Skip to content
GitLab
Explore
Sign in
Primary navigation
Search or go to…
Project
Z
zipr_unpin_plugin
Manage
Activity
Members
Labels
Plan
Issues
Issue boards
Milestones
Wiki
Code
Merge requests
Repository
Branches
Commits
Tags
Repository graph
Compare revisions
Snippets
Build
Pipelines
Jobs
Pipeline schedules
Artifacts
Deploy
Releases
Container Registry
Model registry
Operate
Environments
Monitor
Incidents
Service Desk
Analyze
Value stream analytics
Contributor analytics
CI/CD analytics
Repository analytics
Model experiments
Help
Help
Support
GitLab documentation
Compare GitLab plans
Community forum
Contribute to GitLab
Provide feedback
Keyboard shortcuts
?
Snippets
Groups
Projects
This is an archived project. Repository and other project resources are read-only.
Show more breadcrumbs
Open Source Software
zipr_unpin_plugin
Commits
a7927c3b
Commit
a7927c3b
authored
6 years ago
by
Jason Hiser
Browse files
Options
Downloads
Patches
Plain Diff
fixed issue where ldr <float reg>, [pcrel] would get patched but destroy an int register.
parent
0ea0da05
No related branches found
Branches containing commit
No related tags found
Tags containing commit
No related merge requests found
Changes
1
Hide whitespace changes
Inline
Side-by-side
Showing
1 changed file
unpin.cpp
+216
-47
216 additions, 47 deletions
unpin.cpp
with
216 additions
and
47 deletions
unpin.cpp
+
216
−
47
View file @
a7927c3b
...
...
@@ -334,12 +334,13 @@ void Unpin_t::DoUpdateForInstructions()
{
assert
(
bo_wrt
==
nullptr
);
// not yet imp'd WRT offsetting.
assert
(
to_addr
==
0
);
// not yet imp'd WRT offsetting.
const
auto
mnemonic
=
disasm
.
getMnemonic
();
const
auto
is_adr_type
=
mnemonic
==
"adr"
;
const
auto
is_adrp_type
=
mnemonic
==
"adrp"
;
const
auto
mnemonic
=
disasm
.
getMnemonic
();
const
auto
is_adr_type
=
mnemonic
==
"adr"
;
const
auto
is_adrp_type
=
mnemonic
==
"adrp"
;
const
auto
is_ldr_type
=
mnemonic
==
"ldr"
;
const
auto
is_ldr_int_type
=
is_ldr_type
&&
disasm
.
getOperand
(
0
).
isGeneralPurposeRegister
();
const
auto
is_ldr_fp_type
=
is_ldr_type
&&
disasm
.
getOperand
(
0
).
isFpuRegister
();
const
auto
is_ldrsw_type
=
mnemonic
==
"ldrsw"
;
const
auto
mask1
=
(
1
<<
1
)
-
1
;
const
auto
mask2
=
(
1
<<
2
)
-
1
;
const
auto
mask5
=
(
1
<<
5
)
-
1
;
...
...
@@ -486,56 +487,59 @@ void Unpin_t::DoUpdateForInstructions()
}
else
{
// imm19->64 bit address didn't work. Split it up into two parts.
/* the plan :
* FA: b L0
* FT:
* ..
* L0 adrp dest_reg, <addr-page number>
* L1 ldr dest_reg, [dst_reg, #addr-page offset]
* L2: b ft
*/
const
auto
tramp_size
=
3
*
4
;
// 3 insns, 4 bytes each
const
auto
address_to_generate
=
(
imm19_ext
<<
2
)
+
orig_insn_addr
+
(
int64_t
)
reloc
->
GetAddend
()
+
(
int64_t
)
to_addr
;
const
auto
destreg
=
full_insn
&
mask5
;
const
auto
tramp_range
=
ms
.
GetFreeRange
(
tramp_size
);
const
auto
tramp_start
=
tramp_range
.
GetStart
();
// don't be too fancy, just reserve 12 bytes.
ms
.
