- 20 Nov, 2012 9 commits
-
-
Seth Forshee authored
Currently up to 256 frames can be queued for each DMA ring. This is excessive, and now that we have better flow control we can get by with less. Experimentation has shown 64 to work well. Signed-off-by:
Seth Forshee <seth.forshee@canonical.com> Reviewed-by:
Pieter-Paul Giesberts <pieterpg@broadcom.com> Reviewed-by:
Arend van Spriel <arend@broadcom.com> Tested-by:
Daniel Wagner <wagi@monom.org> Signed-off-by:
John W. Linville <linville@tuxdriver.com>
-
Seth Forshee authored
nextrxd() is calling txd(), which means that the tx descriptor count is used to determine when to wrap for determining the next ring buffer entry. This has worked so far since the driver has been using the same number of rx and tx descriptors, but it's obviously going to be a problem if different numbers of descriptors are used. Acked-by:
-
Seth Forshee authored
The brcmsmac internal tx buffering is problematic. The amount of buffering is excessive (228 packets in addition to the 256 slots in each DMA ring), and frames may be dropped due to a lack of flow control. This patch reworks the transmit code path to remove the internal buffering. Frames are immediately handed off to the DMA support rather than passing through an intermediate queue. Non-aggregate frames are queued immediately into the tx rings, and aggregate frames are queued temporarily in an AMPDU session until ready for transmit. Transmit flow control is also added to avoid dropping packets when the tx rings are full. Conceptually this is a separate change, but it's included in this commit because removing the tx queue without adding flow control could cause significant problems. Signed-off-by:
-
Seth Forshee authored
The mac80211 tx queues and brcmsmac DMA fifos both map directly to AC levels. Therefore it's much more straightforward to queue tx frames and choose the tx fifo based on the mac80211 queue instead of mapping 802.1D priority tags to precedence levels then back to AC levels. mac80211 already maps the 802.1D levels to the appropriate AC levels and queues management frames at the maximum priority, so the results should be identical. One functional change resulting from this patch is that AMPDU retries no longer get a priority boost to queue them ahead of packets with the same priority already in the tx queue. This behavior will be restored (in effect at least) in a later patch when the tx queue is removed. Signed-off-by:
-
Seth Forshee authored
Functions for flow control exist but remain unimplemented. Remove these in advance of adding real flow control. Signed-off-by:
-
Seth Forshee authored
Use this helper function rather than open-coding the same calculation in multiple places. Signed-off-by:
-
Seth Forshee authored
According to the comments this "reduces rate lag," but in reality the only way this value is used is for determining whether or not any frames remain to be transmitted. Therefore there's no reason for AMPDU packets to receive any weighting. Signed-off-by:
-
Seth Forshee authored
AMPDU session allows MPDUs to be temporarily queued until either a full AMPDU has been collected or circumstances dictate that transmission should start with a partial AMPDU. Packets are added to the session by calling brcms_c_ampdu_add_frame(). brcms_c_ampdu_finalize() should be called to fix up the tx headers in the first and last packet before adding the packets to the DMA ring. brmcs_c_sendampdu() is converted to using AMPDU sessions. This patch has no real value on it's own, but is needed in preparation for elimination of the tx packet queue from brcmsmac. Signed-off-by:
-
git://git.kernel.org/pub/scm/linux/kernel/git/sameo/nfc-3.0John W. Linville authored
Samuel says: "This is the 2nd NFC pull request for 3.8. With this one we have: - A few HCI improvements in preparation for an upcoming HCI chipset support. - A pn544 code cleanup after the old driver was removed. - An LLCP improvement for notifying user space when one peer stops ACKing I frames." Signed-off-by:
-
- 19 Nov, 2012 12 commits
-
-
Samuel Ortiz authored
Drivers are allowed to modify the sent skb and thus we need to make a copy of it before passing it to the driver. Without this fix, LLCP Tx skbs were not queued properly as the ptype check was failing due to e.g. the pn533 driver skb_pushing the Tx skb. Reported-by:
Thierry Escande <thierry.escande@linux.intel.com> Signed-off-by:
Samuel Ortiz <sameo@linux.intel.com>
-
Samuel Ortiz authored
When the tx pending queues and/or the socket tx queue is getting too deep, we have to let userspace know. We won't be queueing any more frames until the congestion is fixed. Signed-off-by: -
Samuel Ortiz authored
Using the userspace IO vector directly is wrong, we should copy it from user space first. Signed-off-by: -
Marcel Holtmann authored
The platform data header for PN544 based NFC devices should also be mentioned here. Signed-off-by:
Marcel Holtmann <marcel@holtmann.org> Signed-off-by:
-
Marcel Holtmann authored
The pn544.h just provides the platform data struct and defines and nothing else. So move it to to linux/platform_data/ now. Signed-off-by:
-
Marcel Holtmann authored
The majority of the defines and structures from pn544.h are no longer in use. So just remove them. Signed-off-by:
-
Eric Lapuyade authored
