Kernel or patch suggestions for DMA enables Marvell Driver

Kernel or patch suggestions for DMA enables Marvell Driver

[Terrence Martin]

Hi,

I have tried several development kernels to try to get the DMA version 
of the  Marvell  SATA driver to work. The problem is that other problems 
with the kernels have made them unsuitable for use. For example I used 
the latest 2.6.14git kernel as well as the most recent patch for 
2.6.15-rc6. Both of which had problems.

Could someone suggest a version of the kernel that might work, or 
alternatively is it possible to extract the DMA supported driver itself 
for compilation in another version? For example 2.6.14.4.

The official kernel driver is the closest I have gotten to getting this 
controller supported on my system. The source code from Marvell and 
other sources seems to not want to compile at all.

Here is the card I have from lspci

03:04.0 RAID bus controller: Marvell Technology Group Ltd. MV88SX6041 
4-port SATA II PCI-X Controller (rev 03)

Thanks,

Terrence

Re: Kernel or patch suggestions for DMA enables Marvell Driver

[Sebastian Kuzminsky]

Terrence Martin <tmartin@physics.ucsd.edu> wrote:
> I have tried several development kernels to try to get the DMA version 
> of the  Marvell  SATA driver to work. The problem is that other problems 
> with the kernels have made them unsuitable for use. For example I used 
> the latest 2.6.14git kernel as well as the most recent patch for 
> 2.6.15-rc6. Both of which had problems.
> 
> Could someone suggest a version of the kernel that might work, or 
> alternatively is it possible to extract the DMA supported driver itself 
> for compilation in another version? For example 2.6.14.4.
> 
> The official kernel driver is the closest I have gotten to getting this 
> controller supported on my system. The source code from Marvell and 
> other sources seems to not want to compile at all.
> 
> Here is the card I have from lspci
> 
> 03:04.0 RAID bus controller: Marvell Technology Group Ltd. MV88SX6041 
> 4-port SATA II PCI-X Controller (rev 03)

I have a 6081 (Marvell Technology Group Ltd. MV88SX6081 8-port SATA II
PCI-X Controller (rev 09)), it works fine under 2.6.15-rc5.  No hotplug,
and a few wierd problems here and there.

I'm getting ~55 MB/s/disk off each of 4 disks.  220 MB/s total read.
It slows way down when you start writing to them too, of course.


-- 
Sebastian Kuzminsky

Re: Kernel or patch suggestions for DMA enables Marvell Driver

[Terrence Martin]

Sebastian Kuzminsky wrote:
> Terrence Martin <tmartin@physics.ucsd.edu> wrote:
>   
>> I have tried several development kernels to try to get the DMA version 
>> of the  Marvell  SATA driver to work. The problem is that other problems 
>> with the kernels have made them unsuitable for use. For example I used 
>> the latest 2.6.14git kernel as well as the most recent patch for 
>> 2.6.15-rc6. Both of which had problems.
>>
>> Could someone suggest a version of the kernel that might work, or 
>> alternatively is it possible to extract the DMA supported driver itself 
>> for compilation in another version? For example 2.6.14.4.
>>
>> The official kernel driver is the closest I have gotten to getting this 
>> controller supported on my system. The source code from Marvell and 
>> other sources seems to not want to compile at all.
>>
>> Here is the card I have from lspci
>>
>> 03:04.0 RAID bus controller: Marvell Technology Group Ltd. MV88SX6041 
>> 4-port SATA II PCI-X Controller (rev 03)
>>     
>
> I have a 6081 (Marvell Technology Group Ltd. MV88SX6081 8-port SATA II
> PCI-X Controller (rev 09)), it works fine under 2.6.15-rc5.  No hotplug,
> and a few wierd problems here and there.
>
> I'm getting ~55 MB/s/disk off each of 4 disks.  220 MB/s total read.
> It slows way down when you start writing to them too, of course.
>
>
>   
Excellent, I was just about to try rc5, rc6 did not work very well at 
all. I will post at how it works for me.

