Skip to contentSkip to conversation

your mail

DISCUSSION Gérard Roudier 5 participants 11 messages Dec 14, 2001 – Dec 18, 2001 1 message not in the archive

lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): [email protected]

#1 lore raw
Quoted 10 lines · Kirk Alexander

[cc'ed to lkml and Gerard Roudier]

Hi Jens,

You asked people to send in reports of which drivers were broken by the removal of io_request_lock.

My system is a clunky old Digital Pentium Pro with a NCR53c810 rev 2 scsi controller, so it can't use the sym driver.

Use sym53c8xx_2 instead. This one uses 2 different firmwares,

- one based on sym53c8xx driver scripts called 'LOAD/STORE based' firmware,
- and another one that only uses generic scripts instructions and called 'GENERIC' firmware.

The GENERIC firmware has worked for me witn a 810 rev. 2.

I haven't this controller installed at the moment, but I can test the driver by forcing the driver to use the GENERIC scripts instead for any symbios chip.

You may let me know if sym53c8xx_2 still works with 810 rev 2.

Quoted 4 lines

I fixed the problem by seeing what the sym driver did i.e. the patch below
This may not be right at all, and I haven't had a chance to boot the kernel - but it did build OK.

The ncr53c8xx and sym53c8xx version 1 use the obsolete scsi eh handling. Moving the eh code from sym53c8xx_2 (version 2) to ncr53c8xx/sym53c8xx is quite feasible, but may-be it is just useless given sym53c8xx_2. For now, it seems that sym53c8xx_2 replaces both ncr/sym53c8xx without any loss of reliability and performance.

Gérard.

Quoted 9 lines

Cheers,
Kirk Alexander

P.S.
Please excuse me if this has already been fixed or posted, or if I've broken some lkml etiquette - first post I think after lurking off and on for ages. Also first time I've compiled a kernel that has only been out a few days!

You are welcome and you didn't break any etiquette. The lkml is a very open mailing list but it gets more than 200 postings a day. Thus the [email protected] list should be preferred for topics that address scsi specifically.

Quoted 29 lines

--- linux/drivers/scsi/sym53c8xx_comm.h Fri Dec 14
16:46:45 2001
+++ linux/drivers/scsi/sym53c8xx_comm.h Fri Dec 14
16:49:19 2001

(quoted diff) +9 -0
@@ -438,11 +438,20 @@
 #define	NCR_LOCK_NCB(np, flags)
 spin_lock_irqsave(&np->smp_lock, flags)
 #define	NCR_UNLOCK_NCB(np, flags)
 spin_unlock_irqrestore(&np->smp_lock, flags)
 
+#if LINUX_VERSION_CODE >= LinuxVersionCode(2,5,0)
+
+#define	NCR_LOCK_SCSI_DONE(np, flags) \
+		spin_lock_irqsave((np)->done_list->host, flags)
+#define	NCR_UNLOCK_SCSI_DONE(np, flags) \
+		spin_unlock_irqrestore((np)->done_list->host,
 flags)
+#else
+
 #define	NCR_LOCK_SCSI_DONE(np, flags) \
 		spin_lock_irqsave(&io_request_lock, flags)
 #define	NCR_UNLOCK_SCSI_DONE(np, flags) \
 		spin_unlock_irqrestore(&io_request_lock, flags)
 
+#endif

#else

#define NCR_LOCK_DRIVER(flags) do {
save_flags(flags); cli(); } while (0)

#2 lore raw
Quoted 7 lines · On Fri, 14 Dec 2001 17:46:37 +0100 (CET) Gérard Roudier <[email protected]>…
Quoted 3 lines

My system is a clunky old Digital Pentium Pro with a NCR53c810 rev 2 scsi controller, so it can't use the sym driver.

Use sym53c8xx_2 instead. This one uses 2 different firmwares, [...]
You may let me know if sym53c8xx_2 still works with 810 rev 2.

On my system it does. I have it as a second controller and am using sym-2 without troubles.

Regards,
Stephan

#3 lore raw
Quoted 7 lines · Stephan von Krawczynski
Quoted 4 lines · On Fri, 14 Dec 2001 17:46:37 +0100 (CET) Gérard Roudier <[email protected]>…
Quoted 3 lines

My system is a clunky old Digital Pentium Pro with a NCR53c810 rev 2 scsi controller, so it can't use the sym driver.

Quoted 6 lines
Quoted 3 lines

Use sym53c8xx_2 instead. This one uses 2 different firmwares, [...]
You may let me know if sym53c8xx_2 still works with 810 rev 2.

On my system it does. I have it as a second controller and am using sym-2 without troubles.

Thanks for your report.

Gérard.

#4 lore raw
In reply to Gérard Roudier

--- Gérard_Roudier <[email protected]> wrote: >

Quoted 1 line

[snip]

Quoted 3 lines

You may let me know if sym53c8xx_2 still works with 810 rev 2.

I tried the sym53c8xx_2 driver, put a fair load on the system (lots of sync'ing and swapping) and didn't seem to have any trouble.

