selftests/sgx: Use rip relative addressing for encl_stack
Jarkko Sakkinen
maintainerPATCH v2 1/2 selftests/sgx: Use rip relative addressing for encl_stack
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.
--- tools/testing/selftests/sgx/test_encl_bootstrap.S | 6 +----- 1 file changed, 1 insertion(+), 5 deletions(-)
tools/testing/selftests/sgx/test_encl_bootstrap.S +1 -5
@@ -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
Signature · 1 line
2.35.1
Jarkko Sakkinen
maintainerPATCH v2 2/2 selftests/sgx: Make TCS table relocatable
Add a PT_NOTE section with n_namesz containg "TCS" and n_descz containing 32-bit offset to the TCS table inside the enclave. This allows to place the TCS segment freely, and thereby make the kselftest binary layout way more robust.
---
v2:
* Add RIP relative addressing fix for bootstrap as prepending patch, as this depends on it.
* Moved TCS section as the last so that it is easy to add new TCS's, e.g dynamically with EAUG + EMODT, behind it.
--- tools/testing/selftests/sgx/load.c | 56 ++++++++++++++----- tools/testing/selftests/sgx/main.c | 37 +++--------- tools/testing/selftests/sgx/main.h | 2 + tools/testing/selftests/sgx/test_encl.lds | 17 ++++-- .../selftests/sgx/test_encl_bootstrap.S | 7 +++ 5 files changed, 72 insertions(+), 47 deletions(-)
tools/testing/selftests/sgx/load.c +43 -13
@@ -19,6 +19,9 @@#include "defines.h" #include "main.h" +const char *TCS_NOTE_NAME ="TCS" ; +const unsigned long TCS_NOTE_LEN =4 ; +void encl_delete (struct encl *encl) {struct encl_segment *heap_seg;
@@ -187,11 +190,31 @@ bool encl_load(const char *path, struct encl *encl, unsigned long heap_size) encl->nr_segments =1 ;/* one for the heap */ +/* Count the loadable segments and discover the TCS array. */ for (i =0 ; i < ehdr->e_phnum; i++) {Elf64_Phdr *phdr = &phdr_tbl[i]; +Elf64_Nhdr *note; +char *note_name; -if (phdr->p_type ==PT_LOAD ) +switch (phdr->p_type) { +case PT_LOAD : encl->nr_segments++; +break ; + +case PT_NOTE : + note = encl->bin + (phdr->p_offset &PAGE_MASK ); + note_name = &((char *)note)[sizeof (*note)]; + +if (note->n_namesz ==TCS_NOTE_LEN && + !strncmp (note_name,TCS_NOTE_NAME ,TCS_NOTE_LEN )) { +/* 32-bit address. */ + encl->tcs = (struct sgx_tcs *)(unsigned long )(note->n_descsz); + } +break ; + +default : +break ; + } } encl->segment_tbl = calloc(encl->nr_segments,
@@ -215,31 +238,36 @@ bool encl_load(const char *path, struct encl *encl, unsigned long heap_size) goto err; } -if (j ==0 && flags != (PF_R |PF_W )) { -fprintf (stderr, -"TCS has invalid segment flags 0x%02x.\n" , - phdr->p_flags); - goto err; - } -if (j ==0 ) { src_offset = phdr->p_offset &PAGE_MASK ; encl->src = encl->bin + src_offset; + } + + seg->offset = (phdr->p_offset &PAGE_MASK ) - src_offset; + seg->size = (phdr->p_filesz +PAGE_SIZE -1 ) &PAGE_MASK ; + seg->src = encl->src + seg->offset; + seg->measure = true; + +if (seg->offset == (unsigned long )encl->tcs) { +if (flags != (PF_R |PF_W )) { +fprintf (stderr, +"TCS has invalid segment flags 0x%02x.