sanchayanmaity/minix
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity
minix/kernel/arch/i386/breakpoints.c

58 lines
1.6 KiB
C
Raw Normal View History

Convert kernel over to bsdmake
2010-04-02 00:22:33 +02:00
#include "kernel/kernel.h"
this patch adds access to the debug breakpoints to the kernel. They are not used atm, but having them in trunk allows them to be easily used when needed. To set a breakpoint that triggers when the variable foo is written to (the most common use case), one calls: breakpoint_set(vir2phys((vir_bytes) &foo), 0, BREAKPOINT_FLAG_MODE_GLOBAL | BREAKPOINT_FLAG_RW_WRITE | BREAKPOINT_FLAG_LEN_4); It can later be disabled using: breakpoint_set(vir2phys((vir_bytes) &foo), 0, BREAKPOINT_FLAG_MODE_OFF); There are some limitations: - There are at most four breakpoints (hardware limit); the index of the breakpoint (0-3) is specified as the second parameter of breakpoint_set. - The breakpoint exception in the kernel is not handled and causes a panic; it would be reasonably easy to change this by inspecing DR6, printing a message, disabling the breakpoint and continuing. However, in my experience even just a panic can be very useful. - Breakpoints can be set only in the part of the address space that is in every page table. It is useful for the kernel, but to use this for user processes would require saving and restoring the debug registers as part of the context switch. Although the CPU provides support for local breakpoints (I implemened this as BREAKPOINT_FLAG_LOCAL) they only work if task switching is used.
2010-03-19 20:15:20 +01:00
#include "proto.h"
#include "debugreg.h"
Remove useless symbol declarations from headers, make symbols local where possible, add some explicit initialization to global variables.
2010-04-22 09:49:40 +02:00
PRIVATE int breakpoint_set(phys_bytes linaddr, int bp, const int flags)
this patch adds access to the debug breakpoints to the kernel. They are not used atm, but having them in trunk allows them to be easily used when needed. To set a breakpoint that triggers when the variable foo is written to (the most common use case), one calls: breakpoint_set(vir2phys((vir_bytes) &foo), 0, BREAKPOINT_FLAG_MODE_GLOBAL | BREAKPOINT_FLAG_RW_WRITE | BREAKPOINT_FLAG_LEN_4); It can later be disabled using: breakpoint_set(vir2phys((vir_bytes) &foo), 0, BREAKPOINT_FLAG_MODE_OFF); There are some limitations: - There are at most four breakpoints (hardware limit); the index of the breakpoint (0-3) is specified as the second parameter of breakpoint_set. - The breakpoint exception in the kernel is not handled and causes a panic; it would be reasonably easy to change this by inspecing DR6, printing a message, disabling the breakpoint and continuing. However, in my experience even just a panic can be very useful. - Breakpoints can be set only in the part of the address space that is in every page table. It is useful for the kernel, but to use this for user processes would require saving and restoring the debug registers as part of the context switch. Although the CPU provides support for local breakpoints (I implemened this as BREAKPOINT_FLAG_LOCAL) they only work if task switching is used.
2010-03-19 20:15:20 +01:00
{
u32_t dr7, dr7flags;
if (bp >= BREAKPOINT_COUNT)
return EINVAL;
/* convert flags */
dr7flags = 0;
switch (flags & BREAKPOINT_FLAG_RW_MASK) {
case BREAKPOINT_FLAG_RW_EXEC: dr7flags |= DR7_RW_EXEC(bp); break;
case BREAKPOINT_FLAG_RW_WRITE: dr7flags |= DR7_RW_WRITE(bp); break;
case BREAKPOINT_FLAG_RW_RW: dr7flags |= DR7_RW_RW(bp); break;
default: return EINVAL;
}
switch (flags & BREAKPOINT_FLAG_LEN_MASK) {
case BREAKPOINT_FLAG_LEN_1: dr7flags |= DR7_LN_1(bp); break;
case BREAKPOINT_FLAG_LEN_2: dr7flags |= DR7_LN_2(bp); break;
case BREAKPOINT_FLAG_LEN_4: dr7flags |= DR7_LN_4(bp); break;
default: return EINVAL;
}
switch (flags & BREAKPOINT_FLAG_MODE_MASK) {
case BREAKPOINT_FLAG_MODE_OFF: break;
case BREAKPOINT_FLAG_MODE_LOCAL: dr7flags |= DR7_L(bp); break;
case BREAKPOINT_FLAG_MODE_GLOBAL: dr7flags |= DR7_G(bp); break;
default: return EINVAL;
}
/* disable breakpoint before setting address */
dr7 = st_dr7();
dr7 &= ~(DR7_L(bp) | DR7_G(bp) | DR7_RW_MASK(bp) | DR7_LN_MASK(bp));
ld_dr7(dr7);
/* need to set new breakpoint? */
if ((flags & BREAKPOINT_FLAG_MODE_MASK) == BREAKPOINT_FLAG_MODE_OFF)
return 0;
/* set breakpoint address */
switch (bp) {
case 0: ld_dr0(linaddr); break;
case 1: ld_dr1(linaddr); break;
case 2: ld_dr2(linaddr); break;
case 3: ld_dr3(linaddr); break;
Fixed some incorrect uses of printf-like functions.
2010-04-01 16:30:36 +02:00
default: panic("%s:%d: invalid breakpoint index", __FILE__, __LINE__);
this patch adds access to the debug breakpoints to the kernel. They are not used atm, but having them in trunk allows them to be easily used when needed. To set a breakpoint that triggers when the variable foo is written to (the most common use case), one calls: breakpoint_set(vir2phys((vir_bytes) &foo), 0, BREAKPOINT_FLAG_MODE_GLOBAL | BREAKPOINT_FLAG_RW_WRITE | BREAKPOINT_FLAG_LEN_4); It can later be disabled using: breakpoint_set(vir2phys((vir_bytes) &foo), 0, BREAKPOINT_FLAG_MODE_OFF); There are some limitations: - There are at most four breakpoints (hardware limit); the index of the breakpoint (0-3) is specified as the second parameter of breakpoint_set. - The breakpoint exception in the kernel is not handled and causes a panic; it would be reasonably easy to change this by inspecing DR6, printing a message, disabling the breakpoint and continuing. However, in my experience even just a panic can be very useful. - Breakpoints can be set only in the part of the address space that is in every page table. It is useful for the kernel, but to use this for user processes would require saving and restoring the debug registers as part of the context switch. Although the CPU provides support for local breakpoints (I implemened this as BREAKPOINT_FLAG_LOCAL) they only work if task switching is used.
2010-03-19 20:15:20 +01:00
}
/* set new flags */
dr7 |= dr7flags;
ld_dr7(dr7);
return 0;
}
Reference in a new issue Copy permalink