SplitFreeRange
({
tramp_start
,
tramp_start
+
12
});
const
auto
FA
=
from_insn_location
;
const
auto
FT
=
from_insn_location
+
4
;
const
auto
L0
=
tramp_start
;
const
auto
L1
=
tramp_start
+
4
;
const
auto
L2
=
tramp_start
+
8
;
const
auto
branch_bytes
=
string
(
"
\x00\x00\x00\x14
"
,
4
);
// const auto updated_orig_insn_pageno = orig_insn_addr>>12; // orig_insn_pageno was shifted by 0 for adr
const
auto
relocd_insn_pageno
=
L1
>>
12
;
const
auto
address_to_generate_pageno
=
address_to_generate
>>
12
;
const
auto
address_to_generate_page_offset
=
address_to_generate
&
mask12
;
const
auto
relocd_imm21_ext
=
(
int64_t
)
address_to_generate_pageno
-
(
int64_t
)
relocd_insn_pageno
;
const
auto
relocd_immhi19
=
relocd_imm21_ext
>>
2
;
const
auto
relocd_immlo2
=
relocd_imm21_ext
&
mask2
;
// this should be +/- 4gb, so we shouldn't fail now!
assert
(((
relocd_imm21_ext
<<
43
)
>>
43
)
==
relocd_imm21_ext
);
// put an uncond branch at where the adr was.
// and make it point at L0
ms
.
PlopBytes
(
FA
,
branch_bytes
.
c_str
(),
4
);
zo
->
ApplyPatch
(
FA
,
L0
);
// adrp: 1 imm2lo 1 0000 immhi19 Rd
auto
adrp_bytes
=
string
(
"
\x00\x00\x00\x90
"
,
4
);
auto
adrp_word
=*
(
int
*
)
adrp_bytes
.
c_str
();
adrp_word
|=
destreg
<<
0
;
adrp_word
|=
((
relocd_immlo2
&
mask2
)
<<
29
)
|
((
relocd_immhi19
&
mask19
)
<<
5
);
ms
.
PlopBytes
(
L0
,(
char
*
)
&
adrp_word
,
4
);
if
(
is_ldr_int_type
)
{
// imm19->64 bit address didn't work. Split it up into two parts.
/* the plan :
* FA: b L0
* FT:
* ..
* L0 adrp dest_reg, <addr-page number>
* L1 ldr dest_reg, [dst_reg, #addr-page offset]
* L2: b ft
*/
const
auto
tramp_size
=
3
*
4
;
// 3 insns, 4 bytes each
const
auto
tramp_range
=
ms
.
GetFreeRange
(
tramp_size
);
const
auto
tramp_start
=
tramp_range
.
GetStart
();
// don't be too fancy, just reserve 12 bytes.
ms
.
SplitFreeRange
({
tramp_start
,
tramp_start
+
12
});
// and give the bytes some names
const
auto
L0
=
tramp_start
;
const
auto
L1
=
tramp_start
+
4
;
const
auto
L2
=
tramp_start
+
8
;
// calculate the immediates for the new adr and ldr instruction.
const
auto
relocd_insn_pageno
=
L1
>>
12
;
const
auto
relocd_imm21_ext
=
(
int64_t
)
address_to_generate_pageno
-
(
int64_t
)
relocd_insn_pageno
;
const
auto
relocd_immhi19
=
relocd_imm21_ext
>>
2
;
const
auto
relocd_immlo2
=
relocd_imm21_ext
&
mask2
;
// this should be +/- 4gb, so we shouldn't fail now!
assert
(((
relocd_imm21_ext
<<
43
)
>>
43
)
==
relocd_imm21_ext
);
// put an uncond branch at where the adr was.
// and make it point at L0
ms
.
PlopBytes
(
FA
,
branch_bytes
.
c_str
(),
4
);
zo
->
ApplyPatch
(
FA
,
L0
);
// adrp: 1 imm2lo 1 0000 immhi19 Rd
auto
adrp_bytes
=
string
(
"
\x00\x00\x00\x90
"
,
4
);
auto
adrp_word
=*
(
int
*
)
adrp_bytes
.
c_str
();
adrp_word
|=
destreg
<<
0
;
adrp_word
|=
((
relocd_immlo2
&
mask2
)
<<
29
)
|
((
relocd_immhi19
&
mask19
)
<<
5
);
ms
.
