[Gems-users] Re: The process to get pid and related questions

["Min Xu (Hsu)" <xu@xxxxxxxxxxx>]

I see. Now I know where is your offsets to these data structures came from. The
offsets hard-coded in pseq.C was obtained several years ago. It is possible that
they are obsolete. You perhaps need to obtain the new offsets manually again.

Carl Mauer was the one who got these offsets in the first place. But he is not
with us anymore. I can only try my best to tell you whatever I know about this
process of getting the offsets. You perhaps need to refer to more documents.
Sorry, I wish I could be more helpful.

First of all, the big picture is that we need to know the solaris internel data
structure that stores each process's PID. There is an great book on this:
"Solaris Internals: Core Kernel Architecture". You may want to check it out.

Then, you perhaps need correct version of the kernel source code. Find the
function resume, try to locate the assembly line that access the PID with the
assistant of the source code. Also, locate the declaration of the PID data
structure and compute the offset to the field you are interested in. For
example, if you see this code:

struct foo {
  int a;
  int b;
  int pid;
}

resume () {
 foo f;
 f->pid = ...;
}

Then, the offset of pid field is 0x8 due to the two integers ahead of it. You
need to find out which register stores pointer "f" in the assembly and then
read f->pid accordingly.

Finally, you perhaps can try scan the memory in simics for a known PID value. In
any case, perhaps many experiments, hours and luck will be needed. :-)

On Fri, 03 Jun 2005 Weihang Jiang wrote :
> Sorry I forgot to attach it. Here is the assembly code of "reume".
> 
> resume: save %sp, -0xb0, %sp
> resume+4: membar #Sync
> resume+8: flushw
> resume+0xc: stx %i7, [%g7 + 0x38]
> resume+0x10: stx %fp, [%g7 + 0x40]
> resume+0x14: ldx [%g7 + 0x110], %o4
> resume+0x18: rd %fprs, %g4
> resume+0x1c: brnz,pt %o4, +0x34 <resume+0x50>
> resume+0x20: ldx [%g7 + 0xa8], %i1
> resume+0x24: ldx [%g7 + 8], %i2
> resume+0x28: ldx [%g7 + 0x78], %g3
> resume+0x2c: btst 4, %g4
> resume+0x30: be,pt %icc,+0x48 <resume+0x78>
> resume+0x34: st %g4, [%i2 + 0x1b8]
> resume+0x38: add %i2, 0xb0, %o0
> resume+0x3c: rd %gsr, %g5
> resume+0x40: call -0xe740 <fp_fksave>
> resume+0x44: stx %g5, [%o0 + 0x120]
> resume+0x48: ba,a,pt %icc,+0x30 <resume+0x78>
> resume+0x4c: nop
> resume+0x50: ldx [%o4 + 0x100], %o0
> resume+0x54: stx %fp, [%g7 + 0x40]
> resume+0x58: btst 4, %g4
> resume+0x5c: st %g4, [%o0 + 0x108]
> resume+0x60: be,pt %icc,+0x18 <resume+0x78>
> resume+0x64: ldx [%g7 + 0x78], %g3
> resume+0x68: ldx [%o4 + 0x100], %o0
> resume+0x6c: rd %gsr, %g5
> resume+0x70: call -0xe770 <fp_fksave>
> resume+0x74: stx %g5, [%o0 + 0x120]
> resume+0x78: mov %g0, %fprs
> resume+0x7c: brz,pt %g3, +0x10 <resume+0x8c>
> resume+0x80: ldx [%i0 + 0x118], %i3
> resume+0x84: call +0x10fda0 <savectx>
> resume+0x88: mov %g7, %o0
> resume+0x8c: ldx [%g7 + 0x118], %i2
> resume+0x90: ldx [%i1 + 0x18], %o0
> resume+0x94: ldx [%o0 + 0x40], %o1
> resume+0x98: sub %o1, 0xb0, %sp
> resume+0x9c: clr %fp
> resume+0xa0: mov %g7, %l3
> resume+0xa4: mov %o0, %g7
> resume+0xa8: stx %o0, [%i1 + 0x10]
> resume+0xac: clrb [%l3 + 0xa0]
> 
> I am running solaris 9
> 
> in opal/system/pseq.C, there is 
> "
> // dereference virtual addresses plus offset (Solaris 9)
> la_t procp = M_PSTATE->dereference( thread_p + 0x110, 8 ); 
> la_t pidp = M_PSTATE->dereference( procp + 0xb0, 8 );
> int32 pid = M_PSTATE->dereference( pidp + 0x4, 4 );
> #if 0
> // Solaris 8
> la_t procp = M_PSTATE->dereference( thread + 0x130, 8 );
> la_t pidp = M_PSTATE->dereference( procp + 0xb0, 8 );
> int32 pid = M_PSTATE->dereference( pidp + 0x4, 4 );
> #endif
> "
> In step 6, I try both 
> @thread = SIM_read_phys_memory(conf.cpu0, 0x3000010,8)
> @procp= SIM_read_phys_memory(conf.cpu0, SIM_logical_to_physical(conf.cpu0
> ,Sim_DI_Data,thread+0x110),8)
> ^^^^^
> @pidp= SIM_read_phys_memory(conf.cpu0, SIM_logical_to_physical(conf.cpu0
> ,Sim_DI_Data,procp+0xb0),8)
> @pid= SIM_read_phys_memory(conf.cpu0, SIM_logical_to_physical(conf.cpu0
> ,Sim_DI_Data,pidp+0x4),4)
> @print pid 
> and 
> @thread = SIM_read_phys_memory(conf.cpu0, 0x3000010,8)
> @procp= SIM_read_phys_memory(conf.cpu0, SIM_logical_to_physical(conf.cpu0
> ,Sim_DI_Data,thread+0x130),8)
> ^^^^^^
> @pidp= SIM_read_phys_memory(conf.cpu0, SIM_logical_to_physical(conf.cpu0
> ,Sim_DI_Data,procp+0xb0),8)
> @pid= SIM_read_phys_memory(conf.cpu0, SIM_logical_to_physical(conf.cpu0
> ,Sim_DI_Data,pidp+0x4),4)
> @print pid 
> 
> Using "0x110" offset, I always receive following errero message : 
> Traceback (most recent call last):
> File "<string>", line 1, in ?
> sim_core.SimExc_Memory: No translation
> If using "0x130" offset, as I mentioned, the value of pid is always equal to 
> 1.
> 
> Thanks
> 
> Weihang 
> 
> On 6/3/05, Min Xu (Hsu) <xu@xxxxxxxxxxx> wrote:
> > 
> > Weihang,
> > 
> > I am not sure I fully grasped the problem. Where is your attachment?
> > 
> > I wonder whether you are getting the right kernel variable for PID. Do you 
> > have
> > the source code for the kernel version you are running? It is possible the
> > kernel code changes from version to version.
> > 
> > Please provide more information.
> > 
> > Thanks!
> > 
> > -Min
> > 
> > On Fri, 03 Jun 2005 Weihang Jiang wrote :
> > > The process to get pid from inside simulator:
> > >
> > > 1. find the break address using mdb in the simulation machine
> > > mdb -k
> > > 2. find the instruction that modified the pid "stx ... [%l0 + 0x10]"
> > > >::dis resume (see attachment)
> > > can't not find "stx ... [%l0 + 0x10]", instead, I pick "resume+0xa8: stx
> > > %o0, [%i1 + 0x10]"
> > > 3. find the starting address of the function "resume"
> > > > resume=X
> > > 102da90
> > > now the starting address of resume + offset is the break point = 
> > 0x102da90 +
> > > 0xa8 = 0x102db38
> > > 4. ./simics -c xxx.check
> > > break 0x102db38
> > > c
> > > @cpu = SIM_current_processor()
> > > @cwp = SIM_read_register(cpu, SIM_get_register_number(cpu, "cwp"))
> > > @va = SIM_get_interface(cpu, "sparc-v9").read_window_register(cpu, cwp, 
> > 25)
> > > + 0x10
> > > @pa = SIM_logical_to_physical(cpu, Sim_DI_Data, va)
> > > @print pa
> > > output: 50331664 (0x3000010 ) (this is thread_physical_addr)
> > >
> > > 5. @SIM_breakpoint(SIM_get_object
> > > ("phys_mem0"),Sim_Break_Physical,Sim_Access_Write,0x3000010,4,0)
> > > c
> > >
> > > 6. @thread = SIM_read_phys_memory(conf.cpu0, 0x3000010,8)
> > > (thread = 2890513120576)
> > > @procp= SIM_read_phys_memory(conf.cpu0, SIM_logical_to_physical(
> > conf.cpu0
> > > ,Sim_DI_Data,thread+0x130),8)
> > > (procp = 21012480)
> > > @pidp= SIM_read_phys_memory(conf.cpu0, SIM_logical_to_physical(conf.cpu0
> > > ,Sim_DI_Data,procp+0xb0),8)
> > > (pidp = 21451608)
> > > @pid= SIM_read_phys_memory(conf.cpu0, SIM_logical_to_physical(conf.cpu0
> > > ,Sim_DI_Data,pidp+0x4),4)
> > > (pid = 1)
> > >
> > > The problem is that the pid is always equal to "1" (never changed). Can
> > > anyone point out where I made mistake?
> > >
> > > --
> > > Weihang Jiang
> > 
> 
> 
> 
> -- 
> Weihang Jiang