Skip to contentSkip to conversation

Parallel ATA with libata status with the patches I'm working on

DISCUSSION Alan Cox 11 participants 36 messages Nov 3, 2005 – Nov 6, 2005

lore
Share & tools
lore

Alan Cox

author
#0 lore raw

Core Features Fixed
- Per drive tuning
- Filter quirk lists
- Single channel support

And To Add
- Specify PCI bus speed
- HPA
- IRQ mask
- PIO only LBA48
- Serialize
- CRC downspeed
- Mixed legacy/native mode (most work done)

Drivers so far written for the libata parallel work I'm doing

ALI
Driver written with equivalent support to the drivers/ide one. Needs core changes for LBA48 PIO only

AMD
Driver written, given basic testing and equivalent to current drivers/ide

CS5520
Driver written, some debug work to do. Works unlike the drivers/ide one

CS5530
Works single channel needs core changes for dual

CS5535
Written but untested

HPT34X
Driver written, functionality same as drivers/ide. Needs more testing.

HPT36x
Support done for chips up to 302. No 372N/302N or PLL support yet

IT821x
Written but needs further driver and core work yet

RZ1000
Written, simulation tested

SC1200
Written, needs core changes for dual channel

Serverworks
Written, equivalent functionality to drivers/ide plus some bugs fixed

SIL680
Written, tested, running.

Triflex
Written tested, running

VIA
Written, tested on some controllers. Does not yet support the 6410 (TODO item) and very early VIA (needs core changes)

'Generic'
Written, tested, needs polishing and some core changes ideally for irq mask and for simplex

ISA Legacy
Work in progress

---------------

Unsupported Hardware

ACPI
TODO item - it's possible to drive unknown motherboard chipsets through ACPI. Not supported by drivers/ide either

AEC62xx
ATIIXP
CMD64x
CY82C693
IT8172
NS87415
OPTI621
PDC202xx
SGI IOC4
SIS5513
SL82C105
SLC90E66
TRM290
PCMCIA (needs hotplug merge first)

#1 lore raw
Quoted 6 lines · Alan Cox

AEC62xx
ATIIXP
CMD64x
CY82C693
IT8172
NS87415

Quoted 8 lines

OPTI621
PDC202xx
SGI IOC4
SIS5513
SL82C105
SLC90E66
TRM290
PCMCIA (needs hotplug merge first)

+ icside ?

Signature · 3 lines
Russell King
 Linux kernel    2.6 ARM Linux   - http://www.arm.linux.org.uk/
 maintainer of:  2.6 Serial core
#23 lore raw
In reply to Alan Cox
Quoted 4 lines · Alan Cox

Core Features Fixed
- Per drive tuning
- Filter quirk lists
- Single channel support

Are these the same changes that have been recently pushed into Linus' tree without any previous public review or some new ones?

Quoted 10 lines

And To Add
- Specify PCI bus speed
- HPA
- IRQ mask
- PIO only LBA48
- Serialize
- CRC downspeed
- Mixed legacy/native mode (most work done)

Drivers so far written for the libata parallel work I'm doing

Are the patches available somewhere?

Quoted 7 lines

ALI
Driver written with equivalent support to the drivers/ide one. Needs core changes for LBA48 PIO only

AMD
Driver written, given basic testing and equivalent to current drivers/ide

Functionality can't be the same as drivers/ide because libata lacks some core features.

Quoted 2 lines

CS5520
Driver written, some debug work to do. Works unlike the drivers/ide one

Please fix drivers/ide also if not a big problem.

Quoted 2 lines

HPT34X
Driver written, functionality same as drivers/ide. Needs more testing.

Same comment as for AMD.

Quoted 2 lines

Serverworks
Written, equivalent functionality to drivers/ide plus some bugs fixed

Ditto. Also please backport fixes to drivers/ide.