\n" , + phdr->p_flags); + goto err; + } seg->prot =PROT_READ |PROT_WRITE ; seg->flags =SGX_PAGE_TYPE_TCS <<8 ; }else { +if ((flags & (PF_R |PF_W |PF_X )) == (PF_R |PF_W )) + encl->data_offset = seg->offset; + seg->prot = (phdr->p_flags &PF_R ) ?PROT_READ :0 ; seg->prot |= (phdr->p_flags &PF_W ) ?PROT_WRITE :0 ; seg->prot |= (phdr->p_flags &PF_X ) ?PROT_EXEC :0 ; seg->flags = (SGX_PAGE_TYPE_REG <<8 ) | seg->prot; } - seg->offset = (phdr->p_offset &PAGE_MASK ) - src_offset; - seg->size = (phdr->p_filesz +PAGE_SIZE -1 ) &PAGE_MASK ; - seg->src = encl->src + seg->offset; - seg->measure = true; - j++; }
@@ -322,5 +350,7 @@ bool encl_build(struct encl *encl)return false; } + encl->tcs = (struct sgx_tcs *)((unsigned long )encl->tcs + encl->encl_base); +return true; }
tools/testing/selftests/sgx/main.c +8 -29
@@ -109,25 +109,6 @@ static Elf64_Sym *vdso_symtab_get(struct vdso_symtab *symtab, const char *name)return NULL ; } -/* -* Return the offset in the enclave where the data segment can be found. -* The first RW segment loaded is the TCS, skip that to get info on the -* data segment. -*/ -static off_t encl_get_data_offset (struct encl *encl) -{ -int i; - -for (i =1 ; i < encl->nr_segments; i++) { -struct encl_segment *seg = &encl->segment_tbl[i]; - -if (seg->prot == (PROT_READ |PROT_WRITE )) -return seg->offset; - } - -return -1 ; -} -FIXTURE (enclave) {struct encl encl;struct sgx_enclave_run run;
@@ -248,7 +229,7 @@ TEST_F(enclave, unclobbered_vdso)ASSERT_TRUE (setup_test_encl (ENCL_HEAP_SIZE_DEFAULT , &self->encl, _metadata));memset (&self->run,0 ,sizeof (self->run)); - self->run.tcs = self->encl.encl_base; + self->run.tcs = (__u64 )self->encl.tcs; put_op.header.type =ENCL_OP_PUT_TO_BUFFER ; put_op.value =MAGIC ;
@@ -321,7 +302,7 @@ TEST_F(enclave, unclobbered_vdso_oversubscribed)ASSERT_TRUE (setup_test_encl (total_mem, &self->encl, _metadata));memset (&self->run,0 ,sizeof (self->run)); - self->run.tcs = self->encl.encl_base; + self->run.tcs = (__u64 )self->encl.tcs; put_op.header.type =ENCL_OP_PUT_TO_BUFFER ; put_op.value =MAGIC ;
@@ -350,7 +331,7 @@ TEST_F(enclave, clobbered_vdso)ASSERT_TRUE (setup_test_encl (ENCL_HEAP_SIZE_DEFAULT , &self->encl, _metadata));memset (&self->run,0 ,sizeof (self->run)); - self->run.tcs = self->encl.encl_base; + self->run.tcs = (__u64 )self->encl.tcs; put_op.header.type =ENCL_OP_PUT_TO_BUFFER ; put_op.value =MAGIC ;
@@ -386,7 +367,7 @@ TEST_F(enclave, clobbered_vdso_and_user_function)ASSERT_TRUE (setup_test_encl (ENCL_HEAP_SIZE_DEFAULT , &self->encl, _metadata));memset (&self->run,0 ,sizeof (self->run)); - self->run.tcs = self->encl.encl_base; + self->run.tcs = (__u64 )self->encl.tcs; self->run.user_handler = (__u64 )test_handler; self->run.user_data =0xdeadbeef ;
@@ -419,7 +400,7 @@ TEST_F(enclave, tcs_entry)ASSERT_TRUE (setup_test_encl (ENCL_HEAP_SIZE_DEFAULT , &self->encl, _metadata));memset (&self->run,0 ,sizeof (self->run)); - self->run.tcs = self->encl.encl_base; + self->run.tcs = (__u64 )self->encl.tcs; op.type =ENCL_OP_NOP ;
@@ -431,7 +412,7 @@ TEST_F(enclave, tcs_entry)EXPECT_EQ (self->run.exception_addr,0 );/* Move to the next TCS. */ - self->run.tcs = self->encl.encl_base +PAGE_SIZE ; + self->run.tcs = (__u64 )self->encl.tcs +PAGE_SIZE ;EXPECT_EQ (ENCL_CALL (&op, &self->run, true),0 );
