD again, I think it is very
handy.

Quoted 7 lines

My idea (from xD of course) is that copyonwrite is done this way:

1. read old sector
2. allocate new sector
2. write what was just read to new sector.
3. erase old sector.

This is correct.

Quoted 3 lines

Could you explain when I need to set and reset the MEMSTICK_OVERWRITE_UDST?

UDST flag should be set when you're marking the block for reallocation during the read/modify/write cycle. You read the existing physical block, mark it with UDST flag (setting it to zero), then write different physical block on behalf of the same logical one, then erase the original block. The UDST flag is supposed to guard against a situation, whereupon power fails during the write cycle and you're left with two physical blocks mapped to the same logical one (so the one marked with zero UDST value is supposedly "known good").

Quoted 12 lines

#define MEMSTICK_OVERWRITE_PGST1 0x20 #define MEMSTICK_OVERWRITE_PGST0 0x40
I suppose these indicate that page(sector) contains incorrect data, just
like in xD there is page status?
Again, better explanation is welcome.
Also, should I touch that flag when I update sector?

Quoted 2 lines

#define MEMSTICK_OVERWRITE_BKST  0x80
This marks bad blocks?

BKST set to zero indicates that the whole block is bad and shouldn't be used.

PGST1:0 has several values:
11: default, r/w page
10: reserved value, shouldn't be used
01: page is read-only (soft write-protect)
00: page is accessible, but the value is not guaranteed (faulty page that sort-of works)

That's what the spec says.

Quoted 8 lines

Another question is about write of oob data. When I write it, overwrite flag is updated, or I need to use
MEMSTICK_CP_OVERWRITE to update it?
I think former is true.

As I mentioned above, it can be accessed either as part of extra data or separately.

Quoted 3 lines

When I write a sector, I just write 0 to management flag, right?

You shouldn't touch management_flag at all, as far as I can tell. It's only used to indicate special purpose blocks, such as factory written boot blocks, volatile look-up table blocks (for systems with tight RAM requirements) and DRM marked blocks which I has no info about.

Quoted 6 lines

And last question,
If I use MEMSTICK_CP_BLOCK, can I start reading a block from non-zero
page offset?

Yes, it starts from the user specified page address and auto increments it until the current block end is hit.

Quoted 3 lines

And surely last question, what is 'MS_CMD_BLOCK_END'

This command is used to terminate the currently ongoing block operation. If you are using one of the auto-increment modes (with CP_BLOCK set) but do not want to access all the pages until the block end, you must issue this command after the desired number of pages is transferred to return the media's state machine to the initial state. This command never hurts, as you can guess.

Quoted 4 lines

Thanks again for all help so far,

You're welcome.

#19 lore raw
Quoted 61 lines · Alex Dubov
Quoted 8 lines

Hi,

I have few more questions about memsticks.

First of all, I need an explanation of overwrite flag. You already explained it to me once, but I still not sure about few
things.

Overwrite register can be accessed either as part of extra data access or separately (CP_OVERWRITE access mode).

Quoted 6 lines

#define MEMSTICK_OVERWRITE_UDST 0x10
This one I understand, thinking about xD again, I think it is very
handy.

Quoted 7 lines

My idea (from xD of course) is that copyonwrite is done this way:

1. read old sector
2. allocate new sector
2. write what was just read to new sector.
3. erase old sector.

This is correct.

Quoted 3 lines

Could you explain when I need to set and reset the MEMSTICK_OVERWRITE_UDST?

UDST flag should be set when you're marking the block for reallocation during the read/modify/write cycle. You read the existing physical block, mark it with UDST flag (setting it to zero), then write different physical block on behalf of the same logical one, then erase the original block. The UDST flag is supposed to guard against a situation, whereupon power fails during the write cycle and you're left with two physical blocks mapped to the same logical one (so the one marked with zero UDST value is supposedly "known good").

Quoted 12 lines

#define MEMSTICK_OVERWRITE_PGST1 0x20 #define MEMSTICK_OVERWRITE_PGST0 0x40
I suppose these indicate that page(sector) contains incorrect data, just
like in xD there is page status?
Again, better explanation is welcome.
Also, should I touch that flag when I update sector?

Quoted 2 lines

#define MEMSTICK_OVERWRITE_BKST 0x80
This marks bad blocks?

BKST set to zero indicates that the whole block is bad and shouldn't be used.

