xen/23754-AMD-perdev-vector-map.patch

References: bnc#713503

# HG changeset patch
# User George Dunlap <george.dunlap@eu.citrix.com>
# Date 1311701836 -3600
# Node ID 2e0cf9428554da666616982cd0074024ff85b221
# Parent  ef9ed3d2aa870a37ed5e611be9c524d526a2d604
xen: Option to allow per-device vector maps for MSI IRQs

Add a vector-map to pci_dev, and add an option to point MSI-related
IRQs to the vector-map of the device.

This prevents irqs from the same device from being assigned
the same vector on different pcpus.  This is required for systems
using an AMD IOMMU, since the intremap tables on AMD only look at
vector, and not destination ID.

Signed-off-by: George Dunlap <george.dunlap@eu.citrix.com>

# HG changeset patch
# User George Dunlap <george.dunlap@eu.citrix.com>
# Date 1311701852 -3600
# Node ID fa4e2ca9ecffbc432b451f495ad0a403644a6be8
# Parent  2e0cf9428554da666616982cd0074024ff85b221
xen: AMD IOMMU: Automatically enable per-device vector maps

Automatically enable per-device vector maps when using IOMMU,
unless disabled specifically by an IOMMU parameter.

Signed-off-by: George Dunlap <george.dunlap@eu.citrix.com>

# HG changeset patch
# User George Dunlap <george.dunlap@eu.citrix.com>
# Date 1315231215 -3600
# Node ID 32814ad7458dc842a7c588eee13e5c4ee11709a3
# Parent  f1349a968a5ac5577d67ad4a3f3490c580dbe264
xen: Add global irq_vector_map option, set if using AMD global intremap tables

As mentioned in previous changesets, AMD IOMMU interrupt
remapping tables only look at the vector, not the destination
id of an interrupt.  This means that all IRQs going through
the same interrupt remapping table need to *not* share vectors.

The irq "vector map" functionality was originally introduced
after a patch which disabled global AMD IOMMUs entirely.  That
patch has since been reverted, meaning that AMD intremap tables
can either be per-device or global.

This patch therefore introduces a global irq vector map option,
and enables it if we're using an AMD IOMMU with a global
interrupt remapping table.

This patch removes the "irq-perdev-vector-map" boolean
command-line optino and replaces it with "irq_vector_map",
which can have one of three values: none, global, or per-device.

Setting the irq_vector_map to any value will override the
default that the AMD code sets.

Signed-off-by: George Dunlap <george.dunlap@eu.citrix.com>

# HG changeset patch
# User Jan Beulich <jbeulich@suse.com>
# Date 1317730316 -7200
# Node ID a99d75671a911f9c0d5d11e0fe88a0a65863cb44
# Parent  3d1664cc9e458809e399320204aca8536e401ee1
AMD-IOMMU: remove dead variable references

These got orphaned up by recent changes.

Signed-off-by: Jan Beulich <jbeulich@suse.com>
Acked-by: Keir Fraser <keir@xen.org>

--- a/docs/src/user.tex
+++ b/docs/src/user.tex
@@ -4197,6 +4197,10 @@ writing to the VGA console after domain 
 \item [ vcpu\_migration\_delay=$<$minimum\_time$>$] Set minimum time of 
   vcpu migration in microseconds (default 0). This parameter avoids agressive
   vcpu migration. For example, the linux kernel uses 0.5ms by default.
+\item [ irq_vector_map=xxx ] Enable irq vector non-sharing maps.  Setting 'global'
+  will ensure that no  IRQs will share vectors.  Setting 'per-device' will ensure
+  that no IRQs from the same device will share vectors.  Setting to 'none' will
+  disable it entirely, overriding any defaults the IOMMU code may set.
 \end{description}
 
 In addition, the following options may be specified on the Xen command
--- a/xen/arch/x86/irq.c
+++ b/xen/arch/x86/irq.c
@@ -24,6 +24,8 @@
 #include <asm/mach-generic/mach_apic.h>
 #include <public/physdev.h>
 
+static void parse_irq_vector_map_param(char *s);
+
 /* opt_noirqbalance: If true, software IRQ balancing/affinity is disabled. */
 bool_t __read_mostly opt_noirqbalance = 0;
 boolean_param("noirqbalance", opt_noirqbalance);
@@ -32,6 +34,12 @@ unsigned int __read_mostly nr_irqs_gsi =
 unsigned int __read_mostly nr_irqs;
 integer_param("nr_irqs", nr_irqs);
 
+/* This default may be changed by the AMD IOMMU code */
+int __read_mostly opt_irq_vector_map = OPT_IRQ_VECTOR_MAP_DEFAULT;
+custom_param("irq_vector_map", parse_irq_vector_map_param);
+
+vmask_t global_used_vector_map;
+
 u8 __read_mostly *irq_vector;
 struct irq_desc __read_mostly *irq_desc = NULL;
 