@@ -464,11 +445,9 @@ TEST_F(enclave, pte_permissions)ASSERT_TRUE (setup_test_encl (ENCL_HEAP_SIZE_DEFAULT , &self->encl, _metadata));memset (&self->run,0 ,sizeof (self->run)); - self->run.tcs = self->encl.encl_base; + self->run.tcs = (__u64 )self->encl.tcs; - data_start = self->encl.encl_base + -encl_get_data_offset (&self->encl) + -PAGE_SIZE ; + data_start = self->encl.encl_base + self->encl.data_offset +PAGE_SIZE ;/* * Sanity check to ensure it is possible to write to page that will
tools/testing/selftests/sgx/main.h +2 -0
@@ -29,6 +29,8 @@ struct encl {struct encl_segment *segment_tbl;struct sgx_secs secs;struct sgx_sigstruct sigstruct; +struct sgx_tcs *tcs; +unsigned long data_offset; };extern unsigned char sign_key[];
tools/testing/selftests/sgx/test_encl.lds +12 -5
@@ -2,17 +2,15 @@ OUTPUT_FORMAT(elf64-x86-64) PHDRS { - tcs PT_LOAD; text PT_LOAD; data PT_LOAD; + tcs PT_LOAD; + note PT_NOTE; } SECTIONS { . = 0; - .tcs : { - *(.tcs*) - } : tcs . = ALIGN(4096); .text : {
@@ -24,11 +22,20 @@ SECTIONS .data : { *(.data*) + . = ALIGN(4096); } : data + .tcs : { + *(.tcs*) + } : tcs + + .note : { + *(.note.tcs*) + } : note + /DISCARD/ : { *(.comment*) - *(.note*) + *(.note.gnu.*) *(.debug*) *(.eh_frame*) }
tools/testing/selftests/sgx/test_encl_bootstrap.S +7 -0
@@ -10,6 +10,7 @@ .section ".tcs", "aw" .balign 4096 +encl_tcs: .fill 1, 8, 0 # STATE (set by CPU) .fill 1, 8, 0 # FLAGS .quad encl_ssa_tcs1 # OSSA
@@ -90,3 +91,9 @@ encl_stack: .balign 4096 # Stack of TCS #2 .space 4096 + + .section ".note.tcs", "", @progbits + .long 4 + .long encl_tcs + .long 0 + .string "TCS"
Signature · 1 line
2.35.1
Hi Jarkko,
Quoted 19 lines · Jarkko Sakkinen
Simplify the
test_encl_bootstrap.Sflow 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 <[email protected]>Cc Dave Hansen <[email protected]>Signed-off-by Jarkko Sakkinen <[email protected]>--- tools/testing/selftests/sgx/test_encl_bootstrap.S | 6 +----- 1 file changed, 1 insertion(+), 5 deletions(-)tools/testing/selftests/sgx/test_encl_bootstrap.S +0
-0@@ -40,11 +40,7 @@ .text
tools/testing/selftests/sgx/test_encl_bootstrap.S line 46
Quoted 10 lines
tools/testing/selftests/sgx/test_encl_bootstrap.S +1
-5encl_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
maintainerQuoted 38 lines · Reinette Chatre
Hi Jarkko,
Quoted 19 lines · Jarkko Sakkinen
Simplify the
test_encl_bootstrap.Sflow 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 <[email protected]>Cc Dave Hansen <[email protected]>Signed-off-by Jarkko Sakkinen <[email protected]>--- tools/testing/selftests/sgx/test_encl_bootstrap.S | 6 +----- 1 file changed, 1 insertion(+), 5 deletions(-)tools/testing/selftests/sgx/test_encl_bootstrap.S +0
-0@@ -40,11 +40,7 @@ .texttools/testing/selftests/sgx/test_encl_bootstrap.S line 46
Quoted 10 lines
tools/testing/selftests/sgx/test_encl_bootstrap.S +1
-5encl_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 %raxThe 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_entryto use the same stack, no?
