converting the NCR5380 drivers away from scsi_register
Hi Finn,
now that you've officially taken over NCR5380 is there a chance to look into converting the drivers away from scsi_register to the modern scsi_host_alloc/scsi_add_host method? The NCR5380 drivers are a large armound of the remaining users of this interface, so converting them over would be a great help in killing that 10 year old legacy off.
Quoted 3 lines · Christoph Hellwig
now that you've officially taken over NCR5380 is there a chance to look into converting the drivers away from
scsi_registerto the modern scsi_host_alloc/scsi_add_host method?
Have any other conversions taken place? I'm curious to see what was involved. Hopefully not a re-write, like esp_scsi (?)
I have not yet managed to get mac_scsi working as well as it should: PDMA is not working, either because I'm running it on the wrong (poorly documented) hardware or because the driver has bugs.
I still plan to set up a machine with a DMX3191D pci card for testing. It would be nice to have the driver working on that hardware (at least) before making (or just testing) any substantial changes.
This will take time...
Hi Finn,
Quoted 6 lines · Finn Thain
Quoted 3 lines
now that you've officially taken over NCR5380 is there a chance to look into converting the drivers away from
scsi_registerto the modern scsi_host_alloc/scsi_add_host method?Have any other conversions taken place? I'm curious to see what was involved. Hopefully not a re-write, like
esp_scsi(?)
I did a few of them in 2009:
commit c2a24a4ca1137473971842461612e56a654e7edb ("m68k: amiga - A3000 SCSI platform device conversion") commit c1d288a58936cd0654844d807e53a203f4838fb4 ("m68k: amiga - GVP Series II SCSI zorro_driver conversion") commit c737e22cde37e4e2ad126316e4aab7349a491ab3 ("m68k: amiga - A2091/A590 SCSI zorro_driver conversion")
All of them without access to the hardware. So far I haven't heard any complaints ;-)
Gr{oetje,eeting}s,
Geert
--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- [email protected]
In personal conversations with technical people, I call myself a hacker. But when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds
Quoted 19 lines · Geert Uytterhoeven
Hi Finn,
Quoted 6 lines · Finn Thain
Quoted 3 lines
now that you've officially taken over NCR5380 is there a chance to look into converting the drivers away from
scsi_registerto the modern scsi_host_alloc/scsi_add_host method?Have any other conversions taken place? I'm curious to see what was involved. Hopefully not a re-write, like
esp_scsi(?)I did a few of them in 2009:
commit
c2a24a4ca1137473971842461612e56a654e7edb("m68k: amiga - A3000 SCSI platform device conversion") commitc1d288a58936cd0654844d807e53a203f4838fb4("m68k: amiga - GVP Series II SCSIzorro_driverconversion") commitc737e22cde37e4e2ad126316e4aab7349a491ab3("m68k: amiga - A2091/A590 SCSIzorro_driverconversion")
Quoted 2 lines
All of them without access to the hardware. So far I haven't heard any complaints ;-)
Thanks Geert. You do make it look easy to avoid regressions!
sun3x_esp and mac_esp are platform devices so I guess I should convert sun3_scsi and mac_scsi to platform devices also. I'm not sure about atari_scsi, perhaps there's a different bus that can probe it. Michael?
I see that the PCI driver (dmx3191d) and the ARM drivers (oak and cumana_1) have already been converted. So the remaining NCR5380 drivers that still use scsi_register() are the ISA cards: g_NCR5380, dtc, pas16 and t128.
I don't have any ISA hardware. I certainly don't mind if those drivers ultimately get removed along with scsi_register() itself.
Those ISA drivers depend on functionality in the NCR5380.c core driver that was discarded by the other two core driver variants (atari_NCR5380.c and sun3_NCR5380.c) such as DONT_USE_INTR, UNSAFE, DMA_WORKS_RIGHT, AUTOPROBE_IRQ, USLEEP_POLL, NCR53C400.
If it were not for the ISA cards, we may not need three variations on the core driver.
Hi Finn,
Quoted 19 lines · Geert Uytterhoeven
Quoted 18 lines
Hi Finn,
Quoted 5 lines · Finn Thain
Quoted 3 lines
now that you've officially taken over NCR5380 is there a chance to look into converting the drivers away from
scsi_registerto the modern scsi_host_alloc/scsi_add_host method?Have any other conversions taken place? I'm curious to see what was involved. Hopefully not a re-write, like
esp_scsi(?)I did a few of them in 2009:
commit
c2a24a4ca1137473971842461612e56a654e7edb("m68k: amiga - A3000 SCSI platform device conversion") commitc1d288a58936cd0654844d807e53a203f4838fb4("m68k: amiga - GVP Series II SCSIzorro_driverconversion") commitc737e22cde37e4e2ad126316e4aab7349a491ab3("m68k: amiga - A2091/A590 SCSIzorro_driverconversion")
Quoted 7 lines
Quoted 2 lines
All of them without access to the hardware. So far I haven't heard any complaints ;-)
Thanks Geert. You do make it look easy to avoid regressions!
sun3x_espandmac_espare platform devices so I guess I should convertsun3_scsiandmac_scsito platform devices also. I'm not sure aboutatari_scsi, perhaps there's a different bus that can probe it. Michael?
atari_scsi is only used on TT and Falcon Atari variants. 'Testing' for hardware presence is just checking bits from a bitmap populated at arch_init time, that can easily be moved into a platform device setup, much like I already do for the network drivers. Does that answer your question?
Cheers,
Michael
Quoted 17 lines
I see that the PCI driver (dmx3191d) and the ARM drivers (oak and
cumana_1) have already been converted. So the remaining NCR5380 drivers that still usescsi_register()are the ISA cards:g_NCR5380, dtc, pas16 and t128.I don't have any ISA hardware. I certainly don't mind if those drivers ultimately get removed along with
scsi_register()itself.Those ISA drivers depend on functionality in the
NCR5380.ccore driver that was discarded by the other two core driver variants (atari_NCR5380.candsun3_NCR5380.c) such asDONT_USE_INTR, UNSAFE,DMA_WORKS_RIGHT,AUTOPROBE_IRQ,USLEEP_POLL, NCR53C400.If it were not for the ISA cards, we may not need three variations on the core driver.
Quoted 5 lines · Michael Schmitz
atari_scsiis only used on TT and Falcon Atari variants. 'Testing' for hardware presence is just checking bits from a bitmap populated atarch_inittime, that can easily be moved into a platform device setup, much like I already do for the network drivers. Does that answer your question?
