#127: madwifi-ng: 2.6.15 kernel panic: fatal exception in irq handler
------------------------------+---------------------------------------------
Reporter: schorpp | Owner: mrenzmann
Type: defect | Status: new
Priority: critical | Milestone: version 0.9.0 - move to new
codebase
Component: madwifi: driver | Version: trunk
Resolution: | Keywords: panic
------------------------------+---------------------------------------------
Comment (by schorpp):
with
modprobe ath-pci ath_debug=0x00801000
and some extra printk's here:
static void
ath_uapsd_processtriggers(struct ath_softc *sc)
{
struct ath_hal *ah = sc->sc_ah;
struct ath_buf *bf;
struct ath_desc *ds;
struct sk_buff *skb;
struct ieee80211_node *ni;
struct ath_node *an;
struct ieee80211_qosframe *qwh;
struct ath_txq *uapsd_xmit_q = sc->sc_uapsdq;
struct ieee80211com *ic = &sc->sc_ic;
int ac, retval;
u_int8_t tid;
u_int16_t frame_seq;
u_int64_t tsf;
#define PA2DESC(_sc, _pa) \
((struct ath_desc *)((caddr_t)(_sc)->sc_rxdma.dd_desc + \
((_pa) - (_sc)->sc_rxdma.dd_desc_paddr)))
/* XXXAPSD: build in check against max triggers we could see
* based on ic->ic_uapsdmaxtriggers.
*/
DPRINTF(sc, ATH_DEBUG_UAPSD, "%s: A\n", __func__
);
tsf = ath_hal_gettsf64(ah);
ATH_RXBUF_LOCK(sc);
if (sc->sc_rxbufcur == NULL)
sc->sc_rxbufcur = STAILQ_FIRST(&sc->sc_rxbuf);
for (bf = sc->sc_rxbufcur; bf; bf = STAILQ_NEXT(bf, bf_list)) {
ds = bf->bf_desc;
if (ds->ds_link == bf->bf_daddr) {
/* NB: never process the self-linked entry at the
end */
break;
}
if (bf->bf_status & ATH_BUFSTATUS_DONE) {
/*
* already processed this buffer (shouldn't occur
if
* we change code to always process descriptors in
* rx intr handler - as opposed to sometimes
processing
* in the rx tasklet).
*/
continue;
}
skb = bf->bf_skb;
if (skb == NULL) { /* XXX ??? can this happen
*/
printk("%s: no skbuff\n", __func__);
continue;
}
/*
* XXXAPSD: consider new hal call that does only the
subset
* of ath_hal_rxprocdesc we require for trigger
search.
*/
/*
* NB: descriptor memory doesn't need to be sync'd
* due to the way it was allocated.
*/
/*
* Must provide the virtual address of the current
* descriptor, the physical address, and the virtual
* address of the next descriptor in the h/w chain.
* This allows the HAL to look ahead to see if the
* hardware is done with a descriptor by checking the
* done bit in the following descriptor and the address
* of the current descriptor the DMA engine is working
* on. All this is necessary because of our use of
* a self-linked list to avoid rx overruns.
*/
DPRINTF(sc, ATH_DEBUG_UAPSD, "%s: B\n", __func__
);
retval = ath_hal_rxprocdesc(ah, ds, bf->bf_daddr,
PA2DESC(sc, ds->ds_link),tsf);
if (HAL_EINPROGRESS == retval)
break;
... it runs now for hours without any issue and the hostapd ap this client
is connected does no longer need to force by local deauth requests at the
reauth intervalls.
so theres wether a synchronisation/timing problem in the hal or maschine
specific irq/acpi trouble...
the code part of the 20:09 post is in. commenting it out results in very
earlier crashes.
y
tom
--
Ticket URL: <http://madwifi.org/ticket/127>
MadWifi <http://madwifi.org/>
Multiband Atheros Driver for Wireless Fidelity
