KVM: arm64: vgic-its: Implement basic ITS register handlers
authorAndre Przywara <andre.przywara@arm.com>
Fri, 15 Jul 2016 11:43:32 +0000 (12:43 +0100)
committerMarc Zyngier <marc.zyngier@arm.com>
Mon, 18 Jul 2016 17:14:36 +0000 (18:14 +0100)
Add emulation for some basic MMIO registers used in the ITS emulation.
This includes:
- GITS_{CTLR,TYPER,IIDR}
- ID registers
- GITS_{CBASER,CREADR,CWRITER}
  (which implement the ITS command buffer handling)
- GITS_BASER<n>

Most of the handlers are pretty straight forward, only the CWRITER
handler is a bit more involved by taking the new its_cmd mutex and
then iterating over the command buffer.
The registers holding base addresses and attributes are sanitised before
storing them.

Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Reviewed-by: Marc Zyngier <marc.zyngier@arm.com>
Tested-by: Eric Auger <eric.auger@redhat.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
include/kvm/arm_vgic.h
virt/kvm/arm/vgic/vgic-its.c
virt/kvm/arm/vgic/vgic-mmio-v3.c
virt/kvm/arm/vgic/vgic-mmio.h
virt/kvm/arm/vgic/vgic.c

index 8609faced83e518b3fd5a6a4307248e7d136b461..61867492d361d08e67c9e0150f7cd1c28eb7b990 100644 (file)
@@ -22,6 +22,7 @@
 #include <linux/spinlock.h>
 #include <linux/types.h>
 #include <kvm/iodev.h>
+#include <linux/list.h>
 
 #define VGIC_V3_MAX_CPUS       255
 #define VGIC_V2_MAX_CPUS       8
@@ -136,6 +137,21 @@ struct vgic_its {
        bool                    enabled;
        bool                    initialized;
        struct vgic_io_device   iodev;
+
+       /* These registers correspond to GITS_BASER{0,1} */
+       u64                     baser_device_table;
+       u64                     baser_coll_table;
+
+       /* Protects the command queue */
+       struct mutex            cmd_lock;
+       u64                     cbaser;
+       u32                     creadr;
+       u32                     cwriter;
+
+       /* Protects the device and collection lists */
+       struct mutex            its_lock;
+       struct list_head        device_list;
+       struct list_head        collection_list;
 };
 
 struct vgic_dist {
index 6b47b367469027f9d7855c28bd00770877a09460..11cfe2f12c6cdab0d495f8eb64a25f53e2de75a0 100644 (file)
@@ -21,6 +21,7 @@
 #include <linux/kvm.h>
 #include <linux/kvm_host.h>
 #include <linux/interrupt.h>
+#include <linux/list.h>
 #include <linux/uaccess.h>
 
 #include <linux/irqchip/arm-gic-v3.h>
 #include "vgic.h"
 #include "vgic-mmio.h"
 