Yes. I'll convert {atari,mac,sun3}_scsi to platform devices and scsi_host_alloc/scsi_add_host.
Hi Finn,
Quoted 5 lines
Quoted 2 lines
atari_scsiis only used on TT and Falcon Atari variants. 'Testing' for hardware presence is just checking bits from a bitmap populated atarch_inittime, that can easily be moved into a platform device setup, much like I already do for the network drivers. Does that answer your question?Yes. I'll convert {atari,mac,sun3}_scsi to platform devices and scsi_host_alloc/scsi_add_host.
Thanks, much appreciated. I can supply the platform device setup in arch/m68k/atari/config.c once you have decided what kind of resources or platform data will be needed.
If you rather want to do this - either use the ATARIHW_PRESENT() macros to test for ST_SCSI (Falcon, interrupt no. IRQ_MFP_FSCSI) or TT_SCSI (TT, interrupt no. IRQ_TT_MFP_SCSI). Or else, replicate the logic from config_atari() - the SCSI chip directly mapped only in the TT integration, the Falcon needs to access SCSI registers through the ST-DMA chip, and needs the weird ST-DMA locking scheme plus a few other quirks.
Looking at atari_scsi.c - the code is full of IS_A_TT() macros and other Atari specfic macros that could be replaced by testing bits in a feature map. One bit (TT or Falcon style SCSI integration) rather - that still leaves the register access functions for TT and Falcon to sort out. Do you plan to do all that in one go? Might need another platform device for the ST-DMA as well ...
Cheers,
Michael
Quoted 9 lines · Michael Schmitz
Thanks, much appreciated. I can supply the platform device setup in
arch/m68k/atari/config.conce you have decided what kind of resources or platform data will be needed.If you rather want to do this - either use the
ATARIHW_PRESENT()macros to test forST_SCSI(Falcon, interrupt no.IRQ_MFP_FSCSI) orTT_SCSI(TT, interrupt no.IRQ_TT_MFP_SCSI). Or else, replicate the logic fromconfig_atari()
Yes, that was my plan. A patch that replicates that logic is easier to review and less likely to cause regressions (of course, I'd follow the existing conventions in arch/m68k/atari/config.c).
Converting three drivers at once is a win because the first conversion is always the more difficult one. Therefore I'm planning to use the same logic three times over and therefore I'm not intending to address the peculiarities of different ports (that would be better done by you, me and Sam separately).
However, I'll will need you and Sam to test some patches (if they meet with approval once you and Sam get to review them, of course).
Quoted 9 lines
- the SCSI chip directly mapped only in the TT integration, the Falcon needs to access SCSI registers through the ST-DMA chip, and needs the weird ST-DMA locking scheme plus a few other quirks.
Looking at
atari_scsi.c- the code is full ofIS_A_TT()macros and other Atari specfic macros that could be replaced by testing bits in a feature map. One bit (TT or Falcon style SCSI integration) rather - that still leaves the register access functions for TT and Falcon to sort out. Do you plan to do all that in one go?
No, not in one go. That would be a separate patch, so that each patch has a single well defined purpose.
I'm not particularly concerned about atari_scsi.c. I am concerned about the three forks of the core driver (not least because of the shared header) that's why I've sent patches for sun3_NCR5380 and atari_NCR5380 in the past.
Quoted 1 line
Might need another platform device for the ST-DMA as well ...
I don't think that relates to scsi_register() deprecation.
Hi Finn,
Quoted 6 lines
Quoted 3 lines
If you rather want to do this - either use the
ATARIHW_PRESENT()macros to test forST_SCSI(Falcon, interrupt no.IRQ_MFP_FSCSI) orTT_SCSI(TT, interrupt no.IRQ_TT_MFP_SCSI). Or else, replicate the logic fromconfig_atari()Yes, that was my plan. A patch that replicates that logic is easier to review and less likely to cause regressions (of course, I'd follow the existing conventions in
arch/m68k/atari/config.c).
Sounds good to me.
Quoted 2 lines
Converting three drivers at once is a win because the first conversion is always the more difficult one. Therefore I'm planning to use the same logic three times over and therefore I'm not intending to address the peculiarities of different ports (that would be better done by you, me and Sam separately).
OK, I'll think some more about what could be done to merge the quirks into a shared core.
Quoted 6 lines
However, I'll will need you and Sam to test some patches (if they meet with approval once you and Sam get to review them, of course).
Quoted 3 lines
- the SCSI chip directly mapped only in the TT integration, the Falcon
needs to access SCSI registers through the ST-DMA chip, and needs the weird ST-DMA locking scheme plus a few other quirks.
Quoted 7 lines
Quoted 2 lines
Looking at
atari_scsi.c- the code is full ofIS_A_TT()macros and other Atari specfic macros that could be replaced by testing bits in a feature map. One bit (TT or Falcon style SCSI integration) rather - that still leaves the register access functions for TT and Falcon to sort out. Do you plan to do all that in one go?No, not in one go. That would be a separate patch, so that each patch has a single well defined purpose.
I'm not particularly concerned about
atari_scsi.c. I am concerned about the three forks of the core driver (not least because of the shared header) that's why I've sent patches forsun3_NCR5380andatari_NCR5380in the past.
OK, forget about atari_scsi.c for now. From memory, the major difference between atari_NCR5380.c and the others is handling of the ST-DMA/shared interrupt locking. We need to preserve that pretty much as is, or risk serious regressions. I mean, more serious trouble than I already have with the current driver because of the tendency of my Falcon to muck up the SCSI chip's clock signal under heavy load. Also note that there is still one of my patches unmerged (under review since early this year) that is necessary to avoid 'scheduling in interrupt' style panic.
The rest of the differences were tweaks added by the Atari SCSI author to make the driver behave with borderline standards compliant hardware (long reset delay for some CD drives, tweaks to bus settle delays). Probably not that critical in the first instance.
Quoted 5 lines
Quoted 2 lines
Might need another platform device for the ST-DMA as well ...
I don't think that relates to
scsi_register()deprecation.
Not at all - just to the platform device conversion. Not sure I fully understand what you have in mind, though.
Cheers,
Michael
Quoted 9 lines · Michael Schmitz
Quoted 5 lines
I'm not particularly concerned about
atari_scsi.c. I am concerned about the three forks of the core driver (not least because of the shared header) that's why I've sent patches forsun3_NCR5380andatari_NCR5380in the past.OK, forget about
atari_scsi.cfor now. From memory, the major difference betweenatari_NCR5380.cand the others is handling of the ST-DMA/shared interrupt locking.
Yes, sun3_NCR5380.c and NCR5380.c lack support for the ST-DMA chip.
