Observability of the Java Virtual Machine

-
Image
The JVM is one of the most observable runtimes. It provides us lots of tools for troubleshooting a JVM application in production. 1. Thread observability Threads are how the JVM actually does work. When something is wrong in production, the symptom is almost always a thread: stopped, blocked, leaking etc. Thread dumps work on any JVM with no  instrumentation, no agents, no restarts. <Example project link with /threaddump endpoint>         // (1) Deadlock — two threads grab the same pair of locks in opposite order.         new Thread(() -> grab(LOCK_A, LOCK_B), "deadlock-A-then-B").start();         new Thread(() -> grab(LOCK_B, LOCK_A), "deadlock-B-then-A").start(); http://localhost:8080/actuator/threaddump To list the JVMS, we can use the command below. PS C:\observe-jvm> jps -lv 25296 jdk.jcmd/sun.tools.jps.Jps -Dapplication.home=C:\Program Files\Microsoft\jdk-21.0.3.9-hotspot -Xms8m -Djdk.module.main=...
Post a Comment

Understanding SimpleScalar codebase

-
Among others, sim-safe is the most user-friendly simulator and the implementation is crafted for clarity.

Generally there are two types of simulators:
  • Functional simulators
    • implement architecture (what programmers see)
    • perform the actual execution
    • sim-fast, sim-safe, sim-profile, sim-cache
  • Performance simulators
    • implement micro-architecture
    • model system internals
    • measure time
    • sim-outorder
Question: What is Dlite debugger? How can we debug a simulator??

DLite is a symbolic debugger. To use it, start the simulator with -i option
Then user the debugger commands, for example: "step"

Question: To work with TLB, do I have to fully understand sim-outorder ??

http://www.simplescalar.com/docs/simple_tutorial_v2.pdf
page:14

http://www.simplescalar.com/docs/hack_guide_v2.pdf

Question:
How does sim-safe count number of instructions?
Where is iTLB related code?

In main.c file:

/* options database*/
struct opt_odb_t *sim_odb;

 /* simulated registers */
static struct regs_t regs;

 /* simulated memory */
static struct mem_t *mem = NULL;

 /* stats database */
struct stat_sdb_t *sim_sdb;

 /* execution instruction counter */
counter_t sim_num_insn = 0;

 /* track number of refs */
static counter_t sim_num_refs = 0; 

First of all, all the options are written to an options database. There are struct representations for registers and memory. Statistics and counters are also defined as global variables.

int main(int argc, char **argv, char **envp)
{
/* register global options */
sim_odb = opt_new();

 /* register all simulator-specific options */
sim_reg_options(sim_odb);

/* initialize all simulation modules */
sim_init();

 /* initialize architected state */
sim_load_prog(argv[exec_index], argc - exec_index, argv + exec_index, envp);

/* register all simulator stats */
sim_sdb = stat_new();
sim_reg_stats(sim_sdb);

After filling the variables for global and simulator specific options, the simulation module is initialized. This means, the data structures for register and memory representations are created. After that, the most important step is reading the binary file and putting that ECOFF file to related memory structure:

/* initialize architected state */
sim_load_prog(argv[exec_index], argc - exec_index, argv + exec_index, envp);

Loading the binary is done by "loader.c" file. This process is complicated and related to the internals of ECOFF file format, which is beyond the scope of our research.

/* load the program into memory, try both endians */
#if defined(__CYGWIN32__) || defined(_MSC_VER)
fobj = fopen(argv[0], "rb");
#else
fobj = fopen(argv[0], "r");
#endif
if (!fobj) {
fprintf(stderr, "cannot open executable");

 exit(1);
}

 if (fread(&fhdr, sizeof(struct ecoff_filehdr), 1, fobj) < 1) {

fprintf(stderr, "cannot read header from executable");

 exit(1);
}

Then we go on by creating structures for statistical data.

stat_reg_counter(sdb, "sim_num_insn",
"total number of instructions executed",
&sim_num_insn, sim_num_insn, NULL);
stat_reg_counter(sdb, "sim_num_refs",
"total number of loads and stores executed",
&sim_num_refs, 0, NULL);
stat_reg_int(sdb, "sim_elapsed_time",
"total simulation time in seconds",

&sim_elapsed_time, 0, NULL);

After the preparation, now it's time to call the simulator main function:

/* record start of execution time, used in rate stats */
sim_start_time = time((time_t *)NULL);