Thanks,
Bartlomiej

#2 lore raw
In reply to Russell King
Quoted 16 lines · Russell King
Quoted 6 lines · Alan Cox

AEC62xx
ATIIXP
CMD64x
CY82C693
IT8172
NS87415

Quoted 8 lines

OPTI621
PDC202xx
SGI IOC4
SIS5513
SL82C105
SLC90E66
TRM290
PCMCIA (needs hotplug merge first)

Quoted 1 line

+ icside ?

pmac
rapide
bast-ide
ide_arm
ide-cris
ide-h8300
mpc8xx
legacy stuff (optional?)...

IMO porting/rewriting host-drivers to libata now is just counter-productive waste of time...

Bartlomiej

author
#24 lore raw
In reply to Bartlomiej Zolnierkiewicz
Quoted 8 lines · Bartlomiej Zolnierkiewicz
Quoted 4 lines · Alan Cox

Core Features Fixed
- Per drive tuning
- Filter quirk lists
- Single channel support

Are these the same changes that have been recently pushed into Linus' tree without any previous public review or some new ones?

Some of the changes have gone to Jeff Garzik others are pending improvments/driver fixes for existing drivers and testing before Jeff accepts them. Beyond that its up to Jeff what is merged into -mm and on.

Quoted 3 lines
Quoted 1 line

Drivers so far written for the libata parallel work I'm doing

Are the patches available somewhere?

Yes. I'll post updated sets at some point soon. The patch I posted pointers to before is now well out of date.

Quoted 6 lines
Quoted 3 lines

AMD
Driver written, given basic testing and equivalent to current drivers/ide

Functionality can't be the same as drivers/ide because libata lacks some core features.

Well duh, other than core differences. Functionality is indeed different
- CRC errors dont crash SMP boxes for example ;)

Quoted 4 lines
Quoted 2 lines

CS5520
Driver written, some debug work to do. Works unlike the drivers/ide one

Please fix drivers/ide also if not a big problem.

I started this work in part because it was impossible to work with you. I've no interest in fixing drivers/ide and in fact I'm now running as much as I can with CONFIG_IDE=n, because the only way to really test code is to use it all the time.

The things I've fixed however if you want to go fix them in drivers/ide are

CS5520 is totally broken. You want the fixes I sent you over a year ago and maybe more. Its probably best left as I doubt anyone else has a 5520 any more 8)

Serverworks runs the wrong cable detect routine in some cases (conditions ordered wrongly)

I think the HPT366 driver contains several bugs looking at it in comparison with hpts driver and the data sheets I can get hold of. It's hard to be sure however because at times all three disagree with each other. In particular it doesn't know about 302N but think its a 37x and the PLL tune code appears to be unrelated to either data sheet or hpt code. Can't be sure yet on the PLL until I've got PLL modes working in the libata driver.

Alan (drivers/ide elimination department)

author
#3 lore raw
In reply to Bartlomiej Zolnierkiewicz
Quoted 2 lines · Bartlomiej Zolnierkiewicz

IMO porting/rewriting host-drivers to libata now is just counter-productive waste of time...

Face it, drivers/ide/ is beyong saving, it has entered the software engineering twilight zone that follows 'maintainable'.

Shifting a non-complex driver to libata with a few libata improvements is not a difficult process. Figuring out the extra hooks and bits to make it clean is the difficult bit. Each time another PATA horror is supported by libata it becomes easy for other cards with that horror to be moved over.

Also drivers/ide contains detailed support for things that just aren't worth moving over - like SWDMA.

Once the core support is done then quite frankly I'll be able to port a PATA driver to libata faster than sherlock holmes could deduce the drivers/ide locking "rules" let alone fix them.

Alan

author
#17 lore raw
In reply to Russell King
Quoted 3 lines · Russell King

+ icside ?

I've not really looked much outside of the PCI space yet (my first goal is to rescue the PC world and to get testing it wants x86 users) but Jeffs core libata code is strictly bus agnostic.

Alan

#18 lore raw
Quoted 8 lines · Alan Cox
Quoted 3 lines · Russell King

+ icside ?

I've not really looked much outside of the PCI space yet (my first goal is to rescue the PC world and to get testing it wants x86 users) but Jeffs core libata code is strictly bus agnostic.