PlopBytes
(
L0
,(
char
*
)
&
adrp_word
,
4
);
// convert: ldr w/x reg : 0 x1 011 0 00 ---imm19---- Rt5 x1 indicates size (0,1 -> w/x)
// to : ldr x/w reg : 1 x1 111 0 01 01 imm12 Rn5 Rt5 x1 indciates size (0,1 -> w/x)
auto
new_ldr_bytes
=
string
(
"
\x00\x00\x40\xb9
"
,
4
);
...
...
@@ -548,9 +552,75 @@ void Unpin_t::DoUpdateForInstructions()
new_ldr_word
|=
scaled_page_offset
<<
10
;
// imm12
new_ldr_word
|=
orig_ldr_size_bit
<<
30
;
// x1
ms
.
PlopBytes
(
L1
,(
char
*
)
&
new_ldr_word
,
4
);
// put an uncond branch the end of the trampoline
// and make it jump at FT
ms
.
PlopBytes
(
L2
,
branch_bytes
.
c_str
(),
4
);
zo
->
ApplyPatch
(
L2
,
FT
);
// should be few enough of these to always print
cout
<<
"Had to trampoline "
<<
disasm
.
getDisassembly
()
<<
"@"
<<
FA
<<
" to "
<<
hex
<<
L0
<<
"-"
<<
L0
+
tramp_size
-
1
<<
endl
;
}
else
if
(
is_ldr_fp_type
)
{
/* the scheme for int-type operations doesn't work
* for fp-type because there is no free register.
* use this plan:
* FA: b L0
* FT:
* ..
* L0 str x0, [sp+128] ; 128 for red zoning
* L1 adrp x0, <addr-page number>
* L2 ldr dest_reg, [x0, #addr-page offset]
* L3 ldr x0, [sp+128] ; 128 for red zoning
* L4 b FT
*/
// allocate and reserve space for the code.
const
auto
tramp_size
=
5
*
4
;
// 3 insns, 4 bytes each
const
auto
tramp_range
=
ms
.
GetFreeRange
(
tramp_size
);
const
auto
tramp_start
=
tramp_range
.
GetStart
();
// don't be too fancy, just reserve 12 bytes.
ms
.
SplitFreeRange
({
tramp_start
,
tramp_start
+
12
});
// give the bytes some names
const
auto
L0
=
tramp_start
;
const
auto
L1
=
tramp_start
+
4
;
const
auto
L2
=
tramp_start
+
8
;
const
auto
L3
=
tramp_start
+
12
;
const
auto
L4
=
tramp_start
+
16
;
// calculate the immediates for the new adr and ldr instruction.
const
auto
relocd_insn_pageno
=
L1
>>
12
;
const
auto
relocd_imm21_ext
=
(
int64_t
)
address_to_generate_pageno
-
(
int64_t
)
relocd_insn_pageno
;
const
auto
relocd_immhi19
=
relocd_imm21_ext
>>
2
;
const
auto
relocd_immlo2
=
relocd_imm21_ext
&
mask2
;
// this should be +/- 4gb, so we shouldn't fail now!
assert
(((
relocd_imm21_ext
<<
43
)
>>
43
)
==
relocd_imm21_ext
);
// put an uncond branch at where the adr was.
// and make it point at L0
ms
.
PlopBytes
(
FA
,
branch_bytes
.
c_str
(),
4
);
zo
->
ApplyPatch
(
FA
,
L0
);
// put save of x0 in place.
// diassembly: f81803e0 stur x0, [sp, #-128]
const
auto
strx0_bytes
=
string
(
"
\xe0\x03\x18\xf8
"
,
4
);
ms
.
PlopBytes
(
L0
,
strx0_bytes
.
c_str
(),
4
);
// adrp: 1 imm2lo 1 0000 immhi19 Rd
auto
adrp_bytes
=
string
(
"
\x00\x00\x00\x90
"
,
4
);
auto
adrp_word
=*
(
int
*
)
adrp_bytes
.
c_str
();
// adrp_word|=destreg<<0; ; destreg for this insn is x0.
adrp_word
|=
((
relocd_immlo2
&
mask2
)
<<
29
)
|
((
relocd_immhi19
&
mask19
)
<<
5
);
ms
.