Quoted 8 lines

PGST1:0 has several values:
11: default, r/w page
10: reserved value, shouldn't be used
01: page is read-only (soft write-protect)
00: page is accessible, but the value is not guaranteed (faulty page that sort-of works)

That's what the spec says.

Thank you very much.

Quoted 34 lines
Quoted 9 lines

Another question is about write of oob data. When I write it, overwrite flag is updated, or I need to use
MEMSTICK_CP_OVERWRITE to update it?
I think former is true.

As I mentioned above, it can be accessed either as part of extra data or separately.

Quoted 3 lines

When I write a sector, I just write 0 to management flag, right?

You shouldn't touch management_flag at all, as far as I can tell. It's only used to indicate special purpose blocks, such as factory written boot blocks, volatile look-up table blocks (for systems with tight RAM requirements) and DRM marked blocks which I has no info about.

Quoted 6 lines

And last question,
If I use MEMSTICK_CP_BLOCK, can I start reading a block from non-zero
page offset?

Yes, it starts from the user specified page address and auto increments it until the current block end is hit.

Quoted 2 lines
Quoted 8 lines
Quoted 1 line

And surely last question, what is 'MS_CMD_BLOCK_END'

This command is used to terminate the currently ongoing block operation. If you are using one of the auto-increment modes (with CP_BLOCK set) but do not want to access all the pages until the block end, you must issue this command after the desired number of pages is transferred to return the media's state machine to the initial state. This command never hurts, as you can guess.

That what I expected, thanks!

Quoted 7 lines
Quoted 5 lines

Thanks again for all help so far,

You're welcome.

Thank you very much!

Best regards,
Maxim Levitsky

#20 lore raw
Quoted 5 lines · Maxim Levitsky
Quoted 4 lines · Alex Dubov
Quoted 4 lines

Hi,

I have few more questions about memsticks.

Quoted 8 lines
Quoted 8 lines
Quoted 4 lines

First of all, I need an explanation of overwrite flag. You already explained it to me once, but I still not sure about few
things.

Overwrite register can be accessed either as part of extra data access or separately (CP_OVERWRITE access mode).

Few more questions I gathered today.

I currently assume that if I set req->need_card_int, the driver will wait till device raises MEMSTICK_INT_CED regardless of serial/parallel mode. This bit is always available on serial line.
Is that OK to assume?

Another thing that I want to ask is about writes to overwrite/managment flag.
Common sense tells me that I can write the flash many times, but write can only clear bits. Therefore if I write 0xFF, this just means do nothing.
Probably same applies to data content, but that isn't much of the use. Thats why I see that default (good) value of bits in overwrite flag is 1.
This is correct I assume?

Another interesting thing I observed was that MEMSTICK_INT_ERR can mean a correctable error, therefore it shouldn't alone reject read/write of a sector. (I think that it means that one of MEMSTICK_STATUS1_UCFG..MEMSTICK_STATUS1_DTER is set)
Same question about this.

Best regards,
Maxim Levitsky

author
#21 lore raw
Quoted 7 lines

I currently assume that if I set req->need_card_int, the driver will
wait till device raises MEMSTICK_INT_CED regardless of serial/parallel
mode. This bit is always available on serial line. Is that OK to assume?

I'm not quite sure about this question.
Normally, when you wait for the media interrupt, you want to see the whole value of the INT register (CED by itself doesn't indicate successful command completion; in fact it's value is undefined in presence of other INT flags, like BREQ or CMDNK). In parallel mode, host is required to fetch all meaningful INT bits from the media bus, while in serial mode this is only possible, if host supports automatic INT retrieval (the host will issue GET_INT tpc behind the scenes before alerting the software).

Quoted 11 lines

Another thing that I want to ask is about writes to overwrite/managment
flag.
Common sense tells me that I can write the flash many times, but write
can only clear bits. Therefore if I write 0xFF, this just means do
nothing.
Probably same applies to data content, but that isn't much of the use.

Yes, all memsticks are NAND flash, so writing 1s has no effect whatsoever.

Quoted 4 lines

Thats why I see that default (good) value of bits in overwrite flag is
1.
This is correct I assume?

Yes, a direct consequence of the above.

Quoted 10 lines

Another interesting thing I observed was that MEMSTICK_INT_ERR can mean
a correctable error,  therefore it shouldn't alone  reject read/write of
a sector. (I think that it means that one of MEMSTICK_STATUS1_UCFG..MEMSTICK_STATUS1_DTER is set) Same question about this.