Ok, I look forward to fiddling with libata on ARM. 8)

You mentioned that libata doesn't do SWDMA - does it do MWDMA and PIO?

Signature · 3 lines
Russell King
 Linux kernel    2.6 ARM Linux   - http://www.arm.linux.org.uk/
 maintainer of:  2.6 Serial core
#4 lore raw
In reply to Bartlomiej Zolnierkiewicz

From: Bartlomiej Zolnierkiewicz <[email protected]>
Date: Thu, 3 Nov 2005 16:02:45 +0100

Quoted 2 lines

IMO porting/rewriting host-drivers to libata now is just counter-productive waste of time...

This would be an interesting opinion if you were asked for it. But you were not, yet you keep stating it, therefore you must feel threatened in some way.

I have to admit that your handling of Alan's desire to do this work makes you look very non-transparent as the IDE layer maintainer.

What do you care if Alan converts all the drivers with his own time and effort? If it's a waste of time, it's his time, which he can spend in any way he so chooses. You've already stated that you think it's a waste of time on at least 2 or 3 occaisions, so I think he and everyone else gets your point and you don't need to restate this over and over again.

#25 lore raw
In reply to Alan Cox
Quoted 7 lines · Alan Cox
Quoted 6 lines · Bartlomiej Zolnierkiewicz
Quoted 4 lines · Alan Cox

Core Features Fixed
- Per drive tuning
- Filter quirk lists
- Single channel support

Quoted 6 lines
Quoted 2 lines

Are these the same changes that have been recently pushed into Linus' tree without any previous public review or some new ones?

Some of the changes have gone to Jeff Garzik others are pending improvments/driver fixes for existing drivers and testing before Jeff accepts them. Beyond that its up to Jeff what is merged into -mm and on.

Of course it is up to Jeff but still it would be nice to see them on linux-ide ML before hitting -mm or mainline...

Quoted 17 lines
Quoted 3 lines
Quoted 1 line

Drivers so far written for the libata parallel work I'm doing

Are the patches available somewhere?

Yes. I'll post updated sets at some point soon. The patch I posted pointers to before is now well out of date.

Quoted 6 lines
Quoted 3 lines

AMD
Driver written, given basic testing and equivalent to current drivers/ide

Functionality can't be the same as drivers/ide because libata lacks some core features.

Well duh, other than core differences. Functionality is indeed different
- CRC errors dont crash SMP boxes for example ;)

Quoted 6 lines
Quoted 4 lines
Quoted 2 lines

CS5520
Driver written, some debug work to do. Works unlike the drivers/ide one

Please fix drivers/ide also if not a big problem.

I started this work in part because it was impossible to work with you.

Yes, I don't ack patches only because name of the submitter... ;)

Quoted 7 lines

I've no interest in fixing drivers/ide and in fact I'm now running as much as I can with CONFIG_IDE=n, because the only way to really test code is to use it all the time.

The things I've fixed however if you want to go fix them in drivers/ide are

Quoted 6 lines

CS5520 is totally broken. You want the fixes I sent you over a year ago and maybe more. Its probably best left as I doubt anyone else has a 5520 any more 8)

Serverworks runs the wrong cable detect routine in some cases (conditions ordered wrongly)

OK, thanks for the hint.

Quoted 10 lines

I think the HPT366 driver contains several bugs looking at it in comparison with hpts driver and the data sheets I can get hold of. It's hard to be sure however because at times all three disagree with each other. In particular it doesn't know about 302N but think its a 37x and the PLL tune code appears to be unrelated to either data sheet or hpt code. Can't be sure yet on the PLL until I've got PLL modes working in the libata driver.

Alan (drivers/ide elimination department)

I see... same goal here but in more peaceful way...

Bartlomiej

#5 lore raw
In reply to David S. Miller
Quoted 9 lines · David S. Miller

From: Bartlomiej Zolnierkiewicz <[email protected]>
Date: Thu, 3 Nov 2005 16:02:45 +0100

Quoted 2 lines

IMO porting/rewriting host-drivers to libata now is just counter-productive waste of time...