Terrence

Re: Kernel or patch suggestions for DMA enables Marvell Driver

[Terrence Martin]

Terrence Martin wrote:
> Sebastian Kuzminsky wrote:
>> Terrence Martin <tmartin@physics.ucsd.edu> wrote:
>>  
>>> I have tried several development kernels to try to get the DMA 
>>> version of the  Marvell  SATA driver to work. The problem is that 
>>> other problems with the kernels have made them unsuitable for use. 
>>> For example I used the latest 2.6.14git kernel as well as the most 
>>> recent patch for 2.6.15-rc6. Both of which had problems.
>>>
>>> Could someone suggest a version of the kernel that might work, or 
>>> alternatively is it possible to extract the DMA supported driver 
>>> itself for compilation in another version? For example 2.6.14.4.
>>>
>>> The official kernel driver is the closest I have gotten to getting 
>>> this controller supported on my system. The source code from Marvell 
>>> and other sources seems to not want to compile at all.
>>>
>>> Here is the card I have from lspci
>>>
>>> 03:04.0 RAID bus controller: Marvell Technology Group Ltd. 
>>> MV88SX6041 4-port SATA II PCI-X Controller (rev 03)
>>>     
>>
>> I have a 6081 (Marvell Technology Group Ltd. MV88SX6081 8-port SATA II
>> PCI-X Controller (rev 09)), it works fine under 2.6.15-rc5.  No hotplug,
>> and a few wierd problems here and there.
>>
>> I'm getting ~55 MB/s/disk off each of 4 disks.  220 MB/s total read.
>> It slows way down when you start writing to them too, of course.
>>
>>
>>   
> Excellent, I was just about to try rc5, rc6 did not work very well at 
> all. I will post at how it works for me.
>
> Terrence

Nope it appears I have the same problem, the system hangs on reboot with 
rc5 as well as rc6.

Is it possible to grab the drivers out of the newer kernels and put them 
into one I know works like 2.6.14.4?

Terrence

Re: Kernel or patch suggestions for DMA enables Marvell Driver

[Sebastian Kuzminsky]

Terrence Martin <tmartin@physics.ucsd.edu> wrote:
> Nope it appears I have the same problem, the system hangs on reboot with 
> rc5 as well as rc6.

Are you using Raids groups as PVs with LVM?  I'm seeing some hangs with
that configuration in 2.6.15-rc5.  I'm still chasing after it, might be
operator error...


-- 
Sebastian Kuzminsky

Re: Kernel or patch suggestions for DMA enables Marvell Driver

[Terrence Martin]

Sebastian Kuzminsky wrote:

>Terrence Martin <tmartin@physics.ucsd.edu> wrote:
>  
>
>>Nope it appears I have the same problem, the system hangs on reboot with 
>>rc5 as well as rc6.
>>    
>>
>
>Are you using Raids groups as PVs with LVM?  I'm seeing some hangs with
>that configuration in 2.6.15-rc5.  I'm still chasing after it, might be
>operator error...
>
>
>  
>
I am not use LVM but I do use raid 1 software.

Actually the machine eventually booted, it just took a while.

