diff mbox series

[v2,1/2] selftests/sgx: Use rip relative addressing for encl_stack

Message ID 20220322074313.7444-1-jarkko@kernel.org (mailing list archive)
State New, archived
Headers show
Series [v2,1/2] selftests/sgx: Use rip relative addressing for encl_stack | expand

Commit Message

Jarkko Sakkinen March 22, 2022, 7:43 a.m. UTC
Simplify the test_encl_bootstrap.S flow by using rip-relative addressing.
Compiler does the right thing here, and this removes dependency on where
TCS entries need to be located in the binary, i.e. allows the binary layout
changed freely in the future.

Cc: Reinette Chatre <reinette.chatre@intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Signed-off-by: Jarkko Sakkinen <jarkko@kernel.org>
---
 tools/testing/selftests/sgx/test_encl_bootstrap.S | 6 +-----
 1 file changed, 1 insertion(+), 5 deletions(-)

Comments

Reinette Chatre March 28, 2022, 9:49 p.m. UTC | #1
Hi Jarkko,

On 3/22/2022 12:43 AM, Jarkko Sakkinen wrote:
> Simplify the test_encl_bootstrap.S flow by using rip-relative addressing.
> Compiler does the right thing here, and this removes dependency on where
> TCS entries need to be located in the binary, i.e. allows the binary layout
> changed freely in the future.
> 
> Cc: Reinette Chatre <reinette.chatre@intel.com>
> Cc: Dave Hansen <dave.hansen@linux.intel.com>
> Signed-off-by: Jarkko Sakkinen <jarkko@kernel.org>
> ---
>  tools/testing/selftests/sgx/test_encl_bootstrap.S | 6 +-----
>  1 file changed, 1 insertion(+), 5 deletions(-)
> 
> diff --git a/tools/testing/selftests/sgx/test_encl_bootstrap.S b/tools/testing/selftests/sgx/test_encl_bootstrap.S
> index 82fb0dfcbd23..1c1b5c6c4ffe 100644
> --- a/tools/testing/selftests/sgx/test_encl_bootstrap.S
> +++ b/tools/testing/selftests/sgx/test_encl_bootstrap.S
> @@ -40,11 +40,7 @@
>  	.text
>  
>  encl_entry:
> -	# RBX contains the base address for TCS, which is the first address
> -	# inside the enclave for TCS #1 and one page into the enclave for
> -	# TCS #2. By adding the value of encl_stack to it, we get
> -	# the absolute address for the stack.
> -	lea	(encl_stack)(%rbx), %rax
> +	lea	(encl_stack)(%rip), %rax
>  	xchg	%rsp, %rax
>  	push	%rax
>  

The goal of the above snippet is to set RSP to ensure that each thread has its own stack.

Since EENTER computes RIP as EnclaveBase + TCS.OENTRY, by using offset from RIP this
would result in all TCS with OENTRY of encl_entry to use the same stack, no?

Could you please consider the following as an alternative:
https://lore.kernel.org/lkml/65c137c875bd4da675eaba35316ff43d7cfd52f8.1644274683.git.reinette.chatre@intel.com/

The idea in that patch is that a new TCS would always need to be accompanied by a
dedicated stack so, at least for testing purposes, the TCS and stack can be dynamically
allocated together with the TCS page following its stack.  This seems much simpler
to me and also makes the following patch unnecessary.

