2011-04-13 15:45:13

by Nelson Elhage

[permalink] [raw]
Subject: [PATCH] KVM: emulator: Handle wraparound in (cs_base + offset) when fetching.

Currently, setting a large (i.e. negative) base address for %cs does not work on
a 64-bit host. The "JOS" teaching operating system, used by MIT and other
universities, relies on such segments while bootstrapping its way to full
virtual memory management.

Signed-off-by: Nelson Elhage <[email protected]>
---
arch/x86/kvm/emulate.c | 5 ++++-
1 files changed, 4 insertions(+), 1 deletions(-)

diff --git a/arch/x86/kvm/emulate.c b/arch/x86/kvm/emulate.c
index 0ad47b8..54e84b2 100644
--- a/arch/x86/kvm/emulate.c
+++ b/arch/x86/kvm/emulate.c
@@ -505,9 +505,12 @@ static int do_fetch_insn_byte(struct x86_emulate_ctxt *ctxt,
int size, cur_size;

if (eip == fc->end) {
+ unsigned long linear = eip + ctxt->cs_base;
+ if (ctxt->mode != X86EMUL_MODE_PROT64)
+ linear &= (u32)-1;
cur_size = fc->end - fc->start;
size = min(15UL - cur_size, PAGE_SIZE - offset_in_page(eip));
- rc = ops->fetch(ctxt->cs_base + eip, fc->data + cur_size,
+ rc = ops->fetch(linear, fc->data + cur_size,
size, ctxt->vcpu, &ctxt->exception);
if (rc != X86EMUL_CONTINUE)
return rc;
--
1.7.2.43.g36c08.dirty


2011-04-13 16:06:45

by Avi Kivity

[permalink] [raw]
Subject: Re: [PATCH] KVM: emulator: Handle wraparound in (cs_base + offset) when fetching.

On 04/13/2011 06:44 PM, Nelson Elhage wrote:
> Currently, setting a large (i.e. negative) base address for %cs does not work on
> a 64-bit host. The "JOS" teaching operating system, used by MIT and other
> universities, relies on such segments while bootstrapping its way to full
> virtual memory management.
>
> Signed-off-by: Nelson Elhage<[email protected]>
> ---
> arch/x86/kvm/emulate.c | 5 ++++-
> 1 files changed, 4 insertions(+), 1 deletions(-)
>
> diff --git a/arch/x86/kvm/emulate.c b/arch/x86/kvm/emulate.c
> index 0ad47b8..54e84b2 100644
> --- a/arch/x86/kvm/emulate.c
> +++ b/arch/x86/kvm/emulate.c
> @@ -505,9 +505,12 @@ static int do_fetch_insn_byte(struct x86_emulate_ctxt *ctxt,
> int size, cur_size;
>
> if (eip == fc->end) {
> + unsigned long linear = eip + ctxt->cs_base;
> + if (ctxt->mode != X86EMUL_MODE_PROT64)
> + linear&= (u32)-1;
> cur_size = fc->end - fc->start;
> size = min(15UL - cur_size, PAGE_SIZE - offset_in_page(eip));
> - rc = ops->fetch(ctxt->cs_base + eip, fc->data + cur_size,
> + rc = ops->fetch(linear, fc->data + cur_size,
> size, ctxt->vcpu,&ctxt->exception);
> if (rc != X86EMUL_CONTINUE)
> return rc;

A better fix would be to call linearize() here, which does the necessary
truncation as well as segment checks.

However, this patch is a lot more backportable, so I think it should be
applied, and a conversion to linearize() performed afterwards.

--
error compiling committee.c: too many arguments to function

2011-04-15 03:28:10

by Nelson Elhage

[permalink] [raw]
Subject: [PATCH] KVM: emulator: Use linearize() when fetching instructions.

This means that the truncation behavior in linearize needs to grow an additional
slight piece of complexity: when fetching, truncation is dependent on the
execution mode, instead of the current address size.

Signed-off-by: Nelson Elhage <[email protected]>
---
arch/x86/include/asm/kvm_emulate.h | 1 -
arch/x86/kvm/emulate.c | 23 ++++++++++++-----------
2 files changed, 12 insertions(+), 12 deletions(-)

diff --git a/arch/x86/include/asm/kvm_emulate.h b/arch/x86/include/asm/kvm_emulate.h
index 0818448..9b760c8 100644
--- a/arch/x86/include/asm/kvm_emulate.h
+++ b/arch/x86/include/asm/kvm_emulate.h
@@ -265,7 +265,6 @@ struct x86_emulate_ctxt {
unsigned long eip; /* eip before instruction emulation */
/* Emulated execution mode, represented by an X86EMUL_MODE value. */
int mode;
- u32 cs_base;

/* interruptibility state, as a result of execution of STI or MOV SS */
int interruptibility;
diff --git a/arch/x86/kvm/emulate.c b/arch/x86/kvm/emulate.c
index a5f63d4..d3d43a7 100644
--- a/arch/x86/kvm/emulate.c
+++ b/arch/x86/kvm/emulate.c
@@ -542,7 +542,7 @@ static int emulate_nm(struct x86_emulate_ctxt *ctxt)

static int linearize(struct x86_emulate_ctxt *ctxt,
struct segmented_address addr,
- unsigned size, bool write,
+ unsigned size, bool write, bool fetch,
ulong *linear)
{
struct decode_cache *c = &ctxt->decode;
@@ -602,7 +602,7 @@ static int linearize(struct x86_emulate_ctxt *ctxt,
}
break;
}
- if (c->ad_bytes != 8)
+ if (fetch ? ctxt->mode != X86EMUL_MODE_PROT64 : c->ad_bytes != 8)
la &= (u32)-1;
*linear = la;
return X86EMUL_CONTINUE;
@@ -621,7 +621,7 @@ static int segmented_read_std(struct x86_emulate_ctxt *ctxt,
int rc;
ulong linear;

