Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
From:   Vince Weaver <vincent.weaver@maine.edu>
Date:   Thu, 2 Jan 2020 13:55:47 -0500 (EST)
To:     linux-kernel@vger.kernel.org
cc:     Peter Zijlstra <peterz@infradead.org>,
        Ingo Molnar <mingo@redhat.com>,
        Arnaldo Carvalho de Melo <acme@kernel.org>,
        Mark Rutland <mark.rutland@arm.com>,
        Alexander Shishkin <alexander.shishkin@linux.intel.com>,
        Jiri Olsa <jolsa@redhat.com>,
        Namhyung Kim <namhyung@kernel.org>
Subject: [perf] perf_event_open() sometimes returning 0
Message-ID: <alpine.DEB.2.21.2001021349390.11372@macbook-air>
User-Agent: Alpine 2.21 (DEB 202 2017-01-01)
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

Hello

so I was tracking down some odd behavior in the perf_fuzzer which turns 
out to be because perf_even_open() sometimes returns 0 (indicating a file 
descriptor of 0) even though as far as I can tell stdin is still open.

Before I waste too much time trying to track this down, is this a known 
issue?

Some sample strace output:

perf_event_open({type=PERF_TYPE_RAW, size=0x78 /* PERF_ATTR_SIZE_??? */, config=0x1313, ...}, 0, 3, -1, PERF_FLAG_FD_NO_GROUP) = 0
perf_event_open({type=PERF_TYPE_SOFTWARE, size=0x78 /* PERF_ATTR_SIZE_??? */, config=PERF_COUNT_SW_DUMMY, ...}, -1, 3, -1, PERF_FLAG_FD_NO_GROUP|PERF_FLAG_FD_OUTPUT) = 0
perf_event_open({type=PERF_TYPE_SOFTWARE, size=0x78 /* PERF_ATTR_SIZE_??? */, config=PERF_COUNT_SW_PAGE_FAULTS_MIN, ...}, 158266, 2, -1, PERF_FLAG_FD_CLOEXEC) = 0
perf_event_open({type=PERF_TYPE_HW_CACHE, size=0x78 /* PERF_ATTR_SIZE_??? */, config=PERF_COUNT_HW_CACHE_DTLB|PERF_COUNT_HW_CACHE_OP_READ<<8|PERF_COUNT_HW_CACHE_RESULT_MISS<<16, ...}, 0, 4, -1, PERF_FLAG_FD_NO_GROUP|PERF_FLAG_FD_CLOEXEC) = 0
perf_event_open({type=PERF_TYPE_RAW, size=0x78 /* PERF_ATTR_SIZE_??? */, config=0, ...}, -1, 0, -1, PERF_FLAG_FD_OUTPUT|PERF_FLAG_FD_CLOEXEC) = 0

On my Haswell system (running current git) I can reproduce things with a 
single call:

	memset(&pe[0],0,sizeof(struct perf_event_attr));
	pe[0].type=PERF_TYPE_RAW;
	pe[0].size=120;
	pe[0].config=0x0ULL;
	pe[0].sample_period=0x4777c3ULL;
	pe[0].sample_type=PERF_SAMPLE_STREAM_ID; /* 200 */
	pe[0].read_format=PERF_FORMAT_TOTAL_TIME_RUNNING|PERF_FORMAT_ID|PERF_FORMAT_GROUP; /* e */
	pe[0].inherit=1;
	pe[0].exclude_hv=1;
	pe[0].exclude_idle=1;
	pe[0].enable_on_exec=1;
	pe[0].watermark=1;
	pe[0].precise_ip=0; /* arbitrary skid */
	pe[0].mmap_data=1;
	pe[0].exclude_guest=1;
	pe[0].exclude_callchain_kernel=1;
	pe[0].mmap2=1;
	pe[0].comm_exec=1;
	pe[0].context_switch=1;
	pe[0].bpf_event=1;
	pe[0].wakeup_watermark=47545;
	pe[0].bp_type=HW_BREAKPOINT_EMPTY;
	pe[0].branch_sample_type=PERF_SAMPLE_BRANCH_KERNEL|PERF_SAMPLE_BRANCH_ANY_RETURN|PERF_SAMPLE_BRANCH_COND|0x800ULL;
	pe[0].sample_regs_user=42ULL;
	pe[0].sample_stack_user=0xfffffffd;
	pe[0].aux_watermark=25443;
	pe[0].aux_sample_size=8192;

	fd[0]=perf_event_open(&pe[0],
				-1, /* current thread */
				0, /* Only cpu 0 */
				-1, /* New Group Leader */
				PERF_FLAG_FD_OUTPUT|PERF_FLAG_FD_CLOEXEC /*a*/ );