 |
forums.ps2dev.org
Homebrew PS2, PSP & PS3 Development Discussions
|
| View previous topic :: View next topic |
| Author |
Message |
SANiK
Joined: 05 Jul 2005
Posts: 29
|
 Posted: Tue Aug 14, 2007 5:08 pm Post subject: Bug awareness - finding the address of arguments is a NoNo |
 |
|
Err, to save you guys a bunch of debug time, be aware that PSPlink causes random bugs when trying to print a float.
The bug results in PSPlink crashing.
Usbhostfs reports an invalid magic error.
The bug is caused by TyRaNid attempting to find the address of a function argument... although, the PSP uses registers to store/pass arguments, hence the bug.
The bug is in the float printing code of PSPlink, hence avoid printing floats in PSPlink.
And never try to get the address of a function argument! |
|
| Back to top |
|
 |
Raphael
Joined: 17 Jan 2006
Posts: 646
Location: Germany
|
| Posted: Tue Aug 14, 2007 11:05 pm Post subject: |
|
|
Could you be more specific where that code is that tries to find the adress of an argument/print floats, that is faulty?
Can't seem to find that part.
_________________
<Don't push the river, it flows.>
http://wordpress.fx-world.org - my devblog
http://wiki.fx-world.org - VFPU documentation wiki
Alexander Berl |
|
| Back to top |
|
|
jbit
Site Admin
Joined: 28 May 2005
Posts: 293
Location: København, Danmark
|
| Posted: Wed Aug 15, 2007 12:42 am Post subject: Re: Bug awareness - finding the address of arguments is a No |
|
|
| SANiK wrote: | | And never try to get the address of a function argument! |
Erm, why exactly... GCC will automatically store the register on the stack and give that address if you try to get the address of a function argument. The following output is from ee-gcc, but should be similar on PSP compiler...
Original C code:
| Code: | void somefunc(int blah)
{
someotherfunc(&blah);
} |
Unoptimized assembly code (with some annotations ;)
| Code: | 00000020 <somefunc>:
20: addiu sp,sp,-48
24: sd ra,32(sp)
28: sd s8,16(sp)
2c: move s8,sp
30: sw a0,0(s8) ; value of "blah" is stored on the stack
34: jal 0 <someotherfunc>
38: move a0,s8 ; arg0 for someotherfunc is set to the address of "blah" on the stack
3c: move sp,s8
40: ld ra,32(sp)
44: ld s8,16(sp)
48: jr ra
4c: addiu sp,sp,48 |
EDIT:I should point out the above ASM is raw MIPS asm, so "move a0,s8" is in the branch delay slot for JAL, therefore it's execution is completed before the branch is completed. |
|
| Back to top |
|
|
chp
Joined: 23 Jun 2004
Posts: 313
|
| Posted: Wed Aug 15, 2007 1:32 am Post subject: |
|
|
Also, a printf(char* buf, ...)-style function is a varargs-function, which will store the value before '...' on the stack, because it will need to get the address to 'buf' before it starts processing the format-string. I assume it was a function similar to this that you were using?
_________________
GE Dominator |
|
| Back to top |
|
|
TyRaNiD
Joined: 18 Jan 2004
Posts: 918
|
| Posted: Wed Aug 15, 2007 2:31 am Post subject: |
|
|
| It does seem to be true that the PSP compiler is broken, it tries to store the value onto the stack but doesn't actually do so until after the code which uses the value. However I fail to see why it would actually die completely, and at any rate I don't fucking care. |
|
| Back to top |
|
|
SANiK
Joined: 05 Jul 2005
Posts: 29
|
| Posted: Wed Aug 15, 2007 2:55 am Post subject: |
|
|
jbit - The PSPcompiler has its issues =/
chp/Raphael - The bug is in the PSPLink code which handles the conversion of floats to string, and not within the function printf() itself
/*Taken from util.c*/
| Code: |
static int is_inf(float val)
{
void *p;
unsigned int conv;
int sign;
int exp;
int mantissa;
BUG: p = (void *) &val;
conv = *((unsigned int *) p);
sign = (conv >> 31) & 1;
exp = (conv >> 23) & 0xff;
mantissa = conv & 0x7fffff;
if((exp == 255) && (mantissa == 0))
{
if(sign)
{
return -1;
}
else
{
return 1;
}
}
return 0;
}
|
|
|
| Back to top |