+struct its_device {
+       struct list_head dev_list;
+
+       /* the head for the list of ITTEs */
+       struct list_head itt_head;
+       u32 device_id;
+};
+
+#define COLLECTION_NOT_MAPPED ((u32)~0)
+
+struct its_collection {
+       struct list_head coll_list;
+
+       u32 collection_id;
+       u32 target_addr;
+};
+
+#define its_is_collection_mapped(coll) ((coll) && \
+                               ((coll)->target_addr != COLLECTION_NOT_MAPPED))
+
+struct its_itte {
+       struct list_head itte_list;
+
+       struct its_collection *collection;
+       u32 lpi;
+       u32 event_id;
+};
+
+/*
+ * We only implement 48 bits of PA at the moment, although the ITS
+ * supports more. Let's be restrictive here.
+ */
+#define CBASER_ADDRESS(x)      ((x) & GENMASK_ULL(47, 12))
+
+static unsigned long vgic_mmio_read_its_ctlr(struct kvm *vcpu,
+                                            struct vgic_its *its,
+                                            gpa_t addr, unsigned int len)
+{
+       u32 reg = 0;
+
+       mutex_lock(&its->cmd_lock);
+       if (its->creadr == its->cwriter)
+               reg |= GITS_CTLR_QUIESCENT;
+       if (its->enabled)
+               reg |= GITS_CTLR_ENABLE;
+       mutex_unlock(&its->cmd_lock);
+
+       return reg;
+}
+
+static void vgic_mmio_write_its_ctlr(struct kvm *kvm, struct vgic_its *its,
+                                    gpa_t addr, unsigned int len,
+                                    unsigned long val)
+{
+       its->enabled = !!(val & GITS_CTLR_ENABLE);
+}
+
+static unsigned long vgic_mmio_read_its_typer(struct kvm *kvm,
+                                             struct vgic_its *its,
+                                             gpa_t addr, unsigned int len)
+{
+       u64 reg = GITS_TYPER_PLPIS;
+
+       /*
+        * We use linear CPU numbers for redistributor addressing,
+        * so GITS_TYPER.PTA is 0.
+        * Also we force all PROPBASER registers to be the same, so
+        * CommonLPIAff is 0 as well.
+        * To avoid memory waste in the guest, we keep the number of IDBits and
+        * DevBits low - as least for the time being.
+        */
+       reg |= 0x0f << GITS_TYPER_DEVBITS_SHIFT;
+       reg |= 0x0f << GITS_TYPER_IDBITS_SHIFT;
+
+       return extract_bytes(reg, addr & 7, len);
+}
+
+static unsigned long vgic_mmio_read_its_iidr(struct kvm *kvm,
+                                            struct vgic_its *its,
+                                            gpa_t addr, unsigned int len)
+{
+       return (PRODUCT_ID_KVM << 24) | (IMPLEMENTER_ARM << 0);
+}
+
+static unsigned long vgic_mmio_read_its_idregs(struct kvm *kvm,
+                                              struct vgic_its *its,
+                                              gpa_t addr, unsigned int len)
+{
+       switch (addr & 0xffff) {
+       case GITS_PIDR0:
+               return 0x92;    /* part number, bits[7:0] */
+       case GITS_PIDR1:
+               return 0xb4;    /* part number, bits[11:8] */
+       case GITS_PIDR2:
+               return GIC_PIDR2_ARCH_GICv3 | 0x0b;
+       case GITS_PIDR4:
+               return 0x40;    /* This is a 64K software visible page */
+       /* The following are the ID registers for (any) GIC. */
+       case GITS_CIDR0:
+               return 0x0d;
+       case GITS_CIDR1:
+               return 0xf0;
+       case GITS_CIDR2:
+               return 0x05;
+       case GITS_CIDR3:
+               return 0xb1;
+       }
+
+       return 0;
+}
+
+/* Requires the its_lock to be held. */
+static void its_free_itte(struct kvm *kvm, struct its_itte *itte)
+{
+       list_del(&itte->itte_list);
+       kfree(itte);
+}
+
+static int vgic_its_handle_command(struct kvm *kvm, struct vgic_its *its,
+                                  u64 *its_cmd)
+{
+       return -ENODEV;
+}
+
+static u64 vgic_sanitise_its_baser(u64 reg)
+{
+       reg = vgic_sanitise_field(reg, GITS_BASER_SHAREABILITY_MASK,
+                                 GITS_BASER_SHAREABILITY_SHIFT,
+                                 vgic_sanitise_shareability);
+       reg = vgic_sanitise_field(reg, GITS_BASER_INNER_CACHEABILITY_MASK,
+                                 GITS_BASER_INNER_CACHEABILITY_SHIFT,
+                                 vgic_sanitise_inner_cacheability);
+       reg = vgic_sanitise_field(reg, GITS_BASER_OUTER_CACHEABILITY_MASK,
+                                 GITS_BASER_OUTER_CACHEABILITY_SHIFT,
+                                 vgic_sanitise_outer_cacheability);
+
+       /* Bits 15:12 contain bits 51:48 of the PA, which we don't support. */
+       reg &= ~GENMASK_ULL(15, 12);
+
+       /* We support only one (ITS) page size: 64K */
+       reg = (reg & ~GITS_BASER_PAGE_SIZE_MASK) | GITS_BASER_PAGE_SIZE_64K;
+
+       return reg;
+}
+
+static u64 vgic_sanitise_its_cbaser(u64 reg)
+{
+       reg = vgic_sanitise_field(reg, GITS_CBASER_SHAREABILITY_MASK,
+                                 GITS_CBASER_SHAREABILITY_SHIFT,
+                                 vgic_sanitise_shareability);
+       reg = vgic_sanitise_field(reg, GITS_CBASER_INNER_CACHEABILITY_MASK,
+                                 GITS_CBASER_INNER_CACHEABILITY_SHIFT,
+                                 vgic_sanitise_inner_cacheability);
+       reg = vgic_sanitise_field(reg, GITS_CBASER_OUTER_CACHEABILITY_MASK,
+                                 GITS_CBASER_OUTER_CACHEABILITY_SHIFT,
+                                 vgic_sanitise_outer_cacheability);
+
+       /*
+        * Sanitise the physical address to be 64k aligned.
+        * Also limit the physical addresses to 48 bits.
+        */
+       reg &= ~(GENMASK_ULL(51, 48) | GENMASK_ULL(15, 12));
+
+       return reg;
+}
+
+static unsigned long vgic_mmio_read_its_cbaser(struct kvm *kvm,