Cheers,
Kirk Alexander

#5 lore raw
In reply to a message not in the archive
Quoted 10 lines · Gérard Roudier
Quoted 4 lines

I fixed the problem by seeing what the sym driver did i.e. the patch below
This may not be right at all, and I haven't had a chance to boot the kernel - but it did build OK.

The ncr53c8xx and sym53c8xx version 1 use the obsolete scsi eh handling. Moving the eh code from sym53c8xx_2 (version 2) to ncr53c8xx/sym53c8xx is quite feasible, but may-be it is just useless given sym53c8xx_2. For now, it seems that sym53c8xx_2 replaces both ncr/sym53c8xx without any loss of reliability and performance.

Gerard,

For 2.5, why don't we just yank old sym and ncr out of the kernel? Is there _any_ reason to keep the two older ones given your new driver handles it all?

Signature · 1 line
Jens Axboe
#6 lore raw
Quoted 12 lines · Jens Axboe
Quoted 10 lines · Gérard Roudier
Quoted 4 lines

I fixed the problem by seeing what the sym driver did i.e. the patch below
This may not be right at all, and I haven't had a chance to boot the kernel - but it did build OK.

The ncr53c8xx and sym53c8xx version 1 use the obsolete scsi eh handling. Moving the eh code from sym53c8xx_2 (version 2) to ncr53c8xx/sym53c8xx is quite feasible, but may-be it is just useless given sym53c8xx_2. For now, it seems that sym53c8xx_2 replaces both ncr/sym53c8xx without any loss of reliability and performance.

Quoted 5 lines

Gerard,

For 2.5, why don't we just yank old sym and ncr out of the kernel? Is there _any_ reason to keep the two older ones given your new driver handles it all?

On my side, there is obviously no reason to keep them in 2.5, as sym-2 is intended to replace them both. Personnaly I have switched to sym-2 on my systems since several months.

However, I donnot consider myself as the only owner of these drivers. The owners are all people that may need symbios chips support under Linux. My personnal vote, as a user/owner, is to remove them and rely for symbios chip support on sym-2.

--

Linux stable is a different issue. For this one, I would prefer the old drivers to remain in place for a longer time. However, I personnaly will not track bugs on old drivers if either,

- The problem also shows up in sym-2. Then I will try to fix sym-2,
- Or the problem simply doesn't occur in sym-2.

This will apply to problems reported directly by users or by packagers.

By the way, for now, I haven't received any report about sym-2 failing when sym-1 or ncr succeeds, and my feeling is that this could well be very unlikely.

But I can make mistakes, me too. :-)

Gérard.

#7 lore raw
Quoted 4 lines · On Fri, 14 Dec 2001, [ISO-8859-1] Gérard Roudier wrote:

By the way, for now, I haven't received any report about sym-2 failing when sym-1 or ncr succeeds, and my feeling is that this could well be very unlikely.

Ahemm -- well,
maybe I'm the first one. I have a symbios card, which is recognized by
lspci: SCSI storage controller: LSI Logic Corp. / Symbios Logic Inc. (formerly NCR) 53c810 (rev 23).

This card goes into an endless loop during parity-checking. So tried to disable it for the new sym53cxx in modules.conf:
options sym53c8xx mpar:n spar:n
This did not help in this case, however.

There have been so far three ways to solve this problem:
1. Use the very old ncr53c7,8 or so driver. This is working rather unreliable for me.
2. Use the ncr53c8xx, which works usually well
3. Use sym53c8xx(old) compiled with parity disabled

Probably there is a way around that, but somebody trying to install Linux from a SCSI-CDROM with this card for the first time will very likely not succeed. I have seen this with (for instance) Corel-Linux and FreeBSD (same driver).
NB Parity checking for me is not really all that important as there is no hardrive connected to that card. Only CD and scanner.

Peter B

          .         .
          |\_-^^^-_/|
          / (|)_(|) \
         ( === X === )
          \  ._|_.  /
           ^-_   _-^
              °°°
#8 lore raw
Quoted 22 lines · Peter Bornemann
Quoted 4 lines · On Fri, 14 Dec 2001, [ISO-8859-1] Gérard Roudier wrote:

By the way, for now, I haven't received any report about sym-2 failing when sym-1 or ncr succeeds, and my feeling is that this could well be very unlikely.

Ahemm -- well,
maybe I'm the first one. I have a symbios card, which is recognized by
lspci: SCSI storage controller: LSI Logic Corp. / Symbios Logic Inc. (formerly NCR) 53c810 (rev 23).

This card goes into an endless loop during parity-checking. So tried to disable it for the new sym53cxx in modules.conf:
options sym53c8xx mpar:n spar:n
This did not help in this case, however.

There have been so far three ways to solve this problem:
1. Use the very old ncr53c7,8 or so driver. This is working rather unreliable for me.
2. Use the ncr53c8xx, which works usually well
3. Use sym53c8xx(old) compiled with parity disabled

Quoted 6 lines

Probably there is a way around that, but somebody trying to install Linux from a SCSI-CDROM with this card for the first time will very likely not succeed. I have seen this with (for instance) Corel-Linux and FreeBSD (same driver).
NB Parity checking for me is not really all that important as there is no hardrive connected to that card. Only CD and scanner.