Quoted 7 lines
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
maintainerQuoted 42 lines · Jarkko Sakkinen
Quoted 41 lines · Reinette Chatre
Hi Jarkko,
Quoted 19 lines · Jarkko Sakkinen
Simplify the
test_encl_bootstrap.Sflow 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 <[email protected]>Cc Dave Hansen <[email protected]>Signed-off-by Jarkko Sakkinen <[email protected]>--- tools/testing/selftests/sgx/test_encl_bootstrap.S | 6 +----- 1 file changed, 1 insertion(+), 5 deletions(-)tools/testing/selftests/sgx/test_encl_bootstrap.S +0
-0@@ -40,11 +40,7 @@ .texttools/testing/selftests/sgx/test_encl_bootstrap.S line 46
Quoted 10 lines
tools/testing/selftests/sgx/test_encl_bootstrap.S +1
-5encl_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 %raxThe 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_entryto 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/
Quoted 8 lines
Quoted 4 lines
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
Hi Jarkko,
Quoted 42 lines · Jarkko Sakkinen
Quoted 41 lines · Reinette Chatre
Hi Jarkko,
Quoted 19 lines · Jarkko Sakkinen
Simplify the
test_encl_bootstrap.Sflow 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 <[email protected]>Cc Dave Hansen <[email protected]>Signed-off-by Jarkko Sakkinen <[email protected]>--- tools/testing/selftests/sgx/test_encl_bootstrap.S | 6 +----- 1 file changed, 1 insertion(+), 5 deletions(-)tools/testing/selftests/sgx/test_encl_bootstrap.S +0
-0@@ -40,11 +40,7 @@ .texttools/testing/selftests/sgx/test_encl_bootstrap.S line 46
Quoted 10 lines
tools/testing/selftests/sgx/test_encl_bootstrap.S +1
-5encl_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 %raxThe 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_entryto 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/
Quoted 6 lines
Quoted 4 lines
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?
Quoted 2 lines
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:
tools/testing/selftests/sgx/defines.h +7 -0
@@ -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 */
tools/testing/selftests/sgx/main.c +23 -0
@@ -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); + }/*
tools/testing/selftests/sgx/test_encl.c +12 -0
@@ -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; + + asmvolatile ("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;
A message that is not in the archive (sent off-list or not yet received): [email protected]
Jarkko Sakkinen
maintainerQuoted 17 lines · Jarkko Sakkinen
Quoted 1 line · Reinette Chatre
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.
Quoted 4 lines
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
Quoted 10 lines · Jarkko Sakkinen
Quoted 1 line · Reinette Chatre
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]
Quoted 7 lines
Here's the same with s/RBX/RIP/:
[
0x000021a9]> pi 5
lea rax, loc.encl_stackCompiler 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
maintainerQuoted 15 lines · Reinette Chatre
Quoted 12 lines · Jarkko Sakkinen
Quoted 1 line · Reinette Chatre
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/:
Quoted 8 lines
Quoted 5 lines
[
0x000021a9]> pi 5
lea rax, loc.encl_stackCompiler 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
Hi Jarkko,
Quoted 38 lines · Jarkko Sakkinen
Quoted 15 lines · Reinette Chatre
Quoted 12 lines · Jarkko Sakkinen
Quoted 1 line · Reinette Chatre
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/:
Quoted 8 lines
Quoted 5 lines
[
0x000021a9]> pi 5
lea rax, loc.encl_stackCompiler 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:
Quoted 7 lines
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
maintainerQuoted 54 lines · Reinette Chatre
Hi Jarkko,
Quoted 38 lines · Jarkko Sakkinen
Quoted 15 lines · Reinette Chatre
Quoted 12 lines · Jarkko Sakkinen
Quoted 1 line · Reinette Chatre
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/:
Quoted 8 lines
Quoted 5 lines
[
0x000021a9]> pi 5
lea rax, loc.encl_stackCompiler 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:
Quoted 7 lines
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.
Quoted 6 lines
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/[email protected]/T/#u
BR, Jarkko
BR, Jarkko