Created attachment 72688 [details]
Test patch for dma buffer leak

It turns out that the driver uses pci_map_single() for the beacon buffers, but never unmaps them. Please test this patch.

Seems to have done the trick.  I have tested for over 6 hours without failure.  Thanks for your help and time.

Thanks for testing. Is it OK for me to use your name and E-mail address in the "Reported-and-Tested-By" line of the patch?

Seems like I was premature in my testing or it may be a completely different problem.

Mar 23 04:00:31 server kernel: [    7.751926] rtl8192se 0000:07:00.0: setting latency timer to 64
Mar 23 04:00:31 server kernel: [    7.765404] rtl8192se: rtl8192ce: FW Power Save off (module option)
Mar 23 04:00:31 server kernel: [    7.767471] rtl8192se: Driver for Realtek RTL8192SE/RTL8191SE
Mar 23 04:00:31 server kernel: [    7.767473] Loading firmware rtlwifi/rtl8192sefw.bin
Mar 23 04:00:31 server kernel: [    8.293884] ieee80211 phy0: Selected rate control algorithm 'rtl_rc'
Mar 23 04:01:34 server kernel: [   81.821201] rtlwifi:rtl_cam_get_free_entry():<0-0> -----hwsec_cam_bitmap: 0x0 entry_idx=4
Mar 23 04:01:47 server kernel: [   95.212456] rtlwifi: &&&&&&&&&del entry 4
Mar 23 04:01:47 server kernel: [   95.396828] rtl8192se 0000:07:00.0: PCI INT A disabled
Mar 23 04:02:14 server kernel: [  122.618103] rtl8192se 0000:07:00.0: PCI INT A -> GSI 17 (level, low) -> IRQ 17
Mar 23 04:02:14 server kernel: [  122.618117] rtl8192se 0000:07:00.0: setting latency timer to 64
Mar 23 04:02:14 server kernel: [  122.631570] rtl8192se: rtl8192ce: FW Power Save off (module option)
Mar 23 04:02:14 server kernel: [  122.631793] rtl8192se: Driver for Realtek RTL8192SE/RTL8191SE
Mar 23 04:02:14 server kernel: [  122.631794] Loading firmware rtlwifi/rtl8192sefw.bin
Mar 23 04:02:14 server kernel: [  122.634123] ieee80211 phy1: Selected rate control algorithm 'rtl_rc'
Mar 23 04:02:31 server kernel: [  139.493163] rtlwifi:rtl_cam_get_free_entry():<0-0> -----hwsec_cam_bitmap: 0x0 entry_idx=4
Mar 23 07:36:54 server kernel: [12999.650598] rtlwifi:rtl_cam_get_free_entry():<0-0> -----hwsec_cam_bitmap: 0x10 entry_idx=5
Mar 23 10:28:56 server kernel: [23319.482507] rtlwifi: &&&&&&&&&del entry 5
Mar 23 10:32:32 server kernel: [23535.368946] rtlwifi: &&&&&&&&&del entry 4
Mar 23 12:20:42 server kernel: [30023.680194] rtl8192se 0000:07:00.0: PCI INT A disabled

The last line was caused by rmmod rtl8192se.

This problem with the cam is definitely new. At the moment, I do not understand this section of the code and will likely have to get help from Realtek.

My first reading of the code led me to believe that this was a WEP problem, but I see that you are using WPA2. Interesting - obviouly I missed something.

Below is my hostapd configuration if it helps.

auth_algs=1
beacon_int=100
bridge=br0
channel=11
country_code=US
ctrl_interface_group=0
ctrl_interface=/var/run/hostapd
driver=nl80211
dtim_period=2
dump_file=/tmp/hostapd.dump
fragm_threshold=2346
ht_capab=[HT40-][SHORT-GI-20][SHORT-GI-40][MAX-AMSDU-7935][DSSS_CCK-40]
hw_mode=g
ieee80211d=1
ieee80211n=1
ignore_broadcast_ssid=0
interface=wlan0
logger_stdout=-1
logger_stdout_level=2
logger_syslog=-1
logger_syslog_level=2
macaddr_acl=0
max_num_sta=255
rsn_pairwise=CCMP
rts_threshold=2347
ssid=xxx
wmm_ac_be_acm=0
wmm_ac_be_aifs=3
wmm_ac_be_cwmax=10
wmm_ac_be_cwmin=4
wmm_ac_be_txop_limit=0
wmm_ac_bk_acm=0
wmm_ac_bk_aifs=7
wmm_ac_bk_cwmax=10
wmm_ac_bk_cwmin=4
wmm_ac_bk_txop_limit=0
wmm_ac_vi_acm=0
wmm_ac_vi_aifs=2
wmm_ac_vi_cwmax=4
wmm_ac_vi_cwmin=3
wmm_ac_vi_txop_limit=94
wmm_ac_vo_acm=0
wmm_ac_vo_aifs=2
wmm_ac_vo_cwmax=3
wmm_ac_vo_cwmin=2
wmm_ac_vo_txop_limit=47
wmm_enabled=1
wpa=2
wpa_key_mgmt=WPA-PSK
wpa_passphrase=xxx

I have a script that writes the hostapd.conf file on the fly and sets up NAT between the AP and the network connection. Both that approach and your configuration file (modified for my setup) result in beacons, but I cannot connect. As a result, it seems that I will not be able to test here.

I sent a request for help to Realtek, but I do not expect a response until next week. In the meantime, if you have plenty of disk space on the AP machine, could you load rtl8192se with the "debug=3" option? That should provide more info.

Thanks.

Larry, it occurred that one time.  I haven't been able to replicate it again in the last few days.  I've transferred over 2TB of data and 50H+ of testing.  Yes you can use my name and email in the tested by.  I will let you know if long term testing determines any additional issues.

Thanks for your help
Best

Hi Larry,

Do you plan on submitting the patch for this bug upstream?

Thanks,

Joe

Yes and already done. It is in the wireless-testing tree as commit a75e2ad772b6c26efd702f04be1f9a6414d24f22 with a March 26 date. I added the annotation that it be sent to stable, but I think John Linville will be sending it upstream for kernel 3.5 as it missed the 3.4 merge.

Great! Thanks for the info, Larry.

Well, actually I missed the "stable" tag on that one and another from you.  I will direct them appropriately.

A patch referencing this bug report has been merged in Linux v3.4-rc3:

commit 673f7786e205c87b5d978c62827b9a66d097bebb
Author: Larry Finger <Larry.Finger@lwfinger.net>
Date:   Mon Mar 26 10:48:20 2012 -0500

    rtlwifi: Add missing DMA buffer unmapping for PCI drivers