Skip to content

Conversation

aeruder
Copy link
Contributor

@aeruder aeruder commented Oct 5, 2015

Daniel, talked with you on the openocd channel. Could you please try these changes? backtrace should work on all threads after this adjustment.

Andrew Ruder added 3 commits October 5, 2015 14:54
Some targets (Cortex M) require more complicated calculations for
turning the stored stack pointer back into a process stack pointer.
For example, the Cortex M stores a bit in the auto-stacked xPSR
indicating that alignment had to be performed and an additional 4
byte padding is present before the exception stacking.  This change
only sets up the framework for Cortex-M unstacking and does not
add Cortex-M support.

Note: this also fixes the alignment calculation nearly addressed by
change #2301 entitled rtos/rtos.c: fix stack alignment calculation.
Updated calculation is in rtos_generic_stack_align.

Signed-off-by: Andrew Ruder <andrew.ruder@elecsyscorp.com>
Cc: Paul Fertser <fercerpav@gmail.com>
Cc: Andreas Fritiofson <andreas.fritiofson@gmail.com>
Cc: Evan Hunter <evanhunter920@gmail.com>
Cc: Jon Burgess <jburgess777@gmail.com>
In the case that the STKALIGN bit is set on Cortex M processors, on
entry to an exception - the processor can store an additional 4 bytes
of padding before regular stacking to achieve 8-byte alignment on
exception entry.  In the case that this padding is present, the
processor will set bit (1 << 9) in the stacked xPSR register.  Use the
new calculate_process_stack callback to take into account the xPSR
register and use it on the standard Cortex_M3 stacking.

Note: Change #2301 had some misinformation regarding the padding.  On
Cortex-M the padding is stored BEFORE stacking so xPSR is always
available at a fixed offset.

Tested on a Cortex-M0+ (Atmel SAMR21) board which has STKALIGN fixed
to a '1' such that this alignment always occurs on non-aligned stacks.

Behavior of xPSR verified via the (bad-sorry) assembly program below by
setting a breakpoint on the SVC_Handler symbol.  The first time
SVC_Handler is triggered the stack was 0x20000ff8, the second time
SVC_Handler is triggered the stack was 0x20000ffc.  Note that in both
cases the interrupt handler gets 0x20000fd8 for a stack pointer.

GDB exerpt:

Breakpoint 1, 0x000040b6 in Reset_Handler ()
(gdb) hbreak SVC_Handler
Hardware assisted breakpoint 2 at 0x40f8
(gdb) cont
Continuing.

Breakpoint 2, 0x000040f8 in SVC_Handler ()
(gdb) print $msp
$3 = (void *) 0x20000fd8
(gdb) x/9w $msp
0x20000fd8:     0x1     0x2     0x3     0x4
0x20000fe8:     0x88160082      0xa53   0x40ce  0x21000000
0x20000ff8:     0x0
(gdb) cont
Continuing.

Breakpoint 2, 0x000040f8 in SVC_Handler ()
(gdb) print $msp
$4 = (void *) 0x20000fd8
(gdb) x/9w $msp
0x20000fd8:     0x1     0x2     0x3     0x4
0x20000fe8:     0x88160082      0xa53   0x40e8  0x21000200
0x20000ff8:     0x0

Assembly program:

	.cpu cortex-m0plus
	.fpu softvfp
	.thumb
	.syntax unified

.section .vectors
@ pvStack:
	.word	0x20001000
@ pfnReset_Handler:
	.word	Reset_Handler + 1
@ pfnNMI_Handler:
	.word	0
@ pfnHardFault_Handler:
	.word	0
@ pfnReservedM12:
	.word	0
@ pfnReservedM11:
	.word	0
@ pfnReservedM10:
	.word	0
@ pfnReservedM9:
	.word	0
@ pfnReservedM8:
	.word	0
@ pfnReservedM7:
	.word	0
@ pfnReservedM6:
	.word	0
@ pfnSVC_Handler:
	.word	SVC_Handler + 1

