[PATCH] mesh: target 0 aborted

[PATCH] mesh: target 0 aborted

[Momchil Velikov]

This patch makes mesh.c compile, by adapting it to the new
pmac_feature API (ported from the ppc tree). 

In addition it contains the fix from Thomas Capricelli for the
infamous "mesh: target 0 aborted" error, which I've been personally
observing since 2.1.13x. 

Regards,
-velco


--- 1.1/drivers/scsi/mesh.c	Sat Dec  8 02:14:29 2001
+++ edited/drivers/scsi/mesh.c	Wed Jan  2 00:02:23 2002
@@ -28,7 +28,8 @@
 #include <asm/irq.h>
 #include <asm/hydra.h>
 #include <asm/processor.h>
-#include <asm/feature.h>
+#include <asm/machdep.h>
+#include <asm/pmac_feature.h>
 #ifdef CONFIG_PMAC_PBOOK
 #include <linux/adb.h>
 #include <linux/pmu.h>
@@ -155,7 +156,6 @@
 	struct mesh_target tgts[8];
 	void	*dma_cmd_space;
 	struct device_node *ofnode;
-	u8*	mio_base;
 #ifndef MESH_NEW_STYLE_EH
 	Scsi_Cmnd *completed_q;
 	Scsi_Cmnd *completed_qtail;
@@ -258,8 +258,6 @@
 	if (mesh == 0)
 		mesh = find_compatible_devices("scsi", "chrp,mesh0");
 	for (; mesh != 0; mesh = mesh->next) {
-		struct device_node *mio;
-		
 		if (mesh->n_addrs != 2 || mesh->n_intrs != 2) {
 			printk(KERN_ERR "mesh: expected 2 addrs and 2 intrs"
 			       " (got %d,%d)", mesh->n_addrs, mesh->n_intrs);
@@ -325,12 +323,6 @@
 		if (mesh_sync_period < minper)
 			mesh_sync_period = minper;
 
-		ms->mio_base = 0;
-		for (mio = ms->ofnode->parent; mio; mio = mio->parent)
-			if (strcmp(mio->name, "mac-io") == 0 && mio->n_addrs > 0)
-				break;
-		if (mio)
-			ms->mio_base = (u8 *) ioremap(mio->addrs[0].address, 0x40);
 		set_mesh_power(ms, 1);
 
 		mesh_init(ms);
@@ -363,11 +355,9 @@
 		iounmap((void *) ms->mesh);
 	if (ms->dma)
 		iounmap((void *) ms->dma);
-	if (ms->mio_base)
-		iounmap((void *) ms->mio_base);
 	kfree(ms->dma_cmd_space);
 	free_irq(ms->meshintr, ms);
-	feature_clear(ms->ofnode, FEATURE_MESH_enable);
+	pmac_call_feature(PMAC_FTR_MESH_ENABLE, ms->ofnode, 0, 0);
 	return 0;
 }
 
@@ -377,16 +367,10 @@
 	if (_machine != _MACH_Pmac)
 		return;
 	if (state) {
-		feature_set(ms->ofnode, FEATURE_MESH_enable);
-		/* This seems to enable the termination power. strangely
-		   this doesn't fully agree with OF, but with MacOS */
-		if (ms->mio_base)
-			out_8(ms->mio_base + 0x36, 0x70);
+		pmac_call_feature(PMAC_FTR_MESH_ENABLE, ms->ofnode, 0, 1);
 		mdelay(200);
 	} else {
-		feature_clear(ms->ofnode, FEATURE_MESH_enable);
-		if (ms->mio_base)
-			out_8(ms->mio_base + 0x36, 0x34);
+		pmac_call_feature(PMAC_FTR_MESH_ENABLE, ms->ofnode, 0, 1);
 		mdelay(10);
 	}
 }			
@@ -584,25 +568,17 @@
 
 	udelay(100);
 
