* Re: [PATCH] net: eth: xgene: change APM X-Gene SoC platform ethernet to support ACPI
From: David Miller @ 2014-12-19 17:40 UTC (permalink / raw)
To: fkan; +Cc: netdev, patches, linux-kernel
In-Reply-To: <CAL85gmCOLrcTijpEhXx1x5UjHu0sgS5DdGYJxY4mwPFKCHpU-g@mail.gmail.com>
From: Feng Kan <fkan@apm.com>
Date: Fri, 19 Dec 2014 09:36:13 -0800
> On Fri, Dec 19, 2014 at 9:31 AM, David Miller <davem@davemloft.net> wrote:
>> From: Feng Kan <fkan@apm.com>
>> Date: Thu, 18 Dec 2014 15:39:51 -0800
>>
>>> +#define NO_MAC_FOUND 0
>>> +#define RES_ENET_CSR 0
>>> +#define RES_RING_CSR 1
>>> +#define RES_RING_CMD 2
> Is there a issue with the RES defines, do you prefer enum types?
I don't have any problem with those.
^ permalink raw reply
* RE: [PATCH net-next v3] Add support of Cavium Liquidio ethernet adapters
From: Chickles, Derek @ 2014-12-19 17:54 UTC (permalink / raw)
To: Stephen Hemminger, Vatsavayi, Raghu
Cc: davem@davemloft.net, netdev@vger.kernel.org, Burla, Satananda,
Manlunas, Felix, Vatsavayi, Raghu
In-Reply-To: <20141218225940.0697e3c5@urahara>
> > +static struct {
> > + const char str[ETH_GSTRING_LEN];
> > +} ethtool_stats_keys[] = {
> > + {
> > + "jiffies"
> > + }, {
>
> Jiffies does not belong in ethtool stats. Looks like some developer debug
> option.
The intention here was to have a semi-accurate timestamp taken at the time
the statistics were gathered before handing it over to userspace. This is a
generic requirement though, so it probably belongs in ethtool_get_stats().
Otherwise, we can publish some per-second stats if that's preferable.
^ permalink raw reply
* Re: [PATCH net] r8152: drop the tx packet with invalid length
From: Eric Dumazet @ 2014-12-19 18:13 UTC (permalink / raw)
To: Hayes Wang
Cc: Tom Herbert, David Miller, netdev@vger.kernel.org, nic_swsd,
linux-kernel@vger.kernel.org, linux-usb@vger.kernel.org
In-Reply-To: <1419010603.9773.84.camel@edumazet-glaptop2.roam.corp.google.com>
On Fri, 2014-12-19 at 09:36 -0800, Eric Dumazet wrote:
> On Fri, 2014-12-19 at 06:42 +0000, Hayes Wang wrote:
>
> > Excuse me. I try to implement ndo_gso_check() with kernel 3.18.
> > However, I still get packets with gso and their skb lengths are more
> > than the acceptable one. Do I miss something?
> >
> > +static bool rtl8152_gso_check(struct sk_buff *skb, struct net_device *dev)
> > +{
> > + if ((skb->len + sizeof(struct tx_desc)) > agg_buf_sz)
> > + return false;
> > +
> > + return true;
> > +}
> >
> > @@ -5861,6 +5876,9 @@ static const struct net_device_ops rtl8152_netdev_ops = {
> > .ndo_set_mac_address = rtl8152_set_mac_address,
> > .ndo_change_mtu = rtl8152_change_mtu,
> > .ndo_validate_addr = eth_validate_addr,
> > + .ndo_gso_check = rtl8152_gso_check,
> > };
>
> You are right, it seems ndo_gso_check() is buggy at this moment.
>
> Presumably this method should alter %features so that we really segment
> the packets in software.
>
> features &= ~NETIF_F_GSO_MASK;
Could you try following patch ?
diff --git a/drivers/net/macvtap.c b/drivers/net/macvtap.c
index 7df221788cd4..0346bcfe72a5 100644
--- a/drivers/net/macvtap.c
+++ b/drivers/net/macvtap.c
@@ -314,7 +314,7 @@ static rx_handler_result_t macvtap_handle_frame(struct sk_buff **pskb)
*/
if (q->flags & IFF_VNET_HDR)
features |= vlan->tap_features;
- if (netif_needs_gso(dev, skb, features)) {
+ if (netif_needs_gso(dev, skb, &features)) {
struct sk_buff *segs = __skb_gso_segment(skb, features, false);
if (IS_ERR(segs))
diff --git a/drivers/net/xen-netfront.c b/drivers/net/xen-netfront.c
index 22bcb4e12e2a..9cacabaea175 100644
--- a/drivers/net/xen-netfront.c
+++ b/drivers/net/xen-netfront.c
@@ -578,6 +578,7 @@ static int xennet_start_xmit(struct sk_buff *skb, struct net_device *dev)
unsigned long flags;
struct netfront_queue *queue = NULL;
unsigned int num_queues = dev->real_num_tx_queues;
+ netdev_features_t features;
u16 queue_index;
/* Drop the packet if no queues are set up */
@@ -611,9 +612,10 @@ static int xennet_start_xmit(struct sk_buff *skb, struct net_device *dev)
spin_lock_irqsave(&queue->tx_lock, flags);
+ features = netif_skb_features(skb);
if (unlikely(!netif_carrier_ok(dev) ||
(slots > 1 && !xennet_can_sg(dev)) ||
- netif_needs_gso(dev, skb, netif_skb_features(skb)))) {
+ netif_needs_gso(dev, skb, &features))) {
spin_unlock_irqrestore(&queue->tx_lock, flags);
goto drop;
}
diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index c31f74d76ebd..fb1f8c900df9 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -3608,13 +3608,19 @@ static inline bool skb_gso_ok(struct sk_buff *skb, netdev_features_t features)
}
static inline bool netif_needs_gso(struct net_device *dev, struct sk_buff *skb,
- netdev_features_t features)
+ netdev_features_t *features)
{
- return skb_is_gso(skb) && (!skb_gso_ok(skb, features) ||
- (dev->netdev_ops->ndo_gso_check &&
- !dev->netdev_ops->ndo_gso_check(skb, dev)) ||
- unlikely((skb->ip_summed != CHECKSUM_PARTIAL) &&
- (skb->ip_summed != CHECKSUM_UNNECESSARY)));
+ if (!skb_is_gso(skb))
+ return false;
+ if (!skb_gso_ok(skb, *features))
+ return true;
+ if (dev->netdev_ops->ndo_gso_check &&
+ !dev->netdev_ops->ndo_gso_check(skb, dev)) {
+ *features &= ~NETIF_F_GSO_MASK;
+ return true;
+ }
+ return skb->ip_summed != CHECKSUM_PARTIAL &&
+ skb->ip_summed != CHECKSUM_UNNECESSARY;
}
static inline void netif_set_gso_max_size(struct net_device *dev,
diff --git a/net/core/dev.c b/net/core/dev.c
index f411c28d0a66..b61c26b45bb7 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -2668,7 +2668,7 @@ static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device
if (skb->encapsulation)
features &= dev->hw_enc_features;
- if (netif_needs_gso(dev, skb, features)) {
+ if (netif_needs_gso(dev, skb, &features)) {
struct sk_buff *segs;
segs = skb_gso_segment(skb, features);
^ permalink raw reply related
* Re: [PATCH] sunvnet: fix a memory leak in vnet_handle_offloads
From: David Miller @ 2014-12-19 18:22 UTC (permalink / raw)
To: roy.qing.li; +Cc: netdev
In-Reply-To: <1418966375-23188-1-git-send-email-roy.qing.li@gmail.com>
From: roy.qing.li@gmail.com
Date: Fri, 19 Dec 2014 13:19:35 +0800
> From: Li RongQing <roy.qing.li@gmail.com>
>
> when skb_gso_segment returns error, the original skb should be freed
>
> Signed-off-by: Li RongQing <roy.qing.li@gmail.com>
Applied, thanks.
