Are the cast necessary in function "kfree"?

Are the cast necessary in function "kfree"?

[André Lopes]

Hi. This is my first post :-)

I was reading this doc (/linux-2.6/Documentation/scsi/ChangeLog.lpfc),
when I found this phrase: "Replace some kfree((void*)ptr) with
kfree(ptr)."

After that I created a script to detect this situation, in Linux
source code. The results were:

arch/mips/alchemy/common/dbdma.c:408:        kfree((const void *)desc_base);
arch/mips/alchemy/common/dbdma.c:834:    kfree((void *)ctp->chan_desc_base);
arch/s390/include/asm/idals.h:110:        kfree((void *)(unsigned
long) ccw->cda);
arch/powerpc/mm/pgtable_32.c:90:    kfree((void *)pgd);
drivers/ide/ide-acpi.c:369:    kfree((void *)obj_loc);
drivers/net/vxge/vxge-config.c:4962:        kfree((void *)p);
drivers/net/vxge/vxge-config.h:1970:    kfree((void *)tmp);
drivers/net/wan/pc300_tty.c:699:                kfree((void *)buf);
drivers/net/wireless/zd1211rw/zd_chip.c:149:    kfree((void *)a16);
drivers/net/lance.c:367:    kfree((void*)lp->rx_buffs);
drivers/net/lance.c:742:    kfree((void*)lp->rx_buffs);
drivers/net/ucc_geth.c:1890:                kfree((void
*)ugeth->tx_bd_ring_offset[i]);
drivers/net/ucc_geth.c:1920:                kfree((void
*)ugeth->rx_bd_ring_offset[i]);
drivers/s390/char/vmur.c:188:        kfree((void *)(addr_t) ptr->cda);
drivers/block/paride/bpck6.c:227:    kfree((void *)(pi->private));
drivers/infiniband/hw/cxgb3/iwch_provider.c:449:        kfree((void *)
(unsigned long) mhp->kva);
drivers/staging/epl/EplTarget.h:99:#define EPL_FREE(ptr)
kfree((void *)ptr)
drivers/video/console/fbcon.c:960:                kfree((void *) softback_buf);
drivers/video/console/fbcon.c:3428:    kfree((void *)softback_buf);
drivers/video/au1100fb.c:614:    kfree((void*)fbdev);
drivers/pci/rom.c:259:        kfree((void*)(unsigned long)res->start);
drivers/isdn/hisax/config.c:1272:        kfree((void *) cards[cardnr].cs);
drivers/isdn/hisax/fsm.c:47:    kfree((void *) fsm->jumpmatrix);
drivers/isdn/mISDN/fsm.c:54:    kfree((void *) fsm->jumpmatrix);
net/sched/ematch.c:401:                kfree((void *) em->data);
net/sched/em_meta.c:577:    kfree((void *) v->val);
sound/oss/sh_dac_audio.c:325:    kfree((void *)data_buffer);
sound/pci/ice1712/ak4xxx.c:158:        kfree((void*)ak->private_value[0]);
sound/pci/emu10k1/emufx.c:1724:        kfree((void __force *)icode->gpr_map);
sound/pci/emu10k1/emufx.c:2345:        kfree((void __force *)icode->gpr_map);

The prototype for kfree() is:
#include<linux/slab.h>
void kfree(const void *objp)

These casts are not necessary...I think so. If it's yes, can we apply
patch's to clean-up this?

Thanks.

Re: Are the cast necessary in function "kfree"?

[Christian Bornträger]

Am Mittwoch 20 Mai 2009 12:30:21 schrieb André Lopes:

I only looked at the s390 parts:

> arch/s390/include/asm/idals.h:110: kfree((void *)(unsigned long) ccw->cda);
This cast is necessary since cda is an u32 even on 64bit machines.

> drivers/s390/char/vmur.c:188:        kfree((void *)(addr_t) ptr->cda);
dito.

Christian

Re: Are the cast necessary in function "kfree"?

[Paul Mundt]

On Wed, May 20, 2009 at 11:30:21AM +0100, Andr? Lopes wrote:
> Hi. This is my first post :-)
> 
> I was reading this doc (/linux-2.6/Documentation/scsi/ChangeLog.lpfc),
> when I found this phrase: "Replace some kfree((void*)ptr) with
> kfree(ptr)."
> 
[snip]

> These casts are not necessary...I think so. If it's yes, can we apply
> patch's to clean-up this?
> 
The key thing here is "ptr", as long as these are pointers, then the cast
is superfluous. In the case of casting from a non-pointer type, the cast
is necessary, particularly if the data type and pointer size vary.