And both NCR5380.c and sun3_NCR5380.c dropped the code to support ISA cards, as I mentioned.
Both NCR5380.c and sun3_NCR5380.c lack merge_contiguous_buffers().
NCR5380.c lacks support for tagged queueing, while sun3_NCR5380.c and atari_NCR5380.c have divergent implementations of this.
sun3_NCR5380.c and atari_NCR5380.c have divergent implementations of NCR5380_transfer_dma().
sun3_NCR5380.c and atari_NCR5380.c share the same implementation of the NCR5380_main() co-routine, but it differs significantly from the algorithm in NCR5380.c.
Each of sun3_NCR5380.c, atari_NCR5380.c and NCR5380.c have had some clean up and modernization work that does not appear in any of the others.
Despite all of that, I have a patch series that attempts to unify atari_NCR5380.c and sun3_NCR5380.c. This seems like a sensible idea (in theory) given that probably only the ISA cards actually need NCR5380.c.
At some point I'll send some patches for comment, so that we might discuss the issues with everyone on the same page. Until then, I'd prefer to focus on the scsi_register() conversion.
Quoted 6 lines
We need to preserve that pretty much as is, or risk serious regressions. I mean, more serious trouble than I already have with the current driver because of the tendency of my Falcon to muck up the SCSI chip's clock signal under heavy load. Also note that there is still one of my patches unmerged (under review since early this year) that is necessary to avoid 'scheduling in interrupt' style panic.
Can you re-send it please?
Quoted 5 lines
The rest of the differences were tweaks added by the Atari SCSI author to make the driver behave with borderline standards compliant hardware (long reset delay for some CD drives, tweaks to bus settle delays). Probably not that critical in the first instance.
Quoted 8 lines
Quoted 5 lines
Quoted 2 lines
Might need another platform device for the ST-DMA as well ...
I don't think that relates to
scsi_register()deprecation.Not at all - just to the platform device conversion. Not sure I fully understand what you have in mind, though.
With regards to scsi_register() conversion I'll send some patches for you to review once complete and tested (in mac_scsi). They will probably need revision after review; I'm fine with that.
Hi Finn,
Quoted 5 lines
Quoted 2 lines
OK, forget about
atari_scsi.cfor now. From memory, the major difference betweenatari_NCR5380.cand the others is handling of the ST-DMA/shared interrupt locking.Yes,
sun3_NCR5380.candNCR5380.clack support for the ST-DMA chip.
As should be.
Quoted 4 lines
And both
NCR5380.candsun3_NCR5380.cdropped the code to support ISA cards, as I mentioned.Both
NCR5380.candsun3_NCR5380.clackmerge_contiguous_buffers().
I'm sure that was a tweak added by Roman Hodek to improve performance. Should not hurt to merge this one.
Quoted 2 lines
NCR5380.clacks support for tagged queueing, whilesun3_NCR5380.candatari_NCR5380.chave divergent implementations of this.
Might be equivalent for all I've seen - just array vs. bitmap to store tags..
Quoted 2 lines
sun3_NCR5380.candatari_NCR5380.chave divergent implementations ofNCR5380_transfer_dma().
This is due to different DMA hardware in the three cases covered, so expected. There's quite a bit of additional DMA mumble in NCR5380_information_transfer and NCR5380_reselect.
Quoted 2 lines
sun3_NCR5380.candatari_NCR5380.cshare the same implementation of theNCR5380_main()co-routine, but it differs significantly from the algorithm inNCR5380.c.
atari_NCR5380.c was forked from NCR5380.c as early as 1994, and the changes to the NCR5380.c coroutine will be a later addition. NCR5380.c has seen development for a few years that we've never picked up.
Quoted 4 lines
Each of
sun3_NCR5380.c,atari_NCR5380.candNCR5380.chave had some clean up and modernization work that does not appear in any of the others.Despite all of that, I have a patch series that attempts to unify
atari_NCR5380.candsun3_NCR5380.c. This seems like a sensible idea (in theory) given that probably only the ISA cards actually needNCR5380.c.
The diff between the two was amazingly clean and easy to parse (aside from the above mentioned DMA setup related stuff). Merging the two seems quite possible indeed.
Changes to NCR5380. are rather major though. I need to carefully go through the diff again. ISTR trying to run the coroutine as delayed workqueue instead of immediate but dropped that again for some reason.
Quoted 2 lines
At some point I'll send some patches for comment, so that we might discuss the issues with everyone on the same page. Until then, I'd prefer to focus on the
scsi_register()conversion.
OK, do that first so scsi_register can go away.
Quoted 5 lines
Quoted 2 lines
We need to preserve that pretty much as is, or risk serious regressions. I mean, more serious trouble than I already have with the current driver because of the tendency of my Falcon to muck up the SCSI chip's clock signal under heavy load. Also note that there is still one of my patches unmerged (under review since early this year) that is necessary to avoid 'scheduling in interrupt' style panic.
Can you re-send it please?
I'll do that.
Cheers,
Michael
Quoted 6 lines · Michael Schmitz
Quoted 3 lines
NCR5380.clacks support for tagged queueing, whilesun3_NCR5380.candatari_NCR5380.chave divergent implementations of this.Might be equivalent for all I've seen - just array vs. bitmap to store tags..
It is messier than that. setup_use_tagged_queuing is disabled by default on atari, and enabled by default on sun3. If we don't disable it by default everywhere then we need to resolve a change to the algorithm: the following code appears at the beginning of NCR5380_reselect() in atari_NCR5380.c but appears at the end of that routine in sun3_NCR5380.c. I still have to try to figure out the implications of this change.
#ifdef SUPPORT_TAGS
/* If the phase is still MSGIN, the target wants to send some more
* messages. In case it supports tagged queuing, this is probably a
* SIMPLE_QUEUE_TAG for the I_T_L_Q nexus.
*/
tag = TAG_NONE;
if (phase == PHASE_MSGIN && setup_use_tagged_queuing) {
/* Accept previous IDENTIFY message by clearing ACK */
NCR5380_write( INITIATOR_COMMAND_REG, ICR_BASE );
len = 2;
data = msg+1;
if (!NCR5380_transfer_pio(instance, &phase, &len, &data) &&
msg[1] == SIMPLE_QUEUE_TAG)
tag = msg[2];
dprintk(NDEBUG_TAGS, "scsi%d: target mask %02x, lun %d sent tag %d at "
"reselection\n", HOSTNO, target_mask, lun, tag);
}#endif
Quoted 4 lines
Changes to NCR5380. are rather major though. I need to carefully go through the diff again. ISTR trying to run the coroutine as delayed workqueue instead of immediate but dropped that again for some reason.