-	out_le32(&md->control, (RUN|PAUSE|FLUSH|WAKE) << 16);	/* stop dma */
-	out_8(&mr->exception, 0xff);	/* clear all exception bits */
-	out_8(&mr->error, 0xff);	/* clear all error bits */
+	out_8(&mr->exception, 0xff);	
+	out_8(&mr->error, 0xff);	
 	out_8(&mr->sequence, SEQ_RESETMESH);
 	udelay(10);
+
+	out_8(&mr->interrupt, 0xff);	/* clear all interrupt bits */
 	out_8(&mr->intr_mask, INT_ERROR | INT_EXCEPTION | INT_CMDDONE);
 	out_8(&mr->source_id, ms->host->this_id);
 	out_8(&mr->sel_timeout, 25);	/* 250ms */
-	out_8(&mr->sync_params, ASYNC_PARAMS);
-
-	out_8(&mr->bus_status1, BS1_RST);	/* assert RST */
-	udelay(30);			/* leave it on for >= 25us */
-	out_8(&mr->bus_status1, 0);	/* negate RST */
-
-	out_8(&mr->sequence, SEQ_FLUSHFIFO);
-	udelay(1);
-	out_8(&mr->sync_params, ASYNC_PARAMS);
-	out_8(&mr->sequence, SEQ_ENBRESEL);
-	out_8(&mr->interrupt, 0xff);	/* clear all interrupt bits */
+	out_8(&mr->sync_params, ASYNC_PARAMS);	/* asynchronous initially */
+	out_le32(&md->control, (RUN|PAUSE|FLUSH|WAKE) << 16);
 }
 
 /*

Re: [PATCH] mesh: target 0 aborted

[Tom Rini]

On Wed, Jan 02, 2002 at 12:45:34AM +0200, Momchil Velikov wrote:

> This patch makes mesh.c compile, by adapting it to the new
> pmac_feature API (ported from the ppc tree).
> 
> In addition it contains the fix from Thomas Capricelli for the
> infamous "mesh: target 0 aborted" error, which I've been personally
> observing since 2.1.13x.

Er, what exactly is this against?  If this is just what's in the
linuxppc_2_4 tree against current 2.4.18pre, this is either (or will be
now :)) on BenH's list of things to resend to Marcelo, or there's a
problem with it still.  If you added in another patch, please re-send
this vs the linuxppc_2_4 tree.

-- 
Tom Rini (TR1265)
http://gate.crashing.org/~trini/

Re: [PATCH] mesh: target 0 aborted

[Momchil Velikov]

>>>>> "Tom" == Tom Rini <trini@kernel.crashing.org> writes:

Tom> On Wed, Jan 02, 2002 at 12:45:34AM +0200, Momchil Velikov wrote:

>> This patch makes mesh.c compile, by adapting it to the new
>> pmac_feature API (ported from the ppc tree).
>> 
>> In addition it contains the fix from Thomas Capricelli for the
>> infamous "mesh: target 0 aborted" error, which I've been personally
>> observing since 2.1.13x.

Tom> Er, what exactly is this against?  If this is just what's in the
Tom> linuxppc_2_4 tree against current 2.4.18pre, this is either (or will be
Tom> now :)) on BenH's list of things to resend to Marcelo, or there's a
Tom> problem with it still.  If you added in another patch, please re-send
Tom> this vs the linuxppc_2_4 tree.

This is against 2.4.18pre1. You may resync with it.

Re: [PATCH] mesh: target 0 aborted

[Benjamin Herrenschmidt]

>> This patch makes mesh.c compile, by adapting it to the new
>> pmac_feature API (ported from the ppc tree).
>>
>> In addition it contains the fix from Thomas Capricelli for the
>> infamous "mesh: target 0 aborted" error, which I've been personally
>> observing since 2.1.13x.
>
>Er, what exactly is this against?  If this is just what's in the
>linuxppc_2_4 tree against current 2.4.18pre, this is either (or will be
>now :)) on BenH's list of things to resend to Marcelo, or there's a
>problem with it still.  If you added in another patch, please re-send
>this vs the linuxppc_2_4 tree.

The up to date mesh driver didn't get into 2.4.18pre1, either I forgot
to send it to Marcelo along with the other PPC patches, or he missed it.

I'll take care of this.