@@ -60,6 +68,26 @@ static struct timer irq_ratelimit_timer;
 static unsigned int __read_mostly irq_ratelimit_threshold = 10000;
 integer_param("irq_ratelimit", irq_ratelimit_threshold);
 
+static void __init parse_irq_vector_map_param(char *s)
+{
+    char *ss;
+
+    do {
+        ss = strchr(s, ',');
+        if ( ss )
+            *ss = '\0';
+
+        if ( !strcmp(s, "none"))
+            opt_irq_vector_map=OPT_IRQ_VECTOR_MAP_NONE;
+        else if ( !strcmp(s, "global"))
+            opt_irq_vector_map=OPT_IRQ_VECTOR_MAP_GLOBAL;
+        else if ( !strcmp(s, "per-device"))
+            opt_irq_vector_map=OPT_IRQ_VECTOR_MAP_PERDEV;
+
+        s = ss + 1;
+    } while ( ss );
+}
+
 /* Must be called when irq disabled */
 void lock_vector_lock(void)
 {
@@ -344,6 +372,41 @@ hw_irq_controller no_irq_type = {
     end_none
 };
 
+static vmask_t *irq_get_used_vector_mask(int irq)
+{
+    vmask_t *ret = NULL;
+
+    if ( opt_irq_vector_map == OPT_IRQ_VECTOR_MAP_GLOBAL )
+    {
+        struct irq_desc *desc = irq_to_desc(irq);
+
+        ret = &global_used_vector_map;
+
+        if ( desc->chip_data->used_vectors )
+        {
+            printk(XENLOG_INFO "%s: Strange, unassigned irq %d already has used_vectors!\n",
+                   __func__, irq);
+        }
+        else
+        {
+            int vector;
+
+            vector = irq_to_vector(irq);
+            if ( vector > 0 )
+            {
+                printk(XENLOG_INFO "%s: Strange, irq %d already assigned vector %d!\n",
+                       __func__, irq, vector);
+
+                ASSERT(!test_bit(vector, ret));
+
+                set_bit(vector, ret);
+            }
+        }
+    }
+
+    return ret;
+}
+
 int __assign_irq_vector(int irq, struct irq_cfg *cfg, const cpumask_t *mask)
 {
     /*
@@ -362,6 +425,7 @@ int __assign_irq_vector(int irq, struct 
     int cpu, err;
     unsigned long flags;
     cpumask_t tmp_mask;
+    vmask_t *irq_used_vectors = NULL;
 
     old_vector = irq_to_vector(irq);
     if (old_vector) {
@@ -376,6 +440,17 @@ int __assign_irq_vector(int irq, struct 
         return -EAGAIN;
 
     err = -ENOSPC;
+
+    /* This is the only place normal IRQs are ever marked
+     * as "in use".  If they're not in use yet, check to see
+     * if we need to assign a global vector mask. */
+    if ( irq_status[irq] == IRQ_USED )
+    {
+        irq_used_vectors = cfg->used_vectors;
+    }
+    else
+        irq_used_vectors = irq_get_used_vector_mask(irq);
+
     for_each_cpu_mask(cpu, *mask) {
         int new_cpu;
         int vector, offset;
@@ -401,8 +476,8 @@ next:
         if (test_bit(vector, used_vectors))
             goto next;
 
-        if (cfg->used_vectors
-            && test_bit(vector, cfg->used_vectors) )
+        if (irq_used_vectors
+            && test_bit(vector, irq_used_vectors) )
             goto next;
 
         for_each_cpu_mask(new_cpu, tmp_mask)
@@ -420,15 +495,22 @@ next:
             per_cpu(vector_irq, new_cpu)[vector] = irq;
         cfg->vector = vector;
         cpus_copy(cfg->cpu_mask, tmp_mask);
+
+        irq_status[irq] = IRQ_USED;
+        ASSERT((cfg->used_vectors == NULL)
+               || (cfg->used_vectors == irq_used_vectors));
+        cfg->used_vectors = irq_used_vectors;
+
+        if (IO_APIC_IRQ(irq))
+            irq_vector[irq] = vector;
+
         if ( cfg->used_vectors )
         {
             ASSERT(!test_bit(vector, cfg->used_vectors));
+
             set_bit(vector, cfg->used_vectors);
         }
 
