On Tue, Aug 13, 2019 at 07:01:13PM +0100, Colin King wrote:
> From: Colin Ian King <colin.king@xxxxxxxxxxxxx>
>
> Currently the top 32 bits of a 64 bit address is being calculated
> by shifting a u32 twice by 16 bits and then being cast into a 64
> bit address. Shifting a u32 twice by 16 bits always ends up with
> a zero. Fix this by casting the u32 to a 64 bit address first
> and then shifting it 32 bits.
>
> Addresses-Coverity: ("Operands don't affect result")
> Fixes: f0c568a478f0 ("[SCSI] mvumi: Add Marvell UMI driver")
> Signed-off-by: Colin Ian King <colin.king@xxxxxxxxxxxxx>
> ---
> drivers/scsi/mvumi.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/drivers/scsi/mvumi.c b/drivers/scsi/mvumi.c
> index 8906aceda4c4..62df69f1e71e 100644
> --- a/drivers/scsi/mvumi.c
> +++ b/drivers/scsi/mvumi.c
> @@ -296,7 +296,7 @@ static void mvumi_delete_internal_cmd(struct mvumi_hba *mhba,
> sgd_getsz(mhba, m_sg, size);
>
> phy_addr = (dma_addr_t) m_sg->baseaddr_l |
> - (dma_addr_t) ((m_sg->baseaddr_h << 16) << 16);
> + (dma_addr_t) m_sg->baseaddr_h << 32;
Colin, you've sent this patch before on Feb 16, 2019. If you shift by
32 then it's undefined behavior on 32 bit systems. The correct fix is
to move the cast which was what your original patch did actually.
(((dma_addr_t)m_sg->baseaddr_h << 16) << 16);
My suggestion back then was to introduce a macro:
/*
* The dma_addr_t type can be either 32 or 64 bit. Left shifting a 32
* bit number is undefined so this do two 16 bit left shifts.
*
*/
#define DMA_LSHIFT_32(val) (((dma_addr_t)(val) << 16) << 16)
regards,
dan carpenter