This would be an interesting opinion if you were asked for it. But you were not, yet you keep stating it, therefore you must feel threatened in some way.

I'm worried about crap being pushed into libata, that's all.

Besides does current situation of ALSA and OSS bring any bells?

Quoted 2 lines

I have to admit that your handling of Alan's desire to do this work makes you look very non-transparent as the IDE layer maintainer.

By reading linux-ide ML you will see that actually I help in development of libata PATA support.

Quoted 2 lines

What do you care if Alan converts all the drivers with his own time and effort? If it's a waste of time, it's his time, which he can

Are you familiar with background discussions before Alan started this work? Never explained FUD about current state of drivers/ide, personal attacks and replying to eny post relating to IDE with either FUD or "IDE driver is going away"...

Alan plays dirty game and don't say me that this doesn't influnce my work...

Quoted 4 lines

spend in any way he so chooses. You've already stated that you think it's a waste of time on at least 2 or 3 occaisions, so I think he and everyone else gets your point and you don't need to restate this over and over again.

OK, I'll shut up now.

Bartlomiej (no more mr nice guy)

author
#19 lore raw
In reply to Russell King
Quoted 3 lines · Russell King

Ok, I look forward to fiddling with libata on ARM. 8)

You mentioned that libata doesn't do SWDMA - does it do MWDMA and PIO?

Yes. MWDMA works well, PIO is pretty crap in the base -mm tree but much improved in the devel trees. I've generally been running stuff pio before adding the DMA code in fact.

author
#6 lore raw
In reply to Bartlomiej Zolnierkiewicz
Quoted 1 line · Bartlomiej Zolnierkiewicz

I'm worried about crap being pushed into libata, that's all.

I wouldn't be too worried. Jeff Garzik is quite fussy about what goes in and that it gets designed the right way. Various things aren't currently merged because Jeff wants them redone better - such as the ISA legacy support.

Alan

#7 lore raw
In reply to Bartlomiej Zolnierkiewicz
Quoted 2 lines · Bartlomiej Zolnierkiewicz

IMO porting/rewriting host-drivers to libata now is just counter-productive waste of time...

That would only make sense if you consider all PATA obsolete/dead (do you? I'm sometimes not sure).

I don't and (unable to use old IDE due to hot-plug issues) am thankful for Alan's efforts.

Yes, I think it's similar to OSS-ALSA, WRT - you know, 6-months forward notice etc :-)

Signature · 1 line
Krzysztof Halasa
#8 lore raw
Quoted 7 lines · Krzysztof Halasa
Quoted 2 lines · Bartlomiej Zolnierkiewicz

IMO porting/rewriting host-drivers to libata now is just counter-productive waste of time...

That would only make sense if you consider all PATA obsolete/dead (do you? I'm sometimes not sure).

I want to merge old IDE driver w/ libata... and drop remaining crap on the way. If libata gains full PATA support before I do this - it is even better for me...

Quoted 2 lines

I don't and (unable to use old IDE due to hot-plug issues) am thankful for Alan's efforts.

Do you think that libata is currently so much better wrt to PATA hot-(un)plug support?

If so than dream on...

Quoted 2 lines

Yes, I think it's similar to OSS-ALSA, WRT - you know, 6-months forward notice etc :-)

Ain't going to happen...

Guys, I'm not against libata PATA support - I'm all for it but I want TRANSPARENT development and FAIR look at current state of affairs (there is still a lot of stuff on libata's PATA TODO)...

Plus I don't like needless bashing of IDE driver which is still messy but orders of magnitude less than during 2.4.x days... :-)

Bartlomiej

author
#9 lore raw
Quoted 2 lines · Bartlomiej Zolnierkiewicz

Do you think that libata is currently so much better wrt to PATA hot-(un)plug support?

With the hot plug patches for SATA it certainly is. Mainstream it isn't. Best is still 2.4-ac

Quoted 4 lines
Quoted 2 lines

Yes, I think it's similar to OSS-ALSA, WRT - you know, 6-months forward notice etc :-)

Ain't going to happen...

