From: Don Dugger <ddugger@redhat.com>
Date: Mon, 30 Nov 2009 21:57:21 -0500
Subject: [xen] passthrough msi-x mask bit acceleration
Message-id: <200911302157.nAULvLlG008710@sobek.n0ano.com>
Patchwork-id: 21555
O-Subject: [RHEL5.5 PATCH 3/3] BZ 537734: xen: passthrough MSI-X mask bit
	acceleration
Bugzilla: 537734
RH-Acked-by: Prarit Bhargava <prarit@redhat.com>
RH-Acked-by: Miroslav Rezanina <mrezanin@redhat.com>

backporting xen-unstable changeset 19246, 19306 and
fixes from 19902 and partly 19263.

changeset:   19246:9bc5799566be
    hvm: passthrough MSI-X mask bit acceleration

    Add a new parameter to DOMCTL_bind_pt_irq to allow Xen to know the
    guest physical address of MSI-X table. Also add a new MMIO intercept
    handler to intercept that gpa in order to handle MSI-X vector mask
    bit operation in the hypervisor. This reduces the load of device model
    considerably if the guest does mask and unmask frequently

    Signed-off-by: Qing He <qing.he@intel.com>

changeset:   19306:e6b7b747d122
    passthrough: fix some spinlock issues in vmsi

    Apart from efficiency, I hasten to fix the assertion failure.
    - acquire pcidevs_lock before calling pt_irq_xxx_bind_vtd
    - allocate msixtbl_entry beforehand
    - check return value from domain_spin_lock_irq_desc()
    - typo: spin_unlock(&irq_desc->lock) ->
    - spin_unlock_irq(&irq_desc->lock)
    - acquire msixtbl_list_lock with irq_disabled

    Signed-off-by: Kouya Shimura <kouya@jp.fujitsu.com>

changeset:   19902:d768628c28a4
    x86, hvm:  fix get msix entry error

    There is a mistake to get the msix entry number. It should be
    divide instead of modulus.

    Signed-off-by: Yang Zhang <yang.zhang@intel.com>
    Signed-off-by: Qing He <qing.he@intel.com>

changeset:   19263:9c5b4efc934d
    hvm: passthrough MSI-X: fix ia64 link and MSI-X clean up

    This patch fixes the ia64 link error and some clean up of MSI-X code.
    - add ia64 dummy function to link
    - fix unmatched prototype
    - add error check

    Signed-off-by: Isaku Yamahata <yamahata@valinux.co.jp>

Signed-off-by: Qing He <qing.he@intel.com>
Signed-off-by: Don Dugger <donald.d.dugger@intel.com>

---
 arch/x86/domctl.c            |    8 +
 arch/x86/hvm/hvm.c           |    7 +
 arch/x86/hvm/intercept.c     |    6 +-
 arch/x86/hvm/vmsi.c          |  292 ++++++++++++++++++++++++++++++++++++++++++
 arch/x86/msi.c               |   20 +++
 drivers/passthrough/io.c     |   14 ++-
 include/asm-x86/hvm/domain.h |    4 +
 include/asm-x86/msi.h        |    2 +
 include/public/domctl.h      |    1 +
 include/xen/pci.h            |    6 +-
 10 files changed, 356 insertions(+), 4 deletions(-)

Signed-off-by: Don Zickus <dzickus@redhat.com>

diff --git a/arch/x86/domctl.c b/arch/x86/domctl.c
index e24bfe1..33f4eb0 100644
--- a/arch/x86/domctl.c
+++ b/arch/x86/domctl.c
@@ -568,7 +568,11 @@ long arch_do_domctl(
 
         ret = -ESRCH;
         if ( iommu_enabled )
+        {
+            spin_lock(&pcidevs_lock);
             ret = pt_irq_create_bind_vtd(d, bind);
+            spin_unlock(&pcidevs_lock);
+        }
         if ( ret < 0 )
             gdprintk(XENLOG_ERR, "pt_irq_create_bind failed!\n");
 
@@ -586,7 +590,11 @@ long arch_do_domctl(
             break;
         bind = &(domctl->u.bind_pt_irq);
         if ( iommu_enabled )
+        {
+            spin_lock(&pcidevs_lock);
             ret = pt_irq_destroy_bind_vtd(d, bind);
+            spin_unlock(&pcidevs_lock);
+        }
         if ( ret < 0 )
             gdprintk(XENLOG_ERR, "pt_irq_destroy_bind failed!\n");
         rcu_unlock_domain(d);
diff --git a/arch/x86/hvm/hvm.c b/arch/x86/hvm/hvm.c
index 227b3d8..696d37e 100644
--- a/arch/x86/hvm/hvm.c
+++ b/arch/x86/hvm/hvm.c
@@ -246,6 +246,9 @@ int hvm_domain_initialise(struct domain *d)
     spin_lock_init(&d->arch.hvm_domain.irq_lock);
     spin_lock_init(&d->arch.hvm_domain.vapic_access_lock);
 