PlopBytes
(
L1
,(
char
*
)
&
adrp_word
,
4
);
// convert: ldr s/d/q reg: opc2 01 11 00 imm19 Rt5, opc2 indicate size (00,01,10 -> s/d/q)
// to: ldr b/s/d/q reg: size2 11 11 01 opc2 imm12 Rn Rt
auto
new_ldr_bytes
=
string
(
"
\x00\x00\x00\x3d
"
,
4
);
...
...
@@ -578,17 +648,116 @@ void Unpin_t::DoUpdateForInstructions()
ldr_size
==
16
?
0x3u
:
throw
invalid_argument
(
"cannot decode ldr floating-point access size"
);
// add variable fields to new insn
new_ldr_word
|=
destreg
<<
0
;
// Rt
new_ldr_word
|=
destreg
<<
5
;
// Rn
// add variable fields to new insn
and drop it in the mem space.
new_ldr_word
|=
destreg
<<
0
;
// Rt
-- should be actual dest reg, not x0
//
new_ldr_word|=destreg<<5; // Rn
-- should be x0
new_ldr_word
|=
((
address_to_generate_page_offset
/
ldr_size
)
<<
10
);
// imm12
new_ldr_word
|=
(
new_ldr_size_bits
<<
30
);
// size2
new_ldr_word
|=
(
new_ldr_opc2_bits
<<
22
);
// opc2
ms
.
PlopBytes
(
L2
,(
char
*
)
&
new_ldr_word
,
4
);
// drop in the restore of x0
// disassembly: f85803e0 ldur x0, [sp, #-128]
const
auto
ldrx0_bytes
=
string
(
"
\xe0\x03\x58\xf8
"
,
4
);
ms
.
PlopBytes
(
L3
,
ldrx0_bytes
.
c_str
(),
4
);
// put an uncond branch the end of the trampoline
// and make it jump at FT
ms
.
PlopBytes
(
L4
,
branch_bytes
.
c_str
(),
4
);
zo
->
ApplyPatch
(
L4
,
FT
);
// should be few enough of these to always print
cout
<<
"Had to trampoline "
<<
disasm
.
getDisassembly
()
<<
"@"
<<
FA
<<
" to "
<<
hex
<<
L0
<<
"-"
<<
L0
+
tramp_size
-
1
<<
endl
;
ms
.
PlopBytes
(
L1
,(
char
*
)
&
new_ldr_word
,
4
);
}
else
assert
(
0
);
}
}
else
if
(
is_ldrsw_type
)
{
// ldrsw x reg : 1001 1000 imm19 Rt
const
auto
imm19
=
((
int64_t
)
full_insn
>>
5
)
&
mask19
;
const
auto
imm19_ext
=
(
imm19
<<
45
)
>>
45
;
const
auto
referenced_addr
=
(
imm19_ext
<<
2
)
+
from_insn
->
GetAddress
()
->
GetVirtualOffset
()
+
4
;
const
auto
new_imm19_ext
=
((
int64_t
)
referenced_addr
-
(
int64_t
)
from_insn_location
-
4
+
(
int64_t
)
reloc
->
GetAddend
()
+
(
int64_t
)
to_addr
)
>>
2
;
if
(
((
new_imm19_ext
<<
45
)
>>
45
)
==
new_imm19_ext
)
{
const
auto
clean_new_insn
=
full_insn
&
~
(
mask19
<<
5
);
const
auto
new_insn
=
clean_new_insn
|
((
new_imm19_ext
&
mask19
)
<<
5
);
// put the new instruction in the output
ms
.
PlopBytes
(
from_insn_location
,
(
const
char
*
)
&
new_insn
,
insn_bytes_len
);
if
(
m_verbose
)
{
cout
<<
"Relocating a ldrsw pcrel relocation with orig_addr="
<<
hex
<<
(
referenced_addr
)
<<
" offset=(pc+"
<<
imm19_ext
<<
")"
<<
endl
;
cout
<<
"Based on: "
<<
disasm
.
getDisassembly
()
<<
" now located at : 0x"
<<
hex
<<
from_insn_location
<<
" with offset=(pc + "
<<
new_imm19_ext
<<
")"
<<
endl
;
}
}
else
{
// imm19->64 bit address didn't work. Split it up into two parts.
