Re: pvops acpi and xen 4 issues

[Jeremy Fitzhardinge <jeremy@goop.org>]

On 03/10/2010 11:44 AM, M. Nunberg wrote:
> On Mon, 2010-03-08 at 15:02 -0500, Konrad Rzeszutek Wilk wrote:
>    
>>> make sure it was indeed being called. I think this might have something
>>> to do with what Jeremy mentioned about my SCI being on 9 and not 20, but
>>> honestly, I haven't a clue. I have the full log if you need more.
>>>        
>> I am attaching a debug patch that will help in troubleshooting this and
>> potentially fix this.
>>
>> Couple of things:
>>   1). Turn the Enable ACPI SCI in the BIOS to Enable.
>>
>>   2). Compile your Linux kernel with the attached patch and these two options:
>> CONFIG_ACPI_DEBUG=y
>> CONFIG_ACPI_DEBUG_FUNC_TRACE=y
>>
>> Do use your old .config that was triggering the failure in the first
>> case.
>>
>>   3). Edit your kernel line to have:
>> debug  initcall_debug acpi.debug_level=0xffffffff acpi.debug_layer=0x2 xen_sci=fun
>>
>>   4). Boot your kernel. You should see something like this:
>>
>> ... snip..
>> [    0.000000] acpi_sci_ioapic_setup: 0 ->  9 (3, 1)
>>      
> acpi_sci_ioapic_setup: 0 ->  20 (3, 3)
>    
>> ...
>> [    1.483956] xen_setup_pirqs: FADT.SCI: 9 ACPI.SCI:9. Using 9 IRQ,
>> [    1.490025] xen_setup_pirqs: SCI override: trigger: level->level,
>> polarity: high->high
>> [    1.497824] xen: registering gsi 9 triggering 0 polarity 0
>> [    1.503282] xen_allocate_pirq: returning irq 9 for gsi 9
>> [    1.508572] xen: -->  irq=9
>>      
> NR_IRQS:4352 nr_irqs:4096
> xen: registering gsi 0 triggering 1 polarity 0
> xen: -->  irq=0
> Already setup the GSI :0
> xen: registering gsi 1 triggering 1 polarity 0
> xen: -->  irq=1
> Already setup the GSI :1
> xen: registering gsi 2 triggering 1 polarity 0
> xen: -->  irq=2
> Already setup the GSI :2
> xen: registering gsi 3 triggering 1 polarity 0
> xen: -->  irq=3
> Already setup the GSI :3
> xen: registering gsi 4 triggering 1 polarity 0
> xen: -->  irq=4
> Already setup the GSI :4
> xen: registering gsi 5 triggering 1 polarity 0
> xen: -->  irq=5
> Already setup the GSI :5
> xen: registering gsi 6 triggering 1 polarity 0
> xen: -->  irq=6
> Already setup the GSI :6
> xen: registering gsi 7 triggering 1 polarity 0
> xen: -->  irq=7
> Already setup the GSI :7
> xen: registering gsi 8 triggering 1 polarity 0
> xen: -->  irq=8
> Already setup the GSI :8
> xen: registering gsi 9 triggering 1 polarity 0
> xen: -->  irq=9
> Already setup the GSI :9
> xen: registering gsi 10 triggering 1 polarity 0
> xen: -->  irq=10
> Already setup the GSI :10
> xen: registering gsi 11 triggering 1 polarity 0
> xen: -->  irq=11
> Already setup the GSI :11
> xen: registering gsi 12 triggering 1 polarity 0
> xen: -->  irq=12
> Already setup the GSI :12
> xen: registering gsi 13 triggering 1 polarity 0
> xen: -->  irq=13
> Already setup the GSI :13
> xen: registering gsi 14 triggering 1 polarity 0
> xen: -->  irq=14
> Already setup the GSI :14
> xen: registering gsi 15 triggering 1 polarity 0
> xen: -->  irq=15
> Already setup the GSI :15
> xen_setup_pirqs: FADT.SCI: 9 ACPI.SCI:20. Using IRQ 20.
> xen_setup_pirqs: acpi_get_override_irq fails for 20.
> xen_setup_pirqs: SCI override: trigger: level->edge, polarity: high->low
> xen: registering gsi 20 triggering 0 polarity 1
>    alloc irq_desc for 20 on node 0
>    alloc kstat_irqs on node 0
> xen: -->  irq=20
> (XEN) IOAPIC[0]: Set PCI routing entry (8-20 ->  0x98 ->  IRQ 20 Mode:1 Active:1)
> Console: colour VGA+ 80x25
> ....
>    
>> [    1.511254] Already setup the GSI :9
>>      
> And as you predicted, this line did not show up for 20
>    
>> .. and then later:
>> [    3.058890] ACPI: Core revision 20090903
>> [    3.084701] acpi_early_init: FADT.SCI = 9
>>      
> s/9/20, again
>    
>> .. and then lots of ACPI debug stuff.
>>      
> In the log.
>    
>> Thought in your case, the IRQ that should show up is 20 and the 'Already
>> setup the GSI :20' might not show up.
>>
>> Please e-mail the full serial log (even if it does work).
>>      
> All in all, I think it did work. From a novice's perspective (like mine), This message:
> 	ACPI: Interpreter enabled
> was missing when SCI was enabled in the BIOS, but with your patch, it is present.
> I am attaching the full log as a bz2 compressed textfile, as the last time I tried to send something with wide lines, something gobbled it up, though I think the relevant info is inline
>    

Could you try this patch? (Already committed to xen/next, as it works 
for me.)

Thanks,
     J

 From 06ab9af4e7beaf2bcafae4e390b4a51908dfd29e Mon Sep 17 00:00:00 2001
From: Jeremy Fitzhardinge<jeremy.fitzhardinge@citrix.com>
Date: Wed, 10 Mar 2010 15:44:21 -0800
Subject: [PATCH] Merge branch 'xen/dom0/apic-next' into xen/next

* xen/dom0/apic-next:
   xen: parse MADT to get information to set up ACPI SCI

: abulafia:pts/0; git show d02659fd69312e5ff9890c1c7ca2bea9a782ad57
 From d02659fd69312e5ff9890c1c7ca2bea9a782ad57 Mon Sep 17 00:00:00 2001
From: Jeremy Fitzhardinge<jeremy.fitzhardinge@citrix.com>
Date: Wed, 10 Mar 2010 13:07:15 -0800
Subject: [PATCH] xen: parse MADT to get information to set up ACPI SCI

If the SCI is not in its traditional place of irq 9/gsi 9, then we
need to set it up properly.

Signed-off-by: Jeremy Fitzhardinge<jeremy.fitzhardinge@citrix.com>

diff --git a/arch/x86/xen/pci.c b/arch/x86/xen/pci.c
index f999ad8..a2accf4 100644
--- a/arch/x86/xen/pci.c
+++ b/arch/x86/xen/pci.c
@@ -70,6 +70,97 @@ int xen_register_gsi(u32 gsi, int triggering, int polarity)
  	return irq;
  }

+#ifdef CONFIG_ACPI
+#define BAD_MADT_ENTRY(entry, end) (					    \
+		(!entry) || (unsigned long)entry + sizeof(*entry)>  end ||  \
+		((struct acpi_subtable_header *)entry)->length<  sizeof(*entry))
+
+
+static int __init
+xen_acpi_parse_int_src_ovr(struct acpi_subtable_header * header,
+			   const unsigned long end)
+{
+	struct acpi_madt_interrupt_override *intsrc = NULL;
+
+	intsrc = (struct acpi_madt_interrupt_override *)header;
+
+	if (BAD_MADT_ENTRY(intsrc, end))
+		return -EINVAL;
+
+	acpi_table_print_madt_entry(header);
+
+	if (intsrc->source_irq == acpi_gbl_FADT.sci_interrupt) {
+		int gsi;
+		int trigger, polarity;
+
+		trigger = intsrc->inti_flags&  ACPI_MADT_TRIGGER_MASK;
+		polarity = intsrc->inti_flags&  ACPI_MADT_POLARITY_MASK;
+
+		/* Command-line over-ride via acpi_sci= */
+		if (acpi_sci_flags&  ACPI_MADT_TRIGGER_MASK)
+			trigger = acpi_sci_flags&  ACPI_MADT_TRIGGER_MASK;
+
+		if (acpi_sci_flags&  ACPI_MADT_POLARITY_MASK)
+			polarity = acpi_sci_flags&  ACPI_MADT_POLARITY_MASK;
+
+		printk("xen: sci override: source_irq=%d global_irq=%d trigger=%x polarity=%x\n",
+			intsrc->source_irq, intsrc->global_irq,
+			trigger, polarity);
+
+		switch (polarity) {
+		case ACPI_MADT_POLARITY_CONFORMS:
+		case ACPI_MADT_POLARITY_ACTIVE_LOW:
+			polarity = ACPI_ACTIVE_LOW;
+			break;
+
+		case ACPI_MADT_POLARITY_ACTIVE_HIGH:
+			polarity = ACPI_ACTIVE_HIGH;
+			break;
+
+		default:
+			return 0;
+		}
+
+		switch (trigger) {
+		case ACPI_MADT_TRIGGER_CONFORMS:
+		case ACPI_MADT_TRIGGER_LEVEL:
+			trigger = ACPI_LEVEL_SENSITIVE;
+			break;
+
+		case ACPI_MADT_TRIGGER_EDGE:
+			trigger = ACPI_EDGE_SENSITIVE;
+			break;
+
+		default:
+			return 0;
+		}
+
+		gsi = xen_register_gsi(intsrc->global_irq,
+				       trigger, polarity);
+		/*
+		 * stash over-ride to indicate we've been here
+		 * and for later update of acpi_gbl_FADT
+		 */
+		acpi_sci_override_gsi = gsi;
+
+		printk("xen: acpi sci %d\n", gsi);
+	}
+
+	return 0;
+}
+
+static __init void xen_setup_acpi_sci(void)
+{
+  acpi_table_parse_madt(ACPI_MADT_TYPE_INTERRUPT_OVERRIDE,
+			xen_acpi_parse_int_src_ovr,
+			nr_irqs);
+}
+#else
+static __init void xen_setup_acpi_sci(void)
+{
+}
+#endif
+
  void __init xen_setup_pirqs(void)
  {
  	int irq;
@@ -91,6 +182,8 @@ void __init xen_setup_pirqs(void)
  			trigger ? ACPI_LEVEL_SENSITIVE : ACPI_EDGE_SENSITIVE,
  			polarity ? ACPI_ACTIVE_LOW : ACPI_ACTIVE_HIGH);
  	}
+
+	xen_setup_acpi_sci();
  }

  #ifdef CONFIG_PCI_MSI