The errors I get with 2.6.15-r5 are

sata_mv 0000:03:04.0: version 0.5
ACPI: PCI Interrupt 0000:03:04.0[A] -> GSI 56 (level, low) -> IRQ 185
sata_mv 0000:03:04.0: Applying B2 workarounds to unknown rev
sata_mv 0000:03:04.0: 32 slots 4 ports unknown mode IRQ via MSI
ata6: SATA max UDMA/133 cmd 0x0 ctl 0xF88A2120 bmdma 0x0 irq 185
ata7: SATA max UDMA/133 cmd 0x0 ctl 0xF88A4120 bmdma 0x0 irq 185
ata8: SATA max UDMA/133 cmd 0x0 ctl 0xF88A6120 bmdma 0x0 irq 185
ata9: SATA max UDMA/133 cmd 0x0 ctl 0xF88A8120 bmdma 0x0 irq 185
ata6: no device found (phy stat 00000000)
scsi5 : sata_mv
ata7: no device found (phy stat 00000000)
scsi6 : sata_mv
ata8: dev 0 cfg 49:2f00 82:346b 83:7d01 84:4023 85:3469 86:3c01 87:4023 
88:407f
ata8: dev 0 ATA-7, max UDMA/133, 390721968 sectors: LBA48
ata8: slow completion (cmd ef)
ata8: dev 0 configured for UDMA/133
scsi7 : sata_mv
ata9: dev 0 cfg 49:2f00 82:346b 83:7d01 84:4023 85:3469 86:3c01 87:4023 
88:407f
ata9: dev 0 ATA-7, max UDMA/133, 390721968 sectors: LBA48
ata9: slow completion (cmd ef)
ata9: dev 0 configured for UDMA/133
scsi8 : sata_mv
  Vendor: ATA       Model: ST3200826AS       Rev: 3.03
  Type:   Direct-Access                      ANSI SCSI revision: 05
SCSI device sdc: 390721968 512-byte hdwr sectors (200050 MB)
SCSI device sdc: drive cache: write back
SCSI device sdc: 390721968 512-byte hdwr sectors (200050 MB)
SCSI device sdc: drive cache: write back
 sdc:<3>ata8: Entering mv_eng_timeout
mmio_base f8880000 ap f6b91314 qc f6b917f0 scsi_cmnd f6393e00 &cmnd f6393e44
ata8: status=0x50 { DriveReady SeekComplete }
ata8: error=0x01 { AddrMarkNotFound }
sdc: Current: sense key: No Sense
    Additional sense: No additional sense information
 unknown partition table
sd 7:0:0:0: Attached scsi disk sdc
  Vendor: ATA       Model: ST3200826AS       Rev: 3.03
  Type:   Direct-Access                      ANSI SCSI revision: 05
SCSI device sdd: 390721968 512-byte hdwr sectors (200050 MB)
SCSI device sdd: drive cache: write back
SCSI device sdd: 390721968 512-byte hdwr sectors (200050 MB)
SCSI device sdd: drive cache: write back
 sdd:<3>ata8: Entering mv_eng_timeout
mmio_base f8880000 ap f6b91314 qc f6b917f0 scsi_cmnd f6b96080 &cmnd f6b960c4
ata9: Entering mv_eng_timeout
mmio_base f8880000 ap f679c314 qc f679c7f0 scsi_cmnd f6393e00 &cmnd f6393e44
ata8: status=0x50 { DriveReady SeekComplete }
ata8: error=0x01 { AddrMarkNotFound }
sdc: Current: sense key: No Sense
    Additional sense: No additional sense information
ata9: status=0x50 { DriveReady SeekComplete }
ata9: error=0x01 { AddrMarkNotFound }
sdd: Current: sense key: No Sense
    Additional sense: No additional sense information
 unknown partition table
sd 8:0:0:0: Attached scsi disk sdd

This is two seagate 200GB disks connected to two of the Marvell SATA 
ports on this Supermicro Xeon motherboard. The other two sata 
controllers are Intel which work fine.

The disks otherwise are brand new out of the box. I have never seen that 
error=0x01 { AddrMarkNotFound } before. Any ideas?

Terrence

Re: Kernel or patch suggestions for DMA enables Marvell Driver

[Sebastian Kuzminsky]

Terrence Martin <tmartin@physics.ucsd.edu> wrote:
> ata8: dev 0 ATA-7, max UDMA/133, 390721968 sectors: LBA48
> ata8: slow completion (cmd ef)
> ata8: dev 0 configured for UDMA/133
...
> SCSI device sdc: drive cache: write back
>  sdc:<3>ata8: Entering mv_eng_timeout
> mmio_base f8880000 ap f6b91314 qc f6b917f0 scsi_cmnd f6393e00 &cmnd f6393e44
> ata8: status=0x50 { DriveReady SeekComplete }
> ata8: error=0x01 { AddrMarkNotFound }
> sdc: Current: sense key: No Sense
>     Additional sense: No additional sense information
>  unknown partition table

I've seen this error on my system.  It went away and things work well now.
I did two things at the same time and I'm not sure which one fixed things.

My system has two PCI 32/33 slots and three PCI-X 64/66 slots.


Non-working configuration:

      PCI slot 1: Promise TX2 
      PCI slot 2: Adaptec 2930 (5V card)
    PCI-X slot 1: Promise TX2
    PCI-X slot 2: Marvell 6081
    PCI-X slot 3: empty


Working configuration:

      PCI slot 1: Promise TX2
      PCI slot 2: Promise TX2
    PCI-X slot 1: empty
    PCI-X slot 2: empty
    PCI-X slot 3: Marvell 6081


Could be getting rid of the 5V Adaptec card, could be taking all the
non-PCI-X cards out of the PCI-X slots.  Could be something else.
I'm not sure.  Good luck!


-- 
Sebastian Kuzminsky

Re: Kernel or patch suggestions for DMA enables Marvell Driver

[Terrence Martin]

In my case this is actually an onboard Marvell Sata controller. There is 
nothing in the two PCI-X slots at all. :(

I am going to ask Supermicro what the story is. Even the driver from 
Marvell that I have code for, of which I have found several versions, 
refuses to work correctly either on the Xeon system below or some new 
Opteron systems.

