--- linux/net/irda/qos.c.orig 2005-01-22 19:19:56.013787192 +0100 +++ linux/net/irda/qos.c 2005-01-22 19:21:13.493008560 +0100 @@ -73,13 +73,18 @@ * 1.2, chapt 5.3.2.1, p41). But, this number includes the LAP header * (2 bytes), and CRC (32 bits at 4 Mb/s). So, for the I field (LAP * payload), that's only 2042 bytes. Oups ! - * I've had trouble trouble transmitting 2048 bytes frames with USB - * dongles and nsc-ircc at 4 Mb/s, so adjust to 2042... I don't know - * if this bug applies only for 2048 bytes frames or all negociated - * frame sizes, but all hardware seem to support "2048 bytes" frames. - * You can use the sysctl to play with this value anyway. + * My nsc-ircc hardware has troubles receiving 2048 bytes frames at 4 Mb/s, + * so adjust to 2042... I don't know if this bug applies only for 2048 + * bytes frames or all negociated frame sizes, but you can use the sysctl + * to play with this value anyway. * Jean II */ unsigned sysctl_max_tx_data_size = 2042; +/* + * Maximum transmit window, i.e. number of LAP frames between turn-around. + * This allow to override what the peer told us. Some peers are buggy and + * don't always support what they tell us. + * Jean II */ +unsigned sysctl_max_tx_window = 7; /* * Specific device list limits some negotiation parameters at the connection @@ -227,7 +232,19 @@ { __u16 msb = 0x8000; int index = 15; /* Current MSB */ - + + /* Check for buggy peers. + * Note : there is a small probability that it could be us, but I + * would expect driver authors to catch that pretty early and be + * able to check precisely what's going on. If a end user sees this, + * it's very likely the peer. - Jean II */ + if (word == 0) { + WARNING("%s(), Detected buggy peer, adjust null PV to 0x1!\n", + __FUNCTION__); + /* The only safe choice (we don't know the array size) */ + word = 0x1; + } + while (msb) { if (word & msb) break; /* Found it! */ @@ -378,10 +395,14 @@ /* * Make sure the mintt is sensible. + * Main culprit : Ericsson T39. - Jean II */ if (sysctl_min_tx_turn_time > qos->min_turn_time.value) { int i; + WARNING("%s(), Detected buggy peer, adjust mtt to %dus!\n", + __FUNCTION__, sysctl_min_tx_turn_time); + /* We don't really need bits, but easier this way */ i = value_highest_bit(sysctl_min_tx_turn_time, min_turn_times, 8, &qos->min_turn_time.bits); @@ -443,6 +464,11 @@ if (qos->data_size.value > sysctl_max_tx_data_size) /* Allow non discrete adjustement to avoid loosing capacity */ qos->data_size.value = sysctl_max_tx_data_size; + /* + * Override Tx window if user request it. - Jean II + */ + if (qos->window_size.value > sysctl_max_tx_window) + qos->window_size.value = sysctl_max_tx_window; } /* --- linux/net/irda/irsysctl.c.orig 2005-01-22 19:19:56.006788256 +0100 +++ linux/net/irda/irsysctl.c 2005-01-22 19:24:31.273941288 +0100 @@ -40,8 +40,8 @@ enum { DISCOVERY=1, DEVNAME, DEBUG, FAST_POLL, DISCOVERY_SLOTS, DISCOVERY_TIMEOUT, SLOT_TIMEOUT, MAX_BAUD_RATE, MIN_TX_TURN_TIME, - MAX_TX_DATA_SIZE, MAX_NOREPLY_TIME, WARN_NOREPLY_TIME, LAP_KEEPALIVE_TIME, - SPECIFIC_DEV }; + MAX_TX_DATA_SIZE, MAX_TX_WINDOW, MAX_NOREPLY_TIME, WARN_NOREPLY_TIME, + LAP_KEEPALIVE_TIME, SPECIFIC_DEV }; extern int sysctl_discovery; extern int sysctl_discovery_slots; @@ -53,6 +53,7 @@ extern int sysctl_max_baud_rate; extern int sysctl_min_tx_turn_time; extern int sysctl_max_tx_data_size; +extern int sysctl_max_tx_window; extern int sysctl_max_noreply_time; extern int sysctl_warn_noreply_time; extern int sysctl_lap_keepalive_time; @@ -75,6 +76,8 @@ static int min_min_tx_turn_time = 0; static int max_max_tx_data_size = 2048; /* See qos.c - IrLAP spec */ static int min_max_tx_data_size = 64; +static int max_max_tx_window = 7; /* See qos.c - IrLAP spec */ +static int min_max_tx_window = 1; static int max_max_noreply_time = 40; /* See qos.c - IrLAP spec */ static int min_max_noreply_time = 3; static int max_warn_noreply_time = 3; /* 3s == standard */ @@ -135,6 +138,9 @@ { MAX_TX_DATA_SIZE, "max_tx_data_size", &sysctl_max_tx_data_size, sizeof(int), 0644, NULL, &proc_dointvec_minmax, &sysctl_intvec, NULL, &min_max_tx_data_size, &max_max_tx_data_size }, + { MAX_TX_WINDOW, "max_tx_window", &sysctl_max_tx_window, + sizeof(int), 0644, NULL, &proc_dointvec_minmax, &sysctl_intvec, + NULL, &min_max_tx_window, &max_max_tx_window }, { MAX_NOREPLY_TIME, "max_noreply_time", &sysctl_max_noreply_time, sizeof(int), 0644, NULL, &proc_dointvec_minmax, &sysctl_intvec, NULL, &min_max_noreply_time, &max_max_noreply_time }, --- linux/net/irda/parameters.c.orig 2003-05-13 11:20:16.000000000 +0200 +++ linux/net/irda/parameters.c 2005-01-22 19:21:13.527003392 +0100 @@ -204,11 +204,13 @@ { irda_param_t p; int n = 0; + int extract_len; /* Real lenght we extract */ int err; p.pi = pi; /* In case handler needs to know */ p.pl = buf[1]; /* Extract lenght of value */ p.pv.i = 0; /* Clear value */ + extract_len = p.pl; /* Default : extract all */ /* Check if buffer is long enough for parsing */ if (len < (2+p.pl)) { @@ -220,18 +222,30 @@ /* * Check that the integer length is what we expect it to be. If the * handler want a 16 bits integer then a 32 bits is not good enough + * PV_INTEGER means that the handler is flexible. */ if (((type & PV_MASK) != PV_INTEGER) && ((type & PV_MASK) != p.pl)) { ERROR(__FUNCTION__ "(), invalid parameter length! " "Expected %d bytes, but value had %d bytes!\n", type & PV_MASK, p.pl); - /* Skip parameter */ - return p.pl+2; + /* Most parameters are bit/byte fields or little endian, + * so it's ok to only extract a subset of it (the subset + * that the handler expect). This is necessary, as some + * broken implementations seems to add extra undefined bits. + * If the parameter is shorter than we expect or is big + * endian, we can't play those tricks. Jean II */ + if((p.pl < (type & PV_MASK)) || (type & PV_BIG_ENDIAN)) { + /* Skip parameter */ + return p.pl+2; + } else { + /* Extract subset of it, fallthrough */ + extract_len = type & PV_MASK; + } } - switch (p.pl) { + switch (extract_len) { case 1: n += irda_param_unpack(buf+2, "b", &p.pv.i); break;