+                                              struct vgic_its *its,
+                                              gpa_t addr, unsigned int len)
+{
+       return extract_bytes(its->cbaser, addr & 7, len);
+}
+
+static void vgic_mmio_write_its_cbaser(struct kvm *kvm, struct vgic_its *its,
+                                      gpa_t addr, unsigned int len,
+                                      unsigned long val)
+{
+       /* When GITS_CTLR.Enable is 1, this register is RO. */
+       if (its->enabled)
+               return;
+
+       mutex_lock(&its->cmd_lock);
+       its->cbaser = update_64bit_reg(its->cbaser, addr & 7, len, val);
+       its->cbaser = vgic_sanitise_its_cbaser(its->cbaser);
+       its->creadr = 0;
+       /*
+        * CWRITER is architecturally UNKNOWN on reset, but we need to reset
+        * it to CREADR to make sure we start with an empty command buffer.
+        */
+       its->cwriter = its->creadr;
+       mutex_unlock(&its->cmd_lock);
+}
+
+#define ITS_CMD_BUFFER_SIZE(baser)     ((((baser) & 0xff) + 1) << 12)
+#define ITS_CMD_SIZE                   32
+#define ITS_CMD_OFFSET(reg)            ((reg) & GENMASK(19, 5))
+
+/*
+ * By writing to CWRITER the guest announces new commands to be processed.
+ * To avoid any races in the first place, we take the its_cmd lock, which
+ * protects our ring buffer variables, so that there is only one user
+ * per ITS handling commands at a given time.
+ */
+static void vgic_mmio_write_its_cwriter(struct kvm *kvm, struct vgic_its *its,
+                                       gpa_t addr, unsigned int len,
+                                       unsigned long val)
+{
+       gpa_t cbaser;
+       u64 cmd_buf[4];
+       u32 reg;
+
+       if (!its)
+               return;
+
+       mutex_lock(&its->cmd_lock);
+
+       reg = update_64bit_reg(its->cwriter, addr & 7, len, val);
+       reg = ITS_CMD_OFFSET(reg);
+       if (reg >= ITS_CMD_BUFFER_SIZE(its->cbaser)) {
+               mutex_unlock(&its->cmd_lock);
+               return;
+       }
+
+       its->cwriter = reg;
+       cbaser = CBASER_ADDRESS(its->cbaser);
+
+       while (its->cwriter != its->creadr) {
+               int ret = kvm_read_guest(kvm, cbaser + its->creadr,
+                                        cmd_buf, ITS_CMD_SIZE);
+               /*
+                * If kvm_read_guest() fails, this could be due to the guest
+                * programming a bogus value in CBASER or something else going
+                * wrong from which we cannot easily recover.
+                * According to section 6.3.2 in the GICv3 spec we can just
+                * ignore that command then.
+                */
+               if (!ret)
+                       vgic_its_handle_command(kvm, its, cmd_buf);
+
+               its->creadr += ITS_CMD_SIZE;
+               if (its->creadr == ITS_CMD_BUFFER_SIZE(its->cbaser))
+                       its->creadr = 0;
+       }
+
+       mutex_unlock(&its->cmd_lock);
+}
+
+static unsigned long vgic_mmio_read_its_cwriter(struct kvm *kvm,
+                                               struct vgic_its *its,
+                                               gpa_t addr, unsigned int len)
+{
+       return extract_bytes(its->cwriter, addr & 0x7, len);
+}
+
+static unsigned long vgic_mmio_read_its_creadr(struct kvm *kvm,
+                                              struct vgic_its *its,
+                                              gpa_t addr, unsigned int len)
+{
+       return extract_bytes(its->creadr, addr & 0x7, len);
+}
+
+#define BASER_INDEX(addr) (((addr) / sizeof(u64)) & 0x7)
+static unsigned long vgic_mmio_read_its_baser(struct kvm *kvm,
+                                             struct vgic_its *its,
+                                             gpa_t addr, unsigned int len)
+{
+       u64 reg;
+
+       switch (BASER_INDEX(addr)) {
+       case 0:
+               reg = its->baser_device_table;
+               break;
+       case 1:
+               reg = its->baser_coll_table;
+               break;
+       default:
+               reg = 0;
+               break;
+       }
+
+       return extract_bytes(reg, addr & 7, len);
+}
+
+#define GITS_BASER_RO_MASK     (GENMASK_ULL(52, 48) | GENMASK_ULL(58, 56))
+static void vgic_mmio_write_its_baser(struct kvm *kvm,
+                                     struct vgic_its *its,
+                                     gpa_t addr, unsigned int len,
+                                     unsigned long val)
+{
+       u64 entry_size, device_type;
+       u64 reg, *regptr, clearbits = 0;
+
+       /* When GITS_CTLR.Enable is 1, we ignore write accesses. */
+       if (its->enabled)
+               return;
+
+       switch (BASER_INDEX(addr)) {
+       case 0:
+               regptr = &its->baser_device_table;
+               entry_size = 8;
+               device_type = GITS_BASER_TYPE_DEVICE;
+               break;
+       case 1:
+               regptr = &its->baser_coll_table;
+               entry_size = 8;
+               device_type = GITS_BASER_TYPE_COLLECTION;
+               clearbits = GITS_BASER_INDIRECT;
+               break;
+       default:
+               return;
+       }
+
+       reg = update_64bit_reg(*regptr, addr & 7, len, val);
+       reg &= ~GITS_BASER_RO_MASK;
+       reg &= ~clearbits;
+
+       reg |= (entry_size - 1) << GITS_BASER_ENTRY_SIZE_SHIFT;
+       reg |= device_type << GITS_BASER_TYPE_SHIFT;
+       reg = vgic_sanitise_its_baser(reg);
+
+       *regptr = reg;
+}
+
 #define REGISTER_ITS_DESC(off, rd, wr, length, acc)            \
 {                                                              \
        .reg_offset = off,                                      \
        .its_write = wr,                                        \
 }
 
