Re: [PATCH] block: switch to atomic_t for request references

From: Peter Zijlstra <peterz@infradead.org>
To: Christoph Hellwig <hch@infradead.org>
Cc: Jens Axboe <axboe@kernel.dk>,
	"linux-block@vger.kernel.org" <linux-block@vger.kernel.org>,
	keescook@chromium.org,
	Linus Torvalds <torvalds@linux-foundation.org>,
	linux-kernel@vger.kernel.org
Subject: Re: [PATCH] block: switch to atomic_t for request references
Date: Tue, 7 Dec 2021 16:51:48 +0100	[thread overview]
Message-ID: <Ya+DFKo6JVuu++cX@hirez.programming.kicks-ass.net> (raw)
In-Reply-To: <Ya9hdlBuWYUWRQzs@hirez.programming.kicks-ass.net>

On Tue, Dec 07, 2021 at 02:28:22PM +0100, Peter Zijlstra wrote:
> +static inline void refcount_inc(refcount_t *r)
> +{
> +	int one = 1;
> +
> +	asm_volatile_goto (LOCK_PREFIX "xaddl %%eax, %[var]\n\t"
> +			   "addl $1, %%eax\n\t"
> +			   "je %l[cc_zero]\n\t"
> +			   "js %l[cc_error]"
> +			   : : [var] "m" (r->refs.counter), "a" (one)
> +			   : "memory"
> +			   : cc_zero, cc_error);

+       asm_volatile_goto (LOCK_PREFIX "xaddl %[reg], %[var]\n\t"
+                          "addl $1, %[reg]\n\t"
+                          "jz %l[cc_zero]\n\t"
+                          "js %l[cc_error]"
+                          : : [var] "m" (r->refs.counter), [reg] "r" (1)
+                          : "memory"
+                          : cc_zero, cc_error);

Is of course a better implementation, but I'm not sure I actually
understand this code. Afaict: add $1,%[reg], will only set ZF when
%[reg] was -1 such that the result is now 0.

But that's not what the code said; is this GCC going funny in the head
or should I just stop staring at this...