Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
From:   Jim Cromie <jim.cromie@gmail.com>
To:     linux-mm@kvack.org
Cc:     gregkh@linuxfoundation.org, linux@rasmusvillemoes.dk,
        Jim Cromie <jim.cromie@gmail.com>,
        Jason Baron <jbaron@akamai.com>, linux-kernel@vger.kernel.org
Subject: [PATCH 6/7] dyndbg: add locking around zpool-add loop in zpool-init
Date:   Wed, 25 Nov 2020 12:36:25 -0700
Message-Id: <20201125193626.2266995-7-jim.cromie@gmail.com>
In-Reply-To: <20201125193626.2266995-1-jim.cromie@gmail.com>
References: <20201125193626.2266995-1-jim.cromie@gmail.com>
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Precedence: bulk

This commit doesnt improve things, last commit was working, next one
still breaks, despite this "fix".  I keep it separate to isolate it
and the reasoning for review, which now stated, will be refuted as
needed. ;-)

Locking review:

ddebug_zpool_init(), like other ddebug_*_init() routines, does no
locking itself.  Unlike them, it runs later, at late_init.  This patch
doesnt fix the kernel panic that HEAD+1 does.

ddebug_callsite_get/put() are called as a pair under mutex-lock for
all 3 callsite users.

2 of them; ddebug_change() and dynamic_emit_prefix(), do their own
ABBA-ish LGPU (Lock-Get-Put-Unlock).

ddebug_proc_show() does the GP part, and is wrapped by
ddebug_proc_start|stop() with LU.

ddebug_add_module() does LU to protect list_add(), HEAD~1 added
ddebug_zpool_add() loop inside that protection.

This commit adds locking to ddebug_zpool_init(), around the loop of
ddebug_zpool_add(), to match the locking in ddebug_add_module().

Signed-off-by: Jim Cromie <jim.cromie@gmail.com>
---
 lib/dynamic_debug.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/lib/dynamic_debug.c b/lib/dynamic_debug.c
index 473406b069a7..534d73e45844 100644
--- a/lib/dynamic_debug.c
+++ b/lib/dynamic_debug.c
@@ -1195,8 +1195,10 @@ static void __init ddebug_zpool_init(void)
 	}
 
 	/* add-module normally does this, but not in time for builtins */
+	mutex_lock(&ddebug_lock);
 	for (iter = __start___dyndbg; iter < __stop___dyndbg; iter++)
 		ddebug_zpool_add(iter);
+	mutex_unlock(&ddebug_lock);
 
 	v2pr_info("total pages: %lu compaction: %lu\n",
 		  zs_get_total_pages(dd_callsite_zpool),
-- 
2.28.0