In any case I am happy to work with whoever is making the linux kernel 
driver for marvell to try to get to the bottom of the issue. I could 
even provide a login on a dedicated system if that would help.

Cheers,

Terrence



Sebastian Kuzminsky wrote:

>Terrence Martin <tmartin@physics.ucsd.edu> wrote:
>  
>
>>ata8: dev 0 ATA-7, max UDMA/133, 390721968 sectors: LBA48
>>ata8: slow completion (cmd ef)
>>ata8: dev 0 configured for UDMA/133
>>    
>>
>...
>  
>
>>SCSI device sdc: drive cache: write back
>> sdc:<3>ata8: Entering mv_eng_timeout
>>mmio_base f8880000 ap f6b91314 qc f6b917f0 scsi_cmnd f6393e00 &cmnd f6393e44
>>ata8: status=0x50 { DriveReady SeekComplete }
>>ata8: error=0x01 { AddrMarkNotFound }
>>sdc: Current: sense key: No Sense
>>    Additional sense: No additional sense information
>> unknown partition table
>>    
>>
>
>I've seen this error on my system.  It went away and things work well now.
>I did two things at the same time and I'm not sure which one fixed things.
>
>My system has two PCI 32/33 slots and three PCI-X 64/66 slots.
>
>
>Non-working configuration:
>
>      PCI slot 1: Promise TX2 
>      PCI slot 2: Adaptec 2930 (5V card)
>    PCI-X slot 1: Promise TX2
>    PCI-X slot 2: Marvell 6081
>    PCI-X slot 3: empty
>
>
>Working configuration:
>
>      PCI slot 1: Promise TX2
>      PCI slot 2: Promise TX2
>    PCI-X slot 1: empty
>    PCI-X slot 2: empty
>    PCI-X slot 3: Marvell 6081
>
>
>Could be getting rid of the 5V Adaptec card, could be taking all the
>non-PCI-X cards out of the PCI-X slots.  Could be something else.
>I'm not sure.  Good luck!
>
>
>  
>

Re: Kernel or patch suggestions for DMA enables Marvell Driver

[Bogdan Costescu]

On Tue, 20 Dec 2005, Terrence Martin wrote:

> sata_mv 0000:03:04.0: 32 slots 4 ports unknown mode IRQ via MSI

This line seems to suggest that you have MSI enabled in your kernel 
config. Both me and Jeff Garzik had to disable it in order to get the 
driver to work with our Marvell controllers. However, your log shows 
the driver proceeding further than in my case, so maybe you have a 
setup that would allow using MSI - what mainboard is this ? (you 
mentioned Supermicro, but not the model number, unless I missed it...)

-- 
Bogdan Costescu

IWR - Interdisziplinaeres Zentrum fuer Wissenschaftliches Rechnen
Universitaet Heidelberg, INF 368, D-69120 Heidelberg, GERMANY
Telephone: +49 6221 54 8869, Telefax: +49 6221 54 8868
E-mail: Bogdan.Costescu@IWR.Uni-Heidelberg.De

Re: Kernel or patch suggestions for DMA enables Marvell Driver

[Terrence Martin]

Bogdan Costescu wrote:
> On Tue, 20 Dec 2005, Terrence Martin wrote:
>
>> sata_mv 0000:03:04.0: 32 slots 4 ports unknown mode IRQ via MSI
>
> This line seems to suggest that you have MSI enabled in your kernel 
> config. Both me and Jeff Garzik had to disable it in order to get the 
> driver to work with our Marvell controllers. However, your log shows 
> the driver proceeding further than in my case, so maybe you have a 
> setup that would allow using MSI - what mainboard is this ? (you 
> mentioned Supermicro, but not the model number, unless I missed it...)
>
These are X6DHR-TG and X6DVL-EG2

I also have a similar problem with my Supermicro opterons which as far 
as I can tell have the same controllers. The opterons are H8DAR-T

I can disable MSI and see if that helps.

Terrence

Re: Kernel or patch suggestions for DMA enables Marvell Driver

[Terrence Martin]

Bogdan Costescu wrote:
> On Tue, 20 Dec 2005, Terrence Martin wrote:
>
>> sata_mv 0000:03:04.0: 32 slots 4 ports unknown mode IRQ via MSI
>
> This line seems to suggest that you have MSI enabled in your kernel 
> config. Both me and Jeff Garzik had to disable it in order to get the 
> driver to work with our Marvell controllers. However, your log shows 
> the driver proceeding further than in my case, so maybe you have a 
> setup that would allow using MSI - what mainboard is this ? (you 
> mentioned Supermicro, but not the model number, unless I missed it...)
>
I am not sure if the email I sent got through. I am resending the 
motherboard info as well as I have some good news.