-static unsigned long its_mmio_read_raz(struct kvm *kvm, struct vgic_its *its,
-                                      gpa_t addr, unsigned int len)
-{
-       return 0;
-}
-
 static void its_mmio_write_wi(struct kvm *kvm, struct vgic_its *its,
                              gpa_t addr, unsigned int len, unsigned long val)
 {
@@ -55,28 +373,28 @@ static void its_mmio_write_wi(struct kvm *kvm, struct vgic_its *its,
 
 static struct vgic_register_region its_registers[] = {
        REGISTER_ITS_DESC(GITS_CTLR,
-               its_mmio_read_raz, its_mmio_write_wi, 4,
+               vgic_mmio_read_its_ctlr, vgic_mmio_write_its_ctlr, 4,
                VGIC_ACCESS_32bit),
        REGISTER_ITS_DESC(GITS_IIDR,
-               its_mmio_read_raz, its_mmio_write_wi, 4,
+               vgic_mmio_read_its_iidr, its_mmio_write_wi, 4,
                VGIC_ACCESS_32bit),
        REGISTER_ITS_DESC(GITS_TYPER,
-               its_mmio_read_raz, its_mmio_write_wi, 8,
+               vgic_mmio_read_its_typer, its_mmio_write_wi, 8,
                VGIC_ACCESS_64bit | VGIC_ACCESS_32bit),
        REGISTER_ITS_DESC(GITS_CBASER,
-               its_mmio_read_raz, its_mmio_write_wi, 8,
+               vgic_mmio_read_its_cbaser, vgic_mmio_write_its_cbaser, 8,
                VGIC_ACCESS_64bit | VGIC_ACCESS_32bit),
        REGISTER_ITS_DESC(GITS_CWRITER,
-               its_mmio_read_raz, its_mmio_write_wi, 8,
+               vgic_mmio_read_its_cwriter, vgic_mmio_write_its_cwriter, 8,
                VGIC_ACCESS_64bit | VGIC_ACCESS_32bit),
        REGISTER_ITS_DESC(GITS_CREADR,
-               its_mmio_read_raz, its_mmio_write_wi, 8,
+               vgic_mmio_read_its_creadr, its_mmio_write_wi, 8,
                VGIC_ACCESS_64bit | VGIC_ACCESS_32bit),
        REGISTER_ITS_DESC(GITS_BASER,
-               its_mmio_read_raz, its_mmio_write_wi, 0x40,
+               vgic_mmio_read_its_baser, vgic_mmio_write_its_baser, 0x40,
                VGIC_ACCESS_64bit | VGIC_ACCESS_32bit),
        REGISTER_ITS_DESC(GITS_IDREGS_BASE,
-               its_mmio_read_raz, its_mmio_write_wi, 0x30,
+               vgic_mmio_read_its_idregs, its_mmio_write_wi, 0x30,
                VGIC_ACCESS_32bit),
 };
 