Reinette
Jarkko Sakkinen March 30, 2022, 2:54 p.m. UTC | #2
On Mon, Mar 28, 2022 at 02:49:04PM -0700, Reinette Chatre wrote:
> Hi Jarkko,
> 
> On 3/22/2022 12:43 AM, Jarkko Sakkinen wrote:
> > Simplify the test_encl_bootstrap.S flow by using rip-relative addressing.
> > Compiler does the right thing here, and this removes dependency on where
> > TCS entries need to be located in the binary, i.e. allows the binary layout
> > changed freely in the future.
> > 
> > Cc: Reinette Chatre <reinette.chatre@intel.com>
> > Cc: Dave Hansen <dave.hansen@linux.intel.com>
> > Signed-off-by: Jarkko Sakkinen <jarkko@kernel.org>
> > ---
> >  tools/testing/selftests/sgx/test_encl_bootstrap.S | 6 +-----
> >  1 file changed, 1 insertion(+), 5 deletions(-)
> > 
> > diff --git a/tools/testing/selftests/sgx/test_encl_bootstrap.S b/tools/testing/selftests/sgx/test_encl_bootstrap.S
> > index 82fb0dfcbd23..1c1b5c6c4ffe 100644
> > --- a/tools/testing/selftests/sgx/test_encl_bootstrap.S
> > +++ b/tools/testing/selftests/sgx/test_encl_bootstrap.S
> > @@ -40,11 +40,7 @@
> >  	.text
> >  
> >  encl_entry:
> > -	# RBX contains the base address for TCS, which is the first address
> > -	# inside the enclave for TCS #1 and one page into the enclave for
> > -	# TCS #2. By adding the value of encl_stack to it, we get
> > -	# the absolute address for the stack.
> > -	lea	(encl_stack)(%rbx), %rax
> > +	lea	(encl_stack)(%rip), %rax
> >  	xchg	%rsp, %rax
> >  	push	%rax
> >  
> 
> The goal of the above snippet is to set RSP to ensure that each thread has its own stack.
> 
> Since EENTER computes RIP as EnclaveBase + TCS.OENTRY, by using offset from RIP this
> would result in all TCS with OENTRY of encl_entry to use the same stack, no?
> 
> Could you please consider the following as an alternative:
> https://lore.kernel.org/lkml/65c137c875bd4da675eaba35316ff43d7cfd52f8.1644274683.git.reinette.chatre@intel.com/
> 
> The idea in that patch is that a new TCS would always need to be accompanied by a
> dedicated stack so, at least for testing purposes, the TCS and stack can be dynamically
> allocated together with the TCS page following its stack.  This seems much simpler
> to me and also makes the following patch unnecessary.

There's no better alternative than use rip. Compiler will fix it up.

So, no, I won't consider that. This a dead obvious change.

BR, Jarkko
Jarkko Sakkinen March 30, 2022, 2:56 p.m. UTC | #3
On Wed, Mar 30, 2022 at 05:54:18PM +0300, Jarkko Sakkinen wrote:
> On Mon, Mar 28, 2022 at 02:49:04PM -0700, Reinette Chatre wrote:
> > Hi Jarkko,
> > 
> > On 3/22/2022 12:43 AM, Jarkko Sakkinen wrote:
> > > Simplify the test_encl_bootstrap.S flow by using rip-relative addressing.
> > > Compiler does the right thing here, and this removes dependency on where
> > > TCS entries need to be located in the binary, i.e. allows the binary layout
> > > changed freely in the future.
> > > 
> > > Cc: Reinette Chatre <reinette.chatre@intel.com>
> > > Cc: Dave Hansen <dave.hansen@linux.intel.com>
> > > Signed-off-by: Jarkko Sakkinen <jarkko@kernel.org>
> > > ---
> > >  tools/testing/selftests/sgx/test_encl_bootstrap.S | 6 +-----
> > >  1 file changed, 1 insertion(+), 5 deletions(-)
> > > 
> > > diff --git a/tools/testing/selftests/sgx/test_encl_bootstrap.S b/tools/testing/selftests/sgx/test_encl_bootstrap.S
> > > index 82fb0dfcbd23..1c1b5c6c4ffe 100644
> > > --- a/tools/testing/selftests/sgx/test_encl_bootstrap.S
> > > +++ b/tools/testing/selftests/sgx/test_encl_bootstrap.S
> > > @@ -40,11 +40,7 @@
> > >  	.text
> > >  
> > >  encl_entry:
> > > -	# RBX contains the base address for TCS, which is the first address
> > > -	# inside the enclave for TCS #1 and one page into the enclave for
> > > -	# TCS #2. By adding the value of encl_stack to it, we get
> > > -	# the absolute address for the stack.
> > > -	lea	(encl_stack)(%rbx), %rax
> > > +	lea	(encl_stack)(%rip), %rax
> > >  	xchg	%rsp, %rax
> > >  	push	%rax
> > >  
> > 
> > The goal of the above snippet is to set RSP to ensure that each thread has its own stack.
> > 
> > Since EENTER computes RIP as EnclaveBase + TCS.OENTRY, by using offset from RIP this
> > would result in all TCS with OENTRY of encl_entry to use the same stack, no?
> > 
> > Could you please consider the following as an alternative:
> > https://lore.kernel.org/lkml/65c137c875bd4da675eaba35316ff43d7cfd52f8.1644274683.git.reinette.chatre@intel.com/
> > 
> > The idea in that patch is that a new TCS would always need to be accompanied by a
> > dedicated stack so, at least for testing purposes, the TCS and stack can be dynamically
> > allocated together with the TCS page following its stack.  This seems much simpler
> > to me and also makes the following patch unnecessary.
> 
> There's no better alternative than use rip. Compiler will fix it up.
> 
> So, no, I won't consider that. This a dead obvious change.

