Patchwork [v4] e1000: add interrupt mitigation support

login
register
mail settings
Submitter Vincenzo Maffione
Date July 29, 2013, 3:12 p.m.
Message ID <1375110769-17341-1-git-send-email-v.maffione@gmail.com>
Download mbox | patch
Permalink /patch/262815/
State New
Headers show

Comments

Vincenzo Maffione - July 29, 2013, 3:12 p.m.
This patch partially implements the e1000 interrupt mitigation mechanisms.
Using a single QEMUTimer, it emulates the ITR register (which is the newer
mitigation register, recommended by Intel) and approximately emulates
RADV and TADV registers. TIDV and RDTR register functionalities are not
emulated (RDTR is only used to validate RADV, according to the e1000 specs).

RADV, TADV, TIDV and RDTR registers make up the older e1000 mitigation
mechanism and would need a timer each to be completely emulated. However,
a single timer has been used in order to reach a good compromise between
emulation accuracy and simplicity/efficiency.

The implemented mechanism can be enabled/disabled specifying the command
line e1000-specific boolean parameter "mit", e.g.

    qemu-system-x86_64 -device e1000,mitigation=on,... ...

For more information, see the Software developer's manual at
http://download.intel.com/design/network/manuals/8254x_GBe_SDM.pdf.

Interrupt mitigation boosts performance when the guest suffers from
an high interrupt rate (i.e. receiving short UDP packets at high packet
rate). For some numerical results see the following link
http://info.iet.unipi.it/~luigi/papers/20130520-rizzo-vm.pdf

Signed-off-by: Vincenzo Maffione <v.maffione@gmail.com>
---
I removed "mit_timer_on" and "mit_ide" from the VMstate subsection,
since in the migration scheme they are always loaded as 0 on
the destination QEMU instance. This also ensures that the mitigation
timer is never active across a migration, so that we don't need
to call qemu_mod_timer() in e1000_post_load().

 hw/net/e1000.c       | 131 +++++++++++++++++++++++++++++++++++++++++++++++++--
 include/hw/i386/pc.h |   4 ++
 2 files changed, 132 insertions(+), 3 deletions(-)
Stefan Hajnoczi - July 30, 2013, 10:58 a.m.
On Mon, Jul 29, 2013 at 05:12:49PM +0200, Vincenzo Maffione wrote:
> This patch partially implements the e1000 interrupt mitigation mechanisms.
> Using a single QEMUTimer, it emulates the ITR register (which is the newer
> mitigation register, recommended by Intel) and approximately emulates
> RADV and TADV registers. TIDV and RDTR register functionalities are not
> emulated (RDTR is only used to validate RADV, according to the e1000 specs).
> 
> RADV, TADV, TIDV and RDTR registers make up the older e1000 mitigation
> mechanism and would need a timer each to be completely emulated. However,
> a single timer has been used in order to reach a good compromise between
> emulation accuracy and simplicity/efficiency.
> 
> The implemented mechanism can be enabled/disabled specifying the command
> line e1000-specific boolean parameter "mit", e.g.
> 
>     qemu-system-x86_64 -device e1000,mitigation=on,... ...
> 
> For more information, see the Software developer's manual at
> http://download.intel.com/design/network/manuals/8254x_GBe_SDM.pdf.
> 
> Interrupt mitigation boosts performance when the guest suffers from
> an high interrupt rate (i.e. receiving short UDP packets at high packet
> rate). For some numerical results see the following link
> http://info.iet.unipi.it/~luigi/papers/20130520-rizzo-vm.pdf
> 
> Signed-off-by: Vincenzo Maffione <v.maffione@gmail.com>
> ---
> I removed "mit_timer_on" and "mit_ide" from the VMstate subsection,
> since in the migration scheme they are always loaded as 0 on
> the destination QEMU instance. This also ensures that the mitigation
> timer is never active across a migration, so that we don't need
> to call qemu_mod_timer() in e1000_post_load().
> 
>  hw/net/e1000.c       | 131 +++++++++++++++++++++++++++++++++++++++++++++++++--
>  include/hw/i386/pc.h |   4 ++
>  2 files changed, 132 insertions(+), 3 deletions(-)
...
> diff --git a/include/hw/i386/pc.h b/include/hw/i386/pc.h
> index 7fb97b0..2e35263 100644
> --- a/include/hw/i386/pc.h
> +++ b/include/hw/i386/pc.h
> @@ -235,6 +235,10 @@ int e820_add_entry(uint64_t, uint64_t, uint32_t);
>              .driver   = "virtio-net-pci",\
>              .property = "any_layout",\
>              .value    = "off",\
> +        },{\
> +            .driver   = "e1000",\
> +            .property = "mitigation",\
> +            .value    = "off",\
>          }

Looks good.  I'd like to merge this for QEMU 1.7.  QEMU 1.6 is in hard
freeze right now and we cannot add new features.

