Mailing List Archive

[xen-unstable] x86/hvm: accelerate I/O intercept handling
# HG changeset patch
# User Keir Fraser <keir.fraser@citrix.com>
# Date 1270027279 -3600
# Node ID e5e1e1532b9738f29af947d18e6546d9f0735919
# Parent 4f796e29987c0db1579787fe0b7d3d5af00963ea
x86/hvm: accelerate I/O intercept handling

currently we go through the emulator every time a HVM guest does an
I/O port access (in/out). This is unnecessary most of the times, as
both VMX and SVM provide all the necessary information already in the
VMCS/VMCB. String instructions are not covered by this shortcut, but
they are quite rare and we would need to access the guest memory
anyway. This patch decodes the information from VMCB/VMCS and calls a
simple handle_mmio wrapper. In handle_mmio() itself the emulation part
will simply be skipped, this approach avoids code duplication. Since
the vendor specific part is quite trivial, I implemented both the VMX
and SVM part, please check the VMX part for sanity.

I boot-tested both versions and ran some simple benchmarks. A micro
benchmark (hammering an I/O port in a tight loop) shows a significant
performance improvement (down to 66% of the time needed to handle the
intercept on an AMD K8, measured in the guest with TSC). Even with
reading a 1GB file from an emulated IDE harddisk (Dom0 cached) I could
get a 4-5% improvement. Some guest code (e.g. the TCP stack in some
Windows version) exercises the PM-Timer I/O port (0x1F48) very often
(multiple 10,000 times per second), these workloads also benefit with
up to 5% improvement from this patch.

Signed-off-by: Andre Przywara <andre.przywara@amd.com>
---
xen/arch/x86/hvm/emulate.c | 2 -
xen/arch/x86/hvm/io.c | 63 ++++++++++++++++++++++++++++----------
xen/arch/x86/hvm/svm/emulate.c | 3 +
xen/arch/x86/hvm/svm/svm.c | 26 +++++++++++++++
xen/arch/x86/hvm/vmx/vmx.c | 29 ++++++++++++++++-
xen/include/asm-x86/hvm/emulate.h | 4 ++
xen/include/asm-x86/hvm/io.h | 1
xen/include/asm-x86/hvm/vcpu.h | 4 ++
8 files changed, 113 insertions(+), 19 deletions(-)

diff -r 4f796e29987c -r e5e1e1532b97 xen/arch/x86/hvm/emulate.c
--- a/xen/arch/x86/hvm/emulate.c Wed Mar 31 10:12:29 2010 +0100
+++ b/xen/arch/x86/hvm/emulate.c Wed Mar 31 10:21:19 2010 +0100
@@ -48,7 +48,7 @@ static void hvmtrace_io_assist(int is_mm
trace_var(event, 0/*!cycles*/, size, buffer);
}

-static int hvmemul_do_io(
+int hvmemul_do_io(
int is_mmio, paddr_t addr, unsigned long *reps, int size,
paddr_t ram_gpa, int dir, int df, void *p_data)
{
diff -r 4f796e29987c -r e5e1e1532b97 xen/arch/x86/hvm/io.c
--- a/xen/arch/x86/hvm/io.c Wed Mar 31 10:12:29 2010 +0100
+++ b/xen/arch/x86/hvm/io.c Wed Mar 31 10:21:19 2010 +0100
@@ -171,10 +171,22 @@ int handle_mmio(void)
struct hvm_emulate_ctxt ctxt;
struct vcpu *curr = current;
int rc;
-
- hvm_emulate_prepare(&ctxt, guest_cpu_user_regs());
-
- rc = hvm_emulate_one(&ctxt);
+ unsigned long data, reps = 1;
+
+ if ( curr->arch.hvm_vcpu.io_size == 0 ) {
+ hvm_emulate_prepare(&ctxt, guest_cpu_user_regs());
+ rc = hvm_emulate_one(&ctxt);
+ } else {
+ if ( curr->arch.hvm_vcpu.io_dir == 0 )
+ data = guest_cpu_user_regs()->eax;
+ rc = hvmemul_do_io(0, curr->arch.hvm_vcpu.io_port, &reps,
+ curr->arch.hvm_vcpu.io_size, 0,
+ curr->arch.hvm_vcpu.io_dir, 0, &data);
+ if ( curr->arch.hvm_vcpu.io_dir == 1 && rc == X86EMUL_OKAY ) {
+ memcpy(&(guest_cpu_user_regs()->eax),
+ &data, curr->arch.hvm_vcpu.io_size);
+ }
+ }

