kernel_optimize_test/include/linux/lguest.h
Rusty Russell a32a8813d0 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>
2009-06-12 22:27:03 +09:30

65 lines
2.3 KiB
C

/* Things the lguest guest needs to know. Note: like all lguest interfaces,
* this is subject to wild and random change between versions. */
#ifndef _LINUX_LGUEST_H
#define _LINUX_LGUEST_H
#ifndef __ASSEMBLY__
#include <linux/time.h>
#include <asm/irq.h>
#include <asm/lguest_hcall.h>
#define LG_CLOCK_MIN_DELTA 100UL
#define LG_CLOCK_MAX_DELTA ULONG_MAX
/*G:032 The second method of communicating with the Host is to via "struct
* lguest_data". Once the Guest's initialization hypercall tells the Host where
* this is, the Guest and Host both publish information in it. :*/
struct lguest_data
{
/* 512 == enabled (same as eflags in normal hardware). The Guest
* changes interrupts so often that a hypercall is too slow. */
unsigned int irq_enabled;
/* Fine-grained interrupt disabling by the Guest */
DECLARE_BITMAP(blocked_interrupts, LGUEST_IRQS);
/* The Host writes the virtual address of the last page fault here,
* which saves the Guest a hypercall. CR2 is the native register where
* this address would normally be found. */
unsigned long cr2;
/* Wallclock time set by the Host. */
struct timespec time;
/* Interrupt pending set by the Host. The Guest should do a hypercall
* if it re-enables interrupts and sees this set (to X86_EFLAGS_IF). */
int irq_pending;
/* Async hypercall ring. Instead of directly making hypercalls, we can
* place them in here for processing the next time the Host wants.
* This batching can be quite efficient. */
/* 0xFF == done (set by Host), 0 == pending (set by Guest). */
u8 hcall_status[LHCALL_RING_SIZE];
/* The actual registers for the hypercalls. */
struct hcall_args hcalls[LHCALL_RING_SIZE];
/* Fields initialized by the Host at boot: */
/* Memory not to try to access */
unsigned long reserve_mem;
/* KHz for the TSC clock. */
u32 tsc_khz;
/* Page where the top-level pagetable is */
unsigned long pgdir;
/* Fields initialized by the Guest at boot: */
/* Instruction range to suppress interrupts even if enabled */
unsigned long noirq_start, noirq_end;
/* Address above which page tables are all identical. */
unsigned long kernel_address;
/* The vector to try to use for system calls (0x40 or 0x80). */
unsigned int syscall_vec;
};
extern struct lguest_data lguest_data;
#endif /* __ASSEMBLY__ */
#endif /* _LINUX_LGUEST_H */