|
|
Raphael
Joined: 17 Jan 2006
Posts: 646
Location: Germany
|
| Posted: Wed Aug 15, 2007 11:34 pm Post subject: |
|
|
| Code: |
static int is_inf(float val)
{
void *p;
unsigned int conv;
int sign;
int exp;
int mantissa;
p = (void *) &val;
conv = *((unsigned int *) p);
sign = (conv >> 31) & 1;
exp = (conv >> 23) & 0xff;
mantissa = conv & 0x7fffff;
if((exp == 255) && (mantissa == 0))
{
if(sign)
{
return -1;
}
else
{
return 1;
}
}
return 0;
}
int main(int argc, char *argv[])
{
float x = rand()/32768.f;
if (is_inf(x))
{
printf("IsINF\n");
}
sceKernelExitGame();
return(0);
}
|
Compiled with same flags as psplink results in the following compiler output:
| Code: |
$LC1:
.ascii "IsINF\012\000"
#NO_APP
.section .rodata.cst4,"aM",@progbits,4
.align 2
$LC0:
.word 939524096
.text
.align 2
.globl main
.ent main
main:
.frame $sp,16,$31 # vars= 8, regs= 1/0, args= 0, gp= 0
.mask 0x80000000,-8
.fmask 0x00000000,0
.set noreorder
.set nomacro
addiu $sp,$sp,-16
sw $31,8($sp)
jal rand
nop
mtc1 $2,$f0
lui $2,%hi($LC0)
cvt.s.w $f1,$f0
lwc1 $f0,%lo($LC0)($2)
lw $3,0($sp) # !WRONG
li $2,8323072 # 0x7f0000
mul.s $f1,$f1,$f0
ori $2,$2,0xffff
and $5,$3,$2
ext $3,$3,23,8
li $2,255 # 0xff
bne $3,$2,$L2
swc1 $f1,0($sp) # !WRONG
lui $4,%hi($LC1)
bne $5,$0,$L2
addiu $4,$4,%lo($LC1)
jal printf
nop
$L2:
jal sceKernelExitGame
nop
lw $31,8($sp)
move $2,$0
j $31
addiu $sp,$sp,16
|
As Tyranid said, gcc tries to put the float on stack and pop as integer register to do the conversion. However, it puts it in wrong order, hence this code produces wrong results.
Funnily enough, with -O1 it puts them in correct order, but not with any optimization value.
Correct result is given with this code in any case though:
| Code: |
static int is_inf(float val)
{
unsigned int conv;
int sign;
int exp;
int mantissa;
asm("mfc1 %0, %1\n":"=r"(conv):"f"(val));
sign = (conv >> 31) & 1;
exp = (conv >> 23) & 0xff;
mantissa = conv & 0x7fffff;
if((exp == 255) && (mantissa == 0))
{
if(sign)
{
return -1;
}
else
{
return 1;
}
}
return 0;
}
|
Here's a patch for PSPLINKUSB (plus a small fix for VFPU disasm with prefixes) - I could probably commit it myself, but I don't feel like I should be messing on Tyranids repository:
| Code: |
Index: psplink/util.c
===================================================================
--- psplink/util.c (revision 2280)
+++ psplink/util.c (working copy)
@@ -860,14 +860,16 @@
static int is_nan(float val)
{
- void *p;
+ //void *p;
unsigned int conv;
int sign;
int exp;
int mantissa;
- p = (void *) &val;
- conv = *((unsigned int *) p);
+ /*p = (void *) &val;
+ conv = *((unsigned int *) p);*/
+ // Help stupid GCC compiler
+ asm("mfc1 %0, %1\n":"=r"(conv):"f"(val));
sign = (conv >> 31) & 1;
exp = (conv >> 23) & 0xff;
@@ -883,14 +885,16 @@
static int is_inf(float val)
{
- void *p;
+ //void *p;
unsigned int conv;
int sign;
int exp;
int mantissa;
- p = (void *) &val;
- conv = *((unsigned int *) p);
+ /*p = (void *) &val;
+ conv = *((unsigned int *) p);*/
+ // Help stupid GCC compiler
+ asm("mfc1 %0, %1\n":"=r"(conv):"f"(val));
sign = (conv >> 31) & 1;
exp = (conv >> 23) & 0xff;
Index: pspsh/disasm.C
===================================================================
--- pspsh/disasm.C (revision 2280)
+++ pspsh/disasm.C (working copy)
@@ -1149,7 +1149,7 @@
// [hlide] added pfx_sat_names
static const char * const pfx_sat_names[4] =
{
- "", "[0:1]", "", "[-1:1]"
+ "", "0:1", "", "-1:1"
};
/* VFPU prefix instruction operands. The *_SH_* values really specify where
|
_________________
<Don't push the river, it flows.>
http://wordpress.fx-world.org - my devblog
http://wiki.fx-world.org - VFPU documentation wiki
Alexander Berl |
|
| Back to top |
|
|
|
|
You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot vote in polls in this forum
|
Powered by phpBB © 2001, 2005 phpBB Group
|
FAQ
Search
Memberlist
Usergroups
Register
Profile
Log in to check your private messages
Log in