I don't think this a six month project. Well no getting all the PCI stuff sorted is a six month project. Knocking it into shape and cleaning up corner cases and details is considerably more.

Still I've done atiixp, opt621 and it8172 this afternoon.

Alan

#15 lore raw
In reply to Bartlomiej Zolnierkiewicz
Quoted 2 lines · Bartlomiej Zolnierkiewicz

Do you think that libata is currently so much better wrt to PATA hot-(un)plug support?

For me? Libata works. Old IDE driver doesn't work. To be honest, haven't checked since switched to libata, this problem might be fixed now.

Signature · 1 line
Krzysztof Halasa
#16 lore raw
Quoted 7 lines · Krzysztof Halasa
Quoted 2 lines · Bartlomiej Zolnierkiewicz

Do you think that libata is currently so much better wrt to PATA hot-(un)plug support?

For me? Libata works. Old IDE driver doesn't work. To be honest, haven't checked since switched to libata, this problem might be fixed now.

Are you talking about PATA here?

If no there is no point in further discussion (libata is the right choice)...

Bartlomiej

#26 lore raw
In reply to Alan Cox
Quoted 3 lines

CS5520 is totally broken. You want the fixes I sent you over a year ago and maybe more. Its probably best left as I doubt anyone else has a 5520 any more 8)

Send them to me. I still have the original pre-production CS5520 system that I used to implement the original (working) driver with.

Thanks.

Signature · 3 lines
Mark Lord
Real-Time Remedies Inc.
[email protected]
#27 lore raw
Quoted 6 lines · Mark Lord
Quoted 3 lines

CS5520 is totally broken. You want the fixes I sent you over a year ago and maybe more. Its probably best left as I doubt anyone else has a 5520 any more 8)

Send them to me. I still have the original pre-production CS5520 system that I used to implement the original (working) driver with.

Oh, wait. No, I wrote the CS5530 driver, not the 5520.

Is that one okay, or do you also have patches for it (5530)?

Thanks

Signature · 3 lines
Mark Lord
Real-Time Remedies Inc.
[email protected]
author
#28 lore raw
Quoted 3 lines · Mark Lord

Oh, wait. No, I wrote the CS5530 driver, not the 5520.

Is that one okay, or do you also have patches for it (5530)?

I've got a test pata driver for the 5530 but not yet added MWDMA/UDMA switching when the master and slave are in different modes.

The 5520 is a rather different device and quite complex as it implements virtual DMA (it does PIO0->PIO4 via a PCI IDE DMA interface)

#10 lore raw
In reply to Alan Cox
Quoted 1 line · Alan Cox

Still I've done atiixp, opt621 and it8172 this afternoon.

Do you want a guinea pig? I have a system with an Asus A7V motherboard (with both via82xxxx and PDC20265 controllers) and also a notebook with a 440BX/ZX chipset.

I'm willing to trash some installations of mine.

Hope this helps if you're looking for stupid^w^w^w^w^w^wbrave users. :-)

Thanks, Rogério.

Signature · 3 lines
Rogério Brito : [email protected] : http://www.ime.usp.br/~rbrito
Homepage of the algorithms package : http://algorithms.berlios.de
Homepage on freshmeat:  http://freshmeat.net/projects/algorithms/
#21 lore raw
In reply to Alan Cox
Quoted 3 lines · Alan Cox

I've not really looked much outside of the PCI space yet (my first goal is to rescue the PC world and to get testing it wants x86 users) but Jeffs core libata code is strictly bus agnostic.

Some embedded dude (a Finn, I don't remember his name) wrote a libata SATA driver for his company's embedded SATA controller... on the ARM platform.

Though the driver itself hasn't appeared, libata is apparently working quite well with this embedded, non-PCI, ARM SATA chip.

Jeff

#29 lore raw
In reply to Alan Cox

My general take on things:

* Supporting PATA has always been a long term libata goal (and people have known this for years).

* I try to stay as far as possible away from the fight between Alan and Bart. As the changelog shows, I've merged libata patches from both.

* Alan's patches do tend to come straight to me, and it would be nice if he CC'd them to a list (linux-ide).