if ( curr->arch.hvm_vcpu.io_state == HVMIO_awaiting_completion )
curr->arch.hvm_vcpu.io_state = HVMIO_handle_mmio_awaiting_completion;
@@ -184,14 +196,21 @@ int handle_mmio(void)
switch ( rc )
{
case X86EMUL_UNHANDLEABLE:
- gdprintk(XENLOG_WARNING,
- "MMIO emulation failed @ %04x:%lx: "
- "%02x %02x %02x %02x %02x %02x\n",
- hvmemul_get_seg_reg(x86_seg_cs, &ctxt)->sel,
- ctxt.insn_buf_eip,
- ctxt.insn_buf[0], ctxt.insn_buf[1],
- ctxt.insn_buf[2], ctxt.insn_buf[3],
- ctxt.insn_buf[4], ctxt.insn_buf[5]);
+ if ( curr->arch.hvm_vcpu.io_size == 0 )
+ gdprintk(XENLOG_WARNING,
+ "MMIO emulation failed @ %04x:%lx: "
+ "%02x %02x %02x %02x %02x %02x\n",
+ hvmemul_get_seg_reg(x86_seg_cs, &ctxt)->sel,
+ ctxt.insn_buf_eip,
+ ctxt.insn_buf[0], ctxt.insn_buf[1],
+ ctxt.insn_buf[2], ctxt.insn_buf[3],
+ ctxt.insn_buf[4], ctxt.insn_buf[5]);
+ else
+ gdprintk(XENLOG_WARNING,
+ "I/O emulation failed: %s 0x%04x, %i bytes, data=%08lx\n",
+ curr->arch.hvm_vcpu.io_dir ? "in" : "out",
+ curr->arch.hvm_vcpu.io_port,
+ curr->arch.hvm_vcpu.io_size, data);
return 0;
case X86EMUL_EXCEPTION:
if ( ctxt.exn_pending )
@@ -201,15 +220,29 @@ int handle_mmio(void)
break;
}

- hvm_emulate_writeback(&ctxt);
-
- return 1;
+ if ( curr->arch.hvm_vcpu.io_size == 0 )
+ hvm_emulate_writeback(&ctxt);
+ else
+ curr->arch.hvm_vcpu.io_size = 0;
+
+ if (rc == X86EMUL_RETRY)
+ return rc;
+ else
+ return 1;
}

int handle_mmio_with_translation(unsigned long gva, unsigned long gpfn)
{
current->arch.hvm_vcpu.mmio_gva = gva & PAGE_MASK;
current->arch.hvm_vcpu.mmio_gpfn = gpfn;
+ return handle_mmio();
+}
+
+int handle_mmio_decoded(uint16_t port, int size, int dir)
+{
+ current->arch.hvm_vcpu.io_port = port;
+ current->arch.hvm_vcpu.io_size = size;
+ current->arch.hvm_vcpu.io_dir = dir;
return handle_mmio();
}

diff -r 4f796e29987c -r e5e1e1532b97 xen/arch/x86/hvm/svm/emulate.c
--- a/xen/arch/x86/hvm/svm/emulate.c Wed Mar 31 10:12:29 2010 +0100
+++ b/xen/arch/x86/hvm/svm/emulate.c Wed Mar 31 10:21:19 2010 +0100
@@ -149,6 +149,9 @@ int __get_instruction_length_from_list(s
if ( (inst_len = svm_nextrip_insn_length(v)) != 0 )
return inst_len;

+ if ( vmcb->exitcode == VMEXIT_IOIO )
+ return vmcb->exitinfo2 - vmcb->rip;
+
/* Fetch up to the next page break; we'll fetch from the next page
* later if we have to. */
fetch_addr = svm_rip2pointer(v);
diff -r 4f796e29987c -r e5e1e1532b97 xen/arch/x86/hvm/svm/svm.c
--- a/xen/arch/x86/hvm/svm/svm.c Wed Mar 31 10:12:29 2010 +0100
+++ b/xen/arch/x86/hvm/svm/svm.c Wed Mar 31 10:21:19 2010 +0100
@@ -42,6 +42,7 @@
#include <asm/hvm/hvm.h>
#include <asm/hvm/support.h>
#include <asm/hvm/io.h>
+#include <asm/hvm/emulate.h>
#include <asm/hvm/svm/asid.h>
#include <asm/hvm/svm/svm.h>
#include <asm/hvm/svm/vmcb.h>
@@ -1242,6 +1243,23 @@ static void svm_vmexit_do_invalidate_cac
svm_wbinvd_intercept();

__update_guest_eip(regs, inst_len);
+}
+
+static int svm_vmexit_do_io(struct vmcb_struct *vmcb,
+ struct cpu_user_regs *regs)
+{
+ uint16_t port;
+ int bytes, dir;
+ int rc;
+
+ port = (vmcb->exitinfo1 >> 16) & 0xFFFF;
+ bytes = ((vmcb->exitinfo1 >> 4) & 0x07);
+ dir = (vmcb->exitinfo1 & 1) ? IOREQ_READ : IOREQ_WRITE;
+
+ rc = handle_mmio_decoded(port, bytes, dir);
+ if ( rc != X86EMUL_RETRY )
+ __update_guest_eip(regs, vmcb->exitinfo2 - vmcb->rip);
+ return rc;
}

static void svm_invlpg_intercept(unsigned long vaddr)
@@ -1450,11 +1468,17 @@ asmlinkage void svm_vmexit_handler(struc
svm_vmexit_do_hlt(vmcb, regs);
break;

