PFS3 or PFS3 SCSI Direct

[fgh]

Allright, second attempt:

Device id string: 'Generic- SD/MMC 1.00'
BPS=512 Cyls=1961 TPC=255 SPT=63 Mediatype=11
Device size 16136536064 (0x3c1d00000) bytes

[Toni Wilen]

Originally Posted by fgh

Allright, second attempt:

Device id string: 'Generic- SD/MMC 1.00'
BPS=512 Cyls=1961 TPC=255 SPT=63 Mediatype=11
Device size 16136536064 (0x3c1d00000) bytes

Right, as expected, this is some Windows generated "geometry".

Do you get different values if connected to IDE port? What is exact name/model of the adapter? I need to get one too

[fgh]

I have this Multi-card USB adapter.
Edit: And this SD-IDE adapter

Give me a few minutes, I'll reboot and check the values over IDE..

[fgh]

Device id string: 'Memory Card Adapter 6781306'
BPS=512 Cyls=1961 TPC=255 SPT=63 Mediatype=11
Device size 16136257536 (0x3c1cbc000) bytes

278528 bytes smaller...

[fgh]

By the way - it's the a4000 version of scsi.device that uses LBA, according to your post from 2008..

So with stock scsi.device using LBA, how high should an a4000 be able to go using HD_SCSICMD?
Edit: I see Thomas did some tests in that thread with WinUAE. Should real HW be different in this case or not?

[Toni Wilen]

Originally Posted by fgh

By the way - it's the a4000 version of scsi.device that uses LBA, according to your post from 2008..

So with stock scsi.device using LBA, how high should an a4000 be able to go using HD_SCSICMD?
Edit: I see Thomas did some tests in that thread with WinUAE. Should real HW be different in this case or not?

I am quite sure it uses same method to detect disk size, afaik there wasn't any other methods in first ATA version.

Only (?) difference is use of LBA mode addressing when writing block address to IDE registers (LBA bit set = original CHS registers become linear LBA address registers)

Hardware (as in Amiga hardware) won't make any difference, only hardware difference is again drive's identity data. Basically PIO IDE bus itself is nothing more than address decoder + 16 bit data bus, all the logic is in drives.

But A4000 driver could still have some (possible accidental) differences, even more testing to do..

I also need to make WinUAE HDF CHS geometry fully configurable..

[fgh]

Ok, so it is still limited to the given CHS values then. Too bad
(28bit LBA was introduced in ATA-1. Word 60-61 specifies total amount of sectors, or zero if LBA is not supported)

[Toni Wilen]

I just tested and I guess I used different versions last time because actually LBA mode is used by both A1200 and A4000 drivers but only if new enough rom update is loaded (which also includes NSD support). Updated drivers also return full read capacity size.

Perhaps it is possible to create adapter device that automagically modifies identify device cylinders size value?

[fgh]

Ok, that makes sense in light of this comment from gregdonner.org:

Originally Posted by Olaf Barthel

OS revisions following 40.65 would contain only changes made to the SCSI (and the IDE) drivers.

KS3.1 for 500/600/2000 is v40.63 and 1200/4000 is v40.68.

Any luck figuring out how high you can go with >8GB CHS values?
Do you think this still works on a1200/a4000?

Jens could probably easily update his TrueIDE adapter firmware to do that.

And by the way, doesn't 16383/16/63 put the limit at 7.8745GB, not 7.7GB?

[Toni Wilen]

Originally Posted by fgh

Ok, that makes sense in light of this comment from gregdonner.org:

KS3.1 for 500/600/2000 is v40.63 and 1200/4000 is v40.68.

Yeah, it makes more sense. Perhaps I used some bad rom update that only contained A4000 updates..

Any luck figuring out how high you can go with >8GB CHS values?
Do you think this still works on a1200/a4000?

Not yet. Next few days I hope..

Jens could probably easily update his TrueIDE adapter firmware to do that.

It may not be that easy because new Cylinders value is not static but needs to be recalculated using true size from other identify device replies.

And by the way, doesn't 16383/16/63 put the limit at 7.8745GB, not 7.7GB?

Yes, I used 16383/15/63 for some reason..

btw, I am going to release "all-in-one" pfs3 soon now that read capacity "problem" is solved. Feature list (Did I miss something?)

