Re: E Class and SCSI (53c700 driver)

Re: E Class and SCSI (53c700 driver)

[Helge Deller]

Hi Dmitry,

I've added the parisc-kernel-devel mailing list too, since I expect
that *if* somebody is able to answer your question, it will probably
be on the parisc-kernel-devel list...

On 31.03.2015 00:11, Dmitry the Zuryanovich wrote:
> As I understand, E class (in my case - E25) has LASI based 53c710
> which is handled by 53c700.c? (and in case of workstations wrapped by
> lasi700 - what's for? Just for detecting it?)
>
> If I know that that's
>
> 5. Sahp Baat Kiuh SCSI at 0xfff74000 [56/52] {4, 0x0, 0x044, 0x00039}
>
> , is there a reason to try to
>
> #modprobe 53c700 clock=25  base=0xfff74000

Looking at drivers/scsi/lasi700.c, it seems the lasi700 driver sets
additional flags like force_le_on_be, chip710, burst_length and so on.
In addition, the IRQs gets connected via drivers/parisc/lasi.c (see lasi_choose_irq()).
I assume you would need at least a "case 0x39" in there.
But James is the expert on the 53c700 driver, so he might know more... ?
  
> [correct my syntax] and may I expect this to work somehow?
>
> What's the problem with  SCSI on E class? Is that a need of some
> driver which calls 53c700.c with correct parameters taken from
> hardware? Can it be just passed to module via parameter or the
> problem is deeper?

It might be deeper.
In  arch/parisc/kernel/hardware.c I see:
         {HPHW_A_DMA, 0x044, 0x00039, 0x80, "Sahp Baat Kiuh SCSI"},
while other SCSI drivers on LASI have HPHW_FIO, e.g.:
	{HPHW_FIO, 0x016, 0x00082, 0x0, "Gecko Core SCSI"},


For reference, here is the full dmesg from Dmitry when he tried to boot his E25
machine with a recent Linux kernel:
ftp://parisc.parisc-linux.org/dmesg/E25.dmesg  (http://pastebin.com/sXqUjVub)

Helge

Re: E Class and SCSI (53c700 driver)

[James Bottomley]

On Thu, 2015-04-02 at 10:07 +0200, Helge Deller wrote:
> Hi Dmitry,
> 
> I've added the parisc-kernel-devel mailing list too, since I expect
> that *if* somebody is able to answer your question, it will probably
> be on the parisc-kernel-devel list...
> 
> On 31.03.2015 00:11, Dmitry the Zuryanovich wrote:
> > As I understand, E class (in my case - E25) has LASI based 53c710
> > which is handled by 53c700.c? (and in case of workstations wrapped by
> > lasi700 - what's for? Just for detecting it?)
> >
> > If I know that that's
> >
> > 5. Sahp Baat Kiuh SCSI at 0xfff74000 [56/52] {4, 0x0, 0x044, 0x00039}
> >
> > , is there a reason to try to
> >
> > #modprobe 53c700 clock=25  base=0xfff74000
> 
> Looking at drivers/scsi/lasi700.c, it seems the lasi700 driver sets
> additional flags like force_le_on_be, chip710, burst_length and so on.
> In addition, the IRQs gets connected via drivers/parisc/lasi.c (see lasi_choose_irq()).
> I assume you would need at least a "case 0x39" in there.
> But James is the expert on the 53c700 driver, so he might know more... ?

Well, I can tell you that the modprobe won't work.  The 53c700 is
designed to be a wrapper driver, meaning that it relies on another
driver (like lasi700 or zorro7xxx or something) to set up the interrupt
line and the register mappings.

Probably all that's needed is a tiny wrapper driver for whatever bus
it's on (what bus is this?).  If you look at lasi, it basically pulls
the parameters out of firmware, translates the lasi specific IRQ setting
and enables the GSC interrupt (plus some LED stuff specific to lasi).

> > [correct my syntax] and may I expect this to work somehow?
> >
> > What's the problem with  SCSI on E class? Is that a need of some
> > driver which calls 53c700.c with correct parameters taken from
> > hardware? Can it be just passed to module via parameter or the
> > problem is deeper?
> 
> It might be deeper.
> In  arch/parisc/kernel/hardware.c I see:
>          {HPHW_A_DMA, 0x044, 0x00039, 0x80, "Sahp Baat Kiuh SCSI"},

That tells us it's on the Precision bus, doesn't it?

