RFR: 8225035: Thread stack size issue caused by large TLS size

[Andrew Haley]

On 6/12/19 2:42 PM, Thomas Stüfe wrote:
> What is "static tls", __thread variables? As opposed to what, dynamic tls?
> Would that be Posix TLS (pthread_setspecific and friends)? Are the
> implementations different? If they are, which TLS does this patch address?

Yes, this is __thread variables. It's nothing to do with
pthread_key_create.

The crucial distinction here is position-independent code. If an
executable which uses __thread variables is compiled non-PIC a special
(and fast) code sequence is generated. This means that the __thread
variables are allocated at a simple offset from a base
pointer. Dynamic TLS is more complicated, but much more flexible.

So, there is "dynamic TLS" and "static TLS". Dynamic TLS is used by
PIC-compiled libraries.

For this code:

__thread int boof;
int *barf() {
  return &boof;
}

static TLS:

barf:
.LFB0:
	.cfi_startproc
	movq	%fs:0, %rax
	addq	$boof at tpoff, %rax
	ret

dynamic TLS:

barf:
	subq	$8, %rsp
	.byte	0x66
	leaq	boof at tlsgd(%rip), %rdi
	.value	0x6666
	rex64
	call	__tls_get_addr at PLT
	addq	$8, %rsp
	ret

All of this stuff gets patched by the runtime linker.

-- 
Andrew Haley
Java Platform Lead Engineer
Red Hat UK Ltd. <https://www.redhat.com>
https://keybase.io/andrewhaley
EAC8 43EB D3EF DB98 CC77 2FAD A5CD 6035 332F A671