+    INIT_LIST_HEAD(&d->arch.hvm_domain.msixtbl_list);
+    spin_lock_init(&d->arch.hvm_domain.msixtbl_list_lock);
+
     rc = paging_enable(d, PG_refcounts|PG_translate|PG_external);
     if ( rc != 0 )
         return rc;
@@ -262,11 +265,15 @@ int hvm_domain_initialise(struct domain *d)
     return rc;
 }
 
+extern void msixtbl_pt_cleanup(struct domain *d);
+
 void hvm_domain_relinquish_resources(struct domain *d)
 {
     hvm_destroy_ioreq_page(d, &d->arch.hvm_domain.ioreq);
     hvm_destroy_ioreq_page(d, &d->arch.hvm_domain.buf_ioreq);
 
+    msixtbl_pt_cleanup(d);
+
     pit_deinit(d);
     rtc_deinit(d);
     pmtimer_deinit(d);
diff --git a/arch/x86/hvm/intercept.c b/arch/x86/hvm/intercept.c
index 8e8ee9a..29cd0b1 100644
--- a/arch/x86/hvm/intercept.c
+++ b/arch/x86/hvm/intercept.c
@@ -34,14 +34,16 @@
 extern struct hvm_mmio_handler hpet_mmio_handler;
 extern struct hvm_mmio_handler vlapic_mmio_handler;
 extern struct hvm_mmio_handler vioapic_mmio_handler;
+extern struct hvm_mmio_handler msixtbl_mmio_handler;
 
-#define HVM_MMIO_HANDLER_NR 3
+#define HVM_MMIO_HANDLER_NR 4
 
 static struct hvm_mmio_handler *hvm_mmio_handlers[HVM_MMIO_HANDLER_NR] =
 {
     &hpet_mmio_handler,
     &vlapic_mmio_handler,
-    &vioapic_mmio_handler
+    &vioapic_mmio_handler,
+    &msixtbl_mmio_handler
 };
 
 struct hvm_buffered_io_range {
diff --git a/arch/x86/hvm/vmsi.c b/arch/x86/hvm/vmsi.c
index 2ce5722..a544bc0 100644
--- a/arch/x86/hvm/vmsi.c
+++ b/arch/x86/hvm/vmsi.c
@@ -194,3 +194,295 @@ int vmsi_deliver(struct domain *d, int pirq)
     return 1;
 }
 