^ permalink raw reply
* Re: [bisected] tg3 broken in 3.18.0?
From: Rajat Jain @ 2014-12-19 18:24 UTC (permalink / raw)
To: Marcelo Ricardo Leitner
Cc: Bjorn Helgaas, Prashant Sreedharan, Nils Holland, Michael Chan,
David Miller, netdev, linux-pci@vger.kernel.org, Rafael Wysocki
In-Reply-To: <54945D79.2070008@gmail.com>
One of the reasons to replace the condition (*l == 0xffff0001) with
(*l & 0xffff) == 0x0001) was that some devices apparently returned
0001 for device id to indicate CRS, but returned actual vendor id in
the vendor ID field (hence the need to ignore vendor field).
Are we saying that the tg3 device returns 0x0001 for the device ID and
yet expects it to be treated as a good value (not CRS)?
Thanks,
Rajat
On Fri, Dec 19, 2014 at 9:16 AM, Marcelo Ricardo Leitner
<marcelo.leitner@gmail.com> wrote:
> On 19-12-2014 15:09, Bjorn Helgaas wrote:
>>
>> On Thu, Dec 18, 2014 at 7:10 PM, Prashant Sreedharan
>> <prashant@broadcom.com> wrote:
>>>
>>> On Thu, 2014-12-18 at 21:26 +0100, Nils Holland wrote:
>>>>
>>>> On Thu, Dec 18, 2014 at 11:28:09AM -0800, Prashant Sreedharan wrote:
>>>>>
>>>>> On Thu, 2014-12-18 at 12:15 -0700, Bjorn Helgaas wrote:
>>>>>>
>>>>>>
>>>>>> Any updates from the hardware team?
>>>>>>
>>>>>> This is a pretty serious regression, but as far as I can tell, it is
>>>>>> not a PCI bug. The device should respond to a config read of vendor
>>>>>> ID. If the driver does something that make the read return CRS
>>>>>> status, I think the driver is responsible for doing whatever delay or
>>>>>> other fixup is required.
>>>>>>
>>>>>> I'm inclined to reassign this bug to the tg3 driver unless you think
>>>>>> the PCI core is doing something wrong here.
>>>>>>
>>>>>> Bjorn
>>>>>
>>>>>
>>>>> We were not able to reproduce this issue, could you please check what
>>>>> is
>>>>> the value of reg 0x70, before the pci_device_is_present call is made ?
>>>>> if bit 15 is set config access will be retried.
>>>>>
>>>>> --- a/drivers/net/ethernet/broadcom/tg3.c
>>>>> +++ b/drivers/net/ethernet/broadcom/tg3.c
>>>>> @@ -9025,6 +9025,7 @@ static int tg3_chip_reset(struct tg3 *tp)
>>>>> void (*write_op)(struct tg3 *, u32, u32);
>>>>> int i, err;
>>>>>
>>>>> + printk(KERN_ERR "config state: %x\n", tr32(TG3PCI_PCISTATE));
>>>>> if (!pci_device_is_present(tp->pdev))
>>>>> return -ENODEV;
>>>>
>>>>
>>>> No problem, I gave this a try and here is what I get:
>>>>
>>>> [ 2.185190] libphy: tg3 mdio bus: probed
>>>> [ 2.229357] tsc: Refined TSC clocksource calibration: 2399.999 MHz
>>>> [ 2.244993] config state: 1292
>>>> [ 2.247136] tg3 0000:02:00.0 eth0: Tigon3 [partno(BCM57780) rev
>>>> 57780001]
>>>> (PCI Express) MAC address 00:19:99:ce:13:a6
>>>> [ 2.249279] tg3 0000:02:00.0 eth0: attached PHY driver [Broadcom
>>>> BCM57780]
>>>> (mii_bus:phy_addr=200:01)
>>>> [ 2.251460] tg3 0000:02:00.0 eth0: RXcsums[1] LinkChgREG[0]
>>>> MIirq[0] ASF[0] TSOcap[1]
>>>> [ 2.253672] tg3 0000:02:00.0 eth0: dma_rwctrl[76180000]
>>>> dma_mask[64-bit]
>>>> [...]
>>>> [ 12.204692] tg3 0000:02:00.0
>>>> enp2s0: No firmware running
>>>> [ 12.206653] config state: 1292
>>>> [ 12.208655] config state: 1292
>>>>
>>>> That's all of the three times the new debugging line gets hit when I
>>>> boot my system using the supplied diagnostic patch.
>>>>
>>>> Hope that helps - of course, I'd gladly test any further
>>>> (diagnostic) patches if required! Also, if I can provide any
>>>> additional information that might be of value, just ask:-)
>>>>
>>> Nils/Marcelo thanks for inputs, since reg 0x70 bit 15 is clear it
>>> indicates the chip is not setting the config retry bit. We were hoping
>>> this bit is causing the config access to return CRS but looks like it is
>>> not.
>>>
>>> Btw after forcing the error path (tg3_init_one -> tg3_halt) in the
>>> driver now we are able to reproduce the problem on 5722 in house. We are
>>> working with the HW team to narrow this down.
>>>
>>> Also it is not clear to me how reverting commit cfa6a7877b17a667 fixes
>>> the problem.
>>
>>
>> The full commit is 89665a6a71408796565bfd29cfa6a7877b17a667, and git
>> works with any unique *prefix* of that. The current convention is to
>> use the first 12 characters (I have "[core] abbrev = 12" in my
>> .git/config). Unfortunately, suffixes don't work at all.
>>
>> Anyway, here's why I think 89665a6a7140 makes a difference. We're in this
>> path:
>>
>> pci_device_is_present
>> pci_bus_read_dev_vendor_id(..., crs_timeout = 0)
>> pci_bus_read_config_dword(bus, devfn, PCI_VENDOR_ID, l)
>>
>> and for some reason the chip returns 0x00010001 for that 32-bit read.
>
>
> Actually it returns just 0x00000001, but yeah, that's my understanding too.
>
> Marcelo
>
>
>> Before 89665a6a7140, we compared all 32 bits with "*l == 0xffff0001".
>> This is false, so pci_bus_read_dev_vendor_id() returns true, which
>> means pci_device_is_present() is also true.
>>
>> After 89665a6a7140, we compare only the low 16 bits with ((*l &
>> 0xffff) == 0x0001), which is true, so pci_bus_read_dev_vendor_id()
>> returns false, and pci_device_is_present() is false.
>>
>> Bjorn
>>
>
^ permalink raw reply
* Re: [patches] a bunch of old bluetooth fixes
From: Marcel Holtmann @ 2014-12-19 18:25 UTC (permalink / raw)
To: David S. Miller; +Cc: viro, netdev, linux-bluetooth
In-Reply-To: <20141219.115933.2303527819596860834.davem@davemloft.net>
Hi Dave,
>>> Dave, we can prepare a pull request for these or do you want to take them directly into net tree?
>>
>> and in case you decide to take them directly.
>>
>> Acked-by: Marcel Holtmann <marcel@holtmann.org>
>
> Feel free to create a pull request for me, that works best.
actually Johan already sent one. Should be in your inbox and on netdev.
