All of lore.kernel.org
 help / color / mirror / Atom feed
From: Len Brown <len.brown-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
To: ACPI Developers
	<acpi-devel-5NWGOfrQmneRv+LV9MX5uipxlwaOVQ5f@public.gmane.org>
Cc: linux-acpi <linux-acpi-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Subject: [PATCH] acpi/pci_link.c
Date: 04 Nov 2003 23:50:46 -0500	[thread overview]
Message-ID: <1068007845.6059.33.camel@dhcppc4> (raw)

[-- Attachment #1: Type: text/plain, Size: 1364 bytes --]

Please review this interrupt patch.
It applies to both 2.4 and 2.6.

thanks,
-Len

What the patch does:

Re-enables PCI Link Device programming by default.

This fixes systems that stopped working when we
stopped re-programming active PIRQs:
http://bugzilla.kernel.org/show_bug.cgi?id=1440

For systems that mark their PIRQ links active,
this allows the kernel to again distribute
PIRQs across all available IRQs:
http://bugzilla.kernel.org/show_bug.cgi?id=1391

Fixes an off-by-one bug in the PIRQ distribution
code that caused one of the possible IRQs to be
ignored on every PIRQ link.

To aid in minimizing IRQ sharing, added cmdline
"acpi_irq_pci=" to tell the kernel when it can
use an IRQ for PCI, even if the kernel default
is to reserve that IRQ for ISA use.

Conversely, added "acpi_irq_isa=" cmdline param
to reserve IRQs for ISA.
Workaround for SoundBlaster vs. ACPI PIRQ routing issues:
http://bugzilla.kernel.org/show_bug.cgi?id=430
http://bugzilla.kernel.org/show_bug.cgi?id=1139

Previously we had disabled re-programming active PIRQs to
address http://bugzilla.kernel.org/show_bug.cgi?id=1186
This change will break those broken systems again, unless
they've been fixed in other ways.

Mitigated the impact with a new cmdline parameter
"acpi_irq_static" that restores the previous behavior
of not moving PCI Link devices that are marked active.




[-- Attachment #2: pci_link.patch --]
[-- Type: text/plain, Size: 7820 bytes --]

===== drivers/acpi/pci_link.c 1.20 vs edited =====
--- 1.20/drivers/acpi/pci_link.c	Tue Sep 30 00:50:34 2003
+++ edited/drivers/acpi/pci_link.c	Tue Nov  4 23:40:44 2003
@@ -99,7 +99,7 @@
 	void			*context)
 {
 	struct acpi_pci_link	*link = (struct acpi_pci_link *) context;
-	int			i = 0;
+	u32			i = 0;
 
 	ACPI_FUNCTION_TRACE("acpi_pci_link_check_possible");
 
@@ -294,7 +294,10 @@
 
 	if (!link->irq.active) {
 		ACPI_DEBUG_PRINT((ACPI_DB_ERROR, "No active IRQ resource found\n"));
-		printk(KERN_WARNING "_CRS returns NULL! Using IRQ %d for device (%s [%s]).\n", irq, acpi_device_name(link->device), acpi_device_bid(link->device));
+		printk(KERN_WARNING "_CRS returns NULL! Using IRQ %d for"
+			"device (%s [%s]).\n", irq,
+			acpi_device_name(link->device),
+			acpi_device_bid(link->device));
 		link->irq.active = irq;
 	}
 	
@@ -429,30 +432,61 @@
                             PCI Link IRQ Management
    -------------------------------------------------------------------------- */
 
-#define ACPI_MAX_IRQS		256
-#define ACPI_MAX_ISA_IRQ	16
-
 /*
- * IRQ penalties are used to promote PCI IRQ balancing.  We set each ISA-
- * possible IRQ (0-15) with a default penalty relative to its feasibility
- * for PCI's use:
+ * Interrupt sharing can impact performance, so this code
+ * spreads the PCI link devices across the available IRQs.
+ *
+ * A tables of penalties avoids directing PCI interrupts to well known
+ * ISA IRQs. Boot params are available to over-ride the default table:
  *
- *   Never use:		0, 1, 2 (timer, keyboard, and cascade)
- *   Avoid using:	13, 14, and 15 (FP error and IDE)
- *   Penalize:		3, 4, 6, 7, 12 (known ISA uses)
+ * List interrupts that are free for PCI use.
+ * acpi_irq_pci=n[,m]
  *
- * Thus we're left with IRQs 5, 9, 10, 11, and everything above 15 (IO[S]APIC)
- * as 'best bets' for PCI use.
+ * List interrupts that should not be used for PCI:
+ * acpi_irq_isa=n[,m]
+ *
+ * Note that PCI IRQ routers have a list of possible IRQs,
+ * which may not include the IRQs this table says are available.
+ * 
+ * Since this heuristic can't tell the difference between a link
+ * that no device will attach to, vs. a link which may be shared
+ * by multiple active devices -- it is not optimal.
+ *
+ * If interrupt performance is that important, get an IO-APIC system
+ * with a pin dedicated to each device.  Or for that matter, an MSI
+ * enabled system.
  */
 