@@ -109,6 +427,18 @@ static int vgic_its_init_its(struct kvm *kvm, struct vgic_its *its)
        return ret;
 }
 
+#define INITIAL_BASER_VALUE                                              \
+       (GIC_BASER_CACHEABILITY(GITS_BASER, INNER, RaWb)                | \
+        GIC_BASER_CACHEABILITY(GITS_BASER, OUTER, SameAsInner)         | \
+        GIC_BASER_SHAREABILITY(GITS_BASER, InnerShareable)             | \
+        ((8ULL - 1) << GITS_BASER_ENTRY_SIZE_SHIFT)                    | \
+        GITS_BASER_PAGE_SIZE_64K)
+
+#define INITIAL_PROPBASER_VALUE                                                  \
+       (GIC_BASER_CACHEABILITY(GICR_PROPBASER, INNER, RaWb)            | \
+        GIC_BASER_CACHEABILITY(GICR_PROPBASER, OUTER, SameAsInner)     | \
+        GIC_BASER_SHAREABILITY(GICR_PROPBASER, InnerShareable))
+
 static int vgic_its_create(struct kvm_device *dev, u32 type)
 {
        struct vgic_its *its;
@@ -120,12 +450,24 @@ static int vgic_its_create(struct kvm_device *dev, u32 type)
        if (!its)
                return -ENOMEM;
 
+       mutex_init(&its->its_lock);
+       mutex_init(&its->cmd_lock);
+
        its->vgic_its_base = VGIC_ADDR_UNDEF;
 
+       INIT_LIST_HEAD(&its->device_list);
+       INIT_LIST_HEAD(&its->collection_list);
+
        dev->kvm->arch.vgic.has_its = true;
        its->initialized = false;
        its->enabled = false;
 