.section .text
.global Reset_Handler
Reset_Handler:
    cpsie i
    ldr r0, .stack_start
    ldr r2, .stack_last
    eors r1, r1
.loop_clear:
    str r1, [r0]
    adds r0, r0, #4
    cmp r0, r2
    bne .loop_clear
    subs r2, r2, #4
    mov sp, r2
    movs r0, daniel-k#1
    movs r1, daniel-k#2
    movs r2, daniel-k#3
    movs r3, #4
    svc #0
    ldr r0, .stack_start
    ldr r2, .stack_last
    eors r1, r1
.loop_clear2:
    str r1, [r0]
    adds r0, r0, #4
    cmp r0, r2
    bne .loop_clear2
    mov sp, r2
    movs r0, daniel-k#1
    movs r1, daniel-k#2
    movs r2, daniel-k#3
    movs r3, #4
    svc #0
.loop:
	b .loop
.align 4
.stack_start:
    .word 0x20000f00
.stack_last:
    .word 0x20000ffc

@ first call - 0x2000fff8 -- should already be aligned
@ second call - 0x2000fffc -- should hit the alignment code
.global SVC_Handler
SVC_Handler:
    bx lr

Signed-off-by: Andrew Ruder <andrew.ruder@elecsyscorp.com>
Cc: Paul Fertser <fercerpav@gmail.com>
Cc: Andreas Fritiofson <andreas.fritiofson@gmail.com>
Cc: Evan Hunter <evanhunter920@gmail.com>
Cc: Jon Burgess <jburgess777@gmail.com>
@daniel-k
Copy link
Owner

daniel-k commented Oct 6, 2015

@aeruder
Your patches seem to work.

before:

(gdb) thr 2
[Switching to thread 2 (Thread 2)]
#0  mutex_lock (mutex=0x200001c0 <_rx_mutex>) at [...]
53  }
(gdb) bt
#0  mutex_lock (mutex=0x200001c0 <_rx_mutex>) at [...]
#1  0x20002dfc in ?? ()
Backtrace stopped: previous frame identical to this frame (corrupt stack?)

after:

(gdb) thr 2
[Switching to thread 2 (Thread 2)]
#0  mutex_lock (mutex=mutex@entry=0x200001c0 <_rx_mutex>) at [...]
53  }
(gdb) bt
#0  mutex_lock (mutex=mutex@entry=0x200001c0 <_rx_mutex>) at [...]
#1  0x0000063a in uart_stdio_read (
    buffer=0x200033b0 "~_'!\342\371m\307M\017\321\071\211%\314\377*\272\t\252ak\362\066\036\313\062,\350\035\001\357\v\274̷\346\070\225\347B\264\307E\314\357\347\354G\300\334\017\231\211\275I\033\037\212z$\024\273\035\306\062LH\255L\t\212e\177\025\234\002]\fO\331M\235\253\233\356\271!\321\rܡ\376Z\027I\024\252\221\r\333F\245\337\273\212\177\\m=!\322\366\240\204\226t_\214\206B\234Q\275\256\230go\024\304|\362]\205\303\034\255\311\061t\374\222)\222\321\024B\340\b>\336\341\245\033\347\023?m\233\345\315\327\213\004\264\377a%\016\352\370Wx\006\224\255\066]ê\226\312?\211\314\"\207\371\300\324\354{$\270ڮ\212\257\265U"..., count=1024)
    at [...]
#2  0x000000fe in _read_r (r=<optimized out>, fd=<optimized out>, buffer=<optimized out>, count=<optimized out>)
    at [...]
#3  0x00004cb8 in __sread ()
#4  0x00005b56 in __srefill_r ()
#5  0x00004c40 in __srget_r ()
#6  0x000019be in readline (size=5, buf=0x0 <interrupt_vector>)
    at [...]
#7  shell_run (shell_commands=0xf63 <main+50>, shell_commands@entry=0x0 <interrupt_vector>, 
    line_buf=0x0 <interrupt_vector>, 
    line_buf@entry=0x200007f8 <main_stack+1328> "\376`\346f\376", <incomplete sequence \370>, len=5, len@entry=128)
    at [...]
#8  0x00000f62 in main () at [...]