How you organize TCS and stack is completely unrelated topic.

BR, Jarkko
Reinette Chatre March 30, 2022, 5:40 p.m. UTC | #4
Hi Jarkko,

On 3/30/2022 7:54 AM, Jarkko Sakkinen wrote:
> On Mon, Mar 28, 2022 at 02:49:04PM -0700, Reinette Chatre wrote:
>> Hi Jarkko,
>>
>> On 3/22/2022 12:43 AM, Jarkko Sakkinen wrote:
>>> Simplify the test_encl_bootstrap.S flow by using rip-relative addressing.
>>> Compiler does the right thing here, and this removes dependency on where
>>> TCS entries need to be located in the binary, i.e. allows the binary layout
>>> changed freely in the future.
>>>
>>> Cc: Reinette Chatre <reinette.chatre@intel.com>
>>> Cc: Dave Hansen <dave.hansen@linux.intel.com>
>>> Signed-off-by: Jarkko Sakkinen <jarkko@kernel.org>
>>> ---
>>>  tools/testing/selftests/sgx/test_encl_bootstrap.S | 6 +-----
>>>  1 file changed, 1 insertion(+), 5 deletions(-)
>>>
>>> diff --git a/tools/testing/selftests/sgx/test_encl_bootstrap.S b/tools/testing/selftests/sgx/test_encl_bootstrap.S
>>> index 82fb0dfcbd23..1c1b5c6c4ffe 100644
>>> --- a/tools/testing/selftests/sgx/test_encl_bootstrap.S
>>> +++ b/tools/testing/selftests/sgx/test_encl_bootstrap.S
>>> @@ -40,11 +40,7 @@
>>>  	.text
>>>  
>>>  encl_entry:
>>> -	# RBX contains the base address for TCS, which is the first address
>>> -	# inside the enclave for TCS #1 and one page into the enclave for
>>> -	# TCS #2. By adding the value of encl_stack to it, we get
>>> -	# the absolute address for the stack.
>>> -	lea	(encl_stack)(%rbx), %rax
>>> +	lea	(encl_stack)(%rip), %rax
>>>  	xchg	%rsp, %rax
>>>  	push	%rax
>>>  
>>
>> The goal of the above snippet is to set RSP to ensure that each thread has its own stack.
>>
>> Since EENTER computes RIP as EnclaveBase + TCS.OENTRY, by using offset from RIP this
>> would result in all TCS with OENTRY of encl_entry to use the same stack, no?
>>
>> Could you please consider the following as an alternative:
>> https://lore.kernel.org/lkml/65c137c875bd4da675eaba35316ff43d7cfd52f8.1644274683.git.reinette.chatre@intel.com/
>>
>> The idea in that patch is that a new TCS would always need to be accompanied by a
>> dedicated stack so, at least for testing purposes, the TCS and stack can be dynamically
>> allocated together with the TCS page following its stack.  This seems much simpler
>> to me and also makes the following patch unnecessary.
> 
> There's no better alternative than use rip. Compiler will fix it up.

Could you please elaborate how the compiler will fix it up?

> 
> So, no, I won't consider that. This a dead obvious change.

It is not obvious to me so I attempted to make it obvious by writing a test program that
prints RSP from the two different threads.

test_encl_bootstrap.S gives each thread, TCS #1 and TCS #2, a page of stack.
Before your patch, with the test below printing RSP, this is clear ... the stack used by the
two threads are one page apart:
#  RUN           enclave.tcs_entry ...
rsp TCS #1 = 0X7FD997D97F68
rsp TCS #2 = 0X7FD997D98F68
#            OK  enclave.tcs_entry

After applying this patch both threads use the same stack memory:
#  RUN           enclave.tcs_entry ...
rsp TCS #1 = 0X7FCF778B7F68
rsp TCS #2 = 0X7FCF778B7F68
#            OK  enclave.tcs_entry

Here is the test I used:

diff --git a/tools/testing/selftests/sgx/defines.h b/tools/testing/selftests/sgx/defines.h
index d8587c971941..08b2765dc2f4 100644
--- a/tools/testing/selftests/sgx/defines.h
+++ b/tools/testing/selftests/sgx/defines.h
@@ -27,6 +27,7 @@ enum encl_op_type {
 	ENCL_OP_EACCEPT,
 	ENCL_OP_EMODPE,
 	ENCL_OP_INIT_TCS_PAGE,
+	ENCL_OP_GET_RSP,
 	ENCL_OP_MAX,
 };
 
@@ -76,4 +77,10 @@ struct encl_op_init_tcs_page {
 	uint64_t entry;
 };
 
+struct encl_op_rsp {
+	struct encl_op_header header;
+	uint64_t ret;
+};
+
+
 #endif /* DEFINES_H */
diff --git a/tools/testing/selftests/sgx/main.c b/tools/testing/selftests/sgx/main.c
index a7543e5561a9..2380944dce71 100644
--- a/tools/testing/selftests/sgx/main.c
+++ b/tools/testing/selftests/sgx/main.c
@@ -570,12 +573,14 @@ TEST_F(enclave, clobbered_vdso_and_user_function)
 /*
  * Sanity check that it is possible to enter either of the two hardcoded TCS
  */
 TEST_F(enclave, tcs_entry)
 {
 	struct encl_op_header op;
+	struct encl_op_rsp rsp_op;
 
 	ASSERT_TRUE(setup_test_encl(ENCL_HEAP_SIZE_DEFAULT, &self->encl, _metadata));
 
@@ -591,6 +596,17 @@ TEST_F(enclave, tcs_entry)
 	EXPECT_EQ(self->run.exception_error_code, 0);
 	EXPECT_EQ(self->run.exception_addr, 0);
 
+	rsp_op.ret = 0;
+	rsp_op.header.type = ENCL_OP_GET_RSP;
+
+	EXPECT_EQ(ENCL_CALL(&rsp_op, &self->run, true), 0);
+
+	EXPECT_EEXIT(&self->run);
+	EXPECT_EQ(self->run.exception_vector, 0);
+	EXPECT_EQ(self->run.exception_error_code, 0);
+	EXPECT_EQ(self->run.exception_addr, 0);
+	printf("rsp TCS #1 = 0X%lX \n", rsp_op.ret);
+
 	/* Move to the next TCS. */
 	self->run.tcs = self->encl.encl_base + PAGE_SIZE;
 
@@ -600,6 +616,17 @@ TEST_F(enclave, tcs_entry)
 	EXPECT_EQ(self->run.exception_vector, 0);
 	EXPECT_EQ(self->run.exception_error_code, 0);
 	EXPECT_EQ(self->run.exception_addr, 0);
+	rsp_op.ret = 0;
+	rsp_op.header.type = ENCL_OP_GET_RSP;
+
+	EXPECT_EQ(ENCL_CALL(&rsp_op, &self->run, true), 0);
+
+	EXPECT_EEXIT(&self->run);
+	EXPECT_EQ(self->run.exception_vector, 0);
+	EXPECT_EQ(self->run.exception_error_code, 0);
+	EXPECT_EQ(self->run.exception_addr, 0);
+	printf("rsp TCS #2 = 0X%lX \n", rsp_op.ret);
+
 }
 
 /*
diff --git a/tools/testing/selftests/sgx/test_encl.c b/tools/testing/selftests/sgx/test_encl.c
index c0d6397295e3..b2a94a6d754e 100644
--- a/tools/testing/selftests/sgx/test_encl.c
+++ b/tools/testing/selftests/sgx/test_encl.c
@@ -119,6 +119,17 @@ static void do_encl_op_nop(void *_op)
 
 }
 
+static void do_get_rsp(void *_op)
+{
+	struct encl_op_rsp *op = _op;
+	uint64_t rsp;
+
+	asm volatile("mov %%rsp, %0 \n": "=r"(rsp) ::);
+
+	op->ret = rsp;
+
+}
+
 void encl_body(void *rdi,  void *rsi)
 {
 	const void (*encl_op_array[ENCL_OP_MAX])(void *) = {
@@ -130,6 +141,7 @@ void encl_body(void *rdi,  void *rsi)
 		do_encl_eaccept,
 		do_encl_emodpe,
 		do_encl_init_tcs_page,
+		do_get_rsp,
 	};
 
 	struct encl_op_header *op = (struct encl_op_header *)rdi;
Jarkko Sakkinen March 30, 2022, 7:22 p.m. UTC | #5
On Wed, 2022-03-30 at 22:03 +0300, Jarkko Sakkinen wrote:
> On Wed, 2022-03-30 at 10:40 -0700, Reinette Chatre wrote:
> > Could you please elaborate how the compiler will fix it up?
> 
> Sure.
> 
> Here's the disassembly of the RBX version:
> 
> [0x000021a9]> pi 1
> lea rax, [rbx + loc.encl_stack]
> 
> Here's the same with s/RBX/RIP/:
> 
> [0x000021a9]> pi 5
> lea rax, loc.encl_stack
> Compiler will substitute correct offset relative to the RIP,
> well, because it can and it makes sense.
> 
> It is treated differently than other registers, e.g. when
> LEA is assembled.
> 
> BR, Jarkko


To demonstrate this I did a couple of simple sessions with Rizin
(fork/continuation/something of Radare2):

jarkko@suppilovahvero ~/Downloads (main)> rizin test_encl.rbx.elf
 -- Temporally drop the verbosity prefixing the commands with ':'
[0x00002000]> aaa
[x] Analyze all flags starting with sym. and entry0 (aa)
[x] Analyze function calls (aac)
[x] Analyze len bytes of instructions for references (aar)
[x] Check for classes
[x] Type matching analysis for all functions (aaft)
[x] Propagate noreturn information
[x] Use -AA or aaaa to perform additional experimental analysis.
[0x00002000]> /ad lea
0x0000206e   # 7: lea rax, [rip + 0xf8b]
0x0000206f   # 6: lea eax, [rip + 0xf8b]
0x0000207e   # 1: leave
0x000020a1   # 7: lea rcx, [rip + 0xf58]
0x000020a2   # 6: lea ecx, [rip + 0xf58]
0x000020b4   # 1: leave
0x000020ee   # 1: leave
0x00002128   # 1: leave
0x00002145   # 7: lea rax, [rip - 0x102]
0x00002146   # 6: lea eax, [rip - 0x102]
0x00002150   # 7: lea rax, [rip - 0xd7]
0x00002151   # 6: lea eax, [rip - 0xd7]
0x0000215b   # 7: lea rax, [rip - 0xac]
0x0000215c   # 6: lea eax, [rip - 0xac]
0x00002166   # 7: lea rax, [rip - 0x7d]
0x00002167   # 6: lea eax, [rip - 0x7d]
0x00002171   # 7: lea rax, [rip - 0x4e]
0x00002172   # 6: lea eax, [rip - 0x4e]
0x000021a7   # 1: leave
0x000021a9   # 7: lea rax, [rbx + loc.encl_stack]
0x000021aa   # 6: lea eax, [rbx + loc.encl_stack]
[0x00002000]> s 0x21a9
[0x000021a9]> pi 1
lea rax, [rbx + loc.encl_stack]
[0x000021a9]> 

jarkko@suppilovahvero ~/Downloads (main)> rizin test_encl.elf
 -- Use V! to enter into the visual panels mode (dwm style)
[0x00002000]> aaa
[x] Analyze all flags starting with sym. and entry0 (aa)
[x] Analyze function calls (aac)
[x] Analyze len bytes of instructions for references (aar)
[x] Check for classes
[x] Type matching analysis for all functions (aaft)
[x] Propagate noreturn information
[x] Use -AA or aaaa to perform additional experimental analysis.
[0x00002000]> /ad lea
0x0000206e   # 7: lea rax, [rip + 0xf8b]
0x0000206f   # 6: lea eax, [rip + 0xf8b]
0x0000207e   # 1: leave
0x000020a1   # 7: lea rcx, [rip + 0xf58]
0x000020a2   # 6: lea ecx, [rip + 0xf58]
0x000020b4   # 1: leave
0x000020ee   # 1: leave
0x00002128   # 1: leave
0x00002145   # 7: lea rax, [rip - 0x102]
0x00002146   # 6: lea eax, [rip - 0x102]
0x00002150   # 7: lea rax, [rip - 0xd7]
0x00002151   # 6: lea eax, [rip - 0xd7]
0x0000215b   # 7: lea rax, [rip - 0xac]
0x0000215c   # 6: lea eax, [rip - 0xac]
0x00002166   # 7: lea rax, [rip - 0x7d]
0x00002167   # 6: lea eax, [rip - 0x7d]
0x00002171   # 7: lea rax, [rip - 0x4e]
0x00002172   # 6: lea eax, [rip - 0x4e]
0x000021a7   # 1: leave
0x000021a9   # 7: lea rax, [rip + 0x5e50]
0x000021aa   # 6: lea eax, [rip + 0x5e50]
[0x00002000]> s 0x21a9
[0x000021a9]> pi 1
lea rax, loc.encl_stack
[0x000021a9]> 

BR, Jarkko
Reinette Chatre March 30, 2022, 8:05 p.m. UTC | #6
On 3/30/2022 12:03 PM, Jarkko Sakkinen wrote:
> On Wed, 2022-03-30 at 10:40 -0700, Reinette Chatre wrote:
>> Could you please elaborate how the compiler will fix it up?
> 
> Sure.
> 
> Here's the disassembly of the RBX version:
> 
> [0x000021a9]> pi 1
> lea rax, [rbx + loc.encl_stack]
> 
> Here's the same with s/RBX/RIP/:
> 
> [0x000021a9]> pi 5
> lea rax, loc.encl_stack
> 
> Compiler will substitute correct offset relative to the RIP,
> well, because it can and it makes sense.

It does not make sense to me because, as proven with my test,
the two threads end up sharing the same stack memory.

Reinette
Jarkko Sakkinen March 30, 2022, 8:40 p.m. UTC | #7
On Wed, 2022-03-30 at 13:05 -0700, Reinette Chatre wrote:
> 
> 
> On 3/30/2022 12:03 PM, Jarkko Sakkinen wrote:
> > On Wed, 2022-03-30 at 10:40 -0700, Reinette Chatre wrote:
> > > Could you please elaborate how the compiler will fix it up?
> > 
> > Sure.
> > 
> > Here's the disassembly of the RBX version:
> > 
> > [0x000021a9]> pi 1
> > lea rax, [rbx + loc.encl_stack]
> > 
> > Here's the same with s/RBX/RIP/:
> > 
> > [0x000021a9]> pi 5
> > lea rax, loc.encl_stack
> > 
> > Compiler will substitute correct offset relative to the RIP,
> > well, because it can and it makes sense.
> 
> It does not make sense to me because, as proven with my test,
> the two threads end up sharing the same stack memory.

I see, I need to correct my patch, thanks!

RBX gives correct results because of the binary organization,
i.e. TCS's are placed to zero offset and forward, and 
unrelocated symbol is just compiled in as an untranslated
offset.

RPI is given correct results but how the semantics work
right now is incompatible.

Still, even for kselftest, I would consider a switch
because that way:

1. You can layout binary however you wan and things
   won't break.
2. You can point to any symbol not just stack, if
   ever need.
   
I admit it works semantically but it just super
unrobust.

BR, Jarkko
Reinette Chatre March 30, 2022, 9:29 p.m. UTC | #8
Hi Jarkko,

On 3/30/2022 1:40 PM, Jarkko Sakkinen wrote:
> On Wed, 2022-03-30 at 13:05 -0700, Reinette Chatre wrote:
>>
>>
>> On 3/30/2022 12:03 PM, Jarkko Sakkinen wrote:
>>> On Wed, 2022-03-30 at 10:40 -0700, Reinette Chatre wrote:
>>>> Could you please elaborate how the compiler will fix it up?
>>>
>>> Sure.
>>>
>>> Here's the disassembly of the RBX version:
>>>
>>> [0x000021a9]> pi 1
>>> lea rax, [rbx + loc.encl_stack]
>>>
>>> Here's the same with s/RBX/RIP/:
>>>
>>> [0x000021a9]> pi 5
>>> lea rax, loc.encl_stack
>>>
>>> Compiler will substitute correct offset relative to the RIP,
>>> well, because it can and it makes sense.
>>
>> It does not make sense to me because, as proven with my test,
>> the two threads end up sharing the same stack memory.
> 
> I see, I need to correct my patch, thanks!
> 
> RBX gives correct results because of the binary organization,
> i.e. TCS's are placed to zero offset and forward, and 
> unrelocated symbol is just compiled in as an untranslated
> offset.
> 
> RPI is given correct results but how the semantics work
> right now is incompatible.
> 
> Still, even for kselftest, I would consider a switch
> because that way:
> 
> 1. You can layout binary however you wan and things
>    won't break.
> 2. You can point to any symbol not just stack, if
>    ever need.
>    
> I admit it works semantically but it just super
> unrobust.

I do not think that we need an exceptionally flexible
runtime as part of the SGX selftests but instead something
that is easy(*) to understand while also sufficient to support
the tests.

Reinette

* I do not actually consider the existing enclave test binary
  easy to understand (this thread is proof) but keeping its
  complexity to be minimal would benefit folks needing to
  ramp up on SGX and/or debug kselftest failures.
Jarkko Sakkinen March 30, 2022, 10:30 p.m. UTC | #9
On Wed, 2022-03-30 at 14:29 -0700, Reinette Chatre wrote:
> Hi Jarkko,
> 
> On 3/30/2022 1:40 PM, Jarkko Sakkinen wrote:
> > On Wed, 2022-03-30 at 13:05 -0700, Reinette Chatre wrote:
> > > 
> > > 
> > > On 3/30/2022 12:03 PM, Jarkko Sakkinen wrote:
> > > > On Wed, 2022-03-30 at 10:40 -0700, Reinette Chatre wrote:
> > > > > Could you please elaborate how the compiler will fix it up?
> > > > 
> > > > Sure.
> > > > 
> > > > Here's the disassembly of the RBX version:
> > > > 
> > > > [0x000021a9]> pi 1
> > > > lea rax, [rbx + loc.encl_stack]
> > > > 
> > > > Here's the same with s/RBX/RIP/:
> > > > 
> > > > [0x000021a9]> pi 5
> > > > lea rax, loc.encl_stack
> > > > 
> > > > Compiler will substitute correct offset relative to the RIP,
> > > > well, because it can and it makes sense.
> > > 
> > > It does not make sense to me because, as proven with my test,
> > > the two threads end up sharing the same stack memory.
> > 
> > I see, I need to correct my patch, thanks!
> > 
> > RBX gives correct results because of the binary organization,
> > i.e. TCS's are placed to zero offset and forward, and 
> > unrelocated symbol is just compiled in as an untranslated
> > offset.
> > 
> > RPI is given correct results but how the semantics work
> > right now is incompatible.
> > 
> > Still, even for kselftest, I would consider a switch
> > because that way:
> > 
> > 1. You can layout binary however you wan and things
> >    won't break.
> > 2. You can point to any symbol not just stack, if
> >    ever need.
> >    
> > I admit it works semantically but it just super
> > unrobust.
> 
> I do not think that we need an exceptionally flexible
> runtime as part of the SGX selftests but instead something
> that is easy(*) to understand while also sufficient to support
> the tests.
> 
> Reinette
> 
> * I do not actually consider the existing enclave test binary
>   easy to understand (this thread is proof) but keeping its
>   complexity to be minimal would benefit folks needing to
>   ramp up on SGX and/or debug kselftest failures.

Based on you feedback I refined the patch:

https://lore.kernel.org/linux-sgx/20220330222834.139769-1-jarkko@kernel.org/T/#u

BR, Jarkko



	
	




BR, Jarkko
diff mbox series

Patch

diff --git a/tools/testing/selftests/sgx/test_encl_bootstrap.S b/tools/testing/selftests/sgx/test_encl_bootstrap.S
index 82fb0dfcbd23..1c1b5c6c4ffe 100644
--- a/tools/testing/selftests/sgx/test_encl_bootstrap.S
+++ b/tools/testing/selftests/sgx/test_encl_bootstrap.S
@@ -40,11 +40,7 @@ 
 	.text
 
 encl_entry:
-	# RBX contains the base address for TCS, which is the first address
-	# inside the enclave for TCS #1 and one page into the enclave for
-	# TCS #2. By adding the value of encl_stack to it, we get
-	# the absolute address for the stack.
-	lea	(encl_stack)(%rbx), %rax
+	lea	(encl_stack)(%rip), %rax
 	xchg	%rsp, %rax
 	push	%rax