There are three groups of error flags. While overwrite_flag register is accessible as part of extra data, being the indicator of block goodness it earned its own error flag:

DTER (UCDT) - error (uncorrectable) in data area EXER (UCEX) - error (uncorrectable) in extra data area FGER (UCFG) - error (uncorrectable) in overwrite_flag register

Uncorrectable error means that you've got some bit errors in the data you've obtained from the media. If uncorrectable flag is not set, it means that media's ECC circuit managed to correct the bit flip. The desired way of action is, of course, to reallocate the block before it develops an uncorrectable failure.

#22 lore raw
Quoted 8 lines · Alex Dubov
Quoted 7 lines

I currently assume that if I set req->need_card_int, the driver will
wait till device raises MEMSTICK_INT_CED regardless of serial/parallel
mode. This bit is always available on serial line. Is that OK to assume?

Quoted 8 lines

I'm not quite sure about this question.
Normally, when you wait for the media interrupt, you want to see the whole value of the INT register (CED by itself doesn't indicate successful command completion; in fact it's value is undefined in presence of other INT flags, like BREQ or CMDNK). In parallel mode, host is required to fetch all meaningful INT bits from the media bus, while in serial mode this is only possible, if host supports automatic INT retrieval (the host will issue GET_INT tpc behind the scenes before alerting the software).

Ok, maybe I didn't explain myself correctly.

Device is in serial mode.
I set need_card_int
I send a command.
<here I assume that driver waits for CED bit, which is exposed on data line, if CED won't be set, driver/hardware will timeout>
I send GET_INT _once_, look at flags. If I see CED, no NACK, then ok, otherwise I send error result.

Quoted 13 lines
Quoted 12 lines

Another thing that I want to ask is about writes to overwrite/managment
flag.
Common sense tells me that I can write the flash many times, but write
can only clear bits. Therefore if I write 0xFF, this just means do
nothing.
Probably same applies to data content, but that isn't much of the use.

Quoted 8 lines

Yes, all memsticks are NAND flash, so writing 1s has no effect whatsoever.

Quoted 4 lines

Thats why I see that default (good) value of bits in overwrite flag is
1.
This is correct I assume?

Yes, a direct consequence of the above.

I suspect that managment flag also has default value of 0xFF, and these 'special' features (drm, boot block, temporary table) clear bits out of it.

Quoted 20 lines
Quoted 11 lines

Another interesting thing I observed was that MEMSTICK_INT_ERR can mean
a correctable error, therefore it shouldn't alone reject read/write of
a sector. (I think that it means that one of MEMSTICK_STATUS1_UCFG..MEMSTICK_STATUS1_DTER is set) Same question about this.

There are three groups of error flags. While overwrite_flag register is accessible as part of extra data, being the indicator of block goodness it earned its own error flag:

DTER (UCDT) - error (uncorrectable) in data area EXER (UCEX) - error (uncorrectable) in extra data area FGER (UCFG) - error (uncorrectable) in overwrite_flag register

Quoted 5 lines

Uncorrectable error means that you've got some bit errors in the data you've obtained from the media. If uncorrectable flag is not set, it means that media's ECC circuit managed to correct the bit flip. The desired way of action is, of course, to reallocate the block before it develops an uncorrectable failure.

That I understand clearly.
I ask what the meaning of MEMSTICK_INT_ERR is.
I expect it to be set if any of (un) correctable errors are found. Therefore if MEMSTICK_INT_ERR, I can't report error instantly, but I need to look at status1 to see if error is correctable or not. This is correct?

Best regards,
Maxim Levitsky

#23 lore raw
In reply to Alex Dubov

I have another question.

Looking at ms_block.c, I see that it sometimes changes register window. This doesn't look good.
I see it does put the register window back, but still its a bit obscure.

I added tracking of current register window, so every time I send MS_TPC_SET_RW_REG_ADRS I note the ranges.
And read/write functions now always attempt to send MS_TPC_SET_RW_REG_ADRS. If the window is same as was set last time, TPC is skipped.

However, I am thinking, that maybe I should always write both param and extra register? I just write 0xFF to extra register and thats all. Windows driver does that partially. It writes 0xFF to managmemt and 0xF8 to overwrite flag (why???), but doesn't touch the LBA. I don't think that matters.
It also always sends the MS_TPC_SET_RW_REG_ADRS, which I don't like.