Stackpointer:

(gdb) p $sp
$2 = (void *) 0x20000738 <main_stack+1136>

daniel-k pushed a commit that referenced this pull request Oct 6, 2015
In the case that the STKALIGN bit is set on Cortex M processors, on
entry to an exception - the processor can store an additional 4 bytes
of padding before regular stacking to achieve 8-byte alignment on
exception entry.  In the case that this padding is present, the
processor will set bit (1 << 9) in the stacked xPSR register.  Use the
new calculate_process_stack callback to take into account the xPSR
register and use it on the standard Cortex_M3 stacking.

Note: Change #2301 had some misinformation regarding the padding.  On
Cortex-M the padding is stored BEFORE stacking so xPSR is always
available at a fixed offset.

Tested on a Cortex-M0+ (Atmel SAMR21) board which has STKALIGN fixed
to a '1' such that this alignment always occurs on non-aligned stacks.

Behavior of xPSR verified via the (bad-sorry) assembly program below by
setting a breakpoint on the SVC_Handler symbol.  The first time
SVC_Handler is triggered the stack was 0x20000ff8, the second time
SVC_Handler is triggered the stack was 0x20000ffc.  Note that in both
cases the interrupt handler gets 0x20000fd8 for a stack pointer.

GDB exerpt:

Breakpoint 1, 0x000040b6 in Reset_Handler ()
(gdb) hbreak SVC_Handler
Hardware assisted breakpoint 2 at 0x40f8
(gdb) cont
Continuing.

Breakpoint 2, 0x000040f8 in SVC_Handler ()
(gdb) print $msp
$3 = (void *) 0x20000fd8
(gdb) x/9w $msp
0x20000fd8:     0x1     0x2     0x3     0x4
0x20000fe8:     0x88160082      0xa53   0x40ce  0x21000000
0x20000ff8:     0x0
(gdb) cont
Continuing.

Breakpoint 2, 0x000040f8 in SVC_Handler ()
(gdb) print $msp
$4 = (void *) 0x20000fd8
(gdb) x/9w $msp
0x20000fd8:     0x1     0x2     0x3     0x4
0x20000fe8:     0x88160082      0xa53   0x40e8  0x21000200
0x20000ff8:     0x0

Assembly program:

	.cpu cortex-m0plus
	.fpu softvfp
	.thumb
	.syntax unified

.section .vectors
@ pvStack:
	.word	0x20001000
@ pfnReset_Handler:
	.word	Reset_Handler + 1
@ pfnNMI_Handler:
	.word	0
@ pfnHardFault_Handler:
	.word	0
@ pfnReservedM12:
	.word	0
@ pfnReservedM11:
	.word	0
@ pfnReservedM10:
	.word	0
@ pfnReservedM9:
	.word	0
@ pfnReservedM8:
	.word	0
@ pfnReservedM7:
	.word	0
@ pfnReservedM6:
	.word	0
@ pfnSVC_Handler:
	.word	SVC_Handler + 1

.section .text
.global Reset_Handler
Reset_Handler:
    cpsie i
    ldr r0, .stack_start
    ldr r2, .stack_last
    eors r1, r1
.loop_clear:
    str r1, [r0]
    adds r0, r0, #4
    cmp r0, r2
    bne .loop_clear
    subs r2, r2, #4
    mov sp, r2
    movs r0, #1
    movs r1, #2
    movs r2, #3
    movs r3, #4
    svc #0
    ldr r0, .stack_start
    ldr r2, .stack_last
    eors r1, r1
.loop_clear2:
    str r1, [r0]
    adds r0, r0, #4
    cmp r0, r2
    bne .loop_clear2
    mov sp, r2
    movs r0, #1
    movs r1, #2
    movs r2, #3
    movs r3, #4
    svc #0
.loop:
	b .loop
.align 4
.stack_start:
    .word 0x20000f00
.stack_last:
    .word 0x20000ffc