+#define ACPI_MAX_IRQS		256
+#define ACPI_MAX_ISA_IRQ	16
+
+#define PIRQ_PENALTY_PCI_AVAILABLE	(0)
+#define PIRQ_PENALTY_PCI_POSSIBLE	(16*16)
+#define PIRQ_PENALTY_PCI_USING		(16*16*16)
+#define PIRQ_PENALTY_ISA_TYPICAL	(16*16*16*16)
+#define PIRQ_PENALTY_ISA_USED		(16*16*16*16*16)
+#define PIRQ_PENALTY_ISA_ALWAYS		(16*16*16*16*16*16)
+
 static int acpi_irq_penalty[ACPI_MAX_IRQS] = {
-	1000000,  1000000,  1000000,    10000, 
-	  10000,        0,    10000,    10000,
-	  10000,        0,        0,        0, 
-	  10000,   100000,   100000,   100000,
+	PIRQ_PENALTY_ISA_ALWAYS,	/* IRQ0 timer */
+	PIRQ_PENALTY_ISA_ALWAYS,	/* IRQ1 keyboard */
+	PIRQ_PENALTY_ISA_ALWAYS,	/* IRQ2 cascade */
+	PIRQ_PENALTY_ISA_TYPICAL,	/* IRQ3	serial */
+	PIRQ_PENALTY_ISA_TYPICAL,	/* IRQ4	serial */
+	PIRQ_PENALTY_PCI_AVAILABLE,	/* IRQ5 PCI */
+	PIRQ_PENALTY_ISA_TYPICAL,	/* IRQ6 */
+	PIRQ_PENALTY_ISA_TYPICAL,	/* IRQ7 parallel, spurious */
+	PIRQ_PENALTY_ISA_TYPICAL,	/* IRQ8 rtc, sometimes */
+	PIRQ_PENALTY_PCI_AVAILABLE,	/* IRQ9  PCI, often acpi */
+	PIRQ_PENALTY_PCI_AVAILABLE,	/* IRQ10 PCI */
+	PIRQ_PENALTY_PCI_AVAILABLE,	/* IRQ11 PCI */
+	PIRQ_PENALTY_ISA_TYPICAL,	/* IRQ12 mouse */
+	PIRQ_PENALTY_ISA_USED,	/* IRQ13 fpe, sometimes */
+	PIRQ_PENALTY_ISA_USED,	/* IRQ14 ide0 */
+	PIRQ_PENALTY_ISA_USED,	/* IRQ15 ide1 */
+			/* >IRQ15 */
 };
 
-
 int
 acpi_pci_link_check (void)
 {
@@ -473,20 +507,27 @@
 			continue;
 		}
 
-		if (link->irq.active)
-			acpi_irq_penalty[link->irq.active] += 100;
-		else if (link->irq.possible_count) {
-			int penalty = 100 / link->irq.possible_count;
-			for (i=0; i<link->irq.possible_count; i++) {
+		/*
+		 * reflect the possible and active irqs in the penalty table --
+		 * useful for breaking ties.
+		 */
+		if (link->irq.possible_count) {
+			int penalty = PIRQ_PENALTY_PCI_POSSIBLE / link->irq.possible_count;
+
+			for (i = 0; i < link->irq.possible_count; i++) {
 				if (link->irq.possible[i] < ACPI_MAX_ISA_IRQ)
 					acpi_irq_penalty[link->irq.possible[i]] += penalty;
 			}
+
+		} else if (link->irq.active) {
+			acpi_irq_penalty[link->irq.active] += PIRQ_PENALTY_PCI_POSSIBLE;
 		}
 	}
-
 	return_VALUE(0);
 }
 
+static int acpi_irq_static;	/* don't re-program active IRQs */
+
 static int acpi_pci_link_allocate(struct acpi_pci_link* link) {
 	int irq;
 	int i;
@@ -500,12 +541,14 @@
 		irq = link->irq.active;
 	} else {
 		irq = link->irq.possible[0];
+	}
 
