[PATCH] usb: gadget: s3c-hsotg: Mark s3c_hsotg_map_dma()/s3c_hsotg_unmap_dma() inline

[PATCH] usb: gadget: s3c-hsotg: Mark s3c_hsotg_map_dma()/s3c_hsotg_unmap_dma() inline

[Geert Uytterhoeven]

If NO_DMA=y, depending on the gcc version:

    drivers/built-in.o: In function `s3c_hsotg_map_dma':
    s3c-hsotg.c:(.text+0x375b2c): undefined reference to `usb_gadget_map_request'
    drivers/built-in.o: In function `s3c_hsotg_unmap_dma':
    s3c-hsotg.c:(.text+0x376a32): undefined reference to `usb_gadget_unmap_request'
    make[3]: *** [vmlinux] Error 1

The obvious fix of making USB_S3C_HSOTG depend on HAS_DMA is a bit overkill,
as DMA is never actually used. All (two) calls to s3c_hsotg_map_dma() and
s3c_hsotg_unmap_dma() are protected by a call into the using_dma() checking
function, which always returns false:

    static inline bool using_dma(struct s3c_hsotg *hsotg)
    {
            return false;   /* support is not complete */
    }

Due to the constant return value, gcc optimizes away the calls to
s3c_hsotg_map_dma() and s3c_hsotg_unmap_dma(). But some versions of gcc
(e.g. 4.1.2) still emit the functions themselves, causing link errors.

Marking s3c_hsotg_map_dma()/s3c_hsotg_unmap_dma() inline fixes this.

Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
---
This replaces "[PATCH/RFC] usb: gadget: CONFIG_USB_S3C_HSOTG should depend on
HAS_DMA"

 drivers/usb/gadget/s3c-hsotg.c |   12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

diff --git a/drivers/usb/gadget/s3c-hsotg.c b/drivers/usb/gadget/s3c-hsotg.c
index 2a9cb674926a..2740042228a4 100644
--- a/drivers/usb/gadget/s3c-hsotg.c
+++ b/drivers/usb/gadget/s3c-hsotg.c
@@ -437,9 +437,9 @@ static inline int is_ep_periodic(struct s3c_hsotg_ep *hs_ep)
  * This is the reverse of s3c_hsotg_map_dma(), called for the completion
  * of a request to ensure the buffer is ready for access by the caller.
  */
-static void s3c_hsotg_unmap_dma(struct s3c_hsotg *hsotg,
-				struct s3c_hsotg_ep *hs_ep,
-				struct s3c_hsotg_req *hs_req)
+static inline void s3c_hsotg_unmap_dma(struct s3c_hsotg *hsotg,
+				       struct s3c_hsotg_ep *hs_ep,
+				       struct s3c_hsotg_req *hs_req)
 {
 	struct usb_request *req = &hs_req->req;
 
@@ -861,9 +861,9 @@ static void s3c_hsotg_start_req(struct s3c_hsotg *hsotg,
  * DMA memory, then we map the memory and mark our request to allow us to
  * cleanup on completion.
  */
-static int s3c_hsotg_map_dma(struct s3c_hsotg *hsotg,
-			     struct s3c_hsotg_ep *hs_ep,
-			     struct usb_request *req)
+static inline int s3c_hsotg_map_dma(struct s3c_hsotg *hsotg,
+				    struct s3c_hsotg_ep *hs_ep,
+				    struct usb_request *req)
 {
 	struct s3c_hsotg_req *hs_req = our_req(req);
 	int ret;
-- 
1.7.9.5

Re: [PATCH] usb: gadget: s3c-hsotg: Mark s3c_hsotg_map_dma()/s3c_hsotg_unmap_dma() inline

[Bartlomiej Zolnierkiewicz]

Hi,

On Tuesday, May 06, 2014 07:37:01 PM Geert Uytterhoeven wrote:
> If NO_DMA=y, depending on the gcc version:
> 
>     drivers/built-in.o: In function `s3c_hsotg_map_dma':
>     s3c-hsotg.c:(.text+0x375b2c): undefined reference to `usb_gadget_map_request'
>     drivers/built-in.o: In function `s3c_hsotg_unmap_dma':
>     s3c-hsotg.c:(.text+0x376a32): undefined reference to `usb_gadget_unmap_request'
>     make[3]: *** [vmlinux] Error 1
> 
> The obvious fix of making USB_S3C_HSOTG depend on HAS_DMA is a bit overkill,
> as DMA is never actually used. All (two) calls to s3c_hsotg_map_dma() and
> s3c_hsotg_unmap_dma() are protected by a call into the using_dma() checking
> function, which always returns false:
> 
>     static inline bool using_dma(struct s3c_hsotg *hsotg)
>     {
>             return false;   /* support is not complete */
>     }
> 
> Due to the constant return value, gcc optimizes away the calls to
> s3c_hsotg_map_dma() and s3c_hsotg_unmap_dma(). But some versions of gcc
> (e.g. 4.1.2) still emit the functions themselves, causing link errors.
> 
> Marking s3c_hsotg_map_dma()/s3c_hsotg_unmap_dma() inline fixes this.

How's about just removing the dead code instead?  It has never worked since
the driver introduction back in June 2009 (almost 5 years) and it is trivial
to bring it back from kernel git repository if needed in the future.

PS Could you please fix -next kernels as well (drivers/usb/gadget/s3c-hsotg.c
has moved to drivers/usb/dwc2/gadget.c recently)?

> Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
> ---
> This replaces "[PATCH/RFC] usb: gadget: CONFIG_USB_S3C_HSOTG should depend on
> HAS_DMA"
> 
>  drivers/usb/gadget/s3c-hsotg.c |   12 ++++++------
>  1 file changed, 6 insertions(+), 6 deletions(-)
> 
> diff --git a/drivers/usb/gadget/s3c-hsotg.c b/drivers/usb/gadget/s3c-hsotg.c
> index 2a9cb674926a..2740042228a4 100644
> --- a/drivers/usb/gadget/s3c-hsotg.c
> +++ b/drivers/usb/gadget/s3c-hsotg.c
> @@ -437,9 +437,9 @@ static inline int is_ep_periodic(struct s3c_hsotg_ep *hs_ep)
>   * This is the reverse of s3c_hsotg_map_dma(), called for the completion
>   * of a request to ensure the buffer is ready for access by the caller.
>   */
> -static void s3c_hsotg_unmap_dma(struct s3c_hsotg *hsotg,
> -				struct s3c_hsotg_ep *hs_ep,
> -				struct s3c_hsotg_req *hs_req)
> +static inline void s3c_hsotg_unmap_dma(struct s3c_hsotg *hsotg,
> +				       struct s3c_hsotg_ep *hs_ep,
> +				       struct s3c_hsotg_req *hs_req)
>  {
>  	struct usb_request *req = &hs_req->req;
>  
> @@ -861,9 +861,9 @@ static void s3c_hsotg_start_req(struct s3c_hsotg *hsotg,
>   * DMA memory, then we map the memory and mark our request to allow us to
>   * cleanup on completion.
>   */
> -static int s3c_hsotg_map_dma(struct s3c_hsotg *hsotg,
> -			     struct s3c_hsotg_ep *hs_ep,
> -			     struct usb_request *req)
> +static inline int s3c_hsotg_map_dma(struct s3c_hsotg *hsotg,
> +				    struct s3c_hsotg_ep *hs_ep,
> +				    struct usb_request *req)
>  {
>  	struct s3c_hsotg_req *hs_req = our_req(req);
>  	int ret;

Best regards,
--
Bartlomiej Zolnierkiewicz
Samsung R&D Institute Poland
Samsung Electronics