+       its->baser_device_table = INITIAL_BASER_VALUE                   |
+               ((u64)GITS_BASER_TYPE_DEVICE << GITS_BASER_TYPE_SHIFT);
+       its->baser_coll_table = INITIAL_BASER_VALUE |
+               ((u64)GITS_BASER_TYPE_COLLECTION << GITS_BASER_TYPE_SHIFT);
+       dev->kvm->arch.vgic.propbaser = INITIAL_PROPBASER_VALUE;
+
        dev->private = its;
 
        return 0;
@@ -133,7 +475,36 @@ static int vgic_its_create(struct kvm_device *dev, u32 type)
 
 static void vgic_its_destroy(struct kvm_device *kvm_dev)
 {
+       struct kvm *kvm = kvm_dev->kvm;
        struct vgic_its *its = kvm_dev->private;
+       struct its_device *dev;
+       struct its_itte *itte;
+       struct list_head *dev_cur, *dev_temp;
+       struct list_head *cur, *temp;
+
+       /*
+        * We may end up here without the lists ever having been initialized.
+        * Check this and bail out early to avoid dereferencing a NULL pointer.
+        */
+       if (!its->device_list.next)
+               return;
+
+       mutex_lock(&its->its_lock);
+       list_for_each_safe(dev_cur, dev_temp, &its->device_list) {
+               dev = container_of(dev_cur, struct its_device, dev_list);
+               list_for_each_safe(cur, temp, &dev->itt_head) {
+                       itte = (container_of(cur, struct its_itte, itte_list));
+                       its_free_itte(kvm, itte);
+               }
+               list_del(dev_cur);
+               kfree(dev);
+       }
+
+       list_for_each_safe(cur, temp, &its->collection_list) {
+               list_del(cur);
+               kfree(container_of(cur, struct its_collection, coll_list));
+       }
+       mutex_unlock(&its->its_lock);
 
        kfree(its);
 }
index a5c35050c7864d4ba021233c28c644e1b80e0a59..84a301d789e0158d3e6c4ca9771879cee6e2df83 100644 (file)
 #include "vgic-mmio.h"
 
 /* extract @num bytes at @offset bytes offset in data */
-static unsigned long extract_bytes(unsigned long data, unsigned int offset,
-                                  unsigned int num)
+unsigned long extract_bytes(unsigned long data, unsigned int offset,
+                           unsigned int num)
 {
        return (data >> (offset * 8)) & GENMASK_ULL(num * 8 - 1, 0);
 }
 
 /* allows updates of any half of a 64-bit register (or the whole thing) */
-static u64 update_64bit_reg(u64 reg, unsigned int offset, unsigned int len,
-                           unsigned long val)
+u64 update_64bit_reg(u64 reg, unsigned int offset, unsigned int len,
+                    unsigned long val)
 {
        int lower = (offset & 4) * 8;
        int upper = lower + 8 * len - 1;
index 366d66378732b0eb389f562a59922266ca26fd8a..0b3ecf9d100eca59b27b84f2aa6d1461536443b5 100644 (file)
@@ -96,6 +96,12 @@ unsigned long vgic_data_mmio_bus_to_host(const void *val, unsigned int len);
 void vgic_data_host_to_mmio_bus(void *buf, unsigned int len,
                                unsigned long data);
 
+unsigned long extract_bytes(unsigned long data, unsigned int offset,
+                           unsigned int num);
+
+u64 update_64bit_reg(u64 reg, unsigned int offset, unsigned int len,
+                    unsigned long val);
+
 unsigned long vgic_mmio_read_raz(struct kvm_vcpu *vcpu,
                                 gpa_t addr, unsigned int len);
 
index fb19a554d090c03efa08dddf93019c3bfc385fa9..d3ba1b4227e7d3e37e3186b1aecc30c7a786e990 100644 (file)
@@ -33,10 +33,16 @@ struct vgic_global __section(.hyp.text) kvm_vgic_global_state;
 
 /*
  * Locking order is always:
- *   vgic_cpu->ap_list_lock
- *     vgic_irq->irq_lock
+ * its->cmd_lock (mutex)
+ *   its->its_lock (mutex)
+ *     vgic_cpu->ap_list_lock
+ *       vgic_irq->irq_lock
  *
- * (that is, always take the ap_list_lock before the struct vgic_irq lock).
+ * If you need to take multiple locks, always take the upper lock first,
+ * then the lower ones, e.g. first take the its_lock, then the irq_lock.
+ * If you are already holding a lock and need to take a higher one, you
+ * have to drop the lower ranking lock first and re-aquire it after having
+ * taken the upper one.
  *
  * When taking more than one ap_list_lock at the same time, always take the
  * lowest numbered VCPU's ap_list_lock first, so: