On Tue, 2010-11-23 at 17:30 -0500, Christoph Hellwig wrote:
> plain text document attachment (lio-kill-se_hba_cache)
> There is no need of a slab cache for the hba structures. There's very few of
> them and their allocation it in a slow path.
>
> Signed-off-by: Christoph Hellwig <hch@xxxxxx>
>
> Index: lio-core/drivers/target/target_core_configfs.c
> ===================================================================
> --- lio-core.orig/drivers/target/target_core_configfs.c 2010-11-23 22:12:05.405254018 +0100
> +++ lio-core/drivers/target/target_core_configfs.c 2010-11-23 22:33:25.000253320 +0100
> @@ -3163,7 +3163,7 @@ static struct config_group *target_core_
>
> return &hba->hba_group;
> out:
> - kmem_cache_free(se_hba_cache, hba);
> + kfree(hba);
> return ERR_PTR(ret);
> }
>
> Index: lio-core/drivers/target/target_core_device.c
> ===================================================================
> --- lio-core.orig/drivers/target/target_core_device.c 2010-11-23 22:11:59.006253249 +0100
> +++ lio-core/drivers/target/target_core_device.c 2010-11-23 22:33:25.014252691 +0100
> @@ -1619,7 +1619,7 @@ int core_dev_setup_virtual_lun0(void)
> ret = se_core_add_hba(hba, "rd_dr", 0);
> if (ret < 0) {
> printk("se_core_add_hba() with %d\n", ret);
> - kmem_cache_free(se_hba_cache, hba);
> + kfree(hba);
> return ret;
> }
>
> Index: lio-core/drivers/target/target_core_hba.c
> ===================================================================
> --- lio-core.orig/drivers/target/target_core_hba.c 2010-11-23 22:12:05.407253459 +0100
> +++ lio-core/drivers/target/target_core_hba.c 2010-11-23 22:33:25.026253459 +0100
> @@ -48,8 +48,8 @@ struct se_hba *core_alloc_hba(void)
> {
> struct se_hba *hba;
>
> - hba = kmem_cache_zalloc(se_hba_cache, GFP_KERNEL);
> - if (!(hba)) {
> + hba = kmalloc(sizeof(*hba), GFP_KERNEL);
> + if (!hba) {
> printk(KERN_ERR "Unable to allocate struct se_hba\n");
> return NULL;
> }
Hmmm, actually I assumed you where using kzalloc() here but unfortuately
that is not the case. kmalloc() w/o a trailing memset here creates the
following OOPs in configfs code for the non zero'ed hba->hba_group..
TCM: Registered subsystem plugin: iblock struct module: f09c0118
TCM: Registered subsystem plugin: fileio struct module: f09c71a0
TCM: Registered subsystem plugin: pscsi struct module: f09ced68
TCM: Registered subsystem plugin: stgt struct module: f0a8af40
CORE_HBA[840988271] - TCM FILEIO HBA Driver 4.0 on Generic Target Core Stack v4.0.0-rc5
CORE_HBA[840988271] - Attached FILEIO HBA: 0 to Generic Target Core with TCQ Depth: 256 MaxSectors: 1024
CORE_HBA[1] - Attached HBA to Generic Target Core
BUG: unable to handle kernel paging request at f0041c80
IP: [<c1092f95>] kfree+0x29/0xc3
*pde = 00000000
Oops: 0000 [#1] SMP
last sysfs file: /sys/devices/tcm_loop_0/tcm_loop_adapter_0/host7/target7:0:1/7:0:1:0/type
Modules linked in: target_core_stgt target_core_pscsi target_core_file target_core_iblock target_core_mod iscsi_initiator_mod cryptomgr aead crc32c crypto_hash crypto_algapi scsi_tgt configfs dm_mirror dm_region_hash dm_log dm_snapshot sg pcnet32 mii [last unloaded: target_core_mod]
Pid: 15935, comm: tcm_node Not tainted 2.6.37-rc2+ #52 440BX Desktop Reference Platform/VMware Virtual Platform
EIP: 0060:[<c1092f95>] EFLAGS: 00210282 CPU: 0
EIP is at kfree+0x29/0xc3
EAX: 4556455f EBX: f0041c80 ECX: ee477e70 EDX: cbe92690
ESI: f0cdfc4c EDI: cbe92694 EBP: ee477e48 ESP: ee477e30
DS: 007b ES: 007b FS: 00d8 GS: 0033 SS: 0068
Process tcm_node (pid: 15935, ti=ee476000 task=e1c5b480 task.ti=ee476000)
Stack:
00000000 4556455f cbe92690 cbe92690 f0cdfc4c cbe92694 ee477e60 f0dc59fb
cbe92694 cbe92690 f0cdfc4c ee477ec8 ee477e78 f0dc5a4e cbe92690 ec8b73c0
cbe92690 ee477e8f ee477ed8 f0cbb818 ee477e90 ec8b73c0 656c6966 30006f69
Call Trace:
[<f0dc59fb>] ? config_item_set_name+0x7d/0x8f [configfs]
[<f0dc5a4e>] ? config_group_init_type_name+0x10/0x23 [configfs]
[<f0cbb818>] ? target_core_call_addhbatotarget+0xfe/0x108 [target_core_mod]
[<f0dc4bf9>] ? configfs_mkdir+0x138/0x2d1 [configfs]
[<c109d68a>] ? vfs_mkdir+0x5c/0x88
[<f0dc4ac1>] ? configfs_mkdir+0x0/0x2d1 [configfs]
[<c109f402>] ? sys_mkdirat+0x7a/0xba
[<c1098de9>] ? sys_stat64+0x14/0x28
[<c10888ce>] ? do_mmap_pgoff+0x22f/0x27f
[<c1002903>] ? sysenter_exit+0xf/0x18
[<c109f452>] ? sys_mkdir+0x10/0x12
[<c10028d0>] ? sysenter_do_call+0x12/0x36
Code: 5d c3 55 89 e5 57 56 53 83 ec 0c 83 f8 10 89 45 ec 0f 86 a6 00 00 00 89 c3 81 c3 00 00 00 40 c1 eb 0c c1 e3 05 03 1d 80 44 af c1 <8b> 03 66 85 c0 79 03 8b 5b 0c 8b 03 84 c0 78 14 66 f7 03 00 c0
EIP: [<c1092f95>] kfree+0x29/0xc3 SS:ESP 0068:ee477e30
CR2: 00000000f0041c80
---[ end trace 9a735723aa0c3f01 ]---
Converted to a kzalloc() and rebased into new commit 4a17dbe7bcc.
--nab
--
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
