qemu-devel.nongnu.org archive mirror
 help / color / mirror / Atom feed
From: Vincenzo Maffione <v.maffione@gmail.com>
To: qemu-devel@nongnu.org
Cc: aliguori@us.ibm.com, mst@redhat.com, rizzo@iet.unipi.it,
	jasowang@redhat.com, Vincenzo Maffione <v.maffione@gmail.com>,
	stefanha@redhat.com, pbonzini@redhat.com,
	g.lettieri@iet.unipi.it, lersek@redhat.com, afaerber@suse.de
Subject: [Qemu-devel] [PATCH v5 1/2] e1000: add interrupt mitigation support
Date: Wed, 31 Jul 2013 10:13:49 +0200	[thread overview]
Message-ID: <1375258430-2120-2-git-send-email-v.maffione@gmail.com> (raw)
In-Reply-To: <1375258430-2120-1-git-send-email-v.maffione@gmail.com>

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>
---
 hw/net/e1000.c | 131 +++++++++++++++++++++++++++++++++++++++++++++++++++++++--
 1 file changed, 128 insertions(+), 3 deletions(-)

diff --git a/hw/net/e1000.c b/hw/net/e1000.c
index fdb1f89..32014c2 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(),
 };
 
-- 
1.8.3.4

  reply	other threads:[~2013-07-31 21:40 UTC|newest]

Thread overview: 13+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2013-07-31  8:13 [Qemu-devel] [PATCH v5 0/2] e1000: add interrupt mitigation support Vincenzo Maffione
2013-07-31  8:13 ` Vincenzo Maffione [this message]
2013-07-31 10:05   ` [Qemu-devel] [PATCH v5 1/2] " Andreas Färber
2013-07-31  8:13 ` [Qemu-devel] [PATCH v5 2/2] i386/pc: introducing compat_props for pc-1.6 Vincenzo Maffione
2013-07-31 10:00   ` Andreas Färber
2013-07-31  9:56 ` [Qemu-devel] [PATCH v5 0/2] e1000: add interrupt mitigation support Andreas Färber
2013-07-31 10:19   ` Vincenzo Maffione
2013-07-31 11:20     ` Stefan Hajnoczi
2013-07-31 11:26       ` Andreas Färber
2013-07-31 13:39       ` Vincenzo Maffione
2013-08-01  9:38         ` Stefan Hajnoczi
2013-08-01 14:07           ` Andreas Färber
2013-08-01 20:17             ` Vincenzo Maffione

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=1375258430-2120-2-git-send-email-v.maffione@gmail.com \
    --to=v.maffione@gmail.com \
    --cc=afaerber@suse.de \
    --cc=aliguori@us.ibm.com \
    --cc=g.lettieri@iet.unipi.it \
    --cc=jasowang@redhat.com \
    --cc=lersek@redhat.com \
    --cc=mst@redhat.com \
    --cc=pbonzini@redhat.com \
    --cc=qemu-devel@nongnu.org \
    --cc=rizzo@iet.unipi.it \
    --cc=stefanha@redhat.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).