lguest: improve interrupt handling, speed up stream networking

lguest never checked for pending interrupts when enabling interrupts, and
things still worked.  However, it makes a significant difference to TCP
performance, so it's time we fixed it by introducing a pending_irq flag
and checking it on irq_restore and irq_enable.

These two routines are now too big to patch into the 8/10 bytes
patch space, so we drop that code.

Note: The high latency on interrupt delivery had a very curious
effect: once everything else was optimized, networking without GSO was
faster than networking with GSO, since more interrupts were sent and
hence a greater chance of one getting through to the Guest!

Note2: (Almost) Closing the same loophole for iret doesn't have any
measurable effect, so I'm leaving that patch for the moment.

Before:
	1GB tcpblast Guest->Host:		30.7 seconds
	1GB tcpblast Guest->Host (no GSO):	76.0 seconds

After:
	1GB tcpblast Guest->Host:		6.8 seconds
	1GB tcpblast Guest->Host (no GSO):	27.8 seconds

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

2 files changed, 15 insertions, 8 deletions

diff --git a/arch/x86/lguest/boot.c b/arch/x86/lguest/boot.c
index 2392a7a171c2..37b8c1d3e022 100644
--- a/arch/x86/lguest/boot.c
+++ b/arch/x86/lguest/boot.c
@@ -205,6 +205,12 @@ PV_CALLEE_SAVE_REGS_THUNK(save_fl);
 static void restore_fl(unsigned long flags)
 {
 	lguest_data.irq_enabled = flags;
+	mb();
+	/* Null hcall forces interrupt delivery now, if irq_pending is
+	 * set to X86_EFLAGS_IF (ie. an interrupt is pending, and flags
+	 * enables interrupts. */
+	if (flags & lguest_data.irq_pending)
+		kvm_hypercall0(LHCALL_SEND_INTERRUPTS);
 }
 PV_CALLEE_SAVE_REGS_THUNK(restore_fl);
 
@@ -219,6 +225,11 @@ PV_CALLEE_SAVE_REGS_THUNK(irq_disable);
 static void irq_enable(void)
 {
 	lguest_data.irq_enabled = X86_EFLAGS_IF;
+	mb();
+	/* Null hcall forces interrupt delivery now. */
+	if (lguest_data.irq_pending)
+		kvm_hypercall0(LHCALL_SEND_INTERRUPTS);
+
 }
 PV_CALLEE_SAVE_REGS_THUNK(irq_enable);
 
@@ -972,10 +983,10 @@ static void lguest_restart(char *reason)
  *
  * Our current solution is to allow the paravirt back end to optionally patch
  * over the indirect calls to replace them with something more efficient.  We
- * patch the four most commonly called functions: disable interrupts, enable
- * interrupts, restore interrupts and save interrupts.  We usually have 6 or 10
- * bytes to patch into: the Guest versions of these operations are small enough
- * that we can fit comfortably.
+ * patch two of the simplest of the most commonly called functions: disable
+ * interrupts and save interrupts.  We usually have 6 or 10 bytes to patch
+ * into: the Guest versions of these operations are small enough that we can
+ * fit comfortably.
  *
  * First we need assembly templates of each of the patchable Guest operations,
  * and these are in i386_head.S. */
@@ -986,8 +997,6 @@ static const struct lguest_insns
 	const char *start, *end;
 } lguest_insns[] = {
 	[PARAVIRT_PATCH(pv_irq_ops.irq_disable)] = { lgstart_cli, lgend_cli },
-	[PARAVIRT_PATCH(pv_irq_ops.irq_enable)] = { lgstart_sti, lgend_sti },
-	[PARAVIRT_PATCH(pv_irq_ops.restore_fl)] = { lgstart_popf, lgend_popf },
 	[PARAVIRT_PATCH(pv_irq_ops.save_fl)] = { lgstart_pushf, lgend_pushf },
 };
 
diff --git a/arch/x86/lguest/i386_head.S b/arch/x86/lguest/i386_head.S
index f79541989471..3e0c5545d59c 100644
--- a/arch/x86/lguest/i386_head.S
+++ b/arch/x86/lguest/i386_head.S
@@ -46,8 +46,6 @@ ENTRY(lguest_entry)
 	.globl lgstart_##name; .globl lgend_##name
 
 LGUEST_PATCH(cli, movl $0, lguest_data+LGUEST_DATA_irq_enabled)
-LGUEST_PATCH(sti, movl $X86_EFLAGS_IF, lguest_data+LGUEST_DATA_irq_enabled)
-LGUEST_PATCH(popf, movl %eax, lguest_data+LGUEST_DATA_irq_enabled)
 LGUEST_PATCH(pushf, movl lguest_data+LGUEST_DATA_irq_enabled, %eax)
 /*:*/