linux-kernel - Re: [RFC v2] perf: Rewrite core context handling

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Mon, 13 Jun 2022 16:36:54 +0200
From:   Peter Zijlstra <peterz@...radead.org>
To:     Ravi Bangoria <ravi.bangoria@....com>
Cc:     acme@...nel.org, alexander.shishkin@...ux.intel.com,
        jolsa@...hat.com, namhyung@...nel.org, songliubraving@...com,
        eranian@...gle.com, alexey.budankov@...ux.intel.com,
        ak@...ux.intel.com, mark.rutland@....com, megha.dey@...el.com,
        frederic@...nel.org, maddy@...ux.ibm.com, irogers@...gle.com,
        kim.phillips@....com, linux-kernel@...r.kernel.org,
        santosh.shukla@....com
Subject: Re: [RFC v2] perf: Rewrite core context handling

On Mon, Jun 13, 2022 at 04:35:11PM +0200, Peter Zijlstra wrote:
> 
> 
> Right, so sorry for being incredibly tardy on this. Find below the
> patch fwd ported to something recent.
> 
> I'll reply to this with fixes and comments.

You write:

>> A simple perf stat/record/top survives with the patch but machine
>> crashes with first run of perf test (stale cpc->task_epc causing the
>>crash). Lockdep is also screaming a lot :)


> @@ -7669,20 +7877,15 @@ static void perf_event_addr_filters_exec
>  void perf_event_exec(void)
>  {
>  	struct perf_event_context *ctx;
> -	int ctxn;
> -
> -	for_each_task_context_nr(ctxn) {
> -		perf_event_enable_on_exec(ctxn);
> -		perf_event_remove_on_exec(ctxn);
>  
> -		rcu_read_lock();
> -		ctx = rcu_dereference(current->perf_event_ctxp[ctxn]);
> -		if (ctx) {
> -			perf_iterate_ctx(ctx, perf_event_addr_filters_exec,
> -					 NULL, true);
> -		}
> -		rcu_read_unlock();
> +	rcu_read_lock();
> +	ctx = rcu_dereference(current->perf_event_ctxp);
> +	if (ctx) {
> +		perf_event_enable_on_exec(ctx);
> +		perf_event_remove_on_exec(ctx);
> +		perf_iterate_ctx(ctx, perf_event_addr_filters_exec, NULL, true);
>  	}
> +	rcu_read_unlock();
>  }
>  
>  struct remote_output {

The above goes *bang* because perf_event_remove_on_exec() will take a
mutex, which isn't allowed under rcu_read_lock().

The below cures.

---
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -4384,8 +4384,6 @@ static void perf_event_remove_on_exec(st
 	unsigned long flags;
 	bool modified = false;
 
-	perf_pin_task_context(current);
-
 	mutex_lock(&ctx->mutex);
 
 	if (WARN_ON_ONCE(ctx->task != current))
@@ -4406,13 +4404,11 @@ static void perf_event_remove_on_exec(st
 	raw_spin_lock_irqsave(&ctx->lock, flags);
 	if (modified)
 		clone_ctx = unclone_ctx(ctx);
-	--ctx->pin_count;
 	raw_spin_unlock_irqrestore(&ctx->lock, flags);
 
 unlock:
 	mutex_unlock(&ctx->mutex);
 
-	put_ctx(ctx);
 	if (clone_ctx)
 		put_ctx(clone_ctx);
 }
@@ -7878,14 +7874,16 @@ void perf_event_exec(void)
 {
 	struct perf_event_context *ctx;
 
-	rcu_read_lock();
-	ctx = rcu_dereference(current->perf_event_ctxp);
-	if (ctx) {
-		perf_event_enable_on_exec(ctx);
-		perf_event_remove_on_exec(ctx);
-		perf_iterate_ctx(ctx, perf_event_addr_filters_exec, NULL, true);
-	}
-	rcu_read_unlock();
+	ctx = perf_pin_task_context(current);
+	if (!ctx)
+		return;
+
+	perf_event_enable_on_exec(ctx);
+	perf_event_remove_on_exec(ctx);
+	perf_iterate_ctx(ctx, perf_event_addr_filters_exec, NULL, true);
+
+	perf_unpin_context(ctx);
+	put_ctx(ctx);
 }
 
 struct remote_output {