I don't think it makes sense to unify NCR5380.c and atari_NCR5380.c at this stage. The ISA cards' requirements together are more peculiar than, say, ST-DMA, and are mostly irrelevant to the M68k, ARM and PCI devices.
As an aside, forking NCR5380.c into atari_NCR5380.c, sun3_NCR5380.c and mac_NCR5380.c (deleted in v2.6.3) was clearly misguided as it never actually produced a good alternative to NCR5380.c.
Forking just multiplied the maintenance demands, and diminishing man-power of course lead to neglect and breakage. So I can sympathise with Christoph's comment in NCR5380.c:
* (Note from hch: unfortunately it was not enough for the different
* m68k folks and instead of improving this driver they copied it
* and hacked it up for their needs. As a consequence they lost
* most updates to this driver. Maybe someone will fix all these
* drivers to use a common core one day..)
But one core is hard, even if you have all the hardware. Two cores is better than three.
Hi Finn,
Quoted 11 lines · Michael Schmitz
Quoted 7 lines
Quoted 5 lines
NCR5380.clacks support for tagged queueing, whilesun3_NCR5380.candatari_NCR5380.chave divergent implementations of this.Might be equivalent for all I've seen - just array vs. bitmap to store tags..
It is messier than that.
setup_use_tagged_queuingis disabled by default on atari, and enabled by default on sun3. If we don't disable it by
It's configured but disabled by default on Atari because command_per_lun is set to 1 by default. Can't remember why that was exactly - my best guess would be that queueing up more commands in the lowlevel driver did use to exacerbate timeout problems. But that was over a decade ago. On Sun3, the code is not even configured from what I've seen (SUPPORT_TAGS not defined).
Quoted 2 lines
default everywhere then we need to resolve a change to the algorithm: the following code appears at the beginning of
NCR5380_reselect()inatari_NCR5380.cbut appears at the end of that routine insun3_NCR5380.c.
That's what I meant to say before - sun3_NCR5380.c does peek at the message byte without invoking NCR5380_transfer_pio() (needs to manually set/clear ACK for that resason, finds the disconnected command at issue, does some DMA setup stuff for that command, and then goes ahead to fetch the tag message bytes if the phase is still at message in. No idea where the variable 'phase' gets set in that routine so it might never get to read tag messages.
atari_NCR5390.c does things a bit different - instead of peeking at the message byte, NCR5380_transfer_pio() is used for the first message byte (ACK set there, cleared manually later). NCR5380_transfer_pio() sets the current phase, so the subsequent transfer of the tag message may work there (again ACK raised in NCR5380_transfer_pio(), cleared manually later).
Quoted 2 lines
I still have to try to figure out the implications of this change.
From what I've checked both should work, but the sun3 code never sets 'phase' so it should never actually have worked. It should not even compile.
Quoted 17 lines
#ifdef SUPPORT_TAGS
/* If the phase is still MSGIN, the target wants to send some more
* messages. In case it supports tagged queuing, this is probably a
* SIMPLE_QUEUE_TAG for the I_T_L_Q nexus.
*/
tag = TAG_NONE;
if (phase == PHASE_MSGIN && setup_use_tagged_queuing) {
/* Accept previous IDENTIFY message by clearing ACK */
NCR5380_write( INITIATOR_COMMAND_REG, ICR_BASE );
len = 2;
data = msg+1;
if (!NCR5380_transfer_pio(instance, &phase, &len, &data) &&
msg[1] == SIMPLE_QUEUE_TAG)
tag = msg[2];
dprintk(NDEBUG_TAGS, "scsi%d: target mask %02x, lun %d sent tag %d at "
"reselection\n", HOSTNO, target_mask, lun, tag);
}Quoted 8 lines
#endif
Quoted 2 lines
Changes to NCR5380. are rather major though. I need to carefully go through the diff again. ISTR trying to run the coroutine as delayed workqueue instead of immediate but dropped that again for some reason.
I don't think it makes sense to unify
NCR5380.candatari_NCR5380.cat this stage. The ISA cards' requirements together are more peculiar than, say, ST-DMA, and are mostly irrelevant to the M68k, ARM and PCI devices.
OK, not for today then.
Quoted 2 lines
As an aside, forking
NCR5380.cintoatari_NCR5380.c,sun3_NCR5380.candmac_NCR5380.c(deleted in v2.6.3) was clearly misguided as it never actually produced a good alternative toNCR5380.c.
atari_NCR5380.c was forked at a time when the m68k port was developed entirely independent from the main Linux development (which was i386 only at that stage). Linus did not want to deal with other architectures, and merging m68k code into Linus' code was not something the m68k authors envisaged at that stage. The fork was not meant to produce a good alternative to the main driver, just something that worked. At least that is as I recall it - I had little to do with Linux kernel development at that stage. Merging ongoing development on NCR5380.c into the Atari driver was always something of an afterthought (Geert probably took care of most of that). The sun3 and Mac drivers were then forked off the Atari one. I'm guilty in part of the initial Mac one, but that was done mainly in order to stop Alan Cox from whinging about the lack of a disk driver. (Don't think he liked SCSI in PIO mode much.)
What I mean to say - in hindsight, it was misguided, but at the time we did not have the benefit of a crystal ball (or git, or even fast cross compilers).
Quoted 8 lines
Forking just multiplied the maintenance demands, and diminishing man-power of course lead to neglect and breakage. So I can sympathise with Christoph's comment in
NCR5380.c:* (Note from hch: unfortunately it was not enough for the different
* m68k folks and instead of improving this driver they copied it
* and hacked it up for their needs. As a consequence they lost
* most updates to this driver. Maybe someone will fix all these
* drivers to use a common core one day..)
Yep, I can now sympathise with that. When m68k began to merge with Linus, we had already lost much of the initial man-power and expertise. I'm sure we'd do things different now, but these lessons had to be learned along the way.
Cheers,
Michael
Sun3 SCSI DMA, was Re: converting the NCR5380 drivers away from scsi_register
Quoted 4 lines · Michael Schmitz
On Sun3, the code is not even configured from what I've seen (
SUPPORT_TAGSnot defined)
[...]
Quoted 9 lines
That's what I meant to say before -
sun3_NCR5380.cdoes peek at the message byte without invokingNCR5380_transfer_pio()(needs to manually set/clear ACK for that resason, finds the disconnected command at issue, does some DMA setup stuff for that command, and then goes ahead to fetch the tag message bytes if the phase is still at message in. No idea where the variable 'phase' gets set in that routine so it might never get to read tag messages.