+	if (!(link->irq.active && acpi_irq_static)) {
 		/*
 		 * Select the best IRQ.  This is done in reverse to promote
 		 * the use of IRQs 9, 10, 11, and >15.
 		 */
-		for (i=(link->irq.possible_count-1); i>0; i--) {
+		for (i = (link->irq.possible_count - 1); i >= 0; i--) {
 			if (acpi_irq_penalty[irq] > acpi_irq_penalty[link->irq.possible[i]])
 				irq = link->irq.possible[i];
 		}
@@ -518,13 +561,14 @@
 			acpi_device_bid(link->device));
 		return_VALUE(-ENODEV);
 	} else {
-		acpi_irq_penalty[link->irq.active] += 100;
+		acpi_irq_penalty[link->irq.active] += PIRQ_PENALTY_PCI_USING;
 		printk(PREFIX "%s [%s] enabled at IRQ %d\n", 
 			acpi_device_name(link->device),
 			acpi_device_bid(link->device), link->irq.active);
 	}
 
 	link->irq.setonboot = 1;
+
 	return_VALUE(0);
 }
 
@@ -607,9 +651,12 @@
 	if (result)
 		goto end;
 
+	/* query and set link->irq.active */
 	acpi_pci_link_get_current(link);
 
-	printk(PREFIX "%s [%s] (IRQs", acpi_device_name(device), acpi_device_bid(device));
+//#ifdef CONFIG_ACPI_DEBUG
+	printk(PREFIX "%s [%s] (IRQs", acpi_device_name(device),
+		acpi_device_bid(device));
 	for (i = 0; i < link->irq.possible_count; i++) {
 		if (link->irq.active == link->irq.possible[i]) {
 			printk(" *%d", link->irq.possible[i]);
@@ -619,6 +666,7 @@
 			printk(" %d", link->irq.possible[i]);
 	}
 	printk(")\n");
+//#endif /* CONFIG_ACPI_DEBUG */
 
 	/* TBD: Acquire/release lock */
 	list_add_tail(&link->node, &acpi_link.entries);
@@ -670,5 +718,73 @@
 
 	return_VALUE(0);
 }
+
+/*
+ * modify acpi_irq_penalty[] from cmdline
+ */
+static int __init acpi_irq_penalty_update(char *str, int used)
+{
+	int i;
+
+	for (i = 0; i < 16; i++) {
+		int retval;
+		int irq;
+
+		retval = get_option(&str,&irq);
+
+		if (!retval)
+			break;	/* no number found */
+
+		if (irq < 0)
+			continue;
+		
+		if (irq >= ACPI_MAX_IRQS)
+			continue;
+
+		if (used)
+			acpi_irq_penalty[irq] += PIRQ_PENALTY_ISA_USED;
+		else
+			acpi_irq_penalty[irq] = PIRQ_PENALTY_PCI_AVAILABLE;
+
+		if (retval != 2)	/* no next number */
+			break;
+	}
+	return 1;
+}
+
+/*
+ * Over-ride default table to reserve additional IRQs for use by ISA
+ * e.g. acpi_irq_isa=5
+ * Useful for telling ACPI how not to interfere with your ISA sound card.
+ */
+static int __init acpi_irq_isa(char *str)
+{
+	return(acpi_irq_penalty_update(str, 1));
+}
+__setup("acpi_irq_isa=", acpi_irq_isa);
+
+
+/*
+ * Over-ride default table to free additional IRQs for use by PCI
+ * e.g. acpi_irq_pci=7,15
+ * Useful for freeing up IRQs to reduce PCI IRQ sharing.
+ */
+static int __init acpi_irq_pci(char *str)
+{
+	return(acpi_irq_penalty_update(str, 0));
+}
+__setup("acpi_irq_pci=", acpi_irq_pci);
+
+/*
+ * Do not re-program active IRQs,
+ * keep the static active settings left by the BIOS.
+ * Useful for working around broken BIOS.
+ */
+static int __init acpi_irq_static_set(char *str)
+{
+	acpi_irq_static = 1;
+	return(1);
+}
+__setup("acpi_irq_static", acpi_irq_static_set);
 
 subsys_initcall(acpi_pci_link_init);

                 reply	other threads:[~2003-11-05  4:50 UTC|newest]

Thread overview: [no followups] expand[flat|nested]  mbox.gz  Atom feed

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=1068007845.6059.33.camel@dhcppc4 \
    --to=len.brown-ral2jqcrhueavxtiumwx3w@public.gmane.org \
    --cc=acpi-devel-5NWGOfrQmneRv+LV9MX5uipxlwaOVQ5f@public.gmane.org \
    --cc=linux-acpi-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.