Re: [PATCH] megaraid: kmemleak: Track page allocation for fusion

[Catalin Marinas]

On Thu, Sep 14, 2017 at 02:16:46PM +0800, shuwang@xxxxxxxxxx wrote:
> From: Shu Wang <shuwang@xxxxxxxxxx>
> 
> Kmemleak reports about a thousand false positives for fusion->
> cmd_list[]. Root casue is the cmd_list objects are allocated from
> slab allocator, and stored its pointer in object allocated by page
> allocator. The fix will tell kmemleak to track and scan fusion
> object.
> 
> Before patch:
> kmemleak: 1004 new suspected memory leaks (see /sys/kernel/debug/kmemleak)
> 
> unreferenced object 0xffff88042584e000 (size 8192):
>   backtrace:
>      kmemleak_alloc+0x4a/0xa0
>      __kmalloc+0xec/0x220
>      megasas_alloc_cmdlist_fusion+0x3e/0x140 [megaraid_sas]
>      megasas_alloc_cmds_fusion+0x44/0x450 [megaraid_sas]
>      megasas_init_adapter_fusion+0x21d/0x6e0 [megaraid_sas]
>      megasas_init_fw+0x357/0xd30 [megaraid_sas]
>      megasas_probe_one.part.34+0x5be/0x1040 [megaraid_sas]
>      megasas_probe_one+0x46/0xc0 [megaraid_sas]
>      local_pci_probe+0x45/0xa0
>      work_for_cpu_fn+0x14/0x20
>      process_one_work+0x149/0x360
>      worker_thread+0x1d8/0x3c0
>      kthread+0x109/0x140
>      ret_from_fork+0x25/0x30
> 
> Signed-off-by: Shu Wang <shuwang@xxxxxxxxxx>
> ---
>  drivers/scsi/megaraid/megaraid_sas_fusion.c | 9 +++++++--
>  1 file changed, 7 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/scsi/megaraid/megaraid_sas_fusion.c b/drivers/scsi/megaraid/megaraid_sas_fusion.c
> index 11bd2e698b84..621299edd8de 100644
> --- a/drivers/scsi/megaraid/megaraid_sas_fusion.c
> +++ b/drivers/scsi/megaraid/megaraid_sas_fusion.c
> @@ -48,6 +48,7 @@
>  #include <linux/mutex.h>
>  #include <linux/poll.h>
>  #include <linux/vmalloc.h>
> +#include <linux/kmemleak.h>
>  
>  #include <scsi/scsi.h>
>  #include <scsi/scsi_cmnd.h>
> @@ -4512,7 +4513,9 @@ megasas_alloc_fusion_context(struct megasas_instance *instance)
>  			dev_err(&instance->pdev->dev, "Failed from %s %d\n", __func__, __LINE__);
>  			return -ENOMEM;
>  		}
> -	}
> +	} else
> +		kmemleak_alloc(instance->ctrl_context,
> +			sizeof(struct fusion_context), 1, GFP_KERNEL);
>  
>  	fusion = instance->ctrl_context;
>  
> @@ -4548,9 +4551,11 @@ megasas_free_fusion_context(struct megasas_instance *instance)
>  
>  		if (is_vmalloc_addr(fusion))
>  			vfree(fusion);
> -		else
> +		else {
>  			free_pages((ulong)fusion,
>  				instance->ctrl_context_pages);
> +			kmemleak_free(fusion);
> +		}

Apart from Bart's comments on braces and comment before
kmemleak_alloc(), I'd call kmemleak_free() before free_pages(),
otherwise it may not interact nicely with other tools checking for use
after free. With that:

Reviewed-by: Catalin Marinas <catalin.marinas@xxxxxxx>

-- 
Catalin