+ case VMEXIT_IOIO:
+ if ( ( vmcb->exitinfo1 & ( 1 << 2 ) ) == 0 ) {
+ if ( !svm_vmexit_do_io(vmcb, regs) )
+ hvm_inject_exception(TRAP_gp_fault, 0, 0);
+ break;
+ }
+ /* fallthrough to emulation if a string instruction */
case VMEXIT_CR0_READ ... VMEXIT_CR15_READ:
case VMEXIT_CR0_WRITE ... VMEXIT_CR15_WRITE:
case VMEXIT_INVLPG:
case VMEXIT_INVLPGA:
- case VMEXIT_IOIO:
if ( !handle_mmio() )
hvm_inject_exception(TRAP_gp_fault, 0, 0);
break;
diff -r 4f796e29987c -r e5e1e1532b97 xen/arch/x86/hvm/vmx/vmx.c
--- a/xen/arch/x86/hvm/vmx/vmx.c Wed Mar 31 10:12:29 2010 +0100
+++ b/xen/arch/x86/hvm/vmx/vmx.c Wed Mar 31 10:21:19 2010 +0100
@@ -1721,6 +1721,25 @@ static int vmx_cr_access(unsigned long e
return 1;
}

+static int vmx_io_intercept(unsigned long exit_qualification,
+ struct cpu_user_regs *regs)
+{
+ uint16_t port;
+ int bytes, dir;
+ int rc;
+ int inst_len;
+
+ port = (exit_qualification >> 16) & 0xFFFF;
+ bytes = (exit_qualification & 0x07) + 1;
+ dir = (exit_qualification & 0x08) ? IOREQ_READ : IOREQ_WRITE;
+
+ inst_len = __get_instruction_length();
+ rc = handle_mmio_decoded(port, bytes, dir);
+ if ( rc != X86EMUL_RETRY)
+ __update_guest_eip(inst_len);
+ return rc;
+}
+
static const struct lbr_info {
u32 base, count;
} p4_lbr[] = {
@@ -2589,8 +2608,14 @@ asmlinkage void vmx_vmexit_handler(struc
break;

case EXIT_REASON_IO_INSTRUCTION:
- if ( !handle_mmio() )
- vmx_inject_hw_exception(TRAP_gp_fault, 0);
+ exit_qualification = __vmread(EXIT_QUALIFICATION);
+ if (exit_qualification & 0x10) {
+ if ( !handle_mmio() )
+ vmx_inject_hw_exception(TRAP_gp_fault, 0);
+ } else {
+ if ( !vmx_io_intercept(exit_qualification, regs) )
+ vmx_inject_hw_exception(TRAP_gp_fault, 0);
+ }
break;

case EXIT_REASON_INVD:
diff -r 4f796e29987c -r e5e1e1532b97 xen/include/asm-x86/hvm/emulate.h
--- a/xen/include/asm-x86/hvm/emulate.h Wed Mar 31 10:12:29 2010 +0100
+++ b/xen/include/asm-x86/hvm/emulate.h Wed Mar 31 10:21:19 2010 +0100
@@ -46,4 +46,8 @@ struct segment_register *hvmemul_get_seg
enum x86_segment seg,
struct hvm_emulate_ctxt *hvmemul_ctxt);

+int hvmemul_do_io(
+ int is_mmio, paddr_t addr, unsigned long *reps, int size,
+ paddr_t ram_gpa, int dir, int df, void *p_data);
+
#endif /* __ASM_X86_HVM_EMULATE_H__ */
diff -r 4f796e29987c -r e5e1e1532b97 xen/include/asm-x86/hvm/io.h
--- a/xen/include/asm-x86/hvm/io.h Wed Mar 31 10:12:29 2010 +0100
+++ b/xen/include/asm-x86/hvm/io.h Wed Mar 31 10:21:19 2010 +0100
@@ -100,6 +100,7 @@ void send_invalidate_req(void);
void send_invalidate_req(void);
int handle_mmio(void);
int handle_mmio_with_translation(unsigned long gva, unsigned long gpfn);
+int handle_mmio_decoded(uint16_t port, int size, int dir);
void hvm_interrupt_post(struct vcpu *v, int vector, int type);
void hvm_io_assist(void);
void hvm_dpci_eoi(struct domain *d, unsigned int guest_irq,
diff -r 4f796e29987c -r e5e1e1532b97 xen/include/asm-x86/hvm/vcpu.h
--- a/xen/include/asm-x86/hvm/vcpu.h Wed Mar 31 10:12:29 2010 +0100
+++ b/xen/include/asm-x86/hvm/vcpu.h Wed Mar 31 10:21:19 2010 +0100
@@ -103,6 +103,10 @@ struct hvm_vcpu {
*/
unsigned long mmio_gva;
unsigned long mmio_gpfn;
+ uint16_t io_port;
+ int io_size;
+ unsigned io_dir;
+
/* Callback into x86_emulate when emulating FPU/MMX/XMM instructions. */
void (*fpu_exception_callback)(void *, struct cpu_user_regs *);
void *fpu_exception_callback_arg;

_______________________________________________
Xen-changelog mailing list
Xen-changelog@lists.xensource.com
http://lists.xensource.com/xen-changelog