xilinx_sysace

Hi there,

I am doing some work on a board port (Xilinx Virtex II Pro 405D running 2.4.23) which uses the xilinx_sysace driver on a Xilinx Virtex II Pro system (not the ML300 board - an inhouse custom thing). The driver comes in several pieces because Xilinx supply a (rather unpleasant) HAL which Monta's adapter.c driver code wraps to create a block device.

The driver works fine for me as a read only device but I'm still seeing occasional hard locks on writes (looks to be something not getting io_request_lock at the right moment - currently investigating). They use a kernel thread which just sits and compensates for the hardware not being able to signal when it is ready for a new request - and I think there's a race there.

I looked in to this before but got pulled off to look at some other more pressing bits, and I want to fix this now. Does anyone else have problems writing with it though? Of the people I have spoken to about this most seem to just use System ACE as a configuration device but we use it for configuration, kernel image, userland, system settings...etc.

Cheers,

Jon.

It's as I thought on and off and then on again - the code checks out ok (it's not pretty but it works) - and I seem to be getting unwanted extra unhandled interrupts from the hardware. This driver needs a lot of cleanup anyway - it doesn't handle these kinds of error state, nor does it handle the removal of a mounted CompactFlash, and a dozen other typical problems. I'll post a patch when I've solved the main problem - moan at me by private mail if you're using this, having similar issues, and feel like helping.

Jon.

Hi all,

I have spoken to a number of people about this ongoing issue with the Xilinx System ACE hardware that I am using (I have a Memec board - not the Xilinx one) and the fact that the hardware insists on generating an extra interrupt on SectorWrite operations which is neither documented nor otherwise explainable (but I'd love it if someone would enlighten me).

So I've added a trivial patch to xilinx_sysace to provide a flag for disabling interrupts at least for now. I am working on a bunch of other Virtex II Pro bits and pieces that I'll feed upstream when they're ready but for now xsa_interrupt should probably also check whether the request queue is empty in the case that it's called with no good reason.

The following patch has been briefly tested on a modified inhouse 2.4.23 kernel which uses a few drivers from the Monta tree (like xilinux_sysace) but otherwise differs quite heavily - I don't rely on any generated xparameters and other EDK nonesense and I don't want to rely on that in any way either - least of all for determining whether or not to use interrupts.

Jon.

diff --unified --recursive --new-file xilinx_sysace_mvista/adapter.c xilinx_sysace_nointr/adapter.c

@@ -6,6 +6,10 @@
  * Author: MontaVista Software, Inc.
  *         [email protected]
  *
+ * History:
+ *         22/09/2004 - Added xsa_use_interrupts.
+ *         Jon Masters <[email protected]>.
+ *
  * 2002 (c) MontaVista, Software, Inc.  This file is licensed under the terms
  * of the GNU General Public License version 2.  This program is licensed
  * "as is" without any warranty of any kind, whether express or implied.

@@ -59,9 +63,17 @@
 #include <xbasic_types.h>
 #include "xsysace.h"
 
+/*
+ * We have to disable interrupts on certain boards where writing causes an
+ * additional unexpected hardware interrupt on sector write completion.
+ */
+
+static int xsa_use_interrupts = 0;
+
 MODULE_AUTHOR("MontaVista Software, Inc. <[email protected]>");
 MODULE_DESCRIPTION("Xilinx System ACE block driver");
 MODULE_LICENSE("GPL");
+MODULE_PARM(xsa_enable_interrupts,"i");
 
 /*
  * We have to wait for a lock and for the CompactFlash to not be busy

@@ -497,11 +509,17 @@
 			       DEVICE_NAME, stat, req_str, sector);
 			xsa_complete_request(0);	/* Request failed. */
 		}
+
+		if ((!xsa_use_interrupts) && (stat == XST_SUCCESS)) {
+			printk("RIQC: request complete.\n");
+			xsa_complete_request(1);
+		}
 	}
 
-      exit:
+	
+ exit:
 	remove_wait_queue(&req_wait, &wait);
-
+	
 	xsa_task = NULL;
 	complete_and_exit(&task_sync, 0);
 }

@@ -757,16 +775,25 @@
 		return -ENODEV;
 	}
 
-	i = request_irq(XSA_IRQ, xsysace_interrupt, 0, DEVICE_NAME, NULL);
-	if (i) {
-		printk(KERN_ERR "%s: Could not allocate interrupt %d.\n",
-		       DEVICE_NAME, XSA_IRQ);
-		cleanup();
-		return i;
+	if (xsa_use_interrupts) {
+
+		i = request_irq(XSA_IRQ, xsysace_interrupt, 0, DEVICE_NAME, NULL);
+		if (i) {
+			printk(KERN_ERR "%s: Could not allocate interrupt %d.\n",
+			       DEVICE_NAME, XSA_IRQ);
+			cleanup();
+			return i;
+		}
+		reqirq = 1;
 	}
-	reqirq = 1;
+
 	XSysAce_SetEventHandler(&SysAce, EventHandler, (void *) NULL);
-	XSysAce_EnableInterrupt(&SysAce);
+	
+	if (xsa_use_interrupts) {
+		XSysAce_EnableInterrupt(&SysAce);
+	} else {
+		XSysAce_DisableInterrupt(&SysAce);
+	}
 
 	/* Time to identify the drive. */
 	while (XSysAce_Lock(&SysAce, 0) == XST_DEVICE_BUSY)

@@ -832,10 +859,17 @@
 	register_disk(&xsa_gendisk, MKDEV(xsa_major, 0),
 		      NR_HD << PARTN_BITS, &xsa_fops, size);
 
-	printk(KERN_INFO
-	       "%s at 0x%08X mapped to 0x%08X, irq=%d, %ldKB\n",
-	       DEVICE_NAME, save_BaseAddress, cfg->BaseAddress, XSA_IRQ,
-	       size / 2);
+	if (xsa_use_interrupts) {
+		printk(KERN_INFO
+		       "%s at 0x%08X mapped to 0x%08X, irq=%d, %ldKB\n",
+		       DEVICE_NAME, save_BaseAddress, cfg->BaseAddress, XSA_IRQ,
+		       size / 2);
+	} else {
+		printk(KERN_INFO
+		       "%s at 0x%08X mapped to 0x%08X, polled IO, %ldKB\n",
+		       DEVICE_NAME, save_BaseAddress, cfg->BaseAddress, XSA_IRQ,
+		       size / 2);
+	}
 
 	/* Hook our reset function into system's restart code. */
 	old_restart = ppc_md.restart;

Salut,

Cheers. I also left in a printk that shouldn't be there - just adopting the early and often philosophy since this has been bothering me for a while.

Jon.

We are using the SystemACE on our custom MPC8260 platform and I do not recall seeing extra interrupts on the sector writes.

However, we did uncover some general interrupt issues with interrupt handling in the PPC kernel. We were seeing a large number of BAD interrupts in /proc/interrupts. Unfortunately, the only information I have on the work around is this email snippet that I saved:

<begin snippet>

<end snippet>

Jeff Angielski
The PTR Group

This is on a Virtex II Pro which does not use the interrupt controller that you mention. I'm quite sure *some* people are able to use this just fine - otherwise there wouldn't be interrupt driven IO in the Monta driver, but it doesn't work on all boards, and it's not a simple bug in the interrupt handling code - it's a real extra unwanted interrupt from the chip.

The board in question is the Memec one with lots of custom logic, the xintc and a bunch of other bits running 2.4.23 but without any of the HAL nonesense and Xilinx xparameters stuff.

Cheers,

Jon.