Regards
Marcel
^ permalink raw reply
* Re: [PATCH Iproute2 next] ip netns: 'ip netns id' cmd without argument should not give error
From: Mahesh Bandewar @ 2014-12-19 18:33 UTC (permalink / raw)
To: Vadim Kochan; +Cc: netdev, Stephen Hemminger, Saied Kazemi
In-Reply-To: <20141219083153.GA11360@angus-think>
On Fri, Dec 19, 2014 at 12:31 AM, Vadim Kochan <vadim4j@gmail.com> wrote:
> On Thu, Dec 18, 2014 at 10:05:38PM -0800, Mahesh Bandewar wrote:
>> The command 'ip netns identify' without PID parameter is supposed to
>> use the self PID to identify its ns but a trivial error prevented it
>> from doing so. This patch fixes that error.
>>
>> So before the patch -
>> # ip netns id
>> No pid specified
>> # echo $?
>> 1
>> #
>>
>> After the patch -
>> # ip netns id
>> test-ns
>> # echo $?
>> 0
>> #
>>
>> Signed-off-by: Mahesh Bandewar <maheshb@google.com>
>> Cc: Stephen Hemminger <stephen@networkplumber.org>
>> Cc: Vadim Kochan <vadim4j@gmail.com>
>> Cc: Saied Kazemi <saied@google.com>
>> ---
>> ip/ipnetns.c | 2 +-
>> 1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/ip/ipnetns.c b/ip/ipnetns.c
>> index 1c8aa029073e..ec9afa2177a5 100644
>> --- a/ip/ipnetns.c
>> +++ b/ip/ipnetns.c
>> @@ -298,7 +298,7 @@ static int netns_identify(int argc, char **argv)
>> DIR *dir;
>> struct dirent *entry;
>>
>> - if (argc < 1) {
>> + if (!argc) {
>> pidstr = "self";
>> } else if (argc > 1) {
>> fprintf(stderr, "extra arguments specified\n");
>> --
>> 2.2.0.rc0.207.ga3a616c
>>
>
> Hi Mahesh,
>
> I did not get such error on the current master branch?
>
> Did I miss something ?
>
No you did not! Please ignore this patch. :)
Looks like the tree I was working / using wasn't next and did not have
the patch that made it work (in next). I cherry-picked my fix onto the
next and sent it. My bad!
> Regards,
> Vadim Kochan
^ permalink raw reply
* Re: [bisected] tg3 broken in 3.18.0?
From: Prashant Sreedharan @ 2014-12-19 18:53 UTC (permalink / raw)
To: rajatxjain
Cc: Marcelo Ricardo Leitner, Bjorn Helgaas, Nils Holland,
Michael Chan, David Miller, netdev, linux-pci@vger.kernel.org,
Rafael Wysocki
In-Reply-To: <CAA93t1qOUZEmJEcuUNtkDS0M_pafb_1s==9Yqm9-+-+OmJmVuQ@mail.gmail.com>
On Fri, 2014-12-19 at 10:24 -0800, Rajat Jain wrote:
> One of the reasons to replace the condition (*l == 0xffff0001) with
> (*l & 0xffff) == 0x0001) was that some devices apparently returned
> 0001 for device id to indicate CRS, but returned actual vendor id in
> the vendor ID field (hence the need to ignore vendor field).
>
> Are we saying that the tg3 device returns 0x0001 for the device ID and
> yet expects it to be treated as a good value (not CRS)?
>
No it should not be treated as a good value, this commit has
surfaced/exposed a problem in 5722 chipset which was previously masked.
^ permalink raw reply
* RE: [PATCH net-next v3] Add support of Cavium Liquidio ethernet adapters
From: Chickles, Derek @ 2014-12-19 18:39 UTC (permalink / raw)
To: Stephen Hemminger, Vatsavayi, Raghu
Cc: davem@davemloft.net, netdev@vger.kernel.org, Burla, Satananda,
Manlunas, Felix, Vatsavayi, Raghu
In-Reply-To: <20141218232833.3e640164@urahara>
> > +
> > +static u32 lio_get_link(struct net_device *dev)
> > +{
> > + u32 ret;
> > +
> > + ret = netif_carrier_ok(dev) ? 1 : 0;
> > + return ret;
> > +}
> > +
>
> This is unnecessary, this is what the default ethtool handler already does
> if you give a NULL handler.
Okay
> > + if (linfo->link.s.status) {
> > + ecmd->speed = linfo->link.s.speed;
> > + ecmd->duplex = linfo->link.s.duplex;
> > + } else {
> > + ecmd->speed = -1;
> > + ecmd->duplex = -1;
> > + }
> > +
>
> You should SPEED_UNKNOWN and DUPLEX_UNKNOWN (not -1).
> and don't set ecmd->speed directly, use
> ethtool_cmd_speed_sed(ecmd, speed)
Okay
> > +static void
> > +lio_get_ethtool_stats(struct net_device *netdev,
> > + struct ethtool_stats *stats, u64 *data)
> > +{
> > + struct lio *lio = GET_LIO(netdev);
> > + struct octeon_device *oct_dev = lio->oct_dev;
> > + int i = 0, j;
> > +
> > + data[i++] = jiffies;
> > + data[i++] = lio->stats.tx_packets;
> > + data[i++] = lio->stats.tx_bytes;
> > + data[i++] = lio->stats.rx_packets;
> > + data[i++] = lio->stats.rx_bytes;
> > + data[i++] = lio->stats.tx_errors;
> > + data[i++] = lio->stats.tx_dropped;
> > + data[i++] = lio->stats.rx_dropped;
> > +
>
> Do not mirror existing netdevice statistics in ethtool.
> The purpose of ethtool stats is to provide device specific information.
Okay.
> > +/** \brief Network device get stats
> > + * @param netdev pointer to network device
> > + * @returns pointer stats structure
> > + */
> > +static struct net_device_stats *liquidio_stats(struct net_device *netdev)
> > +{
> > + return &(GET_LIO(netdev)->stats);
> > +}
> > +
>
> Unnecessary function, this is what network core does already
> if .ndo_get_stats is NULL;
Okay.
> > +static int liquidio_ioctl(struct net_device *netdev, struct ifreq *ifr, int
> cmd)
> > +{
> > + switch (cmd) {
> > + case SIOCSHWTSTAMP:
> > + return hwtstamp_ioctl(netdev, ifr, cmd);
> > + default:
> > + return -EINVAL
>
> Standard error code for unsupported ioctl is -EOPNOTSUPP
Okay.
> > +static int __init liquidio_init(void)
> > +{
> > + int i;
> > + struct handshake *hs;
> > +
> > + pr_info("LiquidIO: Starting Network module version %s\n",
> > + LIQUIDIO_VERSION);
> > +
>
> Do you really need to log this. Systems are already too chatty.
Okay. This is available with ethtool anyway.
> > +/**
> > + * \brief Alloc net device
> > + * @param size Size to allocate
> > + * @param nq how many queues
> > + * @returns pointer to net device structure
> > + */
> > +static inline struct net_device *liquidio_alloc_netdev(int size, int nq)
> > +{
> > + if (nq > 1)
> > + return alloc_netdev_mq(size, "lio%d",
> NET_NAME_UNKNOWN,
> > + ether_setup, nq);
> > + else
> > + return alloc_netdev(size, "lio%d", NET_NAME_UNKNOWN,
> > + ether_setup);
> > +}
>
> There is no need for the (nq > 1) test, just do:
> > + return alloc_netdev_mq(size, "lio%d",
> NET_NAME_UNKNOWN,
> > + ether_setup, nq);
>
Okay.
> Also, why does this device not have standard ethernet naming?
No good reason. We will follow the p<slot_number>p<port_number> convention.
> > +void process_noresponse_list(struct octeon_device *oct,
> > + struct octeon_instr_queue *iq)
> > +{
> > + uint32_t get_idx;
> > + struct octeon_soft_command *sc;
> > + struct octeon_instr_ih *ih;
>
> All global function names must start with same prefix, driver may not
> always be built as a module.
>
> In this case oceteon_process_noresponse_list.
Okay. All global symbols will get an octeon_ or lio_ prefix.
We'll reply to the last couple issues separately.
Thanks for all the feedback.
^ permalink raw reply
* Re: [bisected] tg3 broken in 3.18.0?
From: Rajat Jain @ 2014-12-19 19:37 UTC (permalink / raw)
To: Prashant Sreedharan
Cc: Marcelo Ricardo Leitner, Bjorn Helgaas, Nils Holland,
Michael Chan, David Miller, netdev, linux-pci@vger.kernel.org,
Rafael Wysocki
In-Reply-To: <1419015228.27773.3.camel@prashant>
On Fri, Dec 19, 2014 at 10:53 AM, Prashant Sreedharan
<prashant@broadcom.com> wrote:
> On Fri, 2014-12-19 at 10:24 -0800, Rajat Jain wrote:
>> One of the reasons to replace the condition (*l == 0xffff0001) with
>> (*l & 0xffff) == 0x0001) was that some devices apparently returned
>> 0001 for device id to indicate CRS, but returned actual vendor id in
>> the vendor ID field (hence the need to ignore vendor field).
>>
>> Are we saying that the tg3 device returns 0x0001 for the device ID and
>> yet expects it to be treated as a good value (not CRS)?
>>
> No it should not be treated as a good value, this commit has
> surfaced/exposed a problem in 5722 chipset which was previously masked.
>
Got it. Thanks.
I assume you mean its a HW issue, and the workaround if required shall
be taken care in tg3 driver.
Thanks,
Rajat
>
>
^ permalink raw reply
* Re: [PATCH] genirq: set initial affinity when hinting
From: Jesse Brandeburg @ 2014-12-19 19:53 UTC (permalink / raw)
To: Jesse Brandeburg
Cc: Thomas Gleixner, linux-kernel@vger.kernel.org, NetDEV list
In-Reply-To: <20141219012206.4220.27491.stgit@jbrandeb-cp2.jf.intel.com>
+netdev, as network driver developers might be interested in this functionality.
On Thu, Dec 18, 2014 at 5:22 PM, Jesse Brandeburg
<jesse.brandeburg@intel.com> wrote:
> Problem:
> The default behavior of the kernel is somewhat undesirable as all
> requested interrupts end up on CPU0 after registration. A user can
> run irqbalance daemon, or can manually configure smp_affinity via the
> proc filesystem, but the default affinity of the interrupts for all
> devices is always CPU zero, this can cause performance problems or
> very heavy cpu use of only one core if not noticed and fixed by the
> user.
>
> Patch:
> This patch enables the setting of the initial affinity directly
> when the driver sets a hint.
>
> This enabling means that kernel drivers can include an initial
> affinity setting for the interrupt, instead of all interrupts starting
> out life on CPU0. Of course if irqbalance is still running then the
> interrupts will get moved as before.
>
> This function is currently called by drivers in block, crypto,
> infiniband, ethernet and scsi trees, but only a handful, so these will
> be the devices affected by this change.
>
> Tested on i40e, and default interrupts were spread across the CPUs
> according to the hint.
>
> drivers/block/mtip32xx/mtip32xx.c:3
> drivers/block/nvme-core.c:2
> drivers/crypto/qat/qat_dh895xcc/adf_isr.c:3
> drivers/infiniband/hw/qib/qib_iba7322.c:2
> drivers/net/ethernet/intel/i40e/i40e_main.c:3
> drivers/net/ethernet/intel/i40evf/i40evf_main.c:3
> drivers/net/ethernet/intel/ixgbe/ixgbe_main.c:3
> drivers/net/ethernet/mellanox/mlx4/en_cq.c:2
> drivers/scsi/hpsa.c:3
> drivers/scsi/lpfc/lpfc_init.c:3
> drivers/scsi/megaraid/megaraid_sas_base.c:8
> drivers/soc/ti/knav_qmss_acc.c:1
> drivers/soc/ti/knav_qmss_queue.c:2
> drivers/virtio/virtio_pci_common.c:2
>
> Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
> CC: Thomas Gleixner <tglx@linutronix.de>
> ---
> kernel/irq/manage.c | 2 ++
> 1 file changed, 2 insertions(+)
>
> diff --git a/kernel/irq/manage.c b/kernel/irq/manage.c
> index 8069237..f038e58 100644
> --- a/kernel/irq/manage.c
> +++ b/kernel/irq/manage.c
> @@ -243,6 +243,8 @@ int irq_set_affinity_hint(unsigned int irq, const struct cpumask *m)
> return -EINVAL;
> desc->affinity_hint = m;
> irq_put_desc_unlock(desc, flags);
> + /* set the initial affinity to prevent every interrupt being on CPU0 */
> + __irq_set_affinity(irq, m, false);
> return 0;
> }
> EXPORT_SYMBOL_GPL(irq_set_affinity_hint);
>
^ permalink raw reply
* Re: [PATCH 01/10] core: Split out UFO6 support
From: Vlad Yasevich @ 2014-12-19 19:55 UTC (permalink / raw)
To: Michael S. Tsirkin
Cc: netdev, Vladislav Yasevich, virtualization, stefanha, ben,
David Miller
In-Reply-To: <20141218173546.GB5425@redhat.com>
On 12/18/2014 12:35 PM, Michael S. Tsirkin wrote:
> On Thu, Dec 18, 2014 at 10:01:35AM -0500, Vlad Yasevich wrote:
>> On 12/18/2014 02:54 AM, Michael S. Tsirkin wrote:
>>> Cc Dave, pls remember to do this next time otherwise
>>> your patches won't get merged :)
>>>
>>> On Wed, Dec 17, 2014 at 06:31:50PM -0500, Vlad Yasevich wrote:
>>>> On 12/17/2014 05:45 PM, Michael S. Tsirkin wrote:
>>>>> On Wed, Dec 17, 2014 at 01:20:46PM -0500, Vladislav Yasevich wrote:
>>>>>> Split IPv6 support for UFO into its own feature similiar to TSO.
>>>>>> This will later allow us to re-enable UFO support for virtio-net
>>>>>> devices.
>>>>>>
>>>>>> Signed-off-by: Vladislav Yasevich <vyasevic@redhat.com>
>>>>>> ---
>>>>>> include/linux/netdev_features.h | 7 +++++--
>>>>>> include/linux/netdevice.h | 1 +
>>>>>> include/linux/skbuff.h | 1 +
>>>>>> net/core/dev.c | 35 +++++++++++++++++++----------------
>>>>>> net/core/ethtool.c | 2 +-
>>>>>> 5 files changed, 27 insertions(+), 19 deletions(-)
>>>>>>
>>>>>> diff --git a/include/linux/netdev_features.h b/include/linux/netdev_features.h
>>>>>> index dcfdecb..a078945 100644
>>>>>> --- a/include/linux/netdev_features.h
>>>>>> +++ b/include/linux/netdev_features.h
>>>>>> @@ -48,8 +48,9 @@ enum {
>>>>>> NETIF_F_GSO_UDP_TUNNEL_BIT, /* ... UDP TUNNEL with TSO */
>>>>>> NETIF_F_GSO_UDP_TUNNEL_CSUM_BIT,/* ... UDP TUNNEL with TSO & CSUM */
>>>>>> NETIF_F_GSO_MPLS_BIT, /* ... MPLS segmentation */
>>>>>> + NETIF_F_UFO6_BIT, /* ... UDPv6 fragmentation */
>>>>>> /**/NETIF_F_GSO_LAST = /* last bit, see GSO_MASK */
>>>>>> - NETIF_F_GSO_MPLS_BIT,
>>>>>> + NETIF_F_UFO6_BIT,
>>>>>>
>>>>>> NETIF_F_FCOE_CRC_BIT, /* FCoE CRC32 */
>>>>>> NETIF_F_SCTP_CSUM_BIT, /* SCTP checksum offload */
>>>>>> @@ -109,6 +110,7 @@ enum {
>>>>>> #define NETIF_F_TSO_ECN __NETIF_F(TSO_ECN)
>>>>>> #define NETIF_F_TSO __NETIF_F(TSO)
>>>>>> #define NETIF_F_UFO __NETIF_F(UFO)
>>>>>> +#define NETIF_F_UFO6 __NETIF_F(UFO6)
>>>>>> #define NETIF_F_VLAN_CHALLENGED __NETIF_F(VLAN_CHALLENGED)
>>>>>> #define NETIF_F_RXFCS __NETIF_F(RXFCS)
>>>>>> #define NETIF_F_RXALL __NETIF_F(RXALL)
>>>>>> @@ -141,7 +143,7 @@ enum {
>>>>>>
>>>>>> /* List of features with software fallbacks. */
>>>>>> #define NETIF_F_GSO_SOFTWARE (NETIF_F_TSO | NETIF_F_TSO_ECN | \
>>>>>> - NETIF_F_TSO6 | NETIF_F_UFO)
>>>>>> + NETIF_F_TSO6 | NETIF_F_UFO | NETIF_F_UFO6)
>>>>>>
>>>>>> #define NETIF_F_GEN_CSUM NETIF_F_HW_CSUM
>>>>>> #define NETIF_F_V4_CSUM (NETIF_F_GEN_CSUM | NETIF_F_IP_CSUM)
>>>>>> @@ -149,6 +151,7 @@ enum {
>>>>>> #define NETIF_F_ALL_CSUM (NETIF_F_V4_CSUM | NETIF_F_V6_CSUM)
>>>>>>
>>>>>> #define NETIF_F_ALL_TSO (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN)
>>>>>> +#define NETIF_F_ALL_UFO (NETIF_F_UFO | NETIF_F_UFO6)
>>>>>>
>>>>>> #define NETIF_F_ALL_FCOE (NETIF_F_FCOE_CRC | NETIF_F_FCOE_MTU | \
>>>>>> NETIF_F_FSO)
>>>>>> diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
>>>>>> index 74fd5d3..86af10a 100644
>>>>>> --- a/include/linux/netdevice.h
>>>>>> +++ b/include/linux/netdevice.h
>>>>>> @@ -3559,6 +3559,7 @@ static inline bool net_gso_ok(netdev_features_t features, int gso_type)
>>>>>> /* check flags correspondence */
>>>>>> BUILD_BUG_ON(SKB_GSO_TCPV4 != (NETIF_F_TSO >> NETIF_F_GSO_SHIFT));
>>>>>> BUILD_BUG_ON(SKB_GSO_UDP != (NETIF_F_UFO >> NETIF_F_GSO_SHIFT));
>>>>>> + BUILD_BUG_ON(SKB_GSO_UDP6 != (NETIF_F_UFO6 >> NETIF_F_GSO_SHIFT));
>>>>>> BUILD_BUG_ON(SKB_GSO_DODGY != (NETIF_F_GSO_ROBUST >> NETIF_F_GSO_SHIFT));
>>>>>> BUILD_BUG_ON(SKB_GSO_TCP_ECN != (NETIF_F_TSO_ECN >> NETIF_F_GSO_SHIFT));
>>>>>> BUILD_BUG_ON(SKB_GSO_TCPV6 != (NETIF_F_TSO6 >> NETIF_F_GSO_SHIFT));
>>>>>> diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
>>>>>> index 6c8b6f6..8538b67 100644
>>>>>> --- a/include/linux/skbuff.h
>>>>>> +++ b/include/linux/skbuff.h
>>>>>> @@ -372,6 +372,7 @@ enum {
>>>>>>
>>>>>> SKB_GSO_MPLS = 1 << 12,
>>>>>>
>>>>>> + SKB_GSO_UDP6 = 1 << 13
>>>>>> };
>>>>>>
>>>>>> #if BITS_PER_LONG > 32
>>>>>
>>>>> So this implies anything getting GSO packets e.g.
>>>>> from userspace now needs to check IP version to
>>>>> set GSO type correctly.
>>>>>
>>>>> I think you missed some places that do this, e.g. af_packet
>>>>> sockets.
>>>>>
>>>>
>>>> I looked at af_packet sockets and they set this only in the event
>>>> vnet header has been used with a GSO type. In this case, the user
>>>> already knows the the type.
>>>
>>> Imagine you are receiving a packet:
>>>
>>> if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
>>> switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
>>> case VIRTIO_NET_HDR_GSO_TCPV4:
>>> gso_type = SKB_GSO_TCPV4;
>>> break;
>>> case VIRTIO_NET_HDR_GSO_TCPV6:
>>> gso_type = SKB_GSO_TCPV6;
>>> break;
>>> case VIRTIO_NET_HDR_GSO_UDP:
>>> gso_type = SKB_GSO_UDP;
>>> break;
>>> default:
>>> goto out_unlock;
>>> }
>>>
>>> if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
>>> gso_type |= SKB_GSO_TCP_ECN;
>>>
>>> if (vnet_hdr.gso_size == 0)
>>> goto out_unlock;
>>>
>>> }
>>>
>>> we used to report UFO6 as SKB_GSO_UDP, we probably
>>> should keep doing this, with your patch we select the
>>> goto out_unlock path.
>>>
>>>
>>
>> No. The vnet_hdr.gso_type is still going to be VIRTIO_NET_HDR_GSO_UDP
>> since the UDP6 version isn't defined yet. So, it will be marked as
>> GSO_UDP.
>
> I pasted the wrong snippet.
> I meant this:
>
> /* This is a hint as to how much should be * linear. */
> vnet_hdr.hdr_len = __cpu_to_virtio16(false, skb_headlen(skb));
> vnet_hdr.gso_size = __cpu_to_virtio16(false, sinfo->gso_size);
> if (sinfo->gso_type & SKB_GSO_TCPV4)
> vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
> else if (sinfo->gso_type & SKB_GSO_TCPV6)
> vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
> else if (sinfo->gso_type & SKB_GSO_UDP)
> vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
> else if (sinfo->gso_type & SKB_GSO_FCOE)
> goto out_free;
> else
> BUG();
>
> so if we get SKB_GSO_UDP we'll get BUG().
>
>
>
>> This code most likely needs the same workaround as exists in tap and macvtap,
>> i.e select the proxy fragment id and decide what to do with gso_type.
>
> And fixup type to GSO_UDP6 while we are at it.
>
>>>
>>>> It is true that with this series af_packets now can't do IPv6 UFO
>>>> since there is no VIRTIO_NET_HDR_GSO_UDPV6 yet.
>>>
>>> What do you mean by "do".
>>
>> What I mean is that AF_PACKET sockets currently do not do IPv6 UFO
>> correctly, even after Ben's fixes to tap/macvtap. There is no
>> proxy fragment id selection in af_packet case and we have the
>> same problem Ben was trying address for tap/macvtap.
>>> Are we talking about sending or receiving packets?
>>
>> I am talking about sending, see above.
>>
>>> You seem to conflate the two.
>>>
>>> We always discarded ID on RX.
>>>
>>> For tun, this is xmit, so just by saying "this device can
>>> not do UFO" you will never get short packets.
>>>
>>
>> You must mean long packets.
>
> Yes.
>
>> This is actually an issue I've been thinking
>> about. With with your suggestion of switching the GSO type for legacy
>> applications we end up with fragments for IPv6 traffic. As a result,
>> legacy VMs will see a regression for large IPv6 datagrams.
>
> I'm not sure what's meant by my suggestion here :)
What I am referring to here is to change the gso_type to UDPV6 in tap/macvtap
when passing packet to the host.
> It seems clear that legacy applications don't want to get IPv6
> fragment IDs in virtio header. Either we pass them plain ethernet
> packets or assume they are ok with discarding the IDs even
> if we set GSO_UDP.
So, I am not try to pass fragment IDs yet. I am trying to make sure that
older gueest that assume that UFO == UFO4 + UFO6 continue to work and do not
regress. At the same time, I want to enable UFO(4) for new guests.
If we set gso_type to SKB_GSO_UDPV6 before passing the data to the forwarding
path of the host, then this will cause IPv6 fragmentation. If we leave gso_type
as SKB_GSO_UDP, then older VMs will continue to communicate using large UDP
data packets.
Later when VMs support UFO6, when UFO6 is enabled, the fragment id can be communicated.
But that's later.
>
>>>
>>>>
>>>> I suppose we could do something similar there as we do in tun code/macvtap code.
>>>> If that's the case, it currently broken as well.
>>>>
>>>> -vlad
>>>
>>>
>>> Broken is a big word.
>>>
>>> Let's stop conflating two directions.
>>
>> I am not and was talking only about af_packet as that is what you asked about.
>> There is no tun/macvtap in play here. They are handled separately in their
>> respective drivers.
>>
>>>
>>> Here's the way I look at it:
>>>
>>> 1. Userspace doesn't have a way to get fragment IDs
>>> from tun/macvtap/packet sockets.
>>> Presumably, not all users need these IDs.
>>> E.g. old guests clearly don't.
>>>
>>> We should either give them packets stripping the ID,
>>> like we always did, or make sure they never get these packets.
>>> Second option seems achievable for tun if we
>>> never tell linux it can do UFO, but I don't see
>>> how to do it for macvtap and packet socket.
>>>
>>
>> macvtap is slightly problematic, but doable with some tricks.
>> packet socket is an interesting problem. The only way
>> an af_packet socket can receive an skb marked SKB_GSO_UDPV6
>> is if someone else on the host sent it (like another guest).
>
> Or if a NIC set it.
OK, virtio seems to be the only nic doing this... You are right, I
need to cover that scenario.
-vlad
>
>> Since there is are no feature or vnet header length negotiations,
>> it is impossible to tell if an application using this af_packet
>> socket is capable of processing VIRTIO_NET_HDR_GSO_UDPV6
>> type (yet to be defined).
>>
>> So, we can either use existing VIRTIO_NET_HDR_GSO_UDP on receive
>> path, add some kind of negotiation logic to packet socket (like
>> storing the application expected vnet header size), or perform
>> IPv6 fragmentation somehow.
>>
>> Options 1 or 2 are doable.
>
> 1 is using VIRTIO_NET_HDR_GSO_UDP and discarding ID,
> 2 is "some kind of negotiation logic"?
> 2 can't be enough, you will need 1 as well.
>
> So let's start with 1 as a first step.
>
>
>
>
>>>
>>> 2. Userspace doesn't have a way to set fragment IDs
>>> for tun/macvtap/packet sockets.
>>> Presumably, not all users have these IDs. E.g. old
>>> guests clearly don't.
>>>
>>> We should either generate our own ID,
>>> like we always did, or make sure we don't accept
>>> these packets.
>>> Second option is almost sure to break userspace,
>>> so it seems we must do the first one.
>>>
>>
>> Right. This was missing from packet sockets. I can fix it.
>>
>> -vlad
>
> Also, this can't be a patch on top, since we don't
> want bisect to give us configurations which
> can BUG().
>
>
>>>
>>> 3.
>>> Exactly the same two things if you replace userspace
>>> with hypervisor and tun/macvtap/packet socket with
>>> virtio device.
>>>
>>>
>>> 4.
>>> As a next step, we should add a way for userspace
>>> to tell us that it has ids and can handle them.
>>>
>>>
>>>
>>>
>>>
>>>>>
>>>>>> diff --git a/net/core/dev.c b/net/core/dev.c
>>>>>> index 945bbd0..fa4d2ee 100644
>>>>>> --- a/net/core/dev.c
>>>>>> +++ b/net/core/dev.c
>>>>>> @@ -5929,6 +5929,12 @@ static netdev_features_t netdev_fix_features(struct net_device *dev,
>>>>>> features &= ~NETIF_F_ALL_TSO;
>>>>>> }
>>>>>>
>>>>>> + /* UFO requires that SG is present as well */
>>>>>> + if ((features & NETIF_F_ALL_UFO) && !(features & NETIF_F_SG)) {
>>>>>> + netdev_dbg(dev, "Dropping UFO features since no SG feature.\n");
>>>>>> + features &= ~NETIF_F_ALL_UFO;
>>>>>> + }
>>>>>> +
>>>>>> if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
>>>>>> !(features & NETIF_F_IP_CSUM)) {
>>>>>> netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
>>>>>> @@ -5952,24 +5958,21 @@ static netdev_features_t netdev_fix_features(struct net_device *dev,
>>>>>> features &= ~NETIF_F_GSO;
>>>>>> }
>>>>>>
>>>>>> - /* UFO needs SG and checksumming */
>>>>>> - if (features & NETIF_F_UFO) {
>>>>>> - /* maybe split UFO into V4 and V6? */
>>>>>> - if (!((features & NETIF_F_GEN_CSUM) ||
>>>>>> - (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
>>>>>> - == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
>>>>>> - netdev_dbg(dev,
>>>>>> - "Dropping NETIF_F_UFO since no checksum offload features.\n");
>>>>>> - features &= ~NETIF_F_UFO;
>>>>>> - }
>>>>>> -
>>>>>> - if (!(features & NETIF_F_SG)) {
>>>>>> - netdev_dbg(dev,
>>>>>> - "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
>>>>>> - features &= ~NETIF_F_UFO;
>>>>>> - }
>>>>>> + /* UFO also needs checksumming */
>>>>>> + if ((features & NETIF_F_UFO) && !(features & NETIF_F_GEN_CSUM) &&
>>>>>> + !(features & NETIF_F_IP_CSUM)) {
>>>>>> + netdev_dbg(dev,
>>>>>> + "Dropping NETIF_F_UFO since no checksum offload features.\n");
>>>>>> + features &= ~NETIF_F_UFO;
>>>>>> + }
>>>>>> + if ((features & NETIF_F_UFO6) && !(features & NETIF_F_GEN_CSUM) &&
>>>>>> + !(features & NETIF_F_IPV6_CSUM)) {
>>>>>> + netdev_dbg(dev,
>>>>>> + "Dropping NETIF_F_UFO6 since no checksum offload features.\n");
>>>>>> + features &= ~NETIF_F_UFO6;
>>>>>> }
>>>>>>
>>>>>> +
>>>>>> #ifdef CONFIG_NET_RX_BUSY_POLL
>>>>>> if (dev->netdev_ops->ndo_busy_poll)
>>>>>> features |= NETIF_F_BUSY_POLL;
>>>>>> diff --git a/net/core/ethtool.c b/net/core/ethtool.c
>>>>>> index 06dfb29..93eff41 100644
>>>>>> --- a/net/core/ethtool.c
>>>>>> +++ b/net/core/ethtool.c
>>>>>> @@ -223,7 +223,7 @@ static netdev_features_t ethtool_get_feature_mask(u32 eth_cmd)
>>>>>> return NETIF_F_ALL_TSO;
>>>>>> case ETHTOOL_GUFO:
>>>>>> case ETHTOOL_SUFO:
>>>>>> - return NETIF_F_UFO;
>>>>>> + return NETIF_F_ALL_UFO;
>>>>>> case ETHTOOL_GGSO:
>>>>>> case ETHTOOL_SGSO:
>>>>>> return NETIF_F_GSO;
>>>>>> --
>>>>>> 1.9.3
^ permalink raw reply
* Re: [PATCH 01/10] core: Split out UFO6 support
From: Vlad Yasevich @ 2014-12-19 20:13 UTC (permalink / raw)
To: Michael S. Tsirkin
Cc: netdev, Vladislav Yasevich, virtualization, stefanha, ben,
David Miller
In-Reply-To: <20141218175002.GA5539@redhat.com>
On 12/18/2014 12:50 PM, Michael S. Tsirkin wrote:
> On Thu, Dec 18, 2014 at 07:35:46PM +0200, Michael S. Tsirkin wrote:
>>>> We should either generate our own ID,
>>>> like we always did, or make sure we don't accept
>>>> these packets.
>>>> Second option is almost sure to break userspace,
>>>> so it seems we must do the first one.
>>>>
>>>
>>> Right. This was missing from packet sockets. I can fix it.
>>>
>>> -vlad
>>
>> Also, this can't be a patch on top, since we don't
>> want bisect to give us configurations which
>> can BUG().
>
> So how doing this in stages:
>
> 1. add helper that checks skb GSO type.
> If it is SKB_GSO_UDP, check for IPv6, and
> generate the fragment ID.
>
> Call this helper in
> - virtio net rx path
Why do we need id on rx path? Fragment ID should only be generated on tx.
> - tun rx path (get user)
> - macvtap tx patch (get user)
> - packet socket tx patch (get user)
The reset makes sense.
>
> 2. Put back UFO flag everywhere: virtio,tun,macvtap,packet,
> since we can handle IPv6 now even if it's suboptimal.
>
> Above two seem like smalling patches, appropriate for stable.
OK.
>
> Next, work on a new feature to pass
> fragment ID in virtio header:
>
> 3. split UFO/UFO6 as you did here, but add code
> in above 4 places to convert UDP to UDP6,
Doing so will adversely impact IPv6 UFO performance for legacy
guests. I know how many times I've seen mail wrt patch xyz caused
%X regression in my setup and how we've reverted or tried to fixed
things to solve this. If we go with approach, the only "fix' would be
to upgrade the guest and that's not available to some users.
-vlad
> additionally, add code in
> - virtio net tx path
> - tun tx path (get user)
> - macvtap rx patch (put user)
> - packet socket rx patch (put user)
> to convert UDP6 to UDP.
>
> step 3 needs to be bisect-clean, somehow.
>
> 4. add new field in header, new feature bit for virtio net to gate it,
> new ioctls to tun,macvtap,packet socket.
>
> These two are more like optimizations, so not stable material.
>
>
^ permalink raw reply
* Re: [PATCH net-next 2/2] r8152: check the status before submitting rx
From: David Miller @ 2014-12-19 20:44 UTC (permalink / raw)
To: hayeswang; +Cc: netdev, nic_swsd, linux-kernel, linux-usb
In-Reply-To: <1394712342-15778-109-Taiwan-albertk@realtek.com>
From: Hayes Wang <hayeswang@realtek.com>
Date: Fri, 19 Dec 2014 16:56:00 +0800
> Don't submit the rx if the device is unplugged, linking down,
> or stopped.
...
> @@ -1789,6 +1789,11 @@ int r8152_submit_rx(struct r8152 *tp, struct rx_agg *agg, gfp_t mem_flags)
> {
> int ret;
>
> + /* The rx would be stopped, so skip submitting */
> + if (test_bit(RTL8152_UNPLUG, &tp->flags) ||
> + !test_bit(WORK_ENABLE, &tp->flags) || !(tp->speed & LINK_STATUS))
> + return 0;
> +
I think netif_carrier_off() should always be true in all three of those
situations, and would be a much simpler test than what you've coded
here.
^ permalink raw reply
* Re: [PATCH net-next v3] Add support of Cavium Liquidio ethernet adapters
From: Lino Sanfilippo @ 2014-12-19 20:44 UTC (permalink / raw)
To: Raghu Vatsavayi, davem
Cc: netdev, Derek Chickles, Satanand Burla, Felix Manlunas,
Raghu Vatsavayi
In-Reply-To: <1418959519-31681-1-git-send-email-rvatsavayi@caviumnetworks.com>
Hi,
On 19.12.2014 04:25, Raghu Vatsavayi wrote:
> +
> +static int cn6xxx_soft_reset(struct octeon_device *oct)
> +{
> + octeon_write_csr64(oct, CN66XX_WIN_WR_MASK_REG, 0xFF);
> +
> + lio_dev_dbg(oct, "BIST enabled for soft reset\n");
> +
> + OCTEON_PCI_WIN_WRITE(oct, CN66XX_CIU_SOFT_BIST, 1);
> + octeon_write_csr64(oct, CN66XX_SLI_SCRATCH1, 0x1234ULL);
> +
> + OCTEON_PCI_WIN_READ(oct, CN66XX_CIU_SOFT_RST);
> + OCTEON_PCI_WIN_WRITE(oct, CN66XX_CIU_SOFT_RST, 1);
> +
> + /* Wait for 10ms as Octeon resets. */
> + mdelay(10);
> +
> + if (octeon_read_csr64(oct, CN66XX_SLI_SCRATCH1) == 0x1234ULL) {
> + lio_dev_err(oct, "Soft reset failed\n");
> + return 1;
> + }
Before the delay you should probably make sure that the writes are
flushed to avoid pci write posting.
> +
> +static int cn68xx_soft_reset(struct octeon_device *oct)
> +{
> + octeon_write_csr64(oct, CN68XX_WIN_WR_MASK_REG, 0xFF);
> +
> + lio_dev_dbg(oct, "BIST enabled for CN68XX soft reset\n");
> + OCTEON_PCI_WIN_WRITE(oct, CN68XX_CIU_SOFT_BIST, 1);
> +
> + octeon_write_csr64(oct, CN68XX_SLI_SCRATCH1, 0x1234ULL);
> +
> + OCTEON_PCI_WIN_READ(oct, CN68XX_CIU_SOFT_RST);
> + OCTEON_PCI_WIN_WRITE(oct, CN68XX_CIU_SOFT_RST, 1);
> +
> + /* Wait for 100ms as Octeon resets. */
> + mdelay(100);
> +
> + if (octeon_read_csr64(oct, CN68XX_SLI_SCRATCH1) == 0x1234ULL) {
> + lio_dev_err(oct, "Soft reset failed\n");
> + return 1;
> + }
same here.
Regards,
Lino
^ permalink raw reply
* Re: [PATCH net] enic: fix rx skb checksum
From: David Miller @ 2014-12-19 20:45 UTC (permalink / raw)
To: jbenc; +Cc: _govind, netdev, ssujith, benve, sassmann
In-Reply-To: <20141219121144.5f9899ba@griffin>
From: Jiri Benc <jbenc@redhat.com>
Date: Fri, 19 Dec 2014 12:11:44 +0100
> On Thu, 18 Dec 2014 15:58:42 +0530, Govindarajulu Varadarajan wrote:
>> Hardware always provides compliment of IP pseudo checksum. Stack expects
>> whole packet checksum without pseudo checksum if CHECKSUM_COMPLETE is set.
>>
>> This causes checksum error in nf & ovs.
>>
>> [...]
>>
>> Hardware verifies IP & tcp/udp header checksum but does not provide payload
>> checksum, use CHECKSUM_UNNECESSARY. Set it only if its valid IP tcp/udp packet.