In order to merge it into the net-next tree this hunk needs to be moved
from PC_COMPAT_1_5 to PC_COMPAT_1_6 (does not exist yet).  Please
introduce PC_COMPAT_1_6 in include/hw/i386/pc.h (and also update
hw/i386/pc_piix.c and hw/i386/pc_q35.c machine types) in a separate
commit.

Stefan

Patch

diff --git a/hw/net/e1000.c b/hw/net/e1000.c
index b952d8d..ae16b57 100644
--- a/hw/net/e1000.c
+++ b/hw/net/e1000.c
@@ -135,9 +135,16 @@  typedef struct E1000State_st {
 
     QEMUTimer *autoneg_timer;
 
+    QEMUTimer *mit_timer;      /* Mitigation timer. */
+    bool mit_timer_on;         /* Mitigation timer is running. */
+    bool mit_irq_level;        /* Tracks interrupt pin level. */
+    uint32_t mit_ide;          /* Tracks E1000_TXD_CMD_IDE bit. */
+
 /* Compatibility flags for migration to/from qemu 1.3.0 and older */
 #define E1000_FLAG_AUTONEG_BIT 0
+#define E1000_FLAG_MIT_BIT 1
 #define E1000_FLAG_AUTONEG (1 << E1000_FLAG_AUTONEG_BIT)
+#define E1000_FLAG_MIT (1 << E1000_FLAG_MIT_BIT)
     uint32_t compat_flags;
 } E1000State;
 
@@ -158,7 +165,8 @@  enum {
     defreg(TORH),	defreg(TORL),	defreg(TOTH),	defreg(TOTL),
     defreg(TPR),	defreg(TPT),	defreg(TXDCTL),	defreg(WUFC),
     defreg(RA),		defreg(MTA),	defreg(CRCERRS),defreg(VFTA),
-    defreg(VET),
+    defreg(VET),        defreg(RDTR),   defreg(RADV),   defreg(TADV),
+    defreg(ITR),
 };
 
 static void
@@ -245,10 +253,21 @@  static const uint32_t mac_reg_init[] = {
                 E1000_MANC_RMCP_EN,
 };
 
+/* Helper function, *curr == 0 means the value is not set */
+static inline void
+mit_update_delay(uint32_t *curr, uint32_t value)
+{
+    if (value && (*curr == 0 || value < *curr)) {
+        *curr = value;
+    }
+}
+
 static void
 set_interrupt_cause(E1000State *s, int index, uint32_t val)
 {
     PCIDevice *d = PCI_DEVICE(s);
+    uint32_t pending_ints;
+    uint32_t mit_delay;
 
     if (val && (E1000_DEVID >= E1000_DEV_ID_82547EI_MOBILE)) {
         /* Only for 8257x */
@@ -266,7 +285,57 @@  set_interrupt_cause(E1000State *s, int index, uint32_t val)
      */
     s->mac_reg[ICS] = val;
 