-        irq_status[irq] = IRQ_USED;
-            if (IO_APIC_IRQ(irq))
-                    irq_vector[irq] = vector;
         err = 0;
         local_irq_restore(flags);
         break;
@@ -1523,7 +1605,7 @@ int map_domain_pirq(
 
     if ( !IS_PRIV(current->domain) &&
          !(IS_PRIV_FOR(current->domain, d) &&
-          irq_access_permitted(current->domain, pirq)))
+           irq_access_permitted(current->domain, pirq)))
         return -EPERM;
 
     if ( pirq < 0 || pirq >= d->nr_pirqs || irq < 0 || irq >= nr_irqs )
@@ -1571,8 +1653,22 @@ int map_domain_pirq(
 
         if ( desc->handler != &no_irq_type )
             dprintk(XENLOG_G_ERR, "dom%d: irq %d in use\n",
-              d->domain_id, irq);
+                    d->domain_id, irq);
         desc->handler = &pci_msi_type;
+
+        if ( opt_irq_vector_map == OPT_IRQ_VECTOR_MAP_PERDEV
+             && !desc->chip_data->used_vectors )
+        {
+            desc->chip_data->used_vectors = &pdev->info.used_vectors;
+            if ( desc->chip_data->vector != IRQ_VECTOR_UNASSIGNED )
+            {
+                int vector = desc->chip_data->vector;
+                ASSERT(!test_bit(vector, desc->chip_data->used_vectors));
+
+                set_bit(vector, desc->chip_data->used_vectors);
+            }
+        }
+
         d->arch.pirq_irq[pirq] = irq;
         d->arch.irq_pirq[irq] = pirq;
         setup_msi_irq(pdev, msi_desc, irq);
@@ -1583,9 +1679,12 @@ int map_domain_pirq(
         d->arch.pirq_irq[pirq] = irq;
         d->arch.irq_pirq[irq] = pirq;
         spin_unlock_irqrestore(&desc->lock, flags);
+
+        if ( opt_irq_vector_map == OPT_IRQ_VECTOR_MAP_PERDEV )
+            printk(XENLOG_INFO "Per-device vector maps for GSIs not implemented yet.\n");
     }
 
- done:
+done:
     return ret;
 }
 
--- a/xen/drivers/passthrough/amd/pci_amd_iommu.c
+++ b/xen/drivers/passthrough/amd/pci_amd_iommu.c
@@ -166,6 +166,35 @@ int __init amd_iov_detect(void)
         return -ENODEV;
     }
 
+    /*
+     * AMD IOMMUs don't distinguish between vectors destined for
+     * different cpus when doing interrupt remapping.  This means
+     * that interrupts going through the same intremap table
+     * can't share the same vector.
+     *
+     * If irq_vector_map isn't specified, choose a sensible default:
+     * - If we're using per-device interemap tables, per-device
+     *   vector non-sharing maps
+     * - If we're using a global interemap table, global vector
+     *   non-sharing map
+     */
+    if ( opt_irq_vector_map == OPT_IRQ_VECTOR_MAP_DEFAULT )
+    {
+        if ( amd_iommu_perdev_intremap )
+        {
+            printk("AMD-Vi: Enabling per-device vector maps\n");
+            opt_irq_vector_map = OPT_IRQ_VECTOR_MAP_PERDEV;
+        }
+        else
+        {
+            printk("AMD-Vi: Enabling global vector map\n");
+            opt_irq_vector_map = OPT_IRQ_VECTOR_MAP_GLOBAL;
+        }
+    }
+    else
+    {
+        printk("AMD-Vi: Not overriding irq_vector_map setting\n");
+    }
     return scan_pci_devices();
 }
 
--- a/xen/include/asm-x86/irq.h
+++ b/xen/include/asm-x86/irq.h
@@ -45,6 +45,13 @@ extern u8 *irq_vector;
 
 extern bool_t opt_noirqbalance;
 
+#define OPT_IRQ_VECTOR_MAP_DEFAULT 0 /* Do the default thing  */
+#define OPT_IRQ_VECTOR_MAP_NONE    1 /* None */
+#define OPT_IRQ_VECTOR_MAP_GLOBAL  2 /* One global vector map (no vector sharing) */
+#define OPT_IRQ_VECTOR_MAP_PERDEV  3 /* Per-device vetor map (no vector sharing w/in a device) */
+
+extern int opt_irq_vector_map;
+
 /*
  * Per-cpu current frame pointer - the location of the last exception frame on
  * the stack
--- a/xen/include/xen/pci.h
+++ b/xen/include/xen/pci.h
@@ -11,6 +11,7 @@
 #include <xen/types.h>
 #include <xen/list.h>
 #include <xen/spinlock.h>
+#include <xen/irq.h>
 
 /*
  * The PCI interface treats multi-function devices as independent
@@ -38,6 +39,7 @@ struct pci_dev_info {
         u8 bus;
         u8 devfn;
     } physfn;
+    vmask_t used_vectors;
 };
 
 struct pci_dev {