Hi,
i again took a look at the sgiwd93.c wback cache handling - I found some
space for improvement and cleanup:
- init_hpc_chain has to flush to memory anyways - Add an
dma_cache_wback_inv and delete from callers (Also a duplicate one
reported a couple of months back) This one is inlined and only used
in sgiwd93_detect which is __init)
- fill_hpc_chain used to hit "dma_cache_wback_inv" for the area it
sets up the hpc chains. This is not need in case of read as we only
would need to invalidate. Moved this to "dma_setup" and made the
decision to hit dma_cache_wback_inv or dma_cache_inv depending
on dma direction.
I am running this patch now with success. It should improve read
performance as we dont hit writeback on soon to be overwritten pages.
I havent done performance analysis with bonnie etc.
Index: drivers/scsi/sgiwd93.c
===================================================================
RCS file: /cvs/linux/drivers/scsi/sgiwd93.c,v
retrieving revision 1.28.2.1
diff -u -r1.28.2.1 sgiwd93.c
--- drivers/scsi/sgiwd93.c 2001/12/11 18:11:33 1.28.2.1
+++ drivers/scsi/sgiwd93.c 2001/12/13 15:15:45
@@ -4,10 +4,13 @@
* Copyright (C) 1996 David S. Miller (dm@engr.sgi.com)
* 1999 Andrew R. Baker (andrewb@uab.edu)
* - Support for 2nd SCSI controller on Indigo2
+ * 2001 Florian Lohoff (flo@rfc822.org)
+ * - Delete HPC scatter gather (Read corruption on
+ * multiple disks)
+ * - Cleanup wback cache handling
*
* (In all truth, Jed Schimmel wrote all this code.)
*
- * $Id: sgiwd93.c,v 1.19 2000/02/04 07:40:47 ralf Exp $
*/
#include <linux/init.h>
#include <linux/types.h>
@@ -85,7 +88,6 @@
unsigned long physaddr;
unsigned long count;
- dma_cache_wback_inv((unsigned long)addr,len);
physaddr = PHYSADDR(addr);
while (len) {
/*
@@ -104,7 +106,6 @@
static int dma_setup(Scsi_Cmnd *cmd, int datainp)
{
struct WD33C93_hostdata *hdata = (struct WD33C93_hostdata *)cmd->host->hostdata;
- const wd33c93_regs regs = hdata->regs;
struct hpc3_scsiregs *hregs = (struct hpc3_scsiregs *) cmd->host->base;
struct hpc_chunk *hcp = (struct hpc_chunk *) hdata->dma_bounce_buffer;
@@ -124,6 +125,7 @@
*/
if (cmd->SCp.ptr == NULL)
return 1;
+
fill_hpc_entries (&hcp, cmd->SCp.ptr,cmd->SCp.this_residual);
/* To make sure, if we trip an HPC bug, that we transfer
@@ -139,10 +141,14 @@
/* Start up the HPC. */
hregs->ndptr = PHYSADDR(hdata->dma_bounce_buffer);
- if(datainp)
+ if(datainp) {
+ dma_cache_wback_inv((unsigned long) cmd->SCp.ptr, cmd->SCp.this_residual);
hregs->ctrl = (HPC3_SCTRL_ACTIVE);
- else
+ } else {
+ dma_cache_inv((unsigned long) cmd->SCp.ptr, cmd->SCp.this_residual);
hregs->ctrl = (HPC3_SCTRL_ACTIVE | HPC3_SCTRL_DIR);
+ }
+
return 0;
}
@@ -150,7 +156,6 @@
int status)
{
struct WD33C93_hostdata *hdata = (struct WD33C93_hostdata *)instance->hostdata;
- const wd33c93_regs regp = hdata->regs;
struct hpc3_scsiregs *hregs;
if (!SCpnt)
@@ -199,6 +204,9 @@
};
hcp--;
hcp->desc.pnext = PHYSADDR(buf);
+
+ /* Force flush to memory */
+ dma_cache_wback_inv((unsigned long) buf, PAGE_SIZE);
}
int __init sgiwd93_detect(Scsi_Host_Template *SGIblows)
@@ -229,7 +237,6 @@
return 0;
}
init_hpc_chain(buf);
- dma_cache_wback_inv((unsigned long) buf, PAGE_SIZE);
regs.SASR = KSEG1ADDR (0x1fbc0003); /* HPC_SCSI_REG0 | 0x03 | KSEG1 */
regs.SCMD = KSEG1ADDR (0x1fbc0007);
@@ -264,7 +271,6 @@
return 1; /* We registered host0 so return success*/
}
init_hpc_chain(buf);
- dma_cache_wback_inv((unsigned long) buf, PAGE_SIZE);
/* HPC_SCSI_REG1 | 0x03 | KSEG1 */
regs.SASR = KSEG1ADDR(0x1fbc8003);
@@ -275,7 +281,6 @@
hdata1 = (struct WD33C93_hostdata *)sgiwd93_host1->hostdata;
hdata1->no_sync = 0;
hdata1->dma_bounce_buffer = (uchar *) (KSEG1ADDR(buf));
- dma_cache_wback_inv((unsigned long) buf, PAGE_SIZE);
if (request_irq(SGI_WD93_1_IRQ, sgiwd93_intr, 0, "SGI WD93", (void *) sgiwd93_host1)) {
printk(KERN_WARNING "sgiwd93: Could not allocate irq %d (for host1).\n", SGI_WD93_1_IRQ);
--
Florian Lohoff flo@rfc822.org +49-5201-669912
Nine nineth on september the 9th Welcome to the new billenium
PGP signature