-    qemu_set_irq(d->irq[0], (s->mac_reg[IMS] & s->mac_reg[ICR]) != 0);
+    pending_ints = (s->mac_reg[IMS] & s->mac_reg[ICR]);
+    if (!s->mit_irq_level && pending_ints) {
+        /*
+         * Here we detect a potential raising edge. We postpone raising the
+         * interrupt line if we are inside the mitigation delay window
+         * (s->mit_timer_on == 1).
+         * We provide a partial implementation of interrupt mitigation,
+         * emulating only RADV, TADV and ITR (lower 16 bits, 1024ns units for
+         * RADV and TADV, 256ns units for ITR). RDTR is only used to enable
+         * RADV; relative timers based on TIDV and RDTR are not implemented.
+         */
+        if (s->mit_timer_on) {
+            return;
+        }
+        if (s->compat_flags & E1000_FLAG_MIT) {
+            /* Compute the next mitigation delay according to pending
+             * interrupts and the current values of RADV (provided
+             * RDTR!=0), TADV and ITR.
+             * Then rearm the timer.
+             */
+            mit_delay = 0;
+            if (s->mit_ide &&
+                    (pending_ints & (E1000_ICR_TXQE | E1000_ICR_TXDW))) {
+                mit_update_delay(&mit_delay, s->mac_reg[TADV] * 4);
+            }
+            if (s->mac_reg[RDTR] && (pending_ints & E1000_ICS_RXT0)) {
+                mit_update_delay(&mit_delay, s->mac_reg[RADV] * 4);
+            }
+            mit_update_delay(&mit_delay, s->mac_reg[ITR]);
+
+            if (mit_delay) {
+                s->mit_timer_on = 1;
+                qemu_mod_timer(s->mit_timer,
+                        qemu_get_clock_ns(vm_clock) + mit_delay * 256);
+            }
+            s->mit_ide = 0;
+        }
+    }
+
+    s->mit_irq_level = (pending_ints != 0);
+    qemu_set_irq(d->irq[0], s->mit_irq_level);
+}
+
+static void
+e1000_mit_timer(void *opaque)
+{
+    E1000State *s = opaque;
+
+    s->mit_timer_on = 0;
+    /* Call set_interrupt_cause to update the irq level (if necessary). */
+    set_interrupt_cause(s, 0, s->mac_reg[ICR]);
 }
 
 static void
@@ -307,6 +376,10 @@  static void e1000_reset(void *opaque)
     int i;
 
     qemu_del_timer(d->autoneg_timer);
+    qemu_del_timer(d->mit_timer);
+    d->mit_timer_on = 0;
+    d->mit_irq_level = 0;
+    d->mit_ide = 0;
     memset(d->phy_reg, 0, sizeof d->phy_reg);
     memmove(d->phy_reg, phy_reg_init, sizeof phy_reg_init);
     memset(d->mac_reg, 0, sizeof d->mac_reg);
@@ -572,6 +645,7 @@  process_tx_desc(E1000State *s, struct e1000_tx_desc *dp)
     struct e1000_context_desc *xp = (struct e1000_context_desc *)dp;
     struct e1000_tx *tp = &s->tx;
 
+    s->mit_ide |= (txd_lower & E1000_TXD_CMD_IDE);
     if (dtype == E1000_TXD_CMD_DEXT) {	// context descriptor
         op = le32_to_cpu(xp->cmd_and_length);
         tp->ipcss = xp->lower_setup.ip_fields.ipcss;
@@ -1047,7 +1121,8 @@  static uint32_t (*macreg_readops[])(E1000State *, int) = {
     getreg(TORL),	getreg(TOTL),	getreg(IMS),	getreg(TCTL),
     getreg(RDH),	getreg(RDT),	getreg(VET),	getreg(ICS),
     getreg(TDBAL),	getreg(TDBAH),	getreg(RDBAH),	getreg(RDBAL),
-    getreg(TDLEN),	getreg(RDLEN),
+    getreg(TDLEN),      getreg(RDLEN),  getreg(RDTR),   getreg(RADV),
+    getreg(TADV),       getreg(ITR),
 
     [TOTH] = mac_read_clr8,	[TORH] = mac_read_clr8,	[GPRC] = mac_read_clr4,
     [GPTC] = mac_read_clr4,	[TPR] = mac_read_clr4,	[TPT] = mac_read_clr4,
@@ -1069,6 +1144,8 @@  static void (*macreg_writeops[])(E1000State *, int, uint32_t) = {
     [TDH] = set_16bit,	[RDH] = set_16bit,	[RDT] = set_rdt,
     [IMC] = set_imc,	[IMS] = set_ims,	[ICR] = set_icr,
     [EECD] = set_eecd,	[RCTL] = set_rx_control, [CTRL] = set_ctrl,
+    [RDTR] = set_16bit, [RADV] = set_16bit,     [TADV] = set_16bit,
+    [ITR] = set_16bit,
     [RA ... RA+31] = &mac_writereg,
     [MTA ... MTA+127] = &mac_writereg,
     [VFTA ... VFTA+127] = &mac_writereg,
@@ -1150,6 +1227,11 @@  static void e1000_pre_save(void *opaque)
     E1000State *s = opaque;
     NetClientState *nc = qemu_get_queue(s->nic);
 
+    /* If the mitigation timer is active, emulate a timeout now. */
+    if (s->mit_timer_on) {
+        e1000_mit_timer(s);
+    }
+
     if (!(s->compat_flags & E1000_FLAG_AUTONEG)) {
         return;
     }
@@ -1171,6 +1253,14 @@  static int e1000_post_load(void *opaque, int version_id)
     E1000State *s = opaque;
     NetClientState *nc = qemu_get_queue(s->nic);
 