Best regards,
Maxim Levitsky

author
#24 lore raw
Quoted 8 lines

Received: Monday, 9 August, 2010, 8:30 AM
I have another question.

Looking at ms_block.c, I see that it sometimes changes register window.
This doesn't look good.
I see it does put the register window back, but still its a bit obscure.

It looks very good, in fact, it is the Sony specified way to operate the media. MS Pro works quite the same, it just needs fewer operations to actually access data.

Quoted 8 lines

I added tracking of current register window, so every time I send
MS_TPC_SET_RW_REG_ADRS I note the ranges.
And read/write functions now always attempt to send MS_TPC_SET_RW_REG_ADRS. If the window is same as was set last time, TPC
is skipped.

Sure it is. The media will remember the window set.
Media has all its registers in a sort of flat file. SET_RW_REG_ADDR selects the subset of the registers that will receive the data delivered within TPC. This subset is remembered until power off or until changed.

Quoted 5 lines

However, I am thinking, that maybe I should always write both param and
extra register? I just write 0xFF to extra register and thats all.

You should write into a param register when you want to alter the command parameters. You cannot do so during auto incrementing block access, for example.

But, if you're using the auto incrementing write, you will have to write extra register for every page transferred.

That's where changing RW_REG_ADDR comes handy.

Quoted 3 lines

Windows driver does that partially. It writes 0xFF to managmemt and
0xF8 to overwrite flag (why???)

It's a factory default.
Try to read it from some empty block. :-)
(My theory is that missing bits contain invisible ECC data).

Quoted 5 lines

I don't
think that matters.
It also always sends the MS_TPC_SET_RW_REG_ADRS, which I don't like.

This only reduces the performance slightly. SET_RW_REG_ADDR does not influence the media's state machine as far as I can tell, unless you try to push it during the data transfer cycle (whereupon you will end up having a literal value of the tpc in the media data buffer).

#25 lore raw
Quoted 22 lines · Alex Dubov
Quoted 8 lines

Received: Monday, 9 August, 2010, 8:30 AM
I have another question.

Looking at ms_block.c, I see that it sometimes changes register window.
This doesn't look good.
I see it does put the register window back, but still its a bit obscure.

It looks very good, in fact, it is the Sony specified way to operate the media. MS Pro works quite the same, it just needs fewer operations to actually access data.

Quoted 8 lines

I added tracking of current register window, so every time I send
MS_TPC_SET_RW_REG_ADRS I note the ranges.
And read/write functions now always attempt to send MS_TPC_SET_RW_REG_ADRS. If the window is same as was set last time, TPC
is skipped.

Quoted 4 lines

Sure it is. The media will remember the window set.
Media has all its registers in a sort of flat file. SET_RW_REG_ADDR selects the subset of the registers that will receive the data delivered within TPC. This subset is remembered until power off or until changed.

I know everything you have just said.
I just want to point out that code in many places assumes that register window is the same as set on device initialization.

However some code changes the register window, and therefore has to change it back at the end of execution.
If error happens, window can be left changed, while rest of code thinks it isn't changed.
Thats is why a tracking of it would eliminate the problem safely.

Quoted 8 lines
Quoted 7 lines

However, I am thinking, that maybe I should always write both param and
extra register? I just write 0xFF to extra register and thats all.

Quoted 6 lines

You should write into a param register when you want to alter the command parameters. You cannot do so during auto incrementing block access, for example.

But, if you're using the auto incrementing write, you will have to write extra register for every page transferred.

But what if I fill extra register with 0xFF?
And besides on reads, the fact that I *write* the extra register before I execute read command shouldn't matter at all regardless of what I write there.
On writes however I *do* need to write extra register anyway with proper values.

Therefore I see no reason why I can't set write window to cover both param and extra register, and leave it always like that.

Quoted 18 lines

That's where changing RW_REG_ADDR comes handy.

Quoted 3 lines

Windows driver does that partially. It writes 0xFF to managmemt and
0xF8 to overwrite flag (why???)

It's a factory default.
Try to read it from some empty block. :-)
(My theory is that missing bits contain invisible ECC data).

Quoted 5 lines

I don't
think that matters.
It also always sends the MS_TPC_SET_RW_REG_ADRS, which I don't like.

Quoted 4 lines

This only reduces the performance slightly. SET_RW_REG_ADDR does not influence the media's state machine as far as I can tell, unless you try to push it during the data transfer cycle (whereupon you will end up having a literal value of the tpc in the media data buffer).

