Re: kern/50150: wm(4) failure

[Kengo NAKAHARA <k-nakahara%iij.ad.jp@localhost>]

Hi Hauke Fath,

On 2016/05/18 21:05, Kengo NAKAHARA wrote:
> The following reply was made to PR kern/50150; it has been noted by GNATS.
> 
> From: Kengo NAKAHARA <k-nakahara%iij.ad.jp@localhost>
> To: hf%spg.tu-darmstadt.de@localhost
> Cc: gnats-bugs%NetBSD.org@localhost, kern-bug-people%NetBSD.org@localhost, gnats-admin%NetBSD.org@localhost
> Subject: Re: kern/50150: wm(4) failure
> Date: Wed, 18 May 2016 21:00:14 +0900
> 
>  Hi,
>  
>  On 2016/05/18 20:46, Hauke Fath wrote:
>  > On Wed, 18 May 2016 16:20:35 +0900, Kengo NAKAHARA wrote:
>  >> I think if_wm.c:r1.397 might fix this problem. If you have time, Could
>  >> you try if_wm.c:r1.397 or if_wm.c:r1.398 ?
>  > 
>  > Hi Kengo,
>  > 
>  > that's a huge diff against netbsd-7 - almost 40 k in file size alone. 
>  > Rev. 1.391 does not make it a drop-in replacement for the netbsd-7 
>  > version, either, and I don't have any particular knowledge about the 
>  > wm(4) code.
>  > 
>  > Plus, this is about a production machine which cannot run -current.
>  > 
>  > In short: Do you think you could narrow down the relevant changes to a 
>  > netbsd-7 compatible patch?
>  
>  Sorry, I assume you use -current.
>  However, netbsd-7 seems to have a similar problem. As you indicated,
>  there is huge diff between netbsd-7 and -current. So, I will implement
>  new patch for netbsd-7 from scratch. Please give me a time for a few
>  days... or a few weeks.

I implement that patch for netbsd-7. Here is the patch.
====================
diff --git a/sys/dev/pci/if_wm.c b/sys/dev/pci/if_wm.c
index da64cfc..ee4d00b 100644
--- a/sys/dev/pci/if_wm.c
+++ b/sys/dev/pci/if_wm.c
@@ -315,6 +315,7 @@ struct wm_softc {
 
 	/* Control data structures. */
 	int sc_ntxdesc;			/* must be a power of two */
+	int sc_txdesc_size;		/* size of a descriptor */
 	struct wm_control_data_82544 *sc_control_data;
 	bus_dmamap_t sc_cddmamap;	/* control data DMA map */
 	bus_dma_segment_t sc_cd_seg;	/* control data segment */
@@ -483,7 +484,7 @@ do {									\
 	/* If it will wrap around, sync to the end of the ring. */	\
 	if ((__x + __n) > WM_NTXDESC(sc)) {				\
 		bus_dmamap_sync((sc)->sc_dmat, (sc)->sc_cddmamap,	\
-		    WM_CDTXOFF(__x), sizeof(wiseman_txdesc_t) *		\
+		    WM_CDTXOFF(__x), (sc)->sc_txdesc_size *		\
 		    (WM_NTXDESC(sc) - __x), (ops));			\
 		__n -= (WM_NTXDESC(sc) - __x);				\
 		__x = 0;						\
@@ -491,7 +492,7 @@ do {									\
 									\
 	/* Now sync whatever is left. */				\
 	bus_dmamap_sync((sc)->sc_dmat, (sc)->sc_cddmamap,		\
-	    WM_CDTXOFF(__x), sizeof(wiseman_txdesc_t) * __n, (ops));	\
+	    WM_CDTXOFF(__x), (sc)->sc_txdesc_size * __n, (ops));	\
 } while (/*CONSTCOND*/0)
 
 #define	WM_CDRXSYNC(sc, x, ops)						\
@@ -1682,6 +1683,10 @@ wm_attach(device_t parent, device_t self, void *aux)
 	sc->sc_cd_size = sc->sc_type < WM_T_82544 ?
 	    sizeof(struct wm_control_data_82542) :
 	    sizeof(struct wm_control_data_82544);
+	if ((sc->sc_flags & WM_F_NEWQUEUE) != 0)
+		sc->sc_txdesc_size = sizeof(nq_txdesc_t);
+	else
+		sc->sc_txdesc_size = sizeof(wiseman_txdesc_t);
 	if ((error = bus_dmamem_alloc(sc->sc_dmat, sc->sc_cd_size, PAGE_SIZE,
 		    (bus_size_t) 0x100000000ULL, &sc->sc_cd_seg, 1,
 		    &sc->sc_cd_rseg, 0)) != 0) {
====================

I ensured above patch can apply to if_wm.c:r1.289.2.9 and can build.
Could you try this patch?


Thanks,

-- 
//////////////////////////////////////////////////////////////////////
Internet Initiative Japan Inc.

Device Engineering Section,
IoT Platform Development Department,
Network Division,
Technology Unit

Kengo NAKAHARA <k-nakahara%iij.ad.jp@localhost>