Quoted 6 lines
atari_NCR5390.cdoes things a bit different - instead of peeking at the message byte,NCR5380_transfer_pio()is used for the first message byte (ACK set there, cleared manually later).NCR5380_transfer_pio()sets the current phase, so the subsequent transfer of the tag message may work there (again ACK raised inNCR5380_transfer_pio(), cleared manually later).
Yes, but why the discrepancy?
Not unlike the use of DMA in NCR5380_reselect(), the sun3 version of NCR5380_information_transfer() also uses DMA in one situation where the atari version uses PIO. That is, when phase == PHASE_DATA_IN && cmd->device->borken.
Can we just use the PIO code from atari_NCR5380, or should we keep the DMA versions with #ifdef CONFIG_SUN3? Sam?
Quoted 3 lines
From what I've checked both should work, but the sun3 code never sets 'phase' so it should never actually have worked. It should not even compile.
I guess that's why sun3_scsi leaves SUPPORT_TAGS undefined. I suppose we can ignore all the code in sun3_NCR5380.c that is conditional upon #ifdef SUPPORT_TAGS. (That's actually how I did my preliminary merge patches.)
I guess having two versions of the NCR5380_transfer_dma() routine is reasonable but it would be nice if some of the other discrepancies could be resolved. Without knowledge of the Sun 3 DMA hardware it is difficult to avoid some questionable #ifdefs.
Re: Sun3 SCSI DMA, was Re: converting the NCR5380 drivers away from scsi_register
Quoted 26 lines · Finn Thain
Quoted 4 lines · Michael Schmitz
On Sun3, the code is not even configured from what I've seen (
SUPPORT_TAGSnot defined)[...]
Quoted 9 lines
That's what I meant to say before -
sun3_NCR5380.cdoes peek at the message byte without invokingNCR5380_transfer_pio()(needs to manually set/clear ACK for that resason, finds the disconnected command at issue, does some DMA setup stuff for that command, and then goes ahead to fetch the tag message bytes if the phase is still at message in. No idea where the variable 'phase' gets set in that routine so it might never get to read tag messages.Quoted 6 lines
atari_NCR5390.cdoes things a bit different - instead of peeking at the message byte,NCR5380_transfer_pio()is used for the first message byte (ACK set there, cleared manually later).NCR5380_transfer_pio()sets the current phase, so the subsequent transfer of the tag message may work there (again ACK raised inNCR5380_transfer_pio(), cleared manually later).Yes, but why the discrepancy?
Quoted 7 lines
Not unlike the use of DMA in
NCR5380_reselect(), the sun3 version ofNCR5380_information_transfer()also uses DMA in one situation where the atari version uses PIO. That is, when phase ==PHASE_DATA_IN&&cmd->device->borken.Can we just use the PIO code from
atari_NCR5380, or should we keep the DMA versions with #ifdefCONFIG_SUN3? Sam?
Oof, my memory is hazy around this one, but I would strongly hesitate to use the PIO version from the atari driver. IIRC, the DMA controller for the sun3's NCR5380 implementation is extremely fussy about what happens in which phase, and it's quick to anger if you don't handle everything exactly how it expects.
Quoted 4 lines
Quoted 3 lines
From what I've checked both should work, but the sun3 code never sets 'phase' so it should never actually have worked. It should not even compile.
Quoted 8 lines
I guess that's why
sun3_scsileavesSUPPORT_TAGSundefined. I suppose we can ignore all the code insun3_NCR5380.cthat is conditional upon #ifdefSUPPORT_TAGS. (That's actually how I did my preliminary merge patches.)I guess having two versions of the
NCR5380_transfer_dma()routine is reasonable but it would be nice if some of the other discrepancies could be resolved. Without knowledge of the Sun 3 DMA hardware it is difficult to avoid some questionable #ifdefs.
I definitely remember that the DMA setup logic from the atari version did not work at all on sun3. Unfortunatly working from 15 year old memories, I can't quickly recall what the discrepencies were. I never did find any documentation for that DMA chip, so there was a heck of a lot of trial and error on getting to a working driver.
-- Sam
Re: Sun3 SCSI DMA, was Re: converting the NCR5380 drivers away from scsi_register
Hi Sam,
Quoted 6 lines
Quoted 6 lines
Quoted 2 lines
atari_NCR5390.cdoes things a bit different - instead of peeking at the message byte,NCR5380_transfer_pio()is used for the first message byte (ACK set there, cleared manually later).NCR5380_transfer_pio()sets the current phase, so the subsequent transfer of the tag message may work there (again ACK raised inNCR5380_transfer_pio(), cleared manually later).Yes, but why the discrepancy?
Not unlike the use of DMA in
NCR5380_reselect(), the sun3 version ofNCR5380_information_transfer()also uses DMA in one situation where the atari version uses PIO. That is, when phase ==PHASE_DATA_IN&&cmd->device->borken.
Quoted 9 lines
Quoted 2 lines
Can we just use the PIO code from
atari_NCR5380, or should we keep the DMA versions with #ifdefCONFIG_SUN3? Sam?Oof, my memory is hazy around this one, but I would strongly hesitate to use the PIO version from the atari driver. IIRC, the DMA controller for the sun3's NCR5380 implementation is extremely fussy about what happens in which phase, and it's quick to anger if you don't handle everything exactly how it expects.
Does the DMA controller also sit in between the bus and the NCR chip, as on Falcon? Otherwise, I can't see how it would matter if we just bypass it and use PIO instead.
The PIO code from the Atari driver worked just fine for the Mac 5380 (in fact, the first Mac driver used PIO for everything, including data in/out).
Quoted 11 lines
Quoted 8 lines
Quoted 4 lines
From what I've checked both should work, but the sun3 code never sets 'phase' so it should never actually have worked. It should not even compile.
I guess that's why
sun3_scsileavesSUPPORT_TAGSundefined. I suppose we can ignore all the code insun3_NCR5380.cthat is conditional upon #ifdefSUPPORT_TAGS. (That's actually how I did my preliminary merge patches.)I guess having two versions of the
NCR5380_transfer_dma()routine is reasonable but it would be nice if some of the other discrepancies could be resolved. Without knowledge of the Sun 3 DMA hardware it is difficult to avoid some questionable #ifdefs.I definitely remember that the DMA setup logic from the atari version did not work at all on sun3.
No surprise there - DMA is implementation specific and needs to be handled on a per-case basis.