Indeed. Maybe I too should just send the MS_TPC_SET_RW_REG_ADRS at start of command, and know that nothing will go south....

Even if that reduces performance by 0.2%, it isn't big deal. Data corruptions is very big deal.

Best regards,
Maxim Levitsky

author
#26 lore raw
Quoted 4 lines

I know everything you have just said.
I just want to point out that code in many places assumes that register
window is the same as set on device initialization.

It's not.
Can you please point at a particular place?

Quoted 13 lines
Quoted 1 line

But, if you're using the auto incrementing write, you

will have to write

Quoted 1 line

extra register for every page transferred.

But what if I fill extra register with 0xFF?
And besides on reads, the fact that I *write* the extra register before
I execute read command shouldn't matter at all regardless of what I
write there.
On writes however I *do* need to write extra register anyway with proper
values.

Quoted 3 lines

Therefore I see no reason why I can't set write window to cover both
param and extra register, and leave it always like that.

Because when you do autoincrementing write you _can not_ write into param register. You will break the command execution.

On the other thought, it may be unnecessary to write unique extra data to every page, so one full register write at the beginning of the command may do. Considering, that legacy memstick is not going to evolve, this may be reasonable assumption.

#27 lore raw
Quoted 21 lines · Alex Dubov
Quoted 4 lines

I know everything you have just said.
I just want to point out that code in many places assumes that register
window is the same as set on device initialization.

It's not.
Can you please point at a particular place?

Quoted 13 lines
Quoted 1 line

But, if you're using the auto incrementing write, you

will have to write

Quoted 1 line

extra register for every page transferred.

But what if I fill extra register with 0xFF?
And besides on reads, the fact that I *write* the extra register before
I execute read command shouldn't matter at all regardless of what I
write there.
On writes however I *do* need to write extra register anyway with proper
values.

Quoted 5 lines
Quoted 3 lines

Therefore I see no reason why I can't set write window to cover both
param and extra register, and leave it always like that.

Because when you do autoincrementing write you _can not_ write into param register. You will break the command execution.

Thanks, I finally understand you, so you are objection to write of _param_ register, not the _extra_.
I agree with you.

Quoted 2 lines

On the other thought, it may be unnecessary to write unique extra data to every page, so one full register write at the beginning of the command may do. Considering, that legacy memstick is not going to evolve, this may be reasonable assumption.

Indeed that what I do now.
Maximum I might need to clear page status bits, but I can do that later after I write the block.
This won't be any performance impact because amount of bad pages shouldn't be normally greater that zero.
(Otherwise there will be data loss...)

One interesting thing that I just want your opinion on is what to do with correctable errors.
Common sense suggests to relocate the sector + and mark it bad. But I don't know how common such sectors are, and thus I could do more harm that good by marking too many sectors as bad.

Of course all such problems are reason why today flash devices contain the FTL inside, and can improve/define it in the way they want.

No more questions for now, thank you very much for help.

I hope I create ms_block.c soon, and put that old problem to the rest.

As time permits I will also port your driver for xD portion of jMicron device (which I have).

Best regards,
Maxim Levitsky

author
#28 lore raw
Quoted 7 lines

Maximum I might need to clear page status bits, but I can do that later
after I write the block.
This won't be any performance impact because amount of bad pages
shouldn't be normally greater that zero. (Otherwise there will be data loss...)

These things do degrade. I think, memsticks do write-verify, so bad blocks will appear during write and can be marked as such without any data loss.

Quoted 9 lines

One interesting thing that I just want your opinion on is what to do
with correctable errors.
Common sense suggests to relocate the sector + and mark it bad.
But I don't know how common such sectors are, and thus I could do more
harm that good by marking too many sectors as bad.

I agree that number of writes to the media should be kept minimal. So bad (in either way) blocks encountered should be logged, but not touched, unless the error appears during an actual write/modify operation.

Quoted 3 lines

I hope I create ms_block.c soon, and put that old problem to the rest.

If you have time and desire, try to put a low level format option in - some function to erase all blocks, except the system ones.

Quoted 4 lines

As time permits I will also port your driver for xD portion of jMicron
device (which I have).

By the way, I've got some errata for the newer JMicron chipsets. If they have not contacted you yet, I'll forward it to you.

