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:   Rasmus Villemoes <linux@rasmusvillemoes.dk>
To:     Qiang Zhao <qiang.zhao@nxp.com>, Li Yang <leoyang.li@nxp.com>,
        Christophe Leroy <christophe.leroy@c-s.fr>
Cc:     linuxppc-dev@lists.ozlabs.org,
        linux-arm-kernel@lists.infradead.org, linux-kernel@vger.kernel.org,
        Scott Wood <oss@buserror.net>,
        Rasmus Villemoes <linux@rasmusvillemoes.dk>
Subject: [PATCH v4 38/47] soc: fsl: qe: refactor cpm_muram_alloc_common to prevent BUG on error path
Date:   Fri,  8 Nov 2019 14:01:14 +0100
Message-Id: <20191108130123.6839-39-linux@rasmusvillemoes.dk>
In-Reply-To: <20191108130123.6839-1-linux@rasmusvillemoes.dk>
References: <20191108130123.6839-1-linux@rasmusvillemoes.dk>
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

If the kmalloc() fails, we try to undo the gen_pool allocation we've
just done. Unfortunately, start has already been modified to subtract
the GENPOOL_OFFSET bias, so we're freeing something that very likely
doesn't exist in the gen_pool, meaning we hit the

 kernel BUG at lib/genalloc.c:399!
 Internal error: Oops - BUG: 0 [#1] PREEMPT SMP ARM
 ...
 [<803fd0e8>] (gen_pool_free) from [<80426bc8>] (cpm_muram_alloc_common+0xb0/0xc8)
 [<80426bc8>] (cpm_muram_alloc_common) from [<80426c28>] (cpm_muram_alloc+0x48/0x80)
 [<80426c28>] (cpm_muram_alloc) from [<80428214>] (ucc_slow_init+0x110/0x4f0)
 [<80428214>] (ucc_slow_init) from [<8044a718>] (qe_uart_request_port+0x3c/0x1d8)

(this was tested by just injecting a random failure by adding
"|| (get_random_int()&7) == 0" to the "if (!entry)" condition).

Refactor the code so we do the kmalloc() first, meaning that's the
thing that needs undoing in case gen_pool_alloc_algo() then
fails. This allows a later cleanup to move the locking from the
callers into the _common function, keeping the kmalloc() out of the
critical region and then, hopefully (if all the muram_alloc callers
allow) change it to a GFP_KERNEL allocation.

Signed-off-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
---
 drivers/soc/fsl/qe/qe_common.c | 16 +++++++---------
 1 file changed, 7 insertions(+), 9 deletions(-)

diff --git a/drivers/soc/fsl/qe/qe_common.c b/drivers/soc/fsl/qe/qe_common.c
index feb33ec1c3d6..bc9b436684df 100644
--- a/drivers/soc/fsl/qe/qe_common.c
+++ b/drivers/soc/fsl/qe/qe_common.c
@@ -119,23 +119,21 @@ static s32 cpm_muram_alloc_common(unsigned long size,
 	struct muram_block *entry;
 	s32 start;
 
+	entry = kmalloc(sizeof(*entry), GFP_ATOMIC);
+	if (!entry)
+		return -ENOMEM;
 	start = gen_pool_alloc_algo(muram_pool, size, algo, data);
-	if (!start)
-		goto out2;
+	if (!start) {
+		kfree(entry);
+		return -ENOMEM;
+	}
 	start = start - GENPOOL_OFFSET;
 	memset_io(cpm_muram_addr(start), 0, size);
-	entry = kmalloc(sizeof(*entry), GFP_ATOMIC);
-	if (!entry)
-		goto out1;
 	entry->start = start;
 	entry->size = size;
 	list_add(&entry->head, &muram_block_list);
 
 	return start;
-out1:
-	gen_pool_free(muram_pool, start, size);
-out2:
-	return -ENOMEM;
 }
 
 /*
-- 
2.23.0