+/* MSI-X mask bit hypervisor interception */
+struct msixtbl_entry
+{
+    struct list_head list;
+    atomic_t refcnt;    /* how many bind_pt_irq called for the device */
+
+    /* TODO: resolve the potential race by destruction of pdev */
+    struct pci_dev *pdev;
+    unsigned long gtable;       /* gpa of msix table */
+    unsigned long table_len;
+    unsigned long table_flags[MAX_MSIX_TABLE_ENTRIES / BITS_PER_LONG + 1];
+
+    struct rcu_head rcu;
+};
+
+static struct msixtbl_entry *msixtbl_find_entry(
+    struct vcpu *v, unsigned long addr)
+{
+    struct msixtbl_entry *entry;
+    struct domain *d = v->domain;
+
+    list_for_each_entry( entry, &d->arch.hvm_domain.msixtbl_list, list )
+        if ( addr >= entry->gtable &&
+             addr < entry->gtable + entry->table_len )
+            return entry;
+
+    return NULL;
+}
+
+static void __iomem *msixtbl_addr_to_virt(
+    struct msixtbl_entry *entry, unsigned long addr)
+{
+    int idx, nr_page;
+
+    if ( !entry )
+        return NULL;
+
+    nr_page = (addr >> PAGE_SHIFT) -
+              (entry->gtable >> PAGE_SHIFT);
+
+    if ( !entry->pdev )
+        return NULL;
+
+    idx = entry->pdev->msix_table_idx[nr_page];
+    if ( !idx )
+        return NULL;
+
+    return (void *)(fix_to_virt(idx) +
+                    (addr & ((1UL << PAGE_SHIFT) - 1)));
+}
+
+static int msixtbl_read(
+    struct vcpu *v, unsigned long address,
+    unsigned long len, unsigned long *pval)
+{
+    unsigned long offset;
+    struct msixtbl_entry *entry;
+    void *virt;
+    int r = 0;
+
+    rcu_read_lock();
+
+    if ( len != 4 )
+        goto out;
+
+    offset = address & (PCI_MSIX_ENTRY_SIZE - 1);
+    if ( offset != PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET)
+        goto out;
+
+    entry = msixtbl_find_entry(v, address);
+    virt = msixtbl_addr_to_virt(entry, address);
+    if ( !virt )
+        goto out;
+
+    *pval = readl(virt);
+    r = 1;
+
+out:
+    rcu_read_unlock();
+    return r;
+}
+
+static int msixtbl_write(struct vcpu *v, unsigned long address,
+                        unsigned long len, unsigned long val)
+{
+    unsigned long offset;
+    struct msixtbl_entry *entry;
+    void *virt;
+    int nr_entry;
+    int r = 0;
+
+    rcu_read_lock();
+
+    if ( len != 4 )
+        goto out;
+
+    entry = msixtbl_find_entry(v, address);
+    nr_entry = (address - entry->gtable) / PCI_MSIX_ENTRY_SIZE;
+
+    offset = address & (PCI_MSIX_ENTRY_SIZE - 1);
+    if ( offset != PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET)
+    {
+        set_bit(nr_entry, &entry->table_flags);
+        goto out;
+    }
+
+    /* exit to device model if address/data has been modified */
+    if ( test_and_clear_bit(nr_entry, &entry->table_flags) )
+        goto out;
+
+    virt = msixtbl_addr_to_virt(entry, address);
+    if ( !virt )
+        goto out;
+
+    writel(val, virt);
+    r = 1;
+
+out:
+    rcu_read_unlock();
+    return r;
+}
+
+static int msixtbl_range(struct vcpu *v, unsigned long addr)
+{
+    struct msixtbl_entry *entry;
+    void *virt;
+
+    rcu_read_lock();
+
+    entry = msixtbl_find_entry(v, addr);
+    virt = msixtbl_addr_to_virt(entry, addr);
+
+    rcu_read_unlock();
+
+    return !!virt;
+}
+
+struct hvm_mmio_handler msixtbl_mmio_handler = {
+    .check_handler = msixtbl_range,
+    .read_handler = msixtbl_read,
+    .write_handler = msixtbl_write
+};
+
+static void add_msixtbl_entry(struct domain *d,
+                              struct pci_dev *pdev,
+                              uint64_t gtable,
+                              struct msixtbl_entry *entry)
+{
+    u32 len;
+
+    memset(entry, 0, sizeof(struct msixtbl_entry));
+