#29 lore raw
Quoted 34 lines · Alex Dubov
Quoted 7 lines

Maximum I might need to clear page status bits, but I can do that later
after I write the block.
This won't be any performance impact because amount of bad pages
shouldn't be normally greater that zero. (Otherwise there will be data loss...)

These things do degrade. I think, memsticks do write-verify, so bad blocks will appear during write and can be marked as such without any data loss.

Quoted 9 lines

One interesting thing that I just want your opinion on is what to do
with correctable errors.
Common sense suggests to relocate the sector + and mark it bad.
But I don't know how common such sectors are, and thus I could do more
harm that good by marking too many sectors as bad.

I agree that number of writes to the media should be kept minimal. So bad (in either way) blocks encountered should be logged, but not touched, unless the error appears during an actual write/modify operation.

Quoted 3 lines

I hope I create ms_block.c soon, and put that old problem to the rest.

If you have time and desire, try to put a low level format option in - some function to erase all blocks, except the system ones.

Quoted 1 line
Quoted 6 lines
Quoted 3 lines

As time permits I will also port your driver for xD portion of jMicron
device (which I have).

By the way, I've got some errata for the newer JMicron chipsets. If they have not contacted you yet, I'll forward it to you.

Yeah, I didn't done anything at that direction, but eventually I get to it.

Best regards,
Maxim Levitsky

author
#30 lore raw

JMicron chipset update

In reply to Maxim Levitsky

Apparently, the values I was using to configure a jmicron memstick interface only work with some chipset revisions. So they've sent me the following message.

If you've got a working jmicron adapter handy, you may want to try these changes.

-----------------
We found that there is a definition problem in your code about JMicron MS Controller.

    Our new product would get problem with this part.
    Following is the bit definition in our code.

    // --- Clock Control Register - Offset 0x48 --- //
    // D[31:4] Reserved
    // D[3]    Force MMIO Control.  0: Control by PCI CNFG, 1: Control
by MMIO.
    // D[2:0]  Clock MUX Select
    #define MSHC_CLKMUX_CONTROL_BY_MMIO   0x00000008
    #define MSHC_CLKMUX_CLK_40MHZ                  0x00000001
    #define MSHC_CLKMUX_CLK_50MHZ                  0x00000002
    #define MSHC_CLKMUX_CLK_62_5MHZ              0x00000004
    #define MSHC_CLKMUX_CLK_60MHZ                  0x00000010  // Must
set PCICnfg Offset BCh D[0] to 1
    #define MSHC_CLKMUX_CLK_OFF                      0x00000000
    #define MSHC_CLKMUX_CLK_MASK                   0x00000017

    Driver have to set this register to 0x08 to clear default clock
setting.
    And then set its value with specfic clock setting (EX:  40MHz ->

0x09, 50MHz -> 0x0A) to change clock.
(For MS Pro-HG, we suggest to use 50MHz with 8-bit parallel mode.)

Besides, driver have to set the register to 40MHz setting before
identify MS card for safe.

#31 lore raw
In reply to Alex Dubov

About INT bits, I still don't understand them exactly.

First of all we have 4 meaningful bits.
In parallel mode these are exposed on data lines, in serial mode only MEMSTICK_INT_CED is.
And I can always send MS_TPC_GET_INT or just read the registers.

#define MEMSTICK_INT_CMDNAK 0x01

#define MEMSTICK_INT_BREQ   0x20
#define MEMSTICK_INT_ERR    0x40
#define MEMSTICK_INT_CED    0x80

Now, I send a command to device, say MS_CMD_BLOCK_READ.
What bits I need to poll until I can be sure that command is completed?

Also the MEMSTICK_INT_BREQ tells that input is available in firmware buffer (to read using TPC_READ_LONG_DATA)?

Is that true that MEMSTICK_INT_BREQ is a summary of fifo full/empty bits in status0?

And same about MEMSTICK_INT_ERR and status1.

I try my best to create a driver that actually works, simple, and error free, even in unusual conditions.
Thats why I am asking all these questions.

Thanks for help,
Best regards,
Maxim Levitsky

author
#32 lore raw
Quoted 11 lines

About INT bits, I still don't
understand them exactly.

First of all we have 4 meaningful bits.
In parallel mode these are exposed on data lines, in serial mode only
MEMSTICK_INT_CED is.
And I can always send MS_TPC_GET_INT or just read the registers.

#define MEMSTICK_INT_CMDNAK 0x01

This bit means command was not acknowledged by the media (media not ready).