The other patch for getting rid of "target 0 aborted" need some more
review. You seem to just remove the bus reset. That could be made a
driver option in case it really cause trouble, but I suppose the bug
is elsewhere (while beeing triggered by the bus reset).

I'll look into this around next week.

Ben.

Re: [PATCH] mesh: target 0 aborted

[Momchil Velikov]

>>>>> "Benjamin" == Benjamin Herrenschmidt <benh@kernel.crashing.org> writes:

>>> This patch makes mesh.c compile, by adapting it to the new
>>> pmac_feature API (ported from the ppc tree).
>>> 
>>> In addition it contains the fix from Thomas Capricelli for the
>>> infamous "mesh: target 0 aborted" error, which I've been personally
>>> observing since 2.1.13x.
>> 
>> Er, what exactly is this against?  If this is just what's in the
>> linuxppc_2_4 tree against current 2.4.18pre, this is either (or will be
>> now :)) on BenH's list of things to resend to Marcelo, or there's a
>> problem with it still.  If you added in another patch, please re-send
>> this vs the linuxppc_2_4 tree.

Benjamin> The up to date mesh driver didn't get into 2.4.18pre1, either I forgot
Benjamin> to send it to Marcelo along with the other PPC patches, or he missed it.

Benjamin> I'll take care of this.

Benjamin> The other patch for getting rid of "target 0 aborted" need some more
Benjamin> review. You seem to just remove the bus reset. That could be made a
Benjamin> driver option in case it really cause trouble, but I suppose the bug
Benjamin> is elsewhere (while beeing triggered by the bus reset).

Benjamin> I'll look into this around next week.

Thanks a lot.

Regards,
-velco

Re: [PATCH] mesh: target 0 aborted

[Thomas Capricelli]

On Wednesday 02 January 2002 10:17, Benjamin Herrenschmidt wrote:
> The other patch for getting rid of "target 0 aborted" need some more
> review. You seem to just remove the bus reset. That could be made a
> driver option in case it really cause trouble, but I suppose the bug
> is elsewhere (while beeing triggered by the bus reset).
>
> I'll look into this around next week.


	I'm the one responsible for this patch. I can't boot my powerpc7600 whithout 
it. I've been pushing this patch (on the linux-ppc list) for so long (several 
years, don't remember), that I've given up last year.

	Linuxppc people are even worse than Linus. I did not even get an answer 
about a problem with my patch or whatever. Pure 'nothing'. Not even a 'no'.

	I'm still ready to test/discuss about the pb with anybody, anyway.

	Let's get clear about what I've done : I'm using linuxppc for many years, 
and at one point, the kernel refused to boot. I've checked the difference 
between this kernel and the last one I was using, and merely changed back the 
mesh.c so that it works. As I was not following kernel dvpmt very closely by 
then, I can't tell when the change that broke things came in.

	I would be very happy to change my patch so that this change is optional.

happy new year,
Thomas

Re: [PATCH] mesh: target 0 aborted

[Benjamin Herrenschmidt]

>	I'm the one responsible for this patch. I can't boot my powerpc7600
whithout 
>it. I've been pushing this patch (on the linux-ppc list) for so long
(several 
>years, don't remember), that I've given up last year.
>
>	Linuxppc people are even worse than Linus. I did not even get an answer 
>about a problem with my patch or whatever. Pure 'nothing'. Not even a 'no'.
>
>	I'm still ready to test/discuss about the pb with anybody, anyway.
>
>	Let's get clear about what I've done : I'm using linuxppc for many years, 
>and at one point, the kernel refused to boot. I've checked the difference 
>between this kernel and the last one I was using, and merely changed
back the 
>mesh.c so that it works. As I was not following kernel dvpmt very closely by 
>then, I can't tell when the change that broke things came in.

I'm probably responsible for your patch never beeing merged then, sorry
about that. MESH is a weird beast, sometimes, fixing it for one case will
break others, it's pretty timing sensitive and full of interesting HW
bugs. One problem I have currently is the lack of HW. Fortunately, I've
managed to get back an old 8500 class machine (with MESH), this will
allow me to play a bit with it.

I'm away for now, but I'll look into this once I'm back, next week.

Ben.