About what parity sort are you talking about ?
PCI parity ? SCSI parity ?

PCI parity checking is not an option. If it is this one, then your hardware is simply broken. For such broken hardwares that returns such spurious PCI parity error early during HBA probing, sym-2 can detect this and disable PCI parity checking. This has been reported to work well under FreeBSD. And sym-2 is also supposed to accept the manual disabling, either by compiled-in option or using the mpar=n boot-up options.

For SCSI parity, which is different matter, both drivers try to cope and still sym-2 should accept the spar=n boot-up option.

Could you, please, report me more accurate information. TIA,
Gérard.

#9 lore raw
Quoted 10 lines · On Fri, 14 Dec 2001, [ISO-8859-1] Gérard Roudier wrote:
Quoted 4 lines · Peter Bornemann

Ahemm -- well,
maybe I'm the first one. I have a symbios card, which is recognized by
lspci: SCSI storage controller: LSI Logic Corp. / Symbios Logic Inc. (formerly NCR) 53c810 (rev 23).

Could you, please, report me more accurate information. TIA,

Well, it seems I made my intention not very clear: I do not want You to fix something in the driver, I just wanted from You to leave the old ncr-driver in the kernel, just for the situation of a first install. I think no newbie with little knowledge will be able to install Linux (or, maybe, FreeBSD), when he happens to own such an controller. First, he won't be able to read very much on the screen, for the loop runs much too fast and second, he will not understand when he reads something about a sym53c8xx. Exactly for this case I think the old driver should be left in. If You want, You can tell him "Attention! Use of this driver deprecated. Contact Your support." or whatever seems appropriate. It is just about the first step to linuxland :-)

Hope I managed to make myself clear tris time

Peter B

          .         .
          |\_-^^^-_/|
          / (|)_(|) \
         ( === X === )
          \  ._|_.  /
           ^-_   _-^
              °°°
#10 lore raw
Quoted 25 lines · Peter Bornemann
Quoted 10 lines · On Fri, 14 Dec 2001, [ISO-8859-1] Gérard Roudier wrote:
Quoted 4 lines · Peter Bornemann

Ahemm -- well,
maybe I'm the first one. I have a symbios card, which is recognized by
lspci: SCSI storage controller: LSI Logic Corp. / Symbios Logic Inc. (formerly NCR) 53c810 (rev 23).

Could you, please, report me more accurate information. TIA,

Well, it seems I made my intention not very clear: I do not want You to fix something in the driver, I just wanted from You to leave the old ncr-driver in the kernel, just for the situation of a first install. I think no newbie with little knowledge will be able to install Linux (or, maybe, FreeBSD), when he happens to own such an controller. First, he won't be able to read very much on the screen, for the loop runs much too fast and second, he will not understand when he reads something about a sym53c8xx. Exactly for this case I think the old driver should be left in. If You want, You can tell him "Attention! Use of this driver deprecated. Contact Your support." or whatever seems appropriate. It is just about the first step to linuxland :-)

Quoted 1 line

Hope I managed to make myself clear tris time

I have limited time and am very bad in politics. I do prefer to have to deal with accurate technical issues. My english is also limited to this field.

You would have been clear if you had reported:

1) Which of the mpar= spar= and/or corresponding compiled-in options made your broken hardware works (with high risks of silent corruption).

2) If using the corresponding compiled-in option worked with sym-2.

3) Optionnaly, relevant messages printed by sym-2, even taken by hand, when the problem occurs.

+ any other pertinent information I cannot guess about you thought can help.

About FreeBSD, the only information I have is that the (sad) work-around I implemented and that is incorporated in sym-2 _did_ work around the problem of PCI parity error for people that did report results.

Could you be clear, as expected by technical group of discussions, please.

Gérard.

PS: The ncr53c8xx may just work since it does trust POST software to
enable PCI parity checking bit in PCI config space. But it seems that most POST shits donnot do so, leaving systems with the risk of silent data corruption in contradiction with either PCI specifications and user expectation.

#11 lore raw
In reply to a message not in the archive
Quoted 7 lines · Kirk Alexander

[cc'ed to lkml and Gerard Roudier]

Hi Jens,

You asked people to send in reports of which drivers were broken by the removal of io_request_lock.

Quoted 6 lines

My system is a clunky old Digital Pentium Pro with a NCR53c810 rev 2 scsi controller, so it can't use the sym driver. I fixed the problem by seeing what the sym driver did i.e. the patch below
This may not be right at all, and I haven't had a chance to boot the kernel - but it did build OK.

Missed your original post, it had no subject line. At first view, your patch looks correct. However, check the ->detect() routing and verify it's not assuming the lock is held there. That should be the only pitfall.

Minor nit pick -- since this driver is _in_ the 2.5 tree, there's no way the #ifdef would not hit. So the way I've been fixing these is to just always assume latest kernel.

I think this was already fixed though, but at least know you now you did it right :-)

Signature · 1 line
Jens Axboe