14:08 Fri, 22 Feb 2008 PST -0800

SoyLatte: Release 1.0.2

I'm happy to announce another update for SoyLatte, containing a number of minor improvements. Work also progresses on the feature branch, where I'm focusing on native graphics support.

Changes

Bug fixes:

Download

Binaries, source, build, and contribution instructions are all available from SoyLatte Project Page

[/code/java] permanent link

11:29 Sun, 17 Feb 2008 PST -0800

Fixing ptrace(pt_deny_attach, ...) on Mac OS X 10.5 Leopard

Introduction

PT_DENY_ATTACH is a non-standard ptrace() request type that prevents a debugger from attaching to the calling process. Adam Leventhal recently discovered that Leopard extends PT_DENY_ATTACH to prevent introspection into processes using dtrace. I hope Adam will forgive me for quoting him here, but he put it best:

This is antithetical to the notion of systemic tracing, antithetical to the goals of DTrace, and antithetical to the spirit of open source. I'm sure this was inserted under pressure from ISVs, but that makes the pill no easier to swallow.

This article will cover disabling PT_DENY_ATTACH for all processes on Mac OS X 10.5. Over the previous few years, I've provided similar hacks for both Mac OS X 10.4, and 10.3.

To be clear: this work-around is a hack, and I hold that the correct fix is the removal of PT_DENY_ATTACH from Mac OS X.

How it Works

In xnu the sysent array includes function pointers to all system calls. By saving the old function pointer and inserting my own, it's relatively straight-forward to insert code in the ptrace(2) path.

However, with Mac OS X 10.4, Apple introduced official KEXT Programming Interfaces, with the intention of providing kernel binary compatibility between major operating system releases. As a part of this effort, the sysent array's symbol can not be directly resolved from a kernel extension, thus removing the ability to easily override system call. In 10.4, I was able to work-around this with the amusing temp_patch_ptrace() API. This API has disappeared in 10.5.

For Leopard, I decided to find a public symbol that is placed in the data segment, nearby the sysent array. In the kernel's data segment, nsysent is placed (almost) directly before the sysent array. By examining mach_kernel I can determine the offset to the actual sysent array, and then use this in my kext to patch the actual function. To keep things safe, I added sanity checks to verify that I'd found the real sysent array.

Each sysent structure has the following fields:

struct sysent {
	int16_t		sy_narg;		/* number of arguments */
	int8_t		reserved;		/* unused value */
	int8_t		sy_flags;		/* call flags */
	sy_call_t	*sy_call;		/* implementing function */
	sy_munge_t	*sy_arg_munge32;	/* munge system call arguments for 32-bit processes */
	sy_munge_t	*sy_arg_munge64;	/* munge system call arguments for 64-bit processes */
	int32_t		sy_return_type; /* return type */
	uint16_t	sy_arg_bytes;	/* The size of all arguments for 32-bit system calls, in bytes */
};

The "sy_call" field contains a function pointer to the actual implementing function for a given syscall. If we look at the actual sysent table, we'll see that the first entry is "SYS_nosys":

__private_extern__ struct sysent sysent[] = {
    {0, 0, 0, (sy_call_t *)nosys, NULL, NULL, _SYSCALL_RET_INT_T, 0},

To narrow down the haystack, we'll find the address of the nsysent variable, and then search for the nosys function pointer -- as shown above, nosys should be the first entry in the sysent array.

nm /mach_kernel| grep _nsysent
00502780 D _nsysent
nm /mach_kernel| grep T\ _nosys
00388604 T _nosys

Here is a dump of the mach_kernel, starting at 0x502780. You can see the value is 0x01AB, or 427 -- by looking at the kernel headers, we can determine that this is the correct number of syscall entries. 33 bytes after nsysent, we see 0x388604 (in little-endian byte order) -- this is our nosys function pointer. After counting the size of the sysent structure fields, we can determine that the the sysent array is located 32 bytes after the nsysent variable address. (On PPC, it's directly after).

otool -d /mach_kernel
00502780        ab 01 00 00 00 00 00 00 00 00 00 00 00 00 00 00 
00502790        00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 
005027a0        00 00 00 00 04 86 38 00 00 00 00 00 00 00 00 00

Once we have the address of the array, we can find the SYS_ptrace entry and substitute our own ptrace wrapper:

static int our_ptrace (struct proc *p, struct ptrace_args *uap, int *retval)
{
	if (uap->req == PT_DENY_ATTACH) {
		printf("[ptrace] Blocking PT_DENY_ATTACH for pid %d.\n", uap->pid);
		return (0);
	} else {
		return real_ptrace(p, uap, retval);
	}
}
kern_return_t pt_deny_attach_start (kmod_info_t *ki, void *d) {
	...
	real_ptrace = (ptrace_func_t *) _sysent[SYS_ptrace].sy_call;
	_sysent[SYS_ptrace].sy_call = (sy_call_t *) our_ptrace;
	...
}

Download

You can download the kext source here (sig).

Buyer beware: This code has only seen limited testing, and your mileage may vary. If something goes wrong, sanity checks should prevent a panic, and the module will fail to load.

If the module loads correctly, you should see the following in your dmesg output:

[ptrace] Found nsysent at 0x502780 (count 427), calculated sysent location 0x5027a0.
[ptrace] Sanity check 0 1 0 3 4 4: sysent sanity check succeeded.
[ptrace] Patching ptrace(PT_DENY_ATTACH, ...).
[ptrace] Blocking PT_DENY_ATTACH for pid 82248.

Note: To access the nsysent symbol, the kext is required to declare a dependency on a specific version of Mac OS X. When updating to a new minor release, it should be sufficient to change the 'com.apple.kernel' version in the kext's Info.plist. I've uploaded a new version of the kext with this change, but I won't provide future updates unless a code change is required.

<key>OSBundleLibraries</key>
<dict>
    <key>com.apple.kernel</key>
    <string>9.2.0</string>
</dict>

Much thanks to Ryan Chapman for noting this issue, and testing the kext with 10.5.2.

[/code/macosx] permanent link

22:26 Sun, 03 Feb 2008 PST -0800

Porting Java 6 to FreeBSD Sparc

Over the weekend I implemented an initial port of Java 6 to FreeBSD/Sparc64, primarily as a learning exercise -- I wanted to see how difficult it is to port Java to a platform where both the processor and operating system are already independently supported.

landonf@conpanna:bsd-sparc> uname -s -m
FreeBSD sparc64
 
landonf@conpanna:bsd-sparc> ./bin/java -server Hello
Hello, World

I believe this is the first port of the Sparc JVM to a non-Solaris system, and the work should be applicable to supporting other operating systems, such as NetBSD or Linux Sparc systems. This article will discuss the steps I took, with the hope of aiding future porters.

The JRL-licensed code can be downloaded here: patch-java6-freebsd-sparc-1.gz

By downloading these binaries or source code, you certify that you are a Licensee in good standing under the Java Research License of the Java 2 SDK, and that your access, use, and distribution of code and information you may obtain at this site is subject to the License. To ensure compliance, downloading requires "click-through" authentication:

Bootstrap Environment

Building Java requires Java, which is a catch-22 when you're bootstrapping an unsupported system. To work around this, I used an idea (and scripts) suggested by Havard Eidnes: I set up a second Linux system running Sun's Java, and then mounted my FreeBSD build directory at the exact same path on the Linux machine.

Havard's scripts ssh to the bootstrap host and run the Java commands there. The source files are read from the NFS build tree, and the output files are written back.

You can download my slightly modified version of Havard's scripts here: boot-java.tar.gz. Any bugs are surely my own. To use the scripts, set the following environmental variables:

When calling make, you must also ALT_BOOTDIR to the boot-java path (eg, $HOME/boot-java).

Getting it Running

I started by running 'make' and filling in the blanks -- the best approach is to copy liberally from existing platform implementations. In most cases, I borrowed the solaris-sparc implementation, and merged in code from the bsd-amd64 counterpart:

Nearly all the new code needed to be added to hotspot/src/os_cpu/bsd_sparc. I took an iterative approach, starting from the Solaris code, merging in BSD-specific code, and attempting to build the result. Except for the slow machine I was working with (400Mhz!), merging in the BSD code was a fairly swift process.

Sun Studio vs. GCC

Sun builds the Solaris VM using the Sun Studio toolchain, which is not fully compatible with GCC. I had trouble with gcc 3.4, and eventually settled on 4.0, which worked almost perfectly, barring three issues.

First, gcc defines 'sparc' as a standard preprocessor macro. You can guess how well that works while compiling a sparc-related code; passing the -ansi flag disables the define.

Secondly, gcc does not support passing non-const objects as a reference parameter, while Sun Studio allows it. Relying on this is non-standard, but easily fixed -- see 'Reference to a non-const object cannot be initialized with an r-value of that object'.

Lastly, I had to rewrite the Sun Studio inline assembler template (hotspot/src/os_cpu/solaris_sparc/vm/solaris_sparc.il) in standard assembler. A good discussion of the differences between Studio's inline assembler and GCC-style assembly can be found at Alfred Huang's blog. This was straight-forward -- here's an example:

.inline _Atomic_swap32, 2
.volatile
swap    [%o1],%o0
.nonvolatile
.end

Re-written as:

.global _Atomic_swap32
.align 32
_Atomic_swap32:
    swap    [%o1],%o0
    retl
    nop

[/code/java] permanent link

21:04 Sun, 03 Feb 2008 PST -0800

Java Signal Handling: Turning SIGFPE into java.lang.ArithmeticException

Introduction

When implementing a virtual machine such as Java's, it's necessary (and sometimes beneficial) to handle some unexpected conditions by allowing the errors to occur, and then catching the resultant signals delivered by the operating system. Take, for example, divide by zero:

int i = 5 / 0;

Hotspot could generate code to check divisor == 0 before every division operation:

cmpl    $0, %ecx  // Is the divisor 0
je      L2        // Jump to div-by-zero handler
 
movl    %edx, %eax // store in divisor eax
sarl    $31, %edx // clear edx, leaving the sign bit
idivl   %ecx // divide edx:eax / ecx

But instead, Hotspot takes a leap of faith -- since programs should rarely divide by zero, Java emits the division instruction, and if the divisor is 0, relies on its signal handler to interpret the resultant SIGFPE:

if (sig == SIGFPE  && (info->si_code == FPE_INTDIV || info->si_code == FPE_FLTDIV)) {
    stub = SharedRuntime::continuation_for_implicit_exception(thread, pc, SharedRuntime::IMPLICIT_DIVIDE_BY_ZERO);

On Friday, I received a bug report for the x86_64 version of SoyLatte from Jibril Gueye. As it turns out, divide by zero errors were not being handled in the 64-bit VM, and instead of throwing an ArithmeticException, Java was unceremoniously crashing:

landonf@max> /usr/local/soylatte16-amd64-1.0.1/bin/java Test
#
# An unexpected error has been detected by Java Runtime Environment:
#
#  SIGFPE (0x8) at pc=0x0000000101886ba8, pid=35000, tid=0x301000

After fixing the issue, I thought it would be interesting to discuss how Java handles signals, and why the SIGFPE handler didn't work:

Signal Registration and Delivery

After the JVM has parsed its command line arguments, the os::init_2() operating-specific method is called. This method is responsible for performing any remaining OS-specific initialization tasks, such as the registration of signal handlers. The BSD implementation can be found in hotspot/src/os/bsd/vm/os_bsd.cpp.

At this time, an architecture-specific JVM_handle_bsd_signal() function is registered as a handler for SIGSEGV, SIGPIPE, SIGBUS, SIGILL, and SIGFPE. (See signal.h for descriptions.) When a divide by zero error occurs, SIGFPE is delivered to the process, and the JVM's JVM_handle_bsd_signal() is called.

The signal handler is registered using sigaction, with the SA_SIGINFO flag set. According to the Single Unix Specification, "If SA_SIGINFO is set and the signal is caught, the signal-catching function will be entered as:"

void func(int signo, siginfo_t *info, void *context);

Upon a divide by zero, the provided siginfo structure contains a 'si_code' member set to FPE_INTDIV:

typedef struct __siginfo {
    int     si_signo;               /* signal number */
    int     si_errno;               /* errno association */
    int     si_code;                /* signal code */
    ...
} siginfo_t;

With this information, our Java_handle_bsd_signal() implementation can check the signal number and code, and throw an ArithmeticException:

if (sig == SIGFPE  &&
     (info->si_code == FPE_INTDIV || info->si_code == FPE_FLTDIV)) {
    stub = SharedRuntime::continuation_for_implicit_exception(thread,
      pc, SharedRuntime:: IMPLICIT_DIVIDE_BY_ZERO);

SharedRuntime::continuation_for_implicit_exception() returns the entry point to Hotspot-generated code that sets up Java exception dispatching in the current frame. When the signal handler is finished, it saves the program counter and jumps to this stub, which handles setting up the frame and throwing the ArithmeticException.

Mac OS X and FPE_INTDIV

After receiving the bug report, I decided to take a look at Mac OS X's kernel signal handling code. On Darwin, the sendsig function handles creation and dispatch of UNIX signals to user processes. Looking at sendsig, we see that Mac OS X doesn't set si_code to FPE_INTDIV, and as such, JVM_handle_bsd_signal() can't decipher the signal:

    case SIGFPE:
#define FP_IE 0 /* Invalid operation */
#define FP_DE 1 /* Denormalized operand */
#define FP_ZE 2 /* Zero divide */
#define FP_OE 3 /* overflow */
#define FP_UE 4 /* underflow */
#define FP_PE 5 /* precision */
    if (ut->uu_subcode & (1 << FP_ZE)) {
        sinfo64.si_code = FPE_FLTDIV;
    } else if (ut->uu_subcode & (1 << FP_OE)) {
        sinfo64.si_code = FPE_FLTOVF;
    } else if (ut->uu_subcode & (1 << FP_UE)) {
        sinfo64.si_code = FPE_FLTUND;
    } else if (ut->uu_subcode & (1 << FP_PE)) {
        sinfo64.si_code = FPE_FLTRES;
    } else if (ut->uu_subcode & (1 << FP_IE)) {
        sinfo64.si_code = FPE_FLTINV;
    } else {
        printf("unknown SIGFPE code %ld, subcode %lx\n",
              (long) ut->uu_code, (long) ut->uu_subcode);
        sinfo64.si_code = FPE_NOOP;
    }
    break;

As you can see, there's no code to handle FPE_INTDIV, si_code is set to FPE_NOOP, and an error message is printed to the console. A quick check of dmesg shows that our kernel is indeed printing "unknown SIGFPE" when Java attempts a divide by zero:

sudo dmesg | grep SIGFPE
unknown SIGFPE code 1, subcode 0

This is suboptimal behavior, so I've filed a bug (5708523 - xnu sendsig() does not set siginfo->si_code = FPE_INTDIV for SIGFPE). In the meantime, a fix is necessary.

You may recall the 'void *context' argument passed to the signal handler. On Mac OS X, this is actually a pointer to ucontext structure. The ucontext contains the full context of the thread's state, at the time of the exception. This includes the program counter -- a register containing the address of the instruction that caused the exception.

Since we have the address of the instruction, we can determine what the instruction is. Once we know what the instruction is, we determine if it could have caused an integer divide by zero exception. This fix was used previously in Java to support Linux/x86 1.x kernels, which also did not set si_code.

To determine what instruction(s) could cause a FPE_INTDIV on 64-bit x86 machines, I consulted the Intel 64 and IA-32 Architectures Software Developer's Manuals -- the answer is idiv and idivl. Also, on amd64 machines, most operations remain 32-bit, and 64-bit operations require the a REX prefix. We'll need to skip the prefix if it exists.

Now we can add code to examine the program counter in JVM_handle_bsd_signal():

// HACK: si_code == FPE_INTDIV is not supported on Mac OS X (si_code is set to FPE_FPE_NOOP).
// See also xnu-1228 bsd/dev/i386/unix_signal.c, line 365
// Filed as rdar://5708523 - xnu sendsig() does not set siginfo->si_code = FPE_INTDIV for SIGFPE
} else if (sig == SIGFPE && info->si_code == FPE_NOOP) {
    int op = pc[0];
 
    // Skip REX
    if ((pc[0] & 0xf0) == 0x40) {
        op = pc[1];
    } else {
        op = pc[0];
    }
 
    // Check for IDIV
    if (op == 0xF7) {
        stub = SharedRuntime::continuation_for_implicit_exception(thread, pc, SharedRuntime:: IMPLICIT_DIVIDE_BY_ZERO);
    } else {
        // TODO: handle more cases if we are using other x86 instructions
        //   that can generate SIGFPE signal.
        tty->print_cr("unknown opcode 0x%X with SIGFPE.", op);
        fatal("please update this code.");
    }

With the fix in place, Java throws the expected ArithmeticException:

landonf@max:~> java Test
Exception in thread "main" java.lang.ArithmeticException: / by zero
    at Test.main(Test.java:3)

[/code/java] permanent link