Quoted 1 line

#define MEMSTICK_INT_BREQ   0x20

This bit means media is ready for long data transfer (either in or out).

Quoted 1 line

#define MEMSTICK_INT_ERR    0x40

This bit means that some error had occurred during command execution.

Quoted 1 line

#define MEMSTICK_INT_CED    0x80

This bit marks a completion of the command, but it may switch value in between, so it is not that reliable by itself.

Quoted 5 lines

Now, I send a command to device, say MS_CMD_BLOCK_READ. What bits I need to poll until I can be sure that command is completed?

All of them.

Quoted 5 lines

Also the MEMSTICK_INT_BREQ tells that input is available in firmware
buffer (to read using TPC_READ_LONG_DATA)?

Not exactly. BREQ signal indicated that media's state machine is in the mode to pump data in or out. You wait for it, than you do the data transfer (it works the same with READ and WRITE).

Quoted 5 lines

Is that true that MEMSTICK_INT_BREQ is a summary of fifo full/empty bits
in status0?

This can not be relied upon. Sony states, that only BREQ bit must be used in block transfer operations. BE/BF bits are probably of more use in memstick IO cards (there exists such a beast).

Quoted 2 lines

And same about MEMSTICK_INT_ERR and status1.

status1 errors apply only to data errors (bit flips).
There are errors in command execution that do not result in bit flips, rather data can not be delivered at all or command/parameters are invalid.

Quoted 11 lines

I try my best to create a driver that actually works, simple, and error
free, even in unusual conditions.
Thats why I am asking all these questions.

Thanks for help,
Best regards,
Maxim Levitsky

#33 lore raw
Quoted 47 lines · Alex Dubov
Quoted 11 lines

About INT bits, I still don't
understand them exactly.

First of all we have 4 meaningful bits.
In parallel mode these are exposed on data lines, in serial mode only
MEMSTICK_INT_CED is.
And I can always send MS_TPC_GET_INT or just read the registers.

#define MEMSTICK_INT_CMDNAK 0x01

This bit means command was not acknowledged by the media (media not ready).

Quoted 1 line

#define MEMSTICK_INT_BREQ 0x20

This bit means media is ready for long data transfer (either in or out).

Quoted 1 line

#define MEMSTICK_INT_ERR 0x40

This bit means that some error had occurred during command execution.

Quoted 1 line

#define MEMSTICK_INT_CED 0x80

This bit marks a completion of the command, but it may switch value in between, so it is not that reliable by itself.

Quoted 5 lines

Now, I send a command to device, say MS_CMD_BLOCK_READ. What bits I need to poll until I can be sure that command is completed?

All of them.

Quoted 5 lines

Also the MEMSTICK_INT_BREQ tells that input is available in firmware
buffer (to read using TPC_READ_LONG_DATA)?

Not exactly. BREQ signal indicated that media's state machine is in the mode to pump data in or out. You wait for it, than you do the data transfer (it works the same with READ and WRITE).

Quoted 2 lines
Quoted 7 lines
Quoted 3 lines

Is that true that MEMSTICK_INT_BREQ is a summary of fifo full/empty bits
in status0?

This can not be relied upon. Sony states, that only BREQ bit must be used in block transfer operations. BE/BF bits are probably of more use in memstick IO cards (there exists such a beast).

Even better.

Quoted 7 lines
Quoted 3 lines

And same about MEMSTICK_INT_ERR and status1.

status1 errors apply only to data errors (bit flips).
There are errors in command execution that do not result in bit flips, rather data can not be delivered at all or command/parameters are invalid.

Understood.
However if I get MEMSTICK_INT_ERR, I should also look at status1, because there might be correctable error.

Thank you very much.

Best regards,
Maxim Levitsky

author
#34 lore raw
Quoted 10 lines

result in bit flips,

Quoted 1 line

rather data can not be delivered at all or

command/parameters are invalid.
Understood.
However if I get MEMSTICK_INT_ERR, I should also look at status1,
because there might be correctable error.

I don't remember the details, but it may be, that status1 errors do not raise INT_ERR (or do not always raise?). The idea is, first you check INT_ERR. If command completed successfully and you received the data, you may still need to check status1 for possible bit flips.

Some experimentation may be required, especially for the case of correctable errors. After all, they should not prevent the command from successful completion, yet action should be taken on them (at the very least - log warning, just like with normal disks).