- 68000+, full KS1.x/2.x/3.x compatibility.
- TD,TD64,NSD,DirectSCSI autodetection and 4G limit safety check.
- Automatic max transfer scsi.device workaround.

If end of partition is inside of first 128M: use standard 32-bit IO commands. (No useless autodetection when using floppies and other small media)
If end of partition is inside of first 4G: Test standard 32-bit IO, if it fails (for example pre-ROM 7.0 A590/A2091 or CDTV SCSI): test DirectSCSI. If both failed: show requester and don't mount the partition.
If end of partition is outside of first 4G: NSD, TD64 and DS. If all failed, silently don't mount the partition.

If dostype == PDS\x, do DirectSCSI test first.

Test succeeded = last block of partition must be accessible and last block read request must also modify data buffer (fill buffer with static data, read, if data didn't change, fill buffer with static data + 1, read, if data still didn't change = something is wrong, fail test)

DirectSCSI includes extra test: SCSI Read Capacity command must succeed and returned last block value must be same or larger than last block of partition.

[mfilos]

That seem really nice there Toni!!! /respect
Would a 030/040/060 compiled versions would have made any difference for the corresponded systems?

[Toni Wilen]

Any luck figuring out how high you can go with >8GB CHS values?

Yes and result is surprising.

C=65535 H=15 S=255 is the max possible = ~120G! And it does work!

In emulation IDE registers were updated with correct CHS value and exactly correct block in 119G region was read when I tested it.

Cylinders value is WORD in driver (65535 max), Sectors value is handled as a byte in driver (255), Heads is also byte but max is 15, because 15 is max value that fits in IDE heads register.

I guess adapter hack would only need to use static 65535/15/255 fake values because you use (should use!) OS3.9 or UAE to partition it and both ignore CHS values and use LBA values = new enough hdtoolbox shows correct size, not possibly too large ~120G size.

EDIT: It is possible that (some/all) real drives won't accept "bad" CHS values, this may only work with SD adapters that internally translate between two different protocols.

That seem really nice there Toni!!! /respect
Would a 030/040/060 compiled versions would have made any difference for the corresponded systems?

68020+ version that does not use any "slow" 68040+ instructions may make some sense. (But I have personally never seen any point whatsoever to have any other CPU optimized builds for routines that are not CPU critical. To anyone: I don't mind if you don't agree but don't argue in this thread, please)

[fgh]

Sounds like this is shaping up to be the definitive choice for an amiga file system - great work!
Nice that you can work around the maxtransfer issue as well

Can you test if real drives accept bad CHS values without actually building an adapter?
For most, patching scsi.device is as easy as buying a separate adapter, but Jens should at least be interested in faking the CHS values for the TrueIDE and ACA500 products if it's working reliably.

By the way - the SD adapter only supports SDHC (up to 32GB) so I assume FFFF/10/3F is the max you'll get out of that one.

[Toni Wilen]

Originally Posted by fgh

Can you test if real drives accept bad CHS values without actually building an adapter?

This requires some code that pokes IDE hardware directly (It needs to be run on Amiga hardware if we want valid results). Quite boring task.. Perhaps, perhaps not..

--

"Final" version attached (debugging disabled). Any name ideas? It is not just special ks1.x version anymore.. I have been thinking of "pfs3 all in one = pfs3 aio" but it is a bit stupid..

[prowler]

Originally Posted by Toni Wilen

"Final" version attached (debugging disabled). Any name ideas? It is not just special ks1.x version anymore.. I have been thinking of "pfs3 all in one = pfs3 aio" but it is a bit stupid..

Some suggestions to consider: pfs3 universal, pfs3 unified or pfs3 pure.

[Retrofan]

So if I'm using PFS3DS I can rename yours like that and update the existing file system, isn't it?

BTW thanks a lot

[fgh]

Great! What about PFS universal, PFS total, PFS one or PFS forever?

[zipper]

PFS 3d Max

[Bamiga2002]

Could I replace my original PFS 060 18.5-version with this? I use the normal non-DS version on my BPPC SCSI.

[fgh]

Retrofan and Bamiga: This version can replace any other version of PFS3.