+    INIT_LIST_HEAD(&entry->list);
+    INIT_RCU_HEAD(&entry->rcu);
+    atomic_set(&entry->refcnt, 0);
+
+    len = pci_msix_get_table_len(pdev);
+    entry->table_len = len;
+    entry->pdev = pdev;
+    entry->gtable = (unsigned long) gtable;
+
+    list_add_rcu(&entry->list, &d->arch.hvm_domain.msixtbl_list);
+}
+
+static void free_msixtbl_entry(struct rcu_head *rcu)
+{
+    struct msixtbl_entry *entry;
+
+    entry = container_of (rcu, struct msixtbl_entry, rcu);
+
+    xfree(entry);
+}
+
+static void del_msixtbl_entry(struct msixtbl_entry *entry)
+{
+    list_del_rcu(&entry->list);
+    call_rcu(&entry->rcu, free_msixtbl_entry);
+}
+
+int msixtbl_pt_register(struct domain *d, int pirq, uint64_t gtable)
+{
+    irq_desc_t *irq_desc;
+    struct msi_desc *msi_desc;
+    struct pci_dev *pdev;
+    struct msixtbl_entry *entry, *new_entry;
+    int r = -EINVAL;
+
+    ASSERT(spin_is_locked(&pcidevs_lock));
+
+    /*
+     * xmalloc() with irq_disabled causes the failure of check_lock() 
+     * for xenpool->lock. So we allocate an entry beforehand.
+     */
+    new_entry = xmalloc(struct msixtbl_entry);
+    if ( !new_entry )
+        return -ENOMEM;
+
+    irq_desc = domain_spin_lock_irq_desc(d, pirq, NULL);
+    if ( !irq_desc )
+    {
+        xfree(new_entry);
+        return r;
+    }
+
+    if ( irq_desc->handler != &pci_msi_type )
+        goto out;
+
+    msi_desc = irq_desc->msi_desc;
+    if ( !msi_desc )
+        goto out;
+
+    pdev = msi_desc->dev;
+
+    spin_lock(&d->arch.hvm_domain.msixtbl_list_lock);
+
+    list_for_each_entry( entry, &d->arch.hvm_domain.msixtbl_list, list )
+        if ( pdev == entry->pdev )
+            goto found;
+
+    entry = new_entry;
+    new_entry = NULL;
+    add_msixtbl_entry(d, pdev, gtable, entry);
+
+found:
+    atomic_inc(&entry->refcnt);
+    spin_unlock(&d->arch.hvm_domain.msixtbl_list_lock);
+    r = 0;
+
+out:
+    spin_unlock_irq(&irq_desc->lock);
+    xfree(new_entry);
+    return r;
+}
+
+void msixtbl_pt_unregister(struct domain *d, int pirq)
+{
+    irq_desc_t *irq_desc;
+    struct msi_desc *msi_desc;
+    struct pci_dev *pdev;
+    struct msixtbl_entry *entry;
+
+    ASSERT(spin_is_locked(&pcidevs_lock));
+
+    irq_desc = domain_spin_lock_irq_desc(d, pirq, NULL);
+    if ( !irq_desc )
+        return;
+
+    if ( irq_desc->handler != &pci_msi_type )
+        goto out;
+
+    msi_desc = irq_desc->msi_desc;
+    if ( !msi_desc )
+        goto out;
+
+    pdev = msi_desc->dev;
+
+    spin_lock(&d->arch.hvm_domain.msixtbl_list_lock);
+
+    list_for_each_entry( entry, &d->arch.hvm_domain.msixtbl_list, list )
+        if ( pdev == entry->pdev )
+            goto found;
+
+    spin_unlock(&d->arch.hvm_domain.msixtbl_list_lock);
+
+
+out:
+    spin_unlock_irq(&irq_desc->lock);
+    return;
+
+found:
+    if ( !atomic_dec_and_test(&entry->refcnt) )
+        del_msixtbl_entry(entry);
+
+    spin_unlock(&d->arch.hvm_domain.msixtbl_list_lock);
+    spin_unlock_irq(&irq_desc->lock);
+}
+void msixtbl_pt_cleanup(struct domain *d, int pirq)
+{
+    struct msixtbl_entry *entry, *temp;
+    unsigned long flags;
+
+    /* msixtbl_list_lock must be acquired with irq_disabled for check_lock() */
+    local_irq_save(flags);
+    spin_lock(&d->arch.hvm_domain.msixtbl_list_lock);
+
+    list_for_each_entry_safe( entry, temp,
+                              &d->arch.hvm_domain.msixtbl_list, list )
+        del_msixtbl_entry(entry);
+
+    spin_unlock(&d->arch.hvm_domain.msixtbl_list_lock);
+    local_irq_restore(flags);
+}
diff --git a/arch/x86/msi.c b/arch/x86/msi.c
index 4c13ad3..6333e18 100644
--- a/arch/x86/msi.c
+++ b/arch/x86/msi.c
@@ -830,3 +830,23 @@ int pci_restore_msi_state(struct pci_dev *pdev)
     return 0;
 }
 