* Nonetheless, they get exposure in -mm (via libata-dev.git) for a while, before going upstream.

* The non-core changes, i.e. Alan and Albert's PATA drivers, aren't going upstream for a while, and will be instead living in -mm (via my "pata-drivers" libata-dev.git branch) Too much breakage and user confusion will occur if they are pushed {today|soon}.

* CONFIG_IDE=n is still largely for developers and brave souls only (or for lucky owners of the newest boxes, which simply don't have PATA on them at all).

#31 lore raw
In reply to Alan Cox
Quoted 7 lines · Alan Cox

Core Features Fixed
- Per drive tuning
- Filter quirk lists
- Single channel support

And To Add
- Specify PCI bus speed

Please, do the above such a way that we can have an arch hook for the bus speed. It's actually not uncommon to have several busses with different speeds in a machine. In addition, I've seen IDE drivers calculating based on the bus speed while the HW apparently used a clock source that didn't necessarily had to be ... PciClk ... Oh well...

Quoted 2 lines

- HPA
- IRQ mask

Why do we need the above at all ? It always looked to me like a gross hack but then, I don't fully understand what the problem was on those old x86 that needed it :)

Quoted 6 lines

AEC62xx
ATIIXP
CMD64x
CY82C693
IT8172
NS87415

Quoted 8 lines

OPTI621
PDC202xx
SGI IOC4
SIS5513
SL82C105
SLC90E66
TRM290
PCMCIA (needs hotplug merge first)

Add ppc/pmac.c but its up to me to do it :)

Ben.

#32 lore raw
Quoted 7 lines · Benjamin Herrenschmidt
Quoted 2 lines

- HPA
- IRQ mask

Why do we need the above at all ? It always looked to me like a gross hack but then, I don't fully understand what the problem was on those old x86 that needed it :)

Not sure about IRQ mask.

For host-protected area (HPA), I've been thinking about a dm-hpa driver, which libata causes to auto-attach to the libata-discovered disks during probe.

That gives people full access to the disk (and to the HPA), while ensuring that there are no partition mismatch issues.

Not sure how well it will work out in practice, but it's worth thinking about.

If auto-attach/etc. doesn't work, I lean towards defaulting libata to enable access to the HPA, under the philosophy "export 100% of the hardware". Then an interested party could create an optional dm-hpa piece, to split 100%-of-the-hardware into two pieces, one a partitionable device, and the other, the HPA.

Jeff

#22 lore raw
In reply to Jeff Garzik
Quoted 3 lines · Jeff Garzik

Some embedded dude (a Finn, I don't remember his name) wrote a libata SATA driver for his company's embedded SATA controller... on the ARM platform.

Finn Thain ?

author
#33 lore raw
In reply to Benjamin Herrenschmidt
Quoted 6 lines · Benjamin Herrenschmidt
Quoted 2 lines

- HPA
- IRQ mask

Why do we need the above at all ? It always looked to me like a gross hack but then, I don't fully understand what the problem was on those old x86 that needed it :)

You can't do anything useful with some systems without disabling the HPA because it is used to mask most of the drive at boot to hide from old incompatible BIOS.

IRQ mask is on my todo list and looks quite easy. A small number of controllers mishandle the case when the FIFO empties. Instead of stalling the drive they dribble random numbers.

Quoted 3 lines
Quoted 3 lines

ATIIXP
IT8172
OPTI621

Did initial drivers for those three yesterday

Quoted 1 line
Quoted 1 line

SL82C105

And that one last night/this morning although it (and the old one) both need serialize to fix bugs.

Alan

author
#11 lore raw
In reply to Rogério Brito
Quoted 10 lines · Rogério Brito
Quoted 1 line · Alan Cox

Still I've done atiixp, opt621 and it8172 this afternoon.

Do you want a guinea pig? I have a system with an Asus A7V motherboard (with both via82xxxx and PDC20265 controllers) and also a notebook with a 440BX/ZX chipset.

I'm willing to trash some installations of mine.