The are three motherboards

Supermicro X6DVL-EG2 EM64T
Supermicro X6DHR-TG EM64T
Supermicro  H8DAR-T Opteron

So I compiled 2.6.15-r5 without MSI enabled. The system booted and the 
disks are up and running. hdparm gives 65MB/s for a single disk. I am 
going to raid0 these disks and see what I can get when actually writing 
data.

Looking at MSI though it seems to indicate that without it I cannot get 
PCI-Express to work? It is not critical but I would like to be able to 
use PCI-E on the motherboard that supports it (two PCI-E Slots). I am 
happy to help with figuring out why there is this incompatibility with 
this feature.

Terrence

Re: Kernel or patch suggestions for DMA enables Marvell Driver

[Bogdan Costescu]

On Wed, 21 Dec 2005, Terrence Martin wrote:

> So I compiled 2.6.15-r5 without MSI enabled.

Hmm, so far all reports indicated that MSI had to be disabled... so 
maybe the Marvell controller is at fault.

> Looking at MSI though it seems to indicate that without it I cannot 
> get PCI-Express to work?

I think that INTx emulation should be present by default if MSI is not 
enabled, but I'm not sure...

-- 
Bogdan Costescu

IWR - Interdisziplinaeres Zentrum fuer Wissenschaftliches Rechnen
Universitaet Heidelberg, INF 368, D-69120 Heidelberg, GERMANY
Telephone: +49 6221 54 8869, Telefax: +49 6221 54 8868
E-mail: Bogdan.Costescu@IWR.Uni-Heidelberg.De

Re: Kernel or patch suggestions for DMA enables Marvell Driver

[Terrence Martin]

Bogdan Costescu wrote:

> On Wed, 21 Dec 2005, Terrence Martin wrote:
>
>> So I compiled 2.6.15-r5 without MSI enabled.
>
>
> Hmm, so far all reports indicated that MSI had to be disabled... so 
> maybe the Marvell controller is at fault.
>
>> Looking at MSI though it seems to indicate that without it I cannot 
>> get PCI-Express to work?
>
>
> I think that INTx emulation should be present by default if MSI is not 
> enabled, but I'm not sure...
>
Well I can see the PCI-E port on lspci. I can test them out with a PCI-E 
ethernet card on tuesday just ot make sure.

As for the card being a problem, possibly. I know I got one of my 
opterons to work with the marvell controller with the Marvell written 
driver and as far as I can tell RedHat enables MSI according to the 
kernel config in /boot. That was a binary driver though and on an 
Opteron, although the controller is claimed to be the same as what I 
have in the Xeons according to lspci.  I have not done the same steps 
for these Xeons. I have had little or no luck getting the marvell code 
to work with Redhat stock kernels in Centos 4.2 and was only able to use 
the Marvell driver as a binary compiled by supermicro with a 4.0 system.

I also do not see any mention of MSI in the instructions on how to 
install with the Marvell driver in RH FC.  
http://www.keffective.com/mvsata/ . At a glance it does not seem that 
MSI comes into play with the Marvell written driver, unfortunately that 
driver does not seem to like newer kernels, and I have not been able to 
get it to work myself when compiled from source. I am trying to track 
that problem down with Supermicro and my vendor.

Perhaps this might mean something to someone. With the Marvell source 
code I can get one of the versions to build with 2.6.15-rc5, however I 
get an error "insmod: error inserting 'mv_sata.ko': -1 Unknown symbol in 
module". Any ideas why that error?