+unsigned int pci_msix_get_table_len(struct pci_dev *pdev)
+{
+    int pos;
+    u16 control;
+    u8 bus, slot, func;
+    unsigned int len;
+
+    bus = pdev->bus;
+    slot = PCI_SLOT(pdev->devfn);
+    func = PCI_FUNC(pdev->devfn);
+
+    pos = pci_find_cap_offset(bus, slot, func, PCI_CAP_ID_MSIX);
+    if ( !pos )
+        return 0;
+
+    control = pci_conf_read16(bus, slot, func, msix_control_reg(pos));
+    len = msix_table_size(control) * PCI_MSIX_ENTRY_SIZE;
+
+    return len;
+}
diff --git a/drivers/passthrough/io.c b/drivers/passthrough/io.c
index f62421b..8a818c2 100644
--- a/drivers/passthrough/io.c
+++ b/drivers/passthrough/io.c
@@ -90,7 +90,7 @@ int pt_irq_create_bind_vtd(
 
     if ( pt_irq_bind->irq_type == PT_IRQ_TYPE_MSI )
     {
-
+#ifdef SUPPORT_MSI_REMAPPING
         if ( !test_and_set_bit(pirq, hvm_irq_dpci->mapping))
         {
             set_bit(_HVM_IRQ_DPCI_MSI, &hvm_irq_dpci->mirq[pirq].flags);
@@ -99,6 +99,12 @@ int pt_irq_create_bind_vtd(
             hvm_irq_dpci->msi_gvec_pirq[pt_irq_bind->u.msi.gvec] = pirq;
             /* bind after hvm_irq_dpci is setup to avoid race with irq handler*/
             rc = pirq_guest_bind(d->vcpu[0], pirq, 0);
+            if ( rc == 0 && pt_irq_bind->u.msi.gtable )
+            {
+                rc = msixtbl_pt_register(d, pirq, pt_irq_bind->u.msi.gtable);
+                if ( unlikely(rc) )
+                    pirq_guest_unbind(d, pirq);
+            }
             if ( unlikely(rc) )
             {
                 hvm_irq_dpci->msi_gvec_pirq[pt_irq_bind->u.msi.gvec] = 0;
@@ -127,6 +133,9 @@ int pt_irq_create_bind_vtd(
             hvm_irq_dpci->mirq[pirq].gmsi.gflags = pt_irq_bind->u.msi.gflags;
             hvm_irq_dpci->msi_gvec_pirq[pt_irq_bind->u.msi.gvec] = pirq;
         }
+#else
+        return -ENOSYS;
+#endif
     }
     else
     {
@@ -247,6 +256,9 @@ int pt_irq_destroy_bind_vtd(
         if ( list_empty(&hvm_irq_dpci->mirq[machine_gsi].digl_list) )
         {
             pirq_guest_unbind(d, machine_gsi);
+#ifdef SUPPORT_MSI_REMAPPING
+            msixtbl_pt_unregister(d, machine_gsi);
+#endif
             kill_timer(&hvm_irq_dpci->hvm_timer[domain_irq_to_vector(d, machine_gsi)]);
             hvm_irq_dpci->mirq[machine_gsi].dom   = NULL;
             hvm_irq_dpci->mirq[machine_gsi].flags = 0;
diff --git a/include/asm-x86/hvm/domain.h b/include/asm-x86/hvm/domain.h
index f6c4c61..367ebae 100644
--- a/include/asm-x86/hvm/domain.h
+++ b/include/asm-x86/hvm/domain.h
@@ -66,6 +66,10 @@ struct hvm_domain {
     /* Pass-through */
     struct hvm_iommu       hvm_iommu;
 
+    /* hypervisor intercepted msix table */
+    struct list_head       msixtbl_list;
+    spinlock_t             msixtbl_list_lock;
+
 #if CONFIG_PAGING_LEVELS == 3
     bool_t                 amd_npt_4gb_warning;
 #endif
diff --git a/include/asm-x86/msi.h b/include/asm-x86/msi.h
index a5f67e1..56387c6 100644
--- a/include/asm-x86/msi.h
+++ b/include/asm-x86/msi.h
@@ -81,6 +81,8 @@ extern void teardown_msi_vector(int vector);
 extern int msi_free_vector(struct msi_desc *entry);
 extern int pci_restore_msi_state(struct pci_dev *pdev);
 
+extern unsigned int pci_msix_get_table_len(struct pci_dev *pdev);
+
 struct msi_desc {
 	struct {
 		__u8	type	: 5; 	/* {0: unused, 5h:MSI, 11h:MSI-X} */
diff --git a/include/public/domctl.h b/include/public/domctl.h
index 8542381..38ac7a9 100644
--- a/include/public/domctl.h
+++ b/include/public/domctl.h
@@ -480,6 +480,7 @@ struct xen_domctl_bind_pt_irq {
         struct {
             uint8_t gvec;
             uint32_t gflags;
+            uint64_t gtable;
         } msi;
     } u;
 };
diff --git a/include/xen/pci.h b/include/xen/pci.h
index 0fd22f0..120c02c 100644
--- a/include/xen/pci.h
+++ b/include/xen/pci.h
@@ -29,7 +29,8 @@
 #define PCI_BDF(b,d,f)  ((((b) & 0xff) << 8) | PCI_DEVFN(d,f))
 #define PCI_BDF2(b,df)  ((((b) & 0xff) << 8) | ((df) & 0xff))
 
-#define MAX_MSIX_TABLE_PAGES    8    /* 2048 entries */
+#define MAX_MSIX_TABLE_ENTRIES  2048
+#define MAX_MSIX_TABLE_PAGES    8
 struct pci_dev {
     struct list_head alldevs_list;
     struct list_head domain_list;
@@ -84,4 +85,7 @@ void pci_conf_write32(
 int pci_find_cap_offset(u8 bus, u8 dev, u8 func, u8 cap);
 int pci_find_next_cap(u8 bus, unsigned int devfn, u8 pos, int cap);
 
+int msixtbl_pt_register(struct domain *d, int pirq, uint64_t gtable);
+void msixtbl_pt_unregister(struct domain *d, int pirq);
+
 #endif /* __XEN_PCI_H__ */