- 15 Jan, 2009 38 commits
-
-
Eric Dumazet authored
b44 chip has some hardware limitations, that need GFP_DMA bounce buffers in some situations. In order to not deplete DMA zone, we should keep allocated GFP_DMA skb only for driver use. At rx time, we copy such skb to newly allocated skb, reusing existing copybreak infrastructure. On machines with low amount of memory, all skb meet the hardware limitation, so no copy is needed. We detect this situation using a new device flag, set to one if one GFP_DMA skb was ever allocated by b44_alloc_rx_skb(). Previously allocated skb, even outside from DMA zone will then be recycled, to have minimal impact on DMA zone use. Signed-off-by:
Eric Dumazet <dada1@cosmosbay.com> Tested-by:
Ionut Leonte <ionut.leonte@gmail.com> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Phil Sutter authored
As the kernel warning states: "IRQF_DISABLED is not guaranteed on shared IRQs". Since these IRQs' values are hardcoded and my test system doesn't show any shared use of IRQs at all, rather make them non-shared than non-disabled. Signed-off-by:
Phil Sutter <n0-1@freewrt.org> Acked-by:
Florian Fainelli <florian@openwrt.org> Signed-off-by:
-
Phil Sutter authored
Apparently this doesn't make sense. Otherwise the queue gets disabled as soon as it's getting empty and can only be resurrected by a driver restart. Signed-off-by:
-
Phil Sutter authored
Originally this must have been a rewrite error when introducing 'chain_index'. But the original driver did not use the previous chain item everywhere: when altering the address tx_chain_tail points to, it should move forward, not backwards. Also this is not an "index" but rather the penultimate element in the chain, so rename it accordingly. Signed-off-by:
-
Phil Sutter authored
Triggering TX before the write to the DMA status mask register leads to transferring packets with maximum payload no matter what the actual packet size is. While here, also trigger RX scheduling after writing the DMA status mask register, like it was in the original driver before it was sent upstream. Signed-off-by:
-
Phil Sutter authored
The called netif_rx_schedule() does all the work for us: - it checks the return value of netif_rx_schedule_prep() and - if everything is ok calls __netif_rx_schedule(). Before this change, the driver received absolutely nothing. Signed-off-by:
-
Phil Sutter authored
This function needs an early exit condition to function properly, or else caller assumes napi workload wasn't enough to handle all received packets and korina_rx is called again (and again and again and ...). Signed-off-by:
-
Phil Sutter authored
Without this the driver will crash when the NIC is being restarted. Signed-off-by:
-
Phil Sutter authored
The new value is the one used in the external patch before and allows at least a standard MTU of 1500 to be handled correctly. Impact of this change gets visible when bigger packets are to be received, issuing: | ping -s 492 <IP> and bigger payload sized led to 100% packet loss. Signed-off-by:
-
Phil Sutter authored
Using platform_set_drvdata() here makes no sense, since the driver_data field has already been filled with valuable data (i.e. the MAC address). Also having driver_data point to the net_device is rather pointless since struct korina_device contains an apropriate field for it. Signed-off-by:
-
Eilon Greenstein authored
The "read for interrupts" flag must be set before enabling slow-path interrupts as well (and not just before fast-path interrupts) Signed-off-by:
Eilon Greenstein <eilong@broadcom.com> Signed-off-by:
-
Eilon Greenstein authored
Too big packets could pass due to wrong filter size Signed-off-by:
-
Eilon Greenstein authored
Wrong initialization of the multi-queue indirection table - it should be using the function and not the port index Signed-off-by:
-
Eilon Greenstein authored
Calculation bug due to missing brackets Signed-off-by:
-
Eilon Greenstein authored