> while other SCSI drivers on LASI have HPHW_FIO, e.g.:
> 	{HPHW_FIO, 0x016, 0x00082, 0x0, "Gecko Core SCSI"},
> 
> 
> For reference, here is the full dmesg from Dmitry when he tried to boot his E25
> machine with a recent Linux kernel:
> ftp://parisc.parisc-linux.org/dmesg/E25.dmesg  (http://pastebin.com/sXqUjVub)

There's not enough information there to reverse engineer the information
needed for the driver.

The best docs we have:

https://parisc.wiki.kernel.org/images-parisc/5/52/E-class_hpjournal.pdf

Say the SCSI controller here is precision attached not lasi attached
(the ethernet is on the lasi, though).  Unfortunately, we don't have any
documents at all in the archive for the precision bus that I can find.

James

Re: E Class and SCSI (53c700 driver)

[John David Anglin]

On 2015-04-02, at 4:59 AM, James Bottomley wrote:

> The best docs we have:
> 
> https://parisc.wiki.kernel.org/images-parisc/5/52/E-class_hpjournal.pdf
> 
> Say the SCSI controller here is precision attached not lasi attached
> (the ethernet is on the lasi, though).  Unfortunately, we don't have any
> documents at all in the archive for the precision bus that I can find.

It's also known as HP-IB and quite a few devices were made to attach to it.  Maybe something
can be found with Google.

Dave
--
John David Anglin	dave.anglin@bell.net

Re: E Class and SCSI (53c700 driver)

[James Bottomley]

On Thu, 2015-04-02 at 07:43 -0400, John David Anglin wrote:
> On 2015-04-02, at 4:59 AM, James Bottomley wrote:
> 
> > The best docs we have:
> > 
> > https://parisc.wiki.kernel.org/images-parisc/5/52/E-class_hpjournal.pdf
> > 
> > Say the SCSI controller here is precision attached not lasi attached
> > (the ethernet is on the lasi, though).  Unfortunately, we don't have any
> > documents at all in the archive for the precision bus that I can find.
> 
> It's also known as HP-IB and quite a few devices were made to attach to it.  Maybe something
> can be found with Google.

Well, what do you know: HP has a computer museum full of obsolete
documentation.

http://www.hpmuseum.net/collection_document.php

If anyone can identify the programming specs for the HP-PB (or better
still the actual SCSI card) we might be in business.

James

Re: E Class and SCSI (53c700 driver)

[John David Anglin]

On 2015-04-02, at 7:58 AM, James Bottomley wrote:

> On Thu, 2015-04-02 at 07:43 -0400, John David Anglin wrote:
>> On 2015-04-02, at 4:59 AM, James Bottomley wrote:
>> 
>>> The best docs we have:
>>> 
>>> https://parisc.wiki.kernel.org/images-parisc/5/52/E-class_hpjournal.pdf
>>> 
>>> Say the SCSI controller here is precision attached not lasi attached
>>> (the ethernet is on the lasi, though).  Unfortunately, we don't have any
>>> documents at all in the archive for the precision bus that I can find.
>> 
>> It's also known as HP-IB and quite a few devices were made to attach to it.  Maybe something
>> can be found with Google.
> 
> Well, what do you know: HP has a computer museum full of obsolete
> documentation.
> 
> http://www.hpmuseum.net/collection_document.php
> 
> If anyone can identify the programming specs for the HP-PB (or better
> still the actual SCSI card) we might be in business.


It seems to be a standard: IEEE-488-1978.

Dave
--
John David Anglin	dave.anglin@bell.net

Re: E Class and SCSI (53c700 driver)

[Grant Grundler]

On Thu, Apr 2, 2015 at 4:58 AM, James Bottomley
<James.Bottomley@hansenpartnership.com> wrote:
...
> Well, what do you know: HP has a computer museum full of obsolete
> documentation.
>
> http://www.hpmuseum.net/collection_document.php

oh wow...

> If anyone can identify the programming specs for the HP-PB (or better
> still the actual SCSI card) we might be in business.

This doc will help a lot:
    3000-900_9000-800_OnlineDiagnosticsSubManual_VolIII_LAN-SCSI-MUX_09740-90034_324pages_Jul91.pdf

Look at Section 18 "SCSIPBA" - register descriptions starting on page
18-21. Register overview on 18-33 (page 216 of the PDF).

And I finally found the name of the SCSI chip in the above doc: SPIFI
(v3).  The "Fast-Wide HVD" NIO card uses SPIFI v4.