Terrence
 

Re: Kernel or patch suggestions for DMA enables Marvell Driver

[Randy.Dunlap]

On Thu, 22 Dec 2005, Terrence Martin wrote:

> Bogdan Costescu wrote:
>
> > On Wed, 21 Dec 2005, Terrence Martin wrote:
> >
> >> So I compiled 2.6.15-r5 without MSI enabled.
> >
> >
> > Hmm, so far all reports indicated that MSI had to be disabled... so
> > maybe the Marvell controller is at fault.
> >
> >> Looking at MSI though it seems to indicate that without it I cannot
> >> get PCI-Express to work?
> >
> >
> > I think that INTx emulation should be present by default if MSI is not
> > enabled, but I'm not sure...
> >
> Well I can see the PCI-E port on lspci. I can test them out with a PCI-E
> ethernet card on tuesday just ot make sure.
>
> As for the card being a problem, possibly. I know I got one of my
> opterons to work with the marvell controller with the Marvell written
> driver and as far as I can tell RedHat enables MSI according to the
> kernel config in /boot. That was a binary driver though and on an
> Opteron, although the controller is claimed to be the same as what I
> have in the Xeons according to lspci.  I have not done the same steps
> for these Xeons. I have had little or no luck getting the marvell code
> to work with Redhat stock kernels in Centos 4.2 and was only able to use
> the Marvell driver as a binary compiled by supermicro with a 4.0 system.
>
> I also do not see any mention of MSI in the instructions on how to
> install with the Marvell driver in RH FC.
> http://www.keffective.com/mvsata/ . At a glance it does not seem that
> MSI comes into play with the Marvell written driver, unfortunately that
> driver does not seem to like newer kernels, and I have not been able to
> get it to work myself when compiled from source. I am trying to track
> that problem down with Supermicro and my vendor.

/proc/interrupts would show/tell whether MSI interrupts are
being used, in case you are not sure or wondering about that.

> Perhaps this might mean something to someone. With the Marvell source
> code I can get one of the versions to build with 2.6.15-rc5, however I
> get an error "insmod: error inserting 'mv_sata.ko': -1 Unknown symbol in
> module". Any ideas why that error?

Kernel message log (dmesg) should have more info, like
the missing symbol's name.

-- 
~Randy

Re: Kernel or patch suggestions for DMA enables Marvell Driver

[Terrence Martin]

Randy.Dunlap wrote:

>On Thu, 22 Dec 2005, Terrence Martin wrote:
>
>  
>
>>Bogdan Costescu wrote:
>>
>>    
>>
>>>On Wed, 21 Dec 2005, Terrence Martin wrote:
>>>
>>>      
>>>
>>>>So I compiled 2.6.15-r5 without MSI enabled.
>>>>        
>>>>
>>>Hmm, so far all reports indicated that MSI had to be disabled... so
>>>maybe the Marvell controller is at fault.
>>>
>>>      
>>>
>>>>Looking at MSI though it seems to indicate that without it I cannot
>>>>get PCI-Express to work?
>>>>        
>>>>
>>>I think that INTx emulation should be present by default if MSI is not
>>>enabled, but I'm not sure...
>>>
>>>      
>>>
>>Well I can see the PCI-E port on lspci. I can test them out with a PCI-E
>>ethernet card on tuesday just ot make sure.
>>
>>As for the card being a problem, possibly. I know I got one of my
>>opterons to work with the marvell controller with the Marvell written
>>driver and as far as I can tell RedHat enables MSI according to the
>>kernel config in /boot. That was a binary driver though and on an
>>Opteron, although the controller is claimed to be the same as what I
>>have in the Xeons according to lspci.  I have not done the same steps
>>for these Xeons. I have had little or no luck getting the marvell code
>>to work with Redhat stock kernels in Centos 4.2 and was only able to use
>>the Marvell driver as a binary compiled by supermicro with a 4.0 system.
>>
>>I also do not see any mention of MSI in the instructions on how to
>>install with the Marvell driver in RH FC.
>>http://www.keffective.com/mvsata/ . At a glance it does not seem that
>>MSI comes into play with the Marvell written driver, unfortunately that
>>driver does not seem to like newer kernels, and I have not been able to
>>get it to work myself when compiled from source. I am trying to track
>>that problem down with Supermicro and my vendor.
>>    
>>
>
>/proc/interrupts would show/tell whether MSI interrupts are
>being used, in case you are not sure or wondering about that.
>
>  
>
>>Perhaps this might mean something to someone. With the Marvell source
>>code I can get one of the versions to build with 2.6.15-rc5, however I
>>get an error "insmod: error inserting 'mv_sata.ko': -1 Unknown symbol in
>>module". Any ideas why that error?
>>    
>>
>
>Kernel message log (dmesg) should have more info, like
>the missing symbol's name.
>
>  
>
It does

mv_sata: Unknown symbol scsi_set_device

As for /proc/interupts, even if you do not have a PCI-E device would it 
report some numbers or? Both my Centos 4.2 default kernel as well as my 
kernel.org systems have MIS in the file but both are 0.