Quoted 5 lines
Unfortunatly working from 15 year old
memories, I can't quickly recall what the discrepencies were. I never did find any documentation for that DMA chip, so there was a heck of a lot of trial and error on getting to a working driver.
Do you still have the hardware to test (PIO in reselection, to be specific)?
Cheers,
Michael
Re: Sun3 SCSI DMA, was Re: converting the NCR5380 drivers away from scsi_register
Quoted 26 lines · Michael Schmitz
Hi Sam,
Quoted 18 lines
Quoted 9 lines
Quoted 7 lines
atari_NCR5390.cdoes things a bit different - instead of peeking at the message byte,NCR5380_transfer_pio()is used for the first message byte (ACK set there, cleared manually later).NCR5380_transfer_pio()sets the current phase, so the subsequent transfer of the tag message may work there (again ACK raised inNCR5380_transfer_pio(), cleared manually later).Yes, but why the discrepancy?
Quoted 8 lines
Not unlike the use of DMA in
NCR5380_reselect(), the sun3 version ofNCR5380_information_transfer()also uses DMA in one situation where the atari version uses PIO. That is, when phase ==PHASE_DATA_IN&&cmd->device->borken.Can we just use the PIO code from
atari_NCR5380, or should we keep the DMA versions with #ifdefCONFIG_SUN3? Sam?Quoted 6 lines
Oof, my memory is hazy around this one, but I would strongly hesitate to use the PIO version from the atari driver. IIRC, the DMA controller for the sun3's NCR5380 implementation is extremely fussy about what happens in which phase, and it's quick to anger if you don't handle everything exactly how it expects.
Quoted 6 lines
Does the DMA controller also sit in between the bus and the NCR chip, as on Falcon? Otherwise, I can't see how it would matter if we just bypass it and use PIO instead.
The PIO code from the Atari driver worked just fine for the Mac 5380 (in fact, the first Mac driver used PIO for everything, including data in/out).
PIO definitely works on the Sun3 implementation under the right circumstances, I ran it in PIO mode only for a while (much like the Mac driver).
I can't remember the bus configuration offhand.
Quoted 21 lines
Quoted 9 lines
Quoted 9 lines
Quoted 5 lines
From what I've checked both should work, but the sun3 code never sets 'phase' so it should never actually have worked. It should not even compile.
I guess that's why
sun3_scsileavesSUPPORT_TAGSundefined. I suppose we can ignore all the code insun3_NCR5380.cthat is conditional upon #ifdefSUPPORT_TAGS. (That's actually how I did my preliminary merge patches.)Quoted 8 lines
Quoted 5 lines
I guess having two versions of the
NCR5380_transfer_dma()routine is reasonable but it would be nice if some of the other discrepancies could be resolved. Without knowledge of the Sun 3 DMA hardware it is difficult to avoid some questionable #ifdefs.I definitely remember that the DMA setup logic from the atari version did not work at all on sun3.
No surprise there - DMA is implementation specific and needs to be handled on a per-case basis.
Quoted 8 lines
Quoted 5 lines
Unfortunatly working from 15 year old
memories, I can't quickly recall what the discrepencies were. I never did find any documentation for that DMA chip, so there was a heck of a lot of trial and error on getting to a working driver.Do you still have the hardware to test (PIO in reselection, to be specific)?
Yes, I still have hardware. I'll try to dig something out and make sure the last kernel I've actually tested still works before trying anything new...
-- Sam
Quoted 4 lines
Cheers,
Michael
Re: Sun3 SCSI DMA, was Re: converting the NCR5380 drivers away from scsi_register
Quoted 14 lines · Sam Creasey
Quoted 5 lines · Michael Schmitz
Quoted 4 lines
IIRC, the DMA controller for the sun3's NCR5380 implementation is extremely fussy about what happens in which phase, and it's quick to anger if you don't handle everything exactly how it expects.
Quoted 7 lines
Does the DMA controller also sit in between the bus and the NCR chip, as on Falcon? Otherwise, I can't see how it would matter if we just bypass it and use PIO instead.
The PIO code from the Atari driver worked just fine for the Mac 5380 (in fact, the first Mac driver used PIO for everything, including data in/out).
Quoted 5 lines
PIO definitely works on the Sun3 implementation under the right circumstances, I ran it in PIO mode only for a while (much like the Mac driver).
I can't remember the bus configuration offhand.
It might not be a big problem: sun3_scsi does a lot of PIO a lot as it is.
For NCR5380_reselect(), atari_NCR5380.c apparently uses only PIO whereas sun3_NCR5380.c will use PIO up to a 128 byte size limit (beyond which, only DMA is used).
For NCR5380_information_transfer() and PHASE_DATAIN, atari will use PIO for transfersize <= 31 bytes whereas sun3 will use PIO for <= 128 bytes. However, atari never uses DMA here if cmd->device->borken.
When cmd->device->borken, I assume sun3_NCR5380 inhibits PIO because PIO was itself expected to be problematic?
OTOH, if PIO works up to 128 bytes in all cases, why not larger transfers?
Does the sun3_scsi driver still work after removing #define REAL_DMA?
For NCR5380_information_transfer() and PHASE_CMDOUT, the sun3 version applies the same size limit but only does DMA setup if cmd type is REQ_TYPE_FS (i.e. filesystem request). This command is then sent by PIO, so the DMA setup here seems to assume that the next phase will always be PHASE_DATAIN...
This would seem to imply that PIO always works when not REQ_TYPE_FS, such that no size limit is applicable...
BTW, testing sun3_dma_setup_done against cmd pointers looks very unreliable to me: unique cmds only have unique pointers until freed by the scsi mid-layer. After that, the same pointer is likely be re-used.
Quoted 6 lines
Quoted 6 lines
Quoted 3 lines
I definitely remember that the DMA setup logic from the atari version did not work at all on sun3.
No surprise there - DMA is implementation specific and needs to be handled on a per-case basis.
Perhaps atari_NCR5380 could use some of the sun3_NCR5380 code. E.g. atari avoids DMA entirely for reselect, and (only in some phases) when cmd->device->borken.
Re: Sun3 SCSI DMA, was Re: converting the NCR5380 drivers away from scsi_register
Hi Finn,
Quoted 12 lines · Sam Creasey
Quoted 11 lines · Michael Schmitz
Quoted 8 lines
Quoted 4 lines
IIRC, the DMA controller for the sun3's NCR5380 implementation is extremely fussy about what happens in which phase, and it's quick to anger if you don't handle everything exactly how it expects.
Does the DMA controller also sit in between the bus and the NCR chip, as on Falcon? Otherwise, I can't see how it would matter if we just bypass it and use PIO instead.