Knowing that name also helped find the Net BSD driver (again):
    http://lists.parisc-linux.org/pipermail/parisc-linux/2002-June/016557.html
    http://lists.parisc-linux.org/pipermail/parisc-linux/2002-June/016562.html
    http://lists.parisc-linux.org/pipermail/parisc-linux/2002-June/016563.html
    http://lists.parisc-linux.org/pipermail/parisc-linux/2002-June/016576.html

and looks like NetBSD "newsmips" did support SPIFI-3:
   http://netbsd.mirrors.treibsand.com/ports/newsmips/

cheers,
grant

ps. It's embarrassing I couldn't remember the name of the SCSI chip: I
worked on HPUX SPIFI-4 driver between 1994-1996.

Re: E Class and SCSI (53c700 driver)

[Dmitry the Zuryanovich]

> And I finally found the name of the SCSI chip in the above doc: SPIFI
> (v3).  The "Fast-Wide HVD" NIO card uses SPIFI v4.


http://xepb.org/dtz/2015E25.html - here is photos of the boards, if this 
helps.

-- 
Slack@office

Re: E Class and SCSI (53c700 driver)

[James Bottomley]

On Fri, 2015-04-03 at 15:22 +0300, Dmitry the Zuryanovich wrote:
> > And I finally found the name of the SCSI chip in the above doc: SPIFI
> > (v3).  The "Fast-Wide HVD" NIO card uses SPIFI v4.
> 
> 
> http://xepb.org/dtz/2015E25.html - here is photos of the boards, if this 
> helps.

Well, I think that tells us it's not Symbios (or NCR as it was then)
53c700 based.  It looks like there's a custom HP ASIC, probably for the
PB connection and a Xilinx FPGA which may be configured to run the bus.

James

Re: E Class and SCSI (53c700 driver)

[Grant Grundler]

On Thu, Apr 2, 2015 at 1:59 AM, James Bottomley
<James.Bottomley@hansenpartnership.com> wrote:
....
> Probably all that's needed is a tiny wrapper driver for whatever bus
> it's on (what bus is this?).  If you look at lasi, it basically pulls
> the parameters out of firmware, translates the lasi specific IRQ setting
> and enables the GSC interrupt (plus some LED stuff specific to lasi).
...
>> In  arch/parisc/kernel/hardware.c I see:
>>          {HPHW_A_DMA, 0x044, 0x00039, 0x80, "Sahp Baat Kiuh SCSI"},
>
> That tells us it's on the Precision bus, doesn't it?

Yes. "E-class" was one of the "alphabet soup" servers with HP-PB
(IIRC, not HP-IB) slots. Since the system uses PA-7100LC, the "system
interconnect" is GSC and an HP-PB bus converter provided access to the
HP-PB slots.

I don't believe the 53c710 was connected on E-class servers based on this table:
    http://www.parisc-linux.org/documentation/hp9000_models.html


                                      CPU    BUILT-IN      ADD-ON SCSI I/O
MODEL STRING     PROJECT NAME        TYPE    SE, FWD/TYPE  SE, FWD, GSC-FWD
------------     ------------------  -----   ------------  ----------------
...
806/E25          WB Orville (48 Mhz) 7100LC  NIO, None     NIO(1),NIO(2),NA

("BUILT-IN" and "ADD-ON" SCSI both are NIO - aka HP-PB)

...
> The best docs we have:
> https://parisc.wiki.kernel.org/images-parisc/5/52/E-class_hpjournal.pdf

Also check on:
    http://www.openpa.net/systems/index.html

or more specifically:
   http://www.openpa.net/systems/hp-9000_e25_e35_e45_e55.html

> Say the SCSI controller here is precision attached not lasi attached
> (the ethernet is on the lasi, though).  Unfortunately, we don't have any
> documents at all in the archive for the precision bus that I can find.

I don't recall any HP-PB specific docs being released.  What I can say
is the HP-PB devices used transaction based interrupt routing similar
to how Dino and other PCI host controllers generated interrupts. So no
IRQ lines are used for HP-PB. And IIRC, there is also no IOMMU in
D/E/F/H/I series servers (meaning they are NOT DMA/Cache coherent
systems).

Lastly, the HP-PB SCSI adapters used a proprietary HP designed (not
Symbios designed) SCSI controller. No docs were ever published for
those either. The "fast-narrow SE" controller (28655A) was directly
controlled by the host computer and was not particularly efficient
(IIRC: 3 interrupts per transaction). The "fast-Wide HVD" (28696A) had
an i960 on board to manage the "version 4" chip. SCSI bus control was
handed to the host computer if anything happened that the i960
couldn't handle.

cheers,
grant

ps. I also found the manual for 28655A:
   http://pdfs.icecat.biz/pdf/22940335-3766.pdf