@ first call - 0x2000fff8 -- should already be aligned
@ second call - 0x2000fffc -- should hit the alignment code
.global SVC_Handler
SVC_Handler:
    bx lr

Signed-off-by: Andrew Ruder <andrew.ruder@elecsyscorp.com>
Cc: Paul Fertser <fercerpav@gmail.com>
Cc: Andreas Fritiofson <andreas.fritiofson@gmail.com>
Cc: Evan Hunter <evanhunter920@gmail.com>
Cc: Jon Burgess <jburgess777@gmail.com>
daniel-k pushed a commit that referenced this pull request Sep 4, 2016
Add new AARCH64 target and ARMv8 support files.
This is an instantiation from the cortex_a files but modified to support
64bit ARMv8. Not all features are complete, notably breakpts and single
stepping are not yet implemented.
Currently it lets you halt of the processors, resume, dump the cpu registers
and even get a stack trace with gdb.

> halt
invalid mode value encountered 5
target state: halted
unrecognized psr mode: 0x5
target halted in ARM state due to debug-request, current mode: UNRECOGNIZED
cpsr: 0x600001c5 pc: 0x00093528
MMU: disabled, D-Cache: disabled, I-Cache: disabled

> targets
    TargetName         Type       Endian TapName            State
--  ------------------ ---------- ------ ------------------ ------------
 0* cpu0               aarch64    little cpu.dap            halted

> reg
===== arm v8 registers
(0) r0 (/64): 0x00000000FFFFFFED (dirty)
(1) r1 (/64): 0x00000000F76E4000
(2) r2 (/64): 0x0000000000000000
(3) r3 (/64): 0x0000000000010000
(4) r4 (/64): 0xFFFFFFC06E2939E1
(5) r5 (/64): 0x0000000000000018
(6) r6 (/64): 0x003A699CFB3C8480
(7) r7 (/64): 0x0000000053555555
(8) r8 (/64): 0x00FFFFFFFFFFFFFF
(9) r9 (/64): 0x000000001FFEF992
(10) r10 (/64): 0x0000000000000001
(11) r11 (/64): 0x0000000000000000
(12) r12 (/64): 0x00000000000000F0
(13) r13 (/64): 0x00000000EFDFEAC8
(14) r14 (/64): 0x00000000F6DDA659
(15) r15 (/64): 0x0000000000000000
(16) r16 (/64): 0xFFFFFFC0000F9094
(17) r17 (/64): 0x0000000000000000
(18) r18 (/64): 0x0000000000000000
(19) r19 (/64): 0xFFFFFFC00087C000
(20) r20 (/64): 0x0000000000000002
(21) r21 (/64): 0xFFFFFFC000867C28
(22) r22 (/64): 0xFFFFFFC000916A52
(23) r23 (/64): 0xFFFFFFC00116D8B0
(24) r24 (/64): 0xFFFFFFC000774A0C
(25) r25 (/64): 0x000000008007B000
(26) r26 (/64): 0x000000008007D000
(27) r27 (/64): 0xFFFFFFC000080450
(28) r28 (/64): 0x0000004080000000
(29) r29 (/64): 0xFFFFFFC00087FF20
(30) r30 (/64): 0xFFFFFFC000085114
(31) sp (/64): 0xFFFFFFC00087FF20
(32) pc (/64): 0xFFFFFFC000093528
(33) xPSR (/64): 0x00000000600001C5

And from gdb

(gdb) bt
 #0  cpu_do_idle () at /mnt/host/source/src/third_party/kernel/3.14/arch/arm64/mm/proc.S:87
 #1  0xffffffc000085114 in arch_cpu_idle () at /mnt/host/source/src/third_party/kernel/3.14/arch/arm64/kernel/process.c:107
 #2  0x0000000000000000 in ?? ()

Change-Id: Iccb1d15c7d8ace7b9e811dac3c9757ced4d0f618
Signed-off-by: David Ung <davidu@nvidia.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants