ucontext_t fpregs ABI incompatibility

[alexrp]

• Your Windows build number: (Type ver at a Windows Command Prompt)

Microsoft Windows [Version 10.0.15063]

• What you're doing and what's happening: (Copy&paste specific commands and their output, or include screen shots)

Given a ucontext_t *ctx pointer, I'm trying to access ctx->uc_mcontext.fpregs->_xmm[0].

• What's wrong / what should be happening instead:

The ctx->uc_mcontext.fpregs field contains a NULL pointer. On a normal Linux system, the ctx->uc_mcontext.fpregs field is always a non-NULL pointer, presumably pointing to ctx->__fpregs_mem. Indeed, if I dig into ctx->__fpregs_mem, all the data is there as expected. So it seems WSL is simply not setting the ctx->uc_mcontext.fpregs field as it should.

[benhillis]

@alexrp - Thanks for posting, could you share an strace so I can narrow down what's happening?

[therealkenc]

@benhillis - I don't know if a strace will show much, because there won't be a syscall fail.

Richard's tight repro from unrelated #1692 (message) demonstrates the problem. Besides his gregs ask, fpregs are all nada. On Real Linux it's populated, natch.

[...blah blah gdb startup]
(gdb) break 16
Breakpoint 1 at 0x40071d: file ucontext.c, line 16.
(gdb) run
Starting program: /home/ken/Devel/ucontext/ucontext

Program received signal SIGSEGV, Segmentation fault.
0x0000000000400812 in main () at ucontext.c:41
41              *x = 3;
(gdb) cont
Continuing.

Breakpoint 1, segv (sig=11, info=0x7ffffffddd40, context=0x7ffffffdd730) at ucontext.c:16
16              printf("SIGSEGV at 0x%lx, fixing up\n", (unsigned long)fault_addr);
(gdb) list
11              ucontext_t *c = context;
12              mcontext_t m = c->uc_mcontext;
13              void *fault_addr = (void *)m.gregs[REG_CR2];
14              void *addr;
15
16              printf("SIGSEGV at 0x%lx, fixing up\n", (unsigned long)fault_addr);
17
18              addr = mmap(fault_addr, 4096, PROT_READ|PROT_WRITE, MAP_ANONYMOUS|MAP_PRIVATE, 0, 0);
19              if (addr == MAP_FAILED) {
20                      printf("map failed!\n");
(gdb) print m
$1 = {gregs = {0, 255, 8, 8, 4195776, 140737488215856, 0, 0, 11, 140737488215104, 140737488215632, 0, 0, 3735928559,
    8, 140737488215616, 4196370, 66118, 356485103667, 0, 0, 0, 0}, fpregs = 0x0, __reserved1 = {0, 0, 0, 0, 0, 0, 0,
    0}}
(gdb)

[alexrp]

To be clear, the ucontext_t is obtained as the third argument from a signal handler that was installed via sigaction with SA_SIGINFO.

[therealkenc]

To be clear, the ucontext_t is obtained as the third argument from a signal handler that was installed via sigaction with SA_SIGINFO.

Yep. In the test case I cited it is:

static void segv(int sig, siginfo_t *info, void *context)
{
	ucontext_t *c = context;
	mcontext_t m = c->uc_mcontext;
[...]

They were trapping SIGSEGV, but any signal will do. I assume you are trapping SIGFPE. I hope gregs is addressed also because that is valuable on its own merits, as is fpregs. I haven't tested it personally, but I am pretty sure libsigsegv is broken on WSL.

[microsoft-github-policy-service]

This issue has been automatically closed since it has not had any activity for the past year. If you're still experiencing this issue please re-file this as a new issue or feature request.

Thank you!