On Mon, Dec 23, 2013 at 09:47:09AM +0800, Gu Zheng wrote: > Hi Wolfram, > On 12/21/2013 01:31 AM, Wolfram Sang wrote: > > > On Wed, Dec 18, 2013 at 09:18:08AM +0800, Gu Zheng wrote: > >> When running with the latest kernel, we get the following kmemleak message: > >> unreferenced object 0xffff8800c2a36100 (size 256): > >> comm "modprobe", pid 629, jiffies 4294676002 (age 1531.115s) > >> hex dump (first 32 bytes): > >> 00 00 00 00 ad 4e ad de ff ff ff ff 00 00 00 00 .....N.......... > >> ff ff ff ff ff ff ff ff 78 21 85 82 ff ff ff ff ........x!...... > >> backtrace: > >> [<ffffffff815449ce>] kmemleak_alloc+0x73/0x98 > >> [<ffffffff8113fa2b>] slab_post_alloc_hook+0x28/0x2a > >> [<ffffffff81142004>] __kmalloc+0x12c/0x158 > >> [<ffffffff8134c60b>] kzalloc.constprop.6+0xe/0x10 > >> [<ffffffff8134cb71>] device_private_init+0x18/0x66 > >> [<ffffffff8134f1e8>] dev_set_drvdata+0x1e/0x34 > >> [<ffffffffa01e643e>] i801_probe+0x5d/0x447 [i2c_i801] > >> [<ffffffff812a78be>] local_pci_probe+0x41/0x84 > >> [<ffffffff812a7feb>] pci_device_probe+0xdf/0x10c > >> [<ffffffff8134f90d>] driver_probe_device+0x12f/0x2ee > >> [<ffffffff8134fb2b>] __driver_attach+0x5f/0x83 > >> [<ffffffff8134de28>] bus_for_each_dev+0x64/0x90 > >> [<ffffffff8134f36b>] driver_attach+0x1e/0x20 > >> [<ffffffff8134ef34>] bus_add_driver+0xf2/0x1f0 > >> [<ffffffff8135013e>] driver_register+0x8c/0xc3 > >> [<ffffffff812a7bb4>] __pci_register_driver+0x62/0x67 > >> According to the backtrace, it seems that we do not release the > >> device_private field of adapter device in the fail path of i801_probe > >> or the flow of i2c_del_adapter, so add the cleanup step to fix it. > >> > >> Signed-off-by: Gu Zheng <guz.fnst@xxxxxxxxxxxxxx> > > > > ? Why should we free something we don't have allocated? The fix must be > > somewhere else probably... > > We do allocate the device_private in device_private_init() when calling > dev_set_drvdata() dev_set_drvdata only does the following and doesn't call anything: dev->driver_data = data; > to set adapter data in probe flow, and the calltrace > also shows this. Am I missing something? It should be freed by the driver core in device_release() again. Dunno why it does not happen and you are seeing the leak, but the fix is not correct.
Attachment:
signature.asc
Description: Digital signature
