Skip to contentSkip to conversation

Possible memory leak in ata_piix.c

PATCH In review Catalin Marinas 3 participants 6 messages Dec 9, 2006 – Dec 15, 2006 libata

lore
Share & tools
lore
#0 lore raw

Hi,

Kmemleak found a possible memory leak in piix_init_one() in drivers/ata/ata_piix.c. This only appeared after 2.6.19, maybe caused by the recent patches to this area. Kmemleak cannot find any track of the kzalloc'ed piix_host_priv structure allocated in the above function and reports it. The allocation stack trace is below:

unreferenced object 0xde9bca60 (size 4):
[<c018d85d>] __kmalloc_track_caller
[<c0179249>] __kzalloc
[<c02cf33f>] piix_init_one
[<c023dc2d>] pci_call_probe
[<c023dc81>] __pci_device_probe
[<c023dcb9>] pci_device_probe
[<c029b5fc>] really_probe
[<c029b728>] driver_probe_device
[<c029b8c1>] __driver_attach
[<c029a879>] bus_for_each_dev
[<c029b8e9>] driver_attach
[<c029ae9c>] bus_add_driver
[<c029bd27>] driver_register
[<c023e035>] __pci_register_driver
[<c02cf4ef>] piix_init

Thanks.

Signature · 1 line
Catalin
#1 lore raw

PATCH 1/1 ata_piix: use piix_host_stop() in ich_pata_ops

piix_init_one() allocates host private data which should be freed by piix_host_stop(). ich_pata_ops wasn't converted to piix_host_stop() while merging, leaking 4 bytes on driver detach. Fix it.

This was spotted using Kmemleak by Catalin Marinas.

Cc Catalin Marinas <[email protected]>

---

drivers/ata/ata_piix.c +1 -1
index c7de0bb..dfe17e1 100644
@@ -330,7 +330,7 @@ static const struct ata_port_operations ich_pata_ops = {
 
 	.port_start		= ata_port_start,
 	.port_stop		= ata_port_stop,
-	.host_stop		= ata_host_stop,
+	.host_stop		= piix_host_stop,
 };
 
 static const struct ata_port_operations piix_sata_ops = {
Acked-by Alan Cox on v1 1/1
#2 lore raw

Re: [PATCH] ata_piix: use piix_host_stop() in ich_pata_ops

Quoted 8 lines · On Mon, 11 Dec 2006 22:26:25 +0900 Tejun Heo <[email protected]> wrote:

piix_init_one() allocates host private data which should be freed by piix_host_stop(). ich_pata_ops wasn't converted to piix_host_stop() while merging, leaking 4 bytes on driver detach. Fix it.

This was spotted using Kmemleak by Catalin Marinas.

Cc Catalin Marinas <[email protected]>
#3 lore raw

Re: [PATCH] ata_piix: use piix_host_stop() in ich_pata_ops

In reply to Tejun Heo

Tejun,

Quoted 3 lines · Tejun Heo

piix_init_one() allocates host private data which should be freed by piix_host_stop(). ich_pata_ops wasn't converted to piix_host_stop() while merging, leaking 4 bytes on driver detach. Fix it.

I tried your patch last night but the leak is still reported. I need to investigate further and put some printk's in the piix_host_stop function to check whether the freeing really takes place.

What I can't follow is where the ata_port_info.private_data (port_info[] or ppinfo[]) in piix_init_one gets transfered to ata_host.private_data (the "host" argument) that piix_host_stop tries to free.

Signature · 1 line
Catalin
#4 lore raw

Re: [PATCH] ata_piix: use piix_host_stop() in ich_pata_ops

Quoted 8 lines · Catalin Marinas
Quoted 3 lines · Tejun Heo

piix_init_one() allocates host private data which should be freed by piix_host_stop(). ich_pata_ops wasn't converted to piix_host_stop() while merging, leaking 4 bytes on driver detach. Fix it.

I tried your patch last night but the leak is still reported. I need to investigate further and put some printk's in the piix_host_stop function to check whether the freeing really takes place.

The piix_host_stop() isn't actually called on my machine, so this is not the cause of the leak.

What causes the leak seem to be the error returned by ata_pci_init_one() called from piix_init_one(). These are the related kernel messages:

ata_piix 0000:00:1f.1: version 2.00ac7
ACPI: PCI Interrupt 0000:00:1f.1[A] -> GSI 16 (level, low) -> IRQ 18
PCI: Unable to reserve I/O region #1:8@1f0 for device 0000:00:1f.1
ata_piix: probe of 0000:00:1f.1 failed with error -16

I think the call to ata_pci_init_one() should be followed by some clean-up in case it fails. There is also another return without clean-up in piix_init_one() after the call to piix_disable_ahci(). I don't have time to try this now.

Signature · 1 line
Catalin
#5 lore raw

PATCH 1/1 Fix a memory leak in piix_init_one

The piix_init_one() function in drivers/ata/ata_piix.c can return an error without cleaning up the piix_host_priv structure allocation. This patch adds a kfree() call on the error path.

---

 drivers/ata/ata_piix.c |   17 +++++++++++++----
 1 files changed, 13 insertions(+), 4 deletions(-)
drivers/ata/ata_piix.c +13 -4
index dfe17e1..6717891 100644
@@ -1051,6 +1051,7 @@ static int piix_init_one (struct pci_dev
 	struct ata_port_info *ppinfo[2] = { &port_info[0], &port_info[1] };
 	struct piix_host_priv *hpriv;
 	unsigned long port_flags;
+	int ret;
 
 	if (!printed_version++)
 		dev_printk(KERN_DEBUG, &pdev->dev,
@@ -1075,9 +1076,9 @@ static int piix_init_one (struct pci_dev
 		u8 tmp;
 		pci_read_config_byte(pdev, PIIX_SCC, &tmp);
 		if (tmp == PIIX_AHCI_DEVICE) {
-			int rc = piix_disable_ahci(pdev);
-			if (rc)
-				return rc;
+			ret = piix_disable_ahci(pdev);
+			if (ret)
+				goto cleanup;
 		}
 	}
 
@@ -1107,7 +1108,15 @@ static int piix_init_one (struct pci_dev
 		port_info[1].mwdma_mask = 0;
 		port_info[1].udma_mask = 0;
 	}
-	return ata_pci_init_one(pdev, ppinfo, 2);
+	ret = ata_pci_init_one(pdev, ppinfo, 2);
+	if (ret)
+		goto cleanup;
+	goto out;
+
+ cleanup:
+	kfree(hpriv);
+ out:
+	return ret;
 }
 
 static void piix_host_stop(struct ata_host *host)