Terrence

Re: Kernel or patch suggestions for DMA enables Marvell Driver

[Randy.Dunlap]

On Thu, 22 Dec 2005, Terrence Martin wrote:

> Randy.Dunlap wrote:
>
> >On Thu, 22 Dec 2005, Terrence Martin wrote:
> >
> >
> >
> >>Bogdan Costescu wrote:
> >>
> >>
> >>
> >>>On Wed, 21 Dec 2005, Terrence Martin wrote:
> >>>
> >>>
> >>>
> >>>>So I compiled 2.6.15-r5 without MSI enabled.
> >>>>
> >>>>
> >>>Hmm, so far all reports indicated that MSI had to be disabled... so
> >>>maybe the Marvell controller is at fault.
> >>>
> >>>
> >>>
> >>>>Looking at MSI though it seems to indicate that without it I cannot
> >>>>get PCI-Express to work?
> >>>>
> >>>>
> >>>I think that INTx emulation should be present by default if MSI is not
> >>>enabled, but I'm not sure...
> >>>
> >>>
> >>>
> >>Well I can see the PCI-E port on lspci. I can test them out with a PCI-E
> >>ethernet card on tuesday just ot make sure.
> >>
> >>As for the card being a problem, possibly. I know I got one of my
> >>opterons to work with the marvell controller with the Marvell written
> >>driver and as far as I can tell RedHat enables MSI according to the
> >>kernel config in /boot. That was a binary driver though and on an
> >>Opteron, although the controller is claimed to be the same as what I
> >>have in the Xeons according to lspci.  I have not done the same steps
> >>for these Xeons. I have had little or no luck getting the marvell code
> >>to work with Redhat stock kernels in Centos 4.2 and was only able to use
> >>the Marvell driver as a binary compiled by supermicro with a 4.0 system.
> >>
> >>I also do not see any mention of MSI in the instructions on how to
> >>install with the Marvell driver in RH FC.
> >>http://www.keffective.com/mvsata/ . At a glance it does not seem that
> >>MSI comes into play with the Marvell written driver, unfortunately that
> >>driver does not seem to like newer kernels, and I have not been able to
> >>get it to work myself when compiled from source. I am trying to track
> >>that problem down with Supermicro and my vendor.
> >>
> >>
> >
> >/proc/interrupts would show/tell whether MSI interrupts are
> >being used, in case you are not sure or wondering about that.
> >
> >
> >
> >>Perhaps this might mean something to someone. With the Marvell source
> >>code I can get one of the versions to build with 2.6.15-rc5, however I
> >>get an error "insmod: error inserting 'mv_sata.ko': -1 Unknown symbol in
> >>module". Any ideas why that error?
> >>
> >>
> >
> >Kernel message log (dmesg) should have more info, like
> >the missing symbol's name.
> >
> >
> >
> It does
>
> mv_sata: Unknown symbol scsi_set_device

Well, like insmod reported, that symbol is no longer in the
kernel.  They need to provide an updated driver.

> As for /proc/interupts, even if you do not have a PCI-E device would it
> report some numbers or? Both my Centos 4.2 default kernel as well as my
> kernel.org systems have MIS in the file but both are 0.

MIS is not MSI.  Yes, even without a PCI-E device, the kernel
can support MSI interrupts and report them.  E.g., with my
system's ethernet adapter:

           CPU0       CPU1
  0:   77858078   77854028    IO-APIC-edge  timer
  1:     153775     156060    IO-APIC-edge  i8042
  8:          0          0    IO-APIC-edge  rtc
  9:          0          0   IO-APIC-level  acpi
 12:    1365742    1383425    IO-APIC-edge  i8042
 14:    2793679    2794767    IO-APIC-edge  ide0
 66:          0          0   IO-APIC-level  uhci_hcd:usb1, ehci_hcd:usb5
 74:       4252       3930   IO-APIC-level  HDA Intel
 82:    1430988          0         PCI-MSI  eth0
169:          0          0   IO-APIC-level  uhci_hcd:usb4
177:          3          0   IO-APIC-level  ohci1394
185:          0          0   IO-APIC-level  uhci_hcd:usb3
193:     357018     353269   IO-APIC-level  libata, uhci_hcd:usb2
NMI:      83754      30820
LOC:  155714088  155714281
ERR:          1
MIS:          0


-- 
~Randy