Skip to contentSkip to conversation

IA64 Linux issues

Linux-ia64 Van Maren, Kevin 2 participants 4 messages Jul 28, 2001 – Jul 31, 2001

lore
Share & tools
lore
#0 lore raw

I'm running Caldera's 6/20/01 snapshot (modified 2.4.4 kernel) on a 4X, 8GB Lion.

Under even "light" I/O load (mkfs to 2-4 disks) I get horiffic behavior -- basically what is described on this thread:
http://www.cs.helsinki.fi/linux/linux-kernel/2001-20/0763.html
[I/O drops down to < 300 I/Os per second (vmstat), but all the CPUs are pegged 99%+ system time.]

That is after I first used "swiotlb240" at elilo to avoid the panic, and then hacked the qlogicfc driver (patch at bottom) and applied the latest Adaptec driver patch (6.2.0) from freebsd.org to get 39-bit DMA. Problem occurs with 39160 & QL2200.

I tried upgrading to 2.4.7+IA64-010726, but when I did that, "init" dies:
init[1]: IA-64 Illegal operation fault 0
[...]

  Kernel panic: Attempted to kill init!

So i tried 2.4.5+IA64-010626. Once I fixed two bugs in smp.c that prevented compilation, it seemed to work. Except that the I/O peformance was just as bad.

qlogicfc.c patch:
doesn't do this in 2.4.7, which seems odd, since the driver *appears* to support 64-bit DMA.

Anyway, with the following patch I appear to get 64-bit DMA enabled (at least I don't get a panic on the 8GB IA-64 box running QL2200 cards w/o tweaking the swiotlb array size).

qlogicfc.c renamed from qlogicfc.c.orig +3 -0
@@ -2079,6 +2079,9 @@ static int isp2x00_init(struct Scsi_Host
        sh->irq = irq;
        sh->io_port = io_base;
 
+       if (pci_set_dma_mask(pdev, (dma_addr_t) 0xFFFFffffFFFFffffUL))
+               printk("qlogicfc%d: Failed to update dma_mask\n",
+                       hostdata->host_id);
        LEAVE("isp2x00_init");
 
        return 0;
#1 lore raw
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

"Van" = Van Maren, Kevin <[email protected]> writes:

Van> I'm running Caldera's 6/20/01 snapshot (modified 2.4.4 kernel) on Van> a 4X, 8GB Lion.

Van> Under even "light" I/O load (mkfs to 2-4 disks) I get horiffic Van> behavior -- basically what is described on this thread:
Van> http://www.cs.helsinki.fi/linux/linux-kernel/2001-20/0763.html
Van> [I/O drops down to < 300 I/Os per second (vmstat), but all the Van> CPUs are pegged 99%+ system time.]

Van> That is after I first used "swiotlb240" at elilo to avoid the Van> panic, and then hacked the qlogicfc driver (patch at bottom) and Van> applied the latest Adaptec driver patch (6.2.0) from freebsd.org Van> to get 39-bit DMA. Problem occurs with 39160 & QL2200.

I'd stay away from the qlogicfc.c driver, it's got all sorts of limitations.

Jes

#2 lore raw
In reply to Van Maren, Kevin
Quoted 5 lines

I'd stay away from the qlogicfc.c driver, it's got all sorts of limitations.

Jes

Could you be more specific? I have problems with the Adaptec as well (even after using the version with 39-bit DMA), and it [horrific I/O performance] doesn't appear to be a driver issue.

Kevin

#3 lore raw
In reply to Van Maren, Kevin
Quoted 6 lines
Quoted 6 lines
Quoted 1 line
Quoted 1 line
Quoted 1 line

"Van" = Van Maren, Kevin <[email protected]> writes:

I'd stay away from the qlogicfc.c driver, it's got all sorts of
limitations.

Jes

Van> Could you be more specific? I have problems with the Adaptec as Van> well (even after using the version with 39-bit DMA), and it Van> [horrific I/O performance] doesn't appear to be a driver issue.

I can't remember all the details, but it includes bad error handling, interrupt handling, lack of functionality.

Jes