Hope this helps if you're looking for stupid^w^w^w^w^w^wbrave users. :-)

Let me get the patches cleaned up and in order start of next week and I'd be glad of testers. Especially if a 2.6.14-mm1 appears so I've got something nice to diff against 8)

Alan

author
#12 lore raw
In reply to Rogério Brito

While writing the new sl82c05 driver I noticed a real nasty lurking in the old code. According to the errata docs you have to reset the DMA engine every transfer to work around chip errata. It also says that this resets any other ATA transfer in progress.

If both channels are in use there is no locking between the channels to stop a reset on one channel as DMA begins making a mess of the other channel. Looks like serialize should be set on the driver ?

Alan

#34 lore raw
In reply to Alan Cox
Quoted 11 lines · Alan Cox
Quoted 6 lines · Benjamin Herrenschmidt
Quoted 2 lines

- HPA
- IRQ mask

Why do we need the above at all ? It always looked to me like a gross hack but then, I don't fully understand what the problem was on those old x86 that needed it :)

You can't do anything useful with some systems without disabling the HPA because it is used to mask most of the drive at boot to hide from old incompatible BIOS.

I know, I was talking about IRQ Mask :)

Quoted 3 lines

IRQ mask is on my todo list and looks quite easy. A small number of controllers mishandle the case when the FIFO empties. Instead of stalling the drive they dribble random numbers.

OK, but my question why, what is the reason why we need IRQ mask ? Some old non-PCI controllers can't grok un-related ISA IO cycles during a FIFO read/write ? I suppose those would be broken on SMP too (though I suspect then that those don't exist as SMP machines then :)

Ben.

#13 lore raw
In reply to Alan Cox
Quoted 8 lines · Alan Cox

While writing the new sl82c05 driver I noticed a real nasty lurking in the old code. According to the errata docs you have to reset the DMA engine every transfer to work around chip errata. It also says that this resets any other ATA transfer in progress.

If both channels are in use there is no locking between the channels to stop a reset on one channel as DMA begins making a mess of the other channel. Looks like serialize should be set on the driver ?

Possibly, though benh needs to comment. (I think benh has the only hardware which has the possibility of both channels - the NetWinder only has one channel with one disk.)

Signature · 3 lines
Russell King
 Linux kernel    2.6 ARM Linux   - http://www.arm.linux.org.uk/
 maintainer of:  2.6 Serial core
#14 lore raw
Quoted 6 lines · Russell King
Quoted 5 lines · Alan Cox

While writing the new sl82c05 driver I noticed a real nasty lurking in the old code. According to the errata docs you have to reset the DMA engine every transfer to work around chip errata. It also says that this resets any other ATA transfer in progress.

Quoted 7 lines
Quoted 3 lines

If both channels are in use there is no locking between the channels to stop a reset on one channel as DMA begins making a mess of the other channel. Looks like serialize should be set on the driver ?

Possibly, though benh needs to comment. (I think benh has the only hardware which has the possibility of both channels - the NetWinder only has one channel with one disk.)

I don't have this hw anymore, it was a design I worked on for my previous employer, years ago, I don't have access to it anymore and the company doesn't exist anymore.

Ben.

#30 lore raw
In reply to Jeff Garzik
Quoted 2 lines · Jeff Garzik

* CONFIG_IDE=n is still largely for developers and brave souls only (or for lucky owners of the newest boxes, which simply don't have PATA on them at all).

So, basically the average new user of Linux these days, then. Installing on a new machine. No PATA.

Cheers!

author
#35 lore raw
In reply to Benjamin Herrenschmidt
Quoted 4 lines · Benjamin Herrenschmidt

OK, but my question why, what is the reason why we need IRQ mask ? Some old non-PCI controllers can't grok un-related ISA IO cycles during a FIFO read/write ? I suppose those would be broken on SMP too (though I suspect then that those don't exist as SMP machines then :)

There are PCI controllers that are hosed too. They incorrectly handle the situation where the PCI fifo empties/fills and the controller should indicate to the drive to stall the transfer. The chipsets I know that are afflicted with this are all uniprocessor mainboard devices.

Alan