+    if (!(s->compat_flags & E1000_FLAG_MIT)) {
+        s->mac_reg[ITR] = s->mac_reg[RDTR] = s->mac_reg[RADV] =
+            s->mac_reg[TADV] = 0;
+        s->mit_irq_level = false;
+    }
+    s->mit_ide = 0;
+    s->mit_timer_on = false;
+
     /* nc.link_down can't be migrated, so infer link_down according
      * to link status bit in mac_reg[STATUS].
      * Alternatively, restart link negotiation if it was in progress. */
@@ -1190,6 +1280,28 @@  static int e1000_post_load(void *opaque, int version_id)
     return 0;
 }
 
+static bool e1000_mit_state_needed(void *opaque)
+{
+    E1000State *s = opaque;
+
+    return s->compat_flags & E1000_FLAG_MIT;
+}
+
+static const VMStateDescription vmstate_e1000_mit_state = {
+    .name = "e1000/mit_state",
+    .version_id = 1,
+    .minimum_version_id = 1,
+    .minimum_version_id_old = 1,
+    .fields    = (VMStateField[]) {
+        VMSTATE_UINT32(mac_reg[RDTR], E1000State),
+        VMSTATE_UINT32(mac_reg[RADV], E1000State),
+        VMSTATE_UINT32(mac_reg[TADV], E1000State),
+        VMSTATE_UINT32(mac_reg[ITR], E1000State),
+        VMSTATE_BOOL(mit_irq_level, E1000State),
+        VMSTATE_END_OF_LIST()
+    }
+};
+
 static const VMStateDescription vmstate_e1000 = {
     .name = "e1000",
     .version_id = 2,
@@ -1267,6 +1379,14 @@  static const VMStateDescription vmstate_e1000 = {
         VMSTATE_UINT32_SUB_ARRAY(mac_reg, E1000State, MTA, 128),
         VMSTATE_UINT32_SUB_ARRAY(mac_reg, E1000State, VFTA, 128),
         VMSTATE_END_OF_LIST()
+    },
+    .subsections = (VMStateSubsection[]) {
+        {
+            .vmsd = &vmstate_e1000_mit_state,
+            .needed = e1000_mit_state_needed,
+        }, {
+            /* empty */
+        }
     }
 };
 
@@ -1316,6 +1436,8 @@  pci_e1000_uninit(PCIDevice *dev)
 
     qemu_del_timer(d->autoneg_timer);
     qemu_free_timer(d->autoneg_timer);
+    qemu_del_timer(d->mit_timer);
+    qemu_free_timer(d->mit_timer);
     memory_region_destroy(&d->mmio);
     memory_region_destroy(&d->io);
     qemu_del_nic(d->nic);
@@ -1371,6 +1493,7 @@  static int pci_e1000_init(PCIDevice *pci_dev)
     add_boot_device_path(d->conf.bootindex, dev, "/ethernet-phy@0");
 
     d->autoneg_timer = qemu_new_timer_ms(vm_clock, e1000_autoneg_timer, d);
+    d->mit_timer = qemu_new_timer_ns(vm_clock, e1000_mit_timer, d);
 
     return 0;
 }
@@ -1385,6 +1508,8 @@  static Property e1000_properties[] = {
     DEFINE_NIC_PROPERTIES(E1000State, conf),
     DEFINE_PROP_BIT("autonegotiation", E1000State,
                     compat_flags, E1000_FLAG_AUTONEG_BIT, true),
+    DEFINE_PROP_BIT("mitigation", E1000State,
+                    compat_flags, E1000_FLAG_MIT_BIT, true),
     DEFINE_PROP_END_OF_LIST(),
 };
 
diff --git a/include/hw/i386/pc.h b/include/hw/i386/pc.h
index 7fb97b0..2e35263 100644
--- a/include/hw/i386/pc.h
+++ b/include/hw/i386/pc.h
@@ -235,6 +235,10 @@  int e820_add_entry(uint64_t, uint64_t, uint32_t);
             .driver   = "virtio-net-pci",\
             .property = "any_layout",\
             .value    = "off",\
+        },{\
+            .driver   = "e1000",\
+            .property = "mitigation",\
+            .value    = "off",\
         }
 
 #define PC_COMPAT_1_4 \