 /* output simulation conditions */
char *s = ctime(&sim_start_time);

fprintf(stderr, "\nsim: simulation started @ %s, options follow:\n", s);

 sim_main();

Now let's have a look at sim-safe.c file:


/* start simulation, program loaded, processor precise state initialized */
void
sim_main(void)
{

fprintf(stderr, "sim: ** starting functional simulation **\n");

 /* set up initial default next PC */
regs.regs_NPC = regs.regs_PC + sizeof(md_inst_t);

 while (TRUE)
{

 /* get the next instruction to execute */
MD_FETCH_INST(inst, mem, regs.regs_PC);

 /* keep an instruction count */
sim_num_insn++;
/* decode the instruction */
MD_SET_OPCODE(op, inst);

 /* execute the instruction */
switch (op)
{
#define DEFINST(OP,MSK,NAME,OPFORM,RES,FLAGS,O1,O2,I1,I2,I3) \
case OP: \
          SYMCAT(OP,_IMPL); \
          break;

 /* Additional case statements..
}

 /* go to the next instruction */
regs.regs_PC = regs.regs_NPC;
regs.regs_NPC += sizeof(md_inst_t);

 /* finish early? */
if (max_insts && sim_num_insn >= max_insts)

return;
}
}

As the comment indicates, before the simulator function is called, the program is loaded and the processor is initialized. Inside the main loop, the next instruction is fetched, counted, decoded and executed. Then we go to the next instruction if the maximum instruction limit is not reached.

Comments

Post a Comment

Popular posts from this blog

The WeakReference class, monitoring memory leak and garbage collection in a Java application

-
Image
 Below is a Stack implementation that uses an internal resizeable array structure.  public class MyStack< T > implements Stack< T > { private static final int CAPACITY = 100 ; private Object[] array ; private int pos = 0 ; public MyStack () { this . array = new Object[ CAPACITY ] ; } @Override public void push ( T item) { if ( pos >= array . length / 2 ) { Object[] newArray = new Object[ pos * 2 ] ; System. arraycopy ( array , 0 , newArray , 0 , array . length ) ; array = newArray ; } array [ pos ++] = item ; } @Override public T pop () { if (isEmpty()) { throw new RuntimeException( "empty stack" ) ; } @SuppressWarnings ( "unchecked" ) T item = ( T ) array [ pos - 1 ] ; pos -= 1 ; return item ; } @Override @SuppressWarnings ( "unchecked" ) public T peek...
Read more

Simplescalar Simulator - Part 2: sim-outorder.c

-
sim-outorder implements a very detailed out-of-order issue superscalar processor with a two-level memory system and speculative execution support. This simulator is a performance simulator, tracking the latency of all pipeline operations. Let's have a look at SimpleScalar simulator's code related to iTLB. (Instruction Translation Lookaside Buffer) First, there are simulator options related to iTLB: /* register simulator-specific options */ void sim_reg_options(struct opt_odb_t *odb) { /* TLB options */ opt_reg_string(odb, "-tlb:itlb", "instruction TLB config, i.e., { |none}", &itlb_opt, "itlb:16:4096:4:l", /* print */TRUE, NULL); opt_reg_int(odb, "-tlb:lat", "inst/data TLB miss latency (in cycles)", &tlb_miss_lat, /* default */30, /* print */TRUE, /* format */NULL); } /* check simulator-specific option values */ void  sim_check_options(... ) { /* sim-outorder.c  line: 1098*/   /* u...
Read more

Notes on Java Performance

-
Performance monitoring, performance profiling and performance tuning are different activities. For example, monitoring is not an intrusive action but profiling is. Because it may be changing the responsiveness of the running application. User CPU utilization vs Kernel CPU utilization: The ideal situation is when Kernel CPU utilization is 0%. So ideally, all the CPU cycles must be spent on our application code. JVM just in time compiler performs dynamic optimizations. So it makes decisions while the program is running and generates native code with better performance. Java 1.0 was completely interpreted but beginning with Java 1.1.8, there was a just-in-time compiler. It was 8 times faster. After that, paralellizing the garbage collector was a huge improvement. There are hundreds of Java tuning flags. https://github.com/ScottOaks/JavaPerformanceTuning Boolean flags use this syntax: -XX:+FlagName enables the flag, and -XX:-FlagName disables the flag. 32 BIT VS 64 BIT Fr...
Read more