Some HCI drivers will need it. Signed-off-by:
Eric Lapuyade <eric.lapuyade@intel.com> Signed-off-by:
-
Eric Lapuyade authored
Signed-off-by:
-
Eric Lapuyade authored
A chip with pre-opened gates may send events on a gate that nobody has opened in the handset host. Discard those events. Signed-off-by:
-
Eric Lapuyade authored
NFC_HCI_ID_MGMT_VERSION_SW and NFC_HCI_ID_MGMT_VERSION_HW are optional registers for gate NFC_HCI_ID_MGMT_GATE in standard HCI. When chip doesn't implement, just leave all the information as zeros. Signed-off-by:
-
Eric Lapuyade authored
In some cases, pre-opened pipes don't stay open when a clear all pipes command is sent. They stay created however. Therefore, one can never assume that such a pipe is already open. As re-opening a pipe seems not to be a problem, we do that now. Signed-off-by:
-
Sujith Manoharan authored
The debugfs file for dumping btcoex parameters unconditionally assumes a MCI-based device. This will not work for older btcoex chips. Fix this by branching out the routine into separate functions. Signed-off-by:
Sujith Manoharan <c_manoha@qca.qualcomm.com> Signed-off-by:
-
- 16 Nov, 2012 19 commits
-
-
Sujith Manoharan authored
2484 Mhz (Japan) usage requires filter coefficients to be programmed in the CCK TX FIR registers. This is required for AR9331, AR9485 and AR9462. Fix this and also remove a few useless macros and a duplicate variable. Signed-off-by:
-
Johannes Berg authored
There's no need to ask the user about lib80211 since it will be selected by drivers requiring it, hide it from Kconfig. Signed-off-by:
Johannes Berg <johannes.berg@intel.com> Signed-off-by:
-
Johannes Berg authored
The driver doesn't use any lib80211 symbols so it shouldn't select it in Kconfig. Signed-off-by:
Bing Zhao <bzhao@marvell.com> Signed-off-by:
-
Wei Yongjun authored
Using eth_zero_addr() to assign zero address insetad of memset() or an inefficient copy from a static array. Signed-off-by:
Wei Yongjun <yongjun_wei@trendmicro.com.cn> Signed-off-by:
-
Stanislav Yakovlev authored
Remove unnecessary if statements because libipw_set_geo always returns success. Also change function's return value from int to void. Signed-off-by:
Stanislav Yakovlev <stas.yakovlev@gmail.com> Signed-off-by:
-
Hante Meuleman authored
Removing obsolete functions and prototypes. Moving (and renaming) defines to place with similar definitions. Removing unnecessary includes. Reviewed-by:
Hante Meuleman <meuleman@broadcom.com> Signed-off-by:
Franky Lin <frankyl@broadcom.com> Signed-off-by:
-
Hante Meuleman authored
When debug is turned on for fwil then the whole data buffer is dumped. In some cases this gives excessive amount of debug. With this patch the dumps are limited to 64 bytes. Reviewed-by:
-
Hante Meuleman authored
On sdio module unload followed by load (without removing the device) the access window should be moved back to enumeration space. Force this by removing initialisation of sbwad during probe. Reviewed-by:
-
Hante Meuleman authored
brcmf_ops_sdio_probe used the private_date func->card->dev to store device data of brcmfmac sdio. This is not a good place to store the data. Use dev of func and use func->card->sdio_func to group the functions the driver is using. Reviewed-by:
-
Hante Meuleman authored
on sdio remove the bus_if should be configured for close, so new data from higher layers will be blocked. Also the access to bus_if in the watchdog should be checked for null pointer access on sdio remove. Reviewed-by:
-
Hante Meuleman authored
when brcmf_sdbrcm_probe_attach results in error then cleanup will result in null pointer access. In brcmf_sdbrcm_release and in brcmf_ops_sdio_remove. This patch fixes order of init and de-init. Reviewed-by:
-
Hante Meuleman authored
In brcmf_sdbrcm_probe only error ELINK is seen as error. However brcmf_bus_start can return many more error codes and all should result in failed init of driver. Reviewed-by:
-
Arend van Spriel authored
The IF event need special care. It can be either an ADD, DEL, or CHANGE. For an ADD we need to call brcmf_add_if() before the event handler call. Upon a DEL we need to call brcmf_del_if() after the event handler call. CHANGE does not require special attention. Reviewed-by:
-
Arend van Spriel authored
Firmware fires IF event to add the primary interface but that is already created in the driver. Reviewed-by:
-
Arend van Spriel authored
Replaced <linux/unaligned/access_ok.h> by <asm/unaligned.h> to make it work on ARM and other architectures. Reviewed-by:
-
Hante Meuleman authored
brcmf_fweh_detach can be called while ifp is already NULL, due to init error. Fix NULL pointer access by checking ifp. Reviewed-by:
-
Arend van Spriel authored
The mac_addr field in ifp object is always valid so no need to validate. Reviewed-by:
-
Arend van Spriel authored
The function brcmf_add_if() is called with mac address set to NULL for the primary interface. When handling IF ADD events the firmware provides a address mask in the event to derive its mac address from the primary mac address. Rename the parameter and use it as a mask. Reviewed-by:
-
Arend van Spriel authored
Testing revealed the IF ADD event contains the interface index of the new interface. This would result in a NULL pointer access when handling the event. Reviewed-by:
-