The PIO code from the Atari driver worked just fine for the Mac 5380 (in fact, the first Mac driver used PIO for everything, including data in/out).
Quoted 9 lines
Quoted 4 lines
PIO definitely works on the Sun3 implementation under the right circumstances, I ran it in PIO mode only for a while (much like the Mac driver).
I can't remember the bus configuration offhand.
It might not be a big problem:
sun3_scsidoes a lot of PIO a lot as it is.For
NCR5380_reselect(),atari_NCR5380.capparently uses only PIO whereassun3_NCR5380.cwill use PIO up to a 128 byte size limit (beyond which, only DMA is used).
Transfer sizes for message in during reselection are only a few bytes. According to atari_dma_xfer_len(), DMA reads must be multiple of 512 bytes (writes can be rounded up, reads cannot). PIO is the only option for Atari, I'm afraid.
Quoted 2 lines
For
NCR5380_information_transfer()andPHASE_DATAIN, atari will use PIO for transfersize <= 31 bytes whereas sun3 will use PIO for <= 128 bytes.
IIRC the ST-DMA can't be programmed for shorter transfers than 512 bytes. The condition
(transfersize = NCR5380_dma_xfer_len(instance,cmd,phase)) > 31)
enforces that - transfersize will be 0 on reads that are not multiples of 512 bytes (and even some where the advertised transfer size does meet that condition, but the command is not guranteed to be a block mode command (tape reads).
Quoted 2 lines
However, atari never uses DMA here if
cmd->device->borken.
Different semantics of borken, I presume. On Atari, this means a DMA transfer has failed earlier.
Quoted 4 lines
When
cmd->device->borken, I assumesun3_NCR5380inhibits PIO because PIO was itself expected to be problematic?OTOH, if PIO works up to 128 bytes in all cases, why not larger transfers?
Performance, perhaps. PIO would work, but hog the CPU. In fact, PIO for everything _does_ work, on Falcon as well as Mac (I tried the PIO only mode on the Falcon before trying on Mac). Did I eve mention how reallly incredibly painfully slow that is?
Quoted 4 lines
Does the
sun3_scsidriver still work after removing #defineREAL_DMA?For
NCR5380_information_transfer()andPHASE_CMDOUT, the sun3 version applies the same size limit but only does DMA setup if cmd type isREQ_TYPE_FS(i.e. filesystem request). This command is then sent by PIO, so the DMA setup here seems to assume that the next phase will always bePHASE_DATAIN...
Might be a similar limitation to certain transfer sizes here. Block mode vs. char mode.
Quoted 5 lines
This would seem to imply that PIO always works when not
REQ_TYPE_FS, such that no size limit is applicable...BTW, testing
sun3_dma_setup_doneagainst cmd pointers looks very unreliable to me: unique cmds only have unique pointers until freed by the scsi mid-layer. After that, the same pointer is likely be re-used.
Quoted 7 lines
Quoted 4 lines
Quoted 4 lines
Quoted 2 lines
I definitely remember that the DMA setup logic from the atari version did not work at all on sun3.
No surprise there - DMA is implementation specific and needs to be handled on a per-case basis.
Perhaps
atari_NCR5380could use some of thesun3_NCR5380code. E.g. atari avoids DMA entirely for reselect, and (only in some phases) whencmd->device->borken.
I don't think we can use DMA for the reselect message in transfers. Data phase transfer is handled by the coroutine after reselection anyway - that one will use DMA.
Cheers,
Michael
Re: Sun3 SCSI DMA, was Re: converting the NCR5380 drivers away from scsi_register
Quoted 8 lines · Michael Schmitz
Quoted 4 lines
For
NCR5380_information_transfer()andPHASE_DATAIN, atari will use PIO for transfersize <= 31 bytes whereas sun3 will use PIO for <= 128 bytes.IIRC the ST-DMA can't be programmed for shorter transfers than 512 bytes. The condition
Quoted 6 lines
(transfersize =
NCR5380_dma_xfer_len(instance,cmd,phase)) > 31)enforces that - transfersize will be 0 on reads that are not multiples of 512 bytes (and even some where the advertised transfer size does meet that condition, but the command is not guranteed to be a block mode command (tape reads).
I see.
Quoted 6 lines
Quoted 3 lines
However, atari never uses DMA here if
cmd->device->borken.Different semantics of borken, I presume. On Atari, this means a DMA transfer has failed earlier.
Right. I suppose the question of using PIO in this situation on sun3 would have to be put to the test.
Quoted 9 lines
Quoted 7 lines
When
cmd->device->borken, I assumesun3_NCR5380inhibits PIO because PIO was itself expected to be problematic?OTOH, if PIO works up to 128 bytes in all cases, why not larger transfers?
Performance, perhaps.
Well, sun3_scsi has this:
static inline unsigned long sun3scsi_dma_xfer_len(unsigned long wanted,
struct scsi_cmnd *cmd, int write_flag)
{
if (cmd->request->cmd_type == REQ_TYPE_FS)
return wanted;
else
return 0;
}This does suggest that PIO does work beyond 128 bytes, even if it is avoided for filesystem requests. Which suggests that the atari reselect code might work on sun3.
(As for performance, I don't know why the REQ_TYPE_FS is significant: why not just raise the DMA size limit to 512 instead?)
Quoted 14 lines
PIO would work, but hog the CPU. In fact, PIO for everything _does_ work, on Falcon as well as Mac (I tried the PIO only mode on the Falcon before trying on Mac). Did I eve mention how reallly incredibly painfully slow that is?
Quoted 8 lines
Does the
sun3_scsidriver still work after removing #defineREAL_DMA?For
NCR5380_information_transfer()andPHASE_CMDOUT, the sun3 version applies the same size limit but only does DMA setup if cmd type isREQ_TYPE_FS(i.e. filesystem request). This command is then sent by PIO, so the DMA setup here seems to assume that the next phase will always bePHASE_DATAIN...
Quoted 2 lines
Might be a similar limitation to certain transfer sizes here. Block mode vs. char mode.
That suggests that the test for REQ_TYPE_FS in NCR5380_information_transfer() in sun3_NCR5380 could be improved upon by using NCR5380_dma_xfer_len() when checking the size limit. That way it would work on atari as well, without #ifdefs.
Re: Sun3 SCSI DMA, was Re: converting the NCR5380 drivers away from scsi_register
Hi Finn,
Quoted 8 lines
Quoted 5 lines
Quoted 3 lines
However, atari never uses DMA here if
cmd->device->borken.Different semantics of borken, I presume. On Atari, this means a DMA transfer has failed earlier.
