fsm completion event handling from main loop
The main loop will now poll for finished/failed transactions and handle
them, this was previously handled during the last rx interrupt of a
transaction, which was bad for timing. This does also fix malloc/free
while handling interrupts.
Change-Id: I055110720089e20e65db592eccc3ce4d618e8c63
diff --git a/ccid_common/ccid_slot_fsm.c b/ccid_common/ccid_slot_fsm.c
index d2aec26..9be31d6 100644
--- a/ccid_common/ccid_slot_fsm.c
+++ b/ccid_common/ccid_slot_fsm.c
@@ -110,34 +110,76 @@
{
struct iso_fsm_slot *ss = iso7816_fsm_get_user_priv(fi);
struct ccid_slot *cs = ss->cs;
+
+ switch (event) {
+ case ISO7816_E_ATR_DONE_IND:
+ case ISO7816_E_ATR_ERR_IND:
+ case ISO7816_E_TPDU_DONE_IND:
+ case ISO7816_E_TPDU_FAILED_IND:
+ case ISO7816_E_PPS_DONE_IND:
+ case ISO7816_E_PPS_FAILED_IND:
+ cs->event_data = data;
+ asm volatile("dmb st": : :"memory");
+ cs->event = event;
+ break;
+ default:
+ LOGPCS(cs, LOGL_NOTICE, "%s(event=%d, cause=%d, data=%p) unhandled\n",
+ __func__, event, cause, data);
+ break;
+ }
+}
+
+static int iso_handle_fsm_events(struct ccid_slot *cs, bool enable){
+ struct iso_fsm_slot *ss = ccid_slot2iso_fsm_slot(cs);
struct msgb *tpdu, *resp;
+ volatile uint32_t event = cs->event;
+ volatile void * volatile data = cs->event_data;
+
+ if(!event)
+ return 0;
+ if(event && !data)
+ return 0;
switch (event) {
case ISO7816_E_ATR_DONE_IND:
tpdu = data;
- LOGPCS(cs, LOGL_DEBUG, "%s(event=%d, cause=%d, data=%s)\n", __func__, event, cause,
+ LOGPCS(cs, LOGL_DEBUG, "%s(event=%d, data=%s)\n", __func__, event,
msgb_hexdump(tpdu));
resp = ccid_gen_data_block(cs, ss->seq, CCID_CMD_STATUS_OK, 0,
msgb_data(tpdu), msgb_length(tpdu));
ccid_slot_send_unbusy(cs, resp);
/* Don't free "TPDU" here, as the ATR should survive */
+ cs->event = 0;
+ break;
+ case ISO7816_E_ATR_ERR_IND:
+ tpdu = data;
+ LOGPCS(cs, LOGL_DEBUG, "%s(event=%d, data=%s)\n", __func__, event,
+ msgb_hexdump(tpdu));
+ resp = ccid_gen_data_block(cs, ss->seq, CCID_CMD_STATUS_FAILED, CCID_ERR_ICC_MUTE,
+ msgb_data(tpdu), msgb_length(tpdu));
+ ccid_slot_send_unbusy(cs, resp);
+ /* Don't free "TPDU" here, as the ATR should survive */
+ cs->event = 0;
+ break;
break;
case ISO7816_E_TPDU_DONE_IND:
tpdu = data;
- LOGPCS(cs, LOGL_DEBUG, "%s(event=%d, cause=%d, data=%s)\n", __func__, event, cause,
+ LOGPCS(cs, LOGL_DEBUG, "%s(event=%d, data=%s)\n", __func__, event,
msgb_hexdump(tpdu));
resp = ccid_gen_data_block(cs, ss->seq, CCID_CMD_STATUS_OK, 0, msgb_l2(tpdu), msgb_l2len(tpdu));
ccid_slot_send_unbusy(cs, resp);
msgb_free(tpdu);
+ cs->event = 0;
break;
case ISO7816_E_TPDU_FAILED_IND:
tpdu = data;
- LOGPCS(cs, LOGL_DEBUG, "%s(event=%d, cause=%d, data=%s)\n", __func__, event, cause,
+ LOGPCS(cs, LOGL_DEBUG, "%s(event=%d, data=%s)\n", __func__, event,
msgb_hexdump(tpdu));
/* FIXME: other error causes than card removal?*/
resp = ccid_gen_data_block(cs, ss->seq, CCID_CMD_STATUS_FAILED, CCID_ERR_ICC_MUTE, msgb_l2(tpdu), 0);
ccid_slot_send_unbusy(cs, resp);
msgb_free(tpdu);
+ cs->event = 0;
break;
case ISO7816_E_PPS_DONE_IND:
tpdu = data;
@@ -157,6 +199,7 @@
/* this frees the pps req from the host, pps resp buffer stays with the pps fsm */
msgb_free(tpdu);
+ cs->event = 0;
break;
case ISO7816_E_PPS_FAILED_IND:
tpdu = data;
@@ -165,10 +208,13 @@
ccid_slot_send_unbusy(cs, resp);
/* this frees the pps req from the host, pps resp buffer stays with the pps fsm */
msgb_free(tpdu);
+ cs->event = 0;
+ break;
+ case 0:
break;
default:
- LOGPCS(cs, LOGL_NOTICE, "%s(event=%d, cause=%d, data=%p) unhandled\n",
- __func__, event, cause, data);
+ LOGPCS(cs, LOGL_NOTICE, "%s(event=%d, data=%p) unhandled\n",
+ __func__, event, data);
break;
}
}
@@ -177,7 +223,7 @@
const struct ccid_pc_to_rdr_xfr_block *xfb)
{
struct iso_fsm_slot *ss = ccid_slot2iso_fsm_slot(cs);
- struct msgb *tpdu;
+
ss->seq = xfb->hdr.bSeq;
@@ -185,17 +231,10 @@
OSMO_ASSERT(xfb->wLevelParameter == 0x0000);
OSMO_ASSERT(msgb_length(msg) > xfb->hdr.dwLength);
- /* 'msg' contains the raw CCID message as received from USB. We could create
- * a new message buffer for the ISO7816 side here or we could 'strip the CCID
- * header off the start of the message. Let's KISS and do a copy here */
- tpdu = msgb_alloc(512, "TPDU");
- OSMO_ASSERT(tpdu);
- memcpy(msgb_data(tpdu), xfb->abData, xfb->hdr.dwLength);
- msgb_put(tpdu, xfb->hdr.dwLength);
- msgb_free(msg);
+ msgb_pull(msg, 10);
- LOGPCS(cs, LOGL_DEBUG, "scheduling TPDU transfer: %s\n", msgb_hexdump(tpdu));
- osmo_fsm_inst_dispatch(ss->fi, ISO7816_E_XCEIVE_TPDU_CMD, tpdu);
+ LOGPCS(cs, LOGL_DEBUG, "scheduling TPDU transfer: %s\n", msgb_hexdump(msg));
+ osmo_fsm_inst_dispatch(ss->fi, ISO7816_E_XCEIVE_TPDU_CMD, msg);
/* continues in iso_fsm_clot_user_cb once response/error/timeout is received */
return 0;
}
@@ -329,4 +368,5 @@
.set_clock = iso_fsm_slot_set_clock,
.set_params = iso_fsm_slot_set_params,
.set_rate_and_clock = iso_fsm_slot_set_rate_and_clock,
+ .handle_fsm_events = iso_handle_fsm_events,
};