- rc = linearize(ctxt, addr, size, false, &linear);
+ rc = linearize(ctxt, addr, size, false, false, &linear);
if (rc != X86EMUL_CONTINUE)
return rc;
return ctxt->ops->read_std(linear, data, size, ctxt->vcpu,
@@ -637,11 +637,13 @@ static int do_fetch_insn_byte(struct x86_emulate_ctxt *ctxt,
int size, cur_size;

if (eip == fc->end) {
- unsigned long linear = eip + ctxt->cs_base;
- if (ctxt->mode != X86EMUL_MODE_PROT64)
- linear &= (u32)-1;
+ unsigned long linear;
+ struct segmented_address addr = {VCPU_SREG_CS, eip};
cur_size = fc->end - fc->start;
size = min(15UL - cur_size, PAGE_SIZE - offset_in_page(eip));
+ rc = linearize(ctxt, addr, size, false, true, &linear);
+ f (rc != X86EMUL_CONTINUE)
+ return rc;
rc = ops->fetch(linear, fc->data + cur_size,
size, ctxt->vcpu, &ctxt->exception);
if (rc != X86EMUL_CONTINUE)
@@ -1047,7 +1049,7 @@ static int segmented_read(struct x86_emulate_ctxt *ctxt,
int rc;
ulong linear;

- rc = linearize(ctxt, addr, size, false, &linear);
+ rc = linearize(ctxt, addr, size, false, false, &linear);
if (rc != X86EMUL_CONTINUE)
return rc;
return read_emulated(ctxt, ctxt->ops, linear, data, size);
@@ -1061,7 +1063,7 @@ static int segmented_write(struct x86_emulate_ctxt *ctxt,
int rc;
ulong linear;

- rc = linearize(ctxt, addr, size, true, &linear);
+ rc = linearize(ctxt, addr, size, true, false, &linear);
if (rc != X86EMUL_CONTINUE)
return rc;
return ctxt->ops->write_emulated(linear, data, size,
@@ -1076,7 +1078,7 @@ static int segmented_cmpxchg(struct x86_emulate_ctxt *ctxt,
int rc;
ulong linear;

- rc = linearize(ctxt, addr, size, true, &linear);
+ rc = linearize(ctxt, addr, size, true, false, &linear);
if (rc != X86EMUL_CONTINUE)
return rc;
return ctxt->ops->cmpxchg_emulated(linear, orig_data, data,
@@ -2576,7 +2578,7 @@ static int em_invlpg(struct x86_emulate_ctxt *ctxt)
int rc;
ulong linear;

- rc = linearize(ctxt, c->src.addr.mem, 1, false, &linear);
+ rc = linearize(ctxt, c->src.addr.mem, 1, false, false, &linear);
if (rc == X86EMUL_CONTINUE)
emulate_invlpg(ctxt->vcpu, linear);
/* Disable writeback. */
@@ -3154,7 +3156,6 @@ x86_decode_insn(struct x86_emulate_ctxt *ctxt, void *insn, int insn_len)
c->fetch.end = c->fetch.start + insn_len;
if (insn_len > 0)
memcpy(c->fetch.data, insn, insn_len);
- ctxt->cs_base = seg_base(ctxt, ops, VCPU_SREG_CS);

switch (mode) {
case X86EMUL_MODE_REAL:
--
1.7.2.43.g36c08.dirty

2011-04-17 09:44:25

by Avi Kivity

[permalink] [raw]
Subject: Re: [PATCH] KVM: emulator: Handle wraparound in (cs_base + offset) when fetching.

On 04/13/2011 06:44 PM, Nelson Elhage wrote:
> Currently, setting a large (i.e. negative) base address for %cs does not work on
> a 64-bit host. The "JOS" teaching operating system, used by MIT and other
> universities, relies on such segments while bootstrapping its way to full
> virtual memory management.

Applied, thanks.

--
error compiling committee.c: too many arguments to function

2011-04-17 12:26:39

by Avi Kivity

[permalink] [raw]
Subject: Re: [PATCH] KVM: emulator: Use linearize() when fetching instructions.

On 04/15/2011 06:27 AM, Nelson Elhage wrote:
> This means that the truncation behavior in linearize needs to grow an additional
> slight piece of complexity: when fetching, truncation is dependent on the
> execution mode, instead of the current address size.
>
> diff --git a/arch/x86/kvm/emulate.c b/arch/x86/kvm/emulate.c
> index a5f63d4..d3d43a7 100644
> --- a/arch/x86/kvm/emulate.c
> +++ b/arch/x86/kvm/emulate.c
> @@ -542,7 +542,7 @@ static int emulate_nm(struct x86_emulate_ctxt *ctxt)
>
> static int linearize(struct x86_emulate_ctxt *ctxt,
> struct segmented_address addr,
> - unsigned size, bool write,
> + unsigned size, bool write, bool fetch,

Calls to functions with strings of bool arguments are confusing. Please
make this __linearize, and introduce a new linearize() which doesn't
have a fetch argument.

> ulong *linear)
> {
> struct decode_cache *c =&ctxt->decode;
> @@ -602,7 +602,7 @@ static int linearize(struct x86_emulate_ctxt *ctxt,
> }
> break;
> }

linearize() will currently fault on an unreadable code segment. Need to
avoid that on instruction fetches.

> - if (c->ad_bytes != 8)
> + if (fetch ? ctxt->mode != X86EMUL_MODE_PROT64 : c->ad_bytes != 8)
> la&= (u32)-1;
> *linear = la;
> return X86EMUL_CONTINUE;

--
error compiling committee.c: too many arguments to function