Right. I suppose the question of using PIO in this situation on sun3 would have to be put to the test.
Yep - let's see whether Sam has a working test configuration with current kernels.
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
When
cmd->device->borken, I assumesun3_NCR5380inhibits PIO because PIO was itself expected to be problematic?
The sun3 driver does set the borken flag but never uses it later - can't see where it will inhibit PIO.
Quoted 18 lines
Quoted 5 lines
Quoted 3 lines
OTOH, if PIO works up to 128 bytes in all cases, why not larger transfers?
Performance, perhaps.
Well,
sun3_scsihas this:static inline unsigned long sun3scsi_dma_xfer_len(unsigned long wanted,
structscsi_cmnd*cmd, intwrite_flag){ if (cmd->request->cmd_type == REQ_TYPE_FS) return wanted; else return 0; }
Quoted 4 lines
This does suggest that PIO does work beyond 128 bytes, even if it is avoided for filesystem requests. Which suggests that the atari reselect code might work on sun3.
(As for performance, I don't know why the
REQ_TYPE_FSis significant: why not just raise the DMA size limit to 512 instead?)
No idea really - the sun3 DMA allows to read the correct residual so it probably is not necessary to strictly avoid DMA on anything not guaranteed to transfer the requested size. Not sure what commands will transfer between 128 and 512 bytes ...
Quoted 11 lines
Quoted 11 lines
PIO would work, but hog the CPU. In fact, PIO for everything _does_ work, on Falcon as well as Mac (I tried the PIO only mode on the Falcon before trying on Mac). Did I eve mention how reallly incredibly painfully slow that is?
Quoted 5 lines
Does the
sun3_scsidriver still work after removing #defineREAL_DMA?For
NCR5380_information_transfer()andPHASE_CMDOUT, the sun3 version applies the same size limit but only does DMA setup if cmd type isREQ_TYPE_FS(i.e. filesystem request). This command is then sent by PIO, so the DMA setup here seems to assume that the next phase will always bePHASE_DATAIN...Might be a similar limitation to certain transfer sizes here. Block mode vs. char mode.
Commands need to be sent by PIO always because of the DMA size limit. The DMA setup early on may save time if the next phase is indeed DATAIN or DATAOUT - if a command disconnects the DMA just won't be started. I would have thought the DMA setup could just as easily be done inside NCR5380_transfer_dma instead of splitting it in DMA setup and DMA start - any particular reason for this, Sammy?
Quoted 2 lines
That suggests that the test for
REQ_TYPE_FSinNCR5380_information_transfer()insun3_NCR5380could be improved upon by usingNCR5380_dma_xfer_len()when checking the size limit. That way it would work on atari as well, without #ifdefs.
The test for REQ_TYPE_FS could be done in the specific dma_xfer_len() function which would need to be used whenever DMA setup is going to be used. As it is, the separation of DMA setup from NCR5380_transfer_dma is a bit mysterious and makes it hard to merge the drivers without lots of ifdefs or runtime tests (think multiplatform kernels) anyway.
Cheers,
Michael
Re: Sun3 SCSI DMA, was Re: converting the NCR5380 drivers away from scsi_register
Quoted 7 lines · Michael Schmitz
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
When
cmd->device->borken, I assumesun3_NCR5380inhibits PIO because PIO was itself expected to be problematic?The sun3 driver does set the borken flag but never uses it later - can't see where it will inhibit PIO.
It inhibited PIO when it forked the atari code and commented out the test for cmd->device->borken, to ensure that PIO would never be used here.
Hence my suggestion that PIO itself was expected was problematic. OTOH, PIO is used up to 128 bytes, and beyond 128 bytes if !REQ_TYPE_FS...
Re: Sun3 SCSI DMA, was Re: converting the NCR5380 drivers away from scsi_register
Hi Finn,
Quoted 8 lines · Finn Thain
Quoted 5 lines · Michael Schmitz
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
When
cmd->device->borken, I assumesun3_NCR5380inhibits PIO because PIO was itself expected to be problematic?
Quoted 5 lines
Quoted 2 lines
The sun3 driver does set the borken flag but never uses it later - can't see where it will inhibit PIO.
It inhibited PIO when it forked the atari code and commented out the test for
cmd->device->borken, to ensure that PIO would never be used here.
I see. I just tool that to mean the Sun3 driver never had trouble with devices failing at that section, so no need to switch from DMA to PIO (because DMA never fails).
Quoted 2 lines
Hence my suggestion that PIO itself was expected was problematic. OTOH, PIO is used up to 128 bytes, and beyond 128 bytes if !REQ_TYPE_FS...
Up to 128 for DMA limitations - the other bit I don't understand either.
Cheers,
Michael
Quoted 1 line
--
Re: Sun3 SCSI DMA, was Re: converting the NCR5380 drivers away from scsi_register
In reply to Michael SchmitzQuoted 6 lines · Michael Schmitz
The DMA setup early on may save time if the next phase is indeed DATAIN or DATAOUT - if a command disconnects the DMA just won't be started. I would have thought the DMA setup could just as easily be done inside
NCR5380_transfer_dmainstead of splitting it in DMA setup and DMA start
- any particular reason for this, Sammy?
Doing the DMA setup early on may not work for atari_NCR5380 given merge_contiguous_buffers() takes place prior to NCR5380_transfer_dma() and may alter the length of the tranfer. (sun3_NCR5380 doesn't have merge_contiguous_buffers()).
Re: Sun3 SCSI DMA, was Re: converting the NCR5380 drivers away from scsi_register
Hi Finn,
Quoted 10 lines · Finn Thain
Quoted 6 lines · Michael Schmitz
The DMA setup early on may save time if the next phase is indeed DATAIN or DATAOUT - if a command disconnects the DMA just won't be started. I would have thought the DMA setup could just as easily be done inside
NCR5380_transfer_dmainstead of splitting it in DMA setup and DMA start
- any particular reason for this, Sammy?
Quoted 4 lines
Doing the DMA setup early on may not work for
atari_NCR5380givenmerge_contiguous_buffers()takes place prior toNCR5380_transfer_dma()and may alter the length of the tranfer. (sun3_NCR5380doesn't havemerge_contiguous_buffers()).
Not only that - the DMA setup (on Falcon) will be trashed by subsequent accesses to the NCR5380 (registers are accessed through the ST-DMA, the NCR5380 is not directly connected to the CPU bus).
We'd need to do all DMA setup in NCR5380_transfer_dma instead.
Cheers,
Michael
Quoted 2 lines
--