The size of the doorbell is 4KB, this bug become visible when using more than 8 queues Signed-off-by:
-
Eilon Greenstein authored
Adding missing le_to_cpu and disabling wrong HW endianity flag (the two complete each other) Signed-off-by:
-
Eilon Greenstein authored
Wrong handling of tagged packet if VLAN offload is disabled caused packets to get corrupted Signed-off-by:
-
Eilon Greenstein authored
Without this lock, in some race conditions the driver missed link change indication Signed-off-by:
-
Eilon Greenstein authored
Signed-off-by:
-
Eilon Greenstein authored
Signed-off-by:
-
Eilon Greenstein authored
When the page size is not 4KB, the FW must be programmed to work with the right SGE boundaries and fragment list length. To avoid confusion with the BCM_PAGE_SIZE which is set to 4KB for the FW sake, another alias for the system page size was added to explicitly indicate that it is meant for the SGE Signed-off-by:
-
Eilon Greenstein authored
While working on IA64, it became clear that the following memory barriers are missing Signed-off-by:
-
Eilon Greenstein authored
Since slow-path events, including link update, are handled in work-queue, a race condition was introduced in the self-test that sometimes caused the link status to fail: the self-test was running under RTNL lock, and if the link-watch was scheduled it stoped the shared work-queue (waiting for the RTNL lock) and so the link update event was not handled until the self-test ended (releasing the RTNL lock) with failure (since the link status was not updated) Signed-off-by:
-
Oliver Hartkopp authored
Due to the loopback functionality in can_send() we can not invoke it from hardirq context which was done inside the bcm_tx_timeout_handler() hrtimer callback: [ 700.361154] [<c012228c>] warn_slowpath+0x80/0xb6 [ 700.361163] [<c013d559>] valid_state+0x125/0x136 [ 700.361171] [<c013d858>] mark_lock+0x18e/0x332 [ 700.361180] [<c013e300>] __lock_acquire+0x12e/0xb1e [ 700.361189] [<f8ab5915>] bcm_tx_timeout_handler+0x0/0xbc [can_bcm] [ 700.361198] [<c031e20a>] dev_queue_xmit+0x191/0x479 [ 700.361206] [<c01262a7>] __local_bh_disable+0x2b/0x64 [ 700.361213] [<c031e20a>] dev_queue_xmit+0x191/0x479 [ 700.361225] [<f8aa69a1>] can_send+0xd7/0x11a [can] [ 700.361235] [<f8ab522b>] bcm_can_tx+0x9d/0xd9 [can_bcm] [ 700.361245] [<f8ab597f>] bcm_tx_timeout_handler+0x6a/0xbc [can_bcm] [ 700.361255] [<f8ab5915>] bcm_tx_timeout_handler+0x0/0xbc [can_bcm] [ 700.361263] [<c0134143>] __run_hrtimer+0x5a/0x86 [ 700.361273] [<f8ab5915>] bcm_tx_timeout_handler+0x0/0xbc [can_bcm] [ 700.361282] [<c0134a50>] hrtimer_interrupt+0xb9/0x110 This patch moves the rest of the functionality from the hrtimer callback to the already existing tasklet to fix this slowpath problem. Signed-off-by:
Oliver Hartkopp <oliver@hartkopp.net> Signed-off-by:
-
Magnus Damm authored
This patch hooks up the start_xmit/tx_timeout/get_stats callbacks in the ax88796 driver since they no longer are installed by the lib8390 code. Without this patch the function dev_hard_start_xmit() crashes due to a start_xmit callback with the value NULL. While at it, update the ax88796 driver to make use of use of struct net_device_ops. Signed-off-by:
Magnus Damm <damm@igel.co.jp> Signed-off-by:
-
Benjamin Herrenschmidt authored
This adds an init_dummy_netdev() function that gets a network device structure (allocation and lifetime entirely under caller's control) and initialize the minimum amount of fields so it can be used to schedule NAPI polls without registering a full blown interface. This is to be used by drivers that need to tie several hardware interfaces to a single NAPI poll scheduler due to HW limitations. It also updates the ibm_newemac driver to use that, this fixing the oops on 2.6.29 due to passing NULL as "dev" to netif_napi_add() Symbol is exported GPL only a I don't think we want binary drivers doing that sort of acrobatics (if we want them at all). Signed-off-by:
Benjamin Herrenschmidt <benh@kernel.crashing.org> Tested-by:
Geert Uytterhoeven <Geert.Uytterhoeven@sonycom.com> Signed-off-by:
-
Jos-Vicente Gilabert authored
Taken from http://bugzilla.kernel.org/show_bug.cgi?id=12397 We're doing an sprintf of an 11-char string into an 11-char buffer. Whoops. It breaks firmware uploading. Reported-by:
Jos-Vicente Gilabert <josevteg@gmail.com> Signed-off-by:
Andrew Morton <akpm@linux-foundation.org> Signed-off-by:
-
Dhananjay Phadke authored
o Bail out if pci_map_single() fails while replenishing rx ring. o Drop packet if pci_map_{single,page}() fail in tx. Signed-off-by:Dhananjay Phadke <dhananjay@netxen.com> Signed-off-by:
-
Dhananjay Phadke authored
Some firmware commands like mac address addition/deletion are sent on the transmit ring. So need to hold the tx lock before touching tx producer/consumer indices. Signed-off-by:
-
Dhananjay Phadke authored
This fixes a tiny memory leak when driver is unloaded. The mac address list maintained in netxen_adapter needs to deleted when driver is going down. Signed-off-by:
-
Dhananjay Phadke authored
o Fix order or rom register writes. o Reduce udelays when writing rom registers. This cuts the firmware init time by 40%. o Do not reset core/memory clocks when reinitializing driver. Firmware willl handle this when initialized. Signed-off-by:
-
Dhananjay Phadke authored
o Read negotiated link speed when link state changes. o Fix link speed reporting for hybrid nic boards, which have both 1Gbps and 10Gbps ports. Signed-off-by:
-
Dhananjay Phadke authored
o fix the ip/tcp hdr offset in tx descriptors for ipv6. o cleanup xmit function, move the tso checks into separate function, this reduces unnecessary endian conversions back and forth. o optimize macros to initialize tx descriptors. Signed-off-by:
-
Dhananjay Phadke authored
o Set restricted (little endian) data types in firmware command requests and responses. o Remove unnecessary conversion to LE when writing registers. Signed-off-by:
-
Daniele Venzano authored
The attached patch modifies the sis900 driver when the MAC address read from the hardware is invalid. As suggested, the patch now generates a random address so that the user can go on and use the hardware. In any case a message is also shown to warn on the unexpected condition. This seems to happen with newer HW implementation of the sis900 chipset, since this never came up before. Patch is against vanilla 2.6.28 (but the driver doesn't change so often, so it will probably apply to older/newer versions too). See bugzilla ID 10201 and 11649 and ignore the previous patch. Signed-off-by:
Daniele Venzano <venza@brownhat.org> Signed-off-by:
-
Mike Ditto authored
If something goes wrong attaching to phy driver, we weren't freeing the IRQ. Signed-off-by:
Mike Ditto <mditto@consentry.com> Signed-off-by:
-
Herbert Xu authored
When we get a GSO packet from an untrusted source, we need to ensure that it is sufficiently long so that we don't end up crashing. Based on discovery and patch by Ian Campbell. Signed-off-by:
Herbert Xu <herbert@gondor.apana.org.au> Tested-by:
Ian Campbell <ian.campbell@citrix.com> Signed-off-by:
-
Herbert Xu authored
When an skb with page frags is merged into an existing one, we cannibalise its reference count. This is OK when the skb is reused because we set nr_frags to zero in that case. However, for the case where the skb is freed through kfree_skb, we didn't clear nr_frags which causes the page to be freed prematurely. This is fixed by moving the skb resetting into skb_gro_receive. Reported-by:
Jeff Kirsher <jeffrey.t.kirsher@intel.com> Signed-off-by:
-
- 14 Jan, 2009 2 commits
-
-
David S. Miller authored
Reported by Jiri Klimes. Fix suggested by Patrick McHardy. Signed-off-by: -
Paul Bolle authored
Minor cleanups, either made possible or obvious after commit d700555b (I4l: convert to net_device_ops). Signed-off-by:
Paul Bolle <pebolle@tiscali.nl> Signed-off-by:
-