/* the plan :
* FA: b L0
* FT:
* ..
* L0 adrp dest_reg, <addr-page number>
* L1 ldr dest_reg, [dst_reg, #addr-page offset]
* L2: b ft
*/
const
auto
tramp_size
=
3
*
4
;
// 3 insns, 4 bytes each
const
auto
address_to_generate
=
(
imm19_ext
<<
2
)
+
orig_insn_addr
+
(
int64_t
)
reloc
->
GetAddend
()
+
(
int64_t
)
to_addr
;
const
auto
destreg
=
full_insn
&
mask5
;
const
auto
tramp_range
=
ms
.
GetFreeRange
(
tramp_size
);
const
auto
tramp_start
=
tramp_range
.
GetStart
();
// don't be too fancy, just reserve 12 bytes.
ms
.
SplitFreeRange
({
tramp_start
,
tramp_start
+
12
});
const
auto
FA
=
from_insn_location
;
const
auto
FT
=
from_insn_location
+
4
;
const
auto
L0
=
tramp_start
;
const
auto
L1
=
tramp_start
+
4
;
const
auto
L2
=
tramp_start
+
8
;
const
auto
branch_bytes
=
string
(
"
\x00\x00\x00\x14
"
,
4
);
const
auto
relocd_insn_pageno
=
L1
>>
12
;
const
auto
address_to_generate_pageno
=
address_to_generate
>>
12
;
const
auto
address_to_generate_page_offset
=
address_to_generate
&
mask12
;
const
auto
relocd_imm21_ext
=
(
int64_t
)
address_to_generate_pageno
-
(
int64_t
)
relocd_insn_pageno
;
const
auto
relocd_immhi19
=
relocd_imm21_ext
>>
2
;
const
auto
relocd_immlo2
=
relocd_imm21_ext
&
mask2
;
// this should be +/- 4gb, so we shouldn't fail now!
assert
(((
relocd_imm21_ext
<<
43
)
>>
43
)
==
relocd_imm21_ext
);
// put an uncond branch at where the adr was.
// and make it point at L0
ms
.
PlopBytes
(
FA
,
branch_bytes
.
c_str
(),
4
);
zo
->
ApplyPatch
(
FA
,
L0
);
// adrp: 1 imm2lo 1 0000 immhi19 Rd
auto
adrp_bytes
=
string
(
"
\x00\x00\x00\x90
"
,
4
);
auto
adrp_word
=*
(
int
*
)
adrp_bytes
.
c_str
();
adrp_word
|=
destreg
<<
0
;
adrp_word
|=
((
relocd_immlo2
&
mask2
)
<<
29
)
|
((
relocd_immhi19
&
mask19
)
<<
5
);
ms
.
PlopBytes
(
L0
,(
char
*
)
&
adrp_word
,
4
);
// convert: ldrsw x reg : 1001 1000 ---imm19--- Rt
// to : ldrsw x reg : 1011 1001 10 imm12 Rn Rt
auto
new_ldr_bytes
=
string
(
"
\x00\x00\x80\xb9
"
,
4
);
auto
new_ldr_word
=*
(
int
*
)
new_ldr_bytes
.
c_str
();
const
auto
scale
=
0x2
;
const
auto
scaled_page_offset
=
(
address_to_generate_page_offset
>>
scale
)
;
new_ldr_word
|=
destreg
<<
0
;
// Rt
new_ldr_word
|=
destreg
<<
5
;
// Rn
new_ldr_word
|=
scaled_page_offset
<<
10
;
// imm12
ms
.
PlopBytes
(
L1
,(
char
*
)
&
new_ldr_word
,
4
);
// put an uncond branch the end of the trampoline
// and make it jump at FT
...
...
This diff is collapsed.
Click to expand it.
Preview
0%
Loading
Try again
or
attach a new file
.
Cancel
You are about to add
0
people
to the discussion. Proceed with caution.
Finish editing this message first!
Save comment
Cancel
Please
register
or
sign in
to comment