>>
>> Cc: Jiri Benc <jbenc@redhat.com>
>> Cc: Stefan Assmann <sassmann@redhat.com>
>> Reported-by: Sunil Choudhary <schoudha@redhat.com>
>> Signed-off-by: Govindarajulu Varadarajan <_govind@gmx.com>
>
> Reviewed-by: Jiri Benc <jbenc@redhat.com>
Applied, thanks.
^ permalink raw reply
* Re: pull request: bluetooth 2014-12-19
From: David Miller @ 2014-12-19 20:47 UTC (permalink / raw)
To: johan.hedberg; +Cc: netdev, linux-bluetooth
In-Reply-To: <20141219143400.GA20385@t440s.P-661HNU-F1>
From: Johan Hedberg <johan.hedberg@gmail.com>
Date: Fri, 19 Dec 2014 16:34:00 +0200
> Here's one more pull request for 3.19. It contains the socket type
> verification fixes from Al Viro as well as an skb double-free fix for
> 6lowpan from Jukka Rissanen.
>
> Please let me know if there are any issues pulling. Thanks.
Pulled, thanks Johan.
^ permalink raw reply
* Re: [PATCH net-next v3] Add support of Cavium Liquidio ethernet adapters
From: Stephen Hemminger @ 2014-12-19 20:53 UTC (permalink / raw)
To: Chickles, Derek
Cc: Vatsavayi, Raghu, davem@davemloft.net, netdev@vger.kernel.org,
Burla, Satananda, Manlunas, Felix
In-Reply-To: <CO2PR07MB490962E1E05D146BFD67B66FE6B0@CO2PR07MB490.namprd07.prod.outlook.com>
On Fri, 19 Dec 2014 18:39:44 +0000
"Chickles, Derek" <Derek.Chickles@caviumnetworks.com> wrote:
> > Also, why does this device not have standard ethernet naming?
>
> No good reason. We will follow the p<slot_number>p<port_number> convention.
Use alloc_etherdev_mq and let it choose.
Policy about slot/port is done in userspace (udev/systemd).
^ permalink raw reply
* Re: [PATCH net-next v3] Add support of Cavium Liquidio ethernet adapters
From: Stephen Hemminger @ 2014-12-19 20:54 UTC (permalink / raw)
To: Lino Sanfilippo
Cc: Raghu Vatsavayi, davem, netdev, Derek Chickles, Satanand Burla,
Felix Manlunas, Raghu Vatsavayi
In-Reply-To: <54948E1F.2080405@gmx.de>
On Fri, 19 Dec 2014 21:44:15 +0100
Lino Sanfilippo <LinoSanfilippo@gmx.de> wrote:
> > + OCTEON_PCI_WIN_READ(oct, CN66XX_CIU_SOFT_RST);
> > + OCTEON_PCI_WIN_WRITE(oct, CN66XX_CIU_SOFT_RST, 1);
Missing a following PCI read to force PCI posting?
^ permalink raw reply
* Re: [PATCH net-next v3] Add support of Cavium Liquidio ethernet adapters
From: Lino Sanfilippo @ 2014-12-19 21:00 UTC (permalink / raw)
To: Stephen Hemminger
Cc: Raghu Vatsavayi, davem, netdev, Derek Chickles, Satanand Burla,
Felix Manlunas, Raghu Vatsavayi
In-Reply-To: <20141219125434.594f4d46@urahara>
On 19.12.2014 21:54, Stephen Hemminger wrote:
> On Fri, 19 Dec 2014 21:44:15 +0100
> Lino Sanfilippo <LinoSanfilippo@gmx.de> wrote:
>
>> > + OCTEON_PCI_WIN_READ(oct, CN66XX_CIU_SOFT_RST);
>> > + OCTEON_PCI_WIN_WRITE(oct, CN66XX_CIU_SOFT_RST, 1);
>
> Missing a following PCI read to force PCI posting?
>
Yes, I think so.
^ permalink raw reply
* Re: [PATCH net-next v3] Add support of Cavium Liquidio ethernet adapters
From: Lino Sanfilippo @ 2014-12-19 21:07 UTC (permalink / raw)
To: Stephen Hemminger
Cc: Raghu Vatsavayi, davem, netdev, Derek Chickles, Satanand Burla,
Felix Manlunas, Raghu Vatsavayi
In-Reply-To: <549491D2.1090504@gmx.de>
On 19.12.2014 22:00, Lino Sanfilippo wrote:
> On 19.12.2014 21:54, Stephen Hemminger wrote:
>> On Fri, 19 Dec 2014 21:44:15 +0100
>> Lino Sanfilippo <LinoSanfilippo@gmx.de> wrote:
>>
>>> > + OCTEON_PCI_WIN_READ(oct, CN66XX_CIU_SOFT_RST);
>>> > + OCTEON_PCI_WIN_WRITE(oct, CN66XX_CIU_SOFT_RST, 1);
>>
>> Missing a following PCI read to force PCI posting?
>>
>
> Yes, I think so.
I thought posting means "writes are delayed", so I wrote it should be
avoided (instead of forced). However I think there should be a PCI read
before the call to mdelay.
Regards,
Lino
^ permalink raw reply
* Re: [PATCH net-next v3] Add support of Cavium Liquidio ethernet adapters
From: Lino Sanfilippo @ 2014-12-19 21:14 UTC (permalink / raw)
To: Raghu Vatsavayi, davem
Cc: netdev, Derek Chickles, Satanand Burla, Felix Manlunas,
Raghu Vatsavayi
In-Reply-To: <1418959519-31681-1-git-send-email-rvatsavayi@caviumnetworks.com>
On 19.12.2014 04:25, Raghu Vatsavayi wrote:
> desc_ring[i].info_ptr =
> + (uint64_t)pci_map_single(oct->pci_dev,
> + &droq->info_list[i],
> + OCT_DROQ_INFO_SIZE,
> + PCI_DMA_FROMDEVICE);
> +
> + desc_ring[i].buffer_ptr =
> + (uint64_t)pci_map_single(oct->pci_dev,
> + droq->recv_buf_list[i].
> + data, droq->buffer_size,
> + PCI_DMA_FROMDEVICE);
> + }
You should also check if pci mapping was successful...
Regards,
Lino
^ permalink raw reply
* Re: [PATCH net-next v3] Add support of Cavium Liquidio ethernet adapters
From: Lino Sanfilippo @ 2014-12-19 21:48 UTC (permalink / raw)
To: Raghu Vatsavayi, davem
Cc: netdev, Derek Chickles, Satanand Burla, Felix Manlunas,
Raghu Vatsavayi
In-Reply-To: <1418959519-31681-1-git-send-email-rvatsavayi@caviumnetworks.com>
On 19.12.2014 04:25, Raghu Vatsavayi wrote:
> +static int setup_nic_devices(struct octeon_device *octeon_dev)
> +{
> + struct lio *lio = NULL;
> + struct net_device *netdev;
> + uint8_t macaddr[6], i, j;
> + int octeon_id;
> + struct octeon_soft_command *sc;
> + struct liquidio_if_cfg_resp *resp;
> + int retval;
> + int num_iqueues;
> + int num_oqueues;
> + int q_no;
> + uint64_t q_mask;
> + int num_cpus = num_online_cpus();
> +
> + if (num_cpus & (num_cpus - 1))
> + /* numcpus is not a power of 2 */
> + num_cpus = 1U << (fls(num_cpus) - 1);
> + octeon_id = octeon_dev->octeon_id;
> +
> + sc = kmalloc(sizeof(*sc), GFP_ATOMIC);
> + if (!sc)
> + return -ENOMEM;
> +
> + resp = kmalloc(sizeof(*resp), GFP_KERNEL);
> + if (!resp) {
> + kfree(sc);
> + return -ENOMEM;
> + }
> +
> + for (i = 0; i < octeon_dev->props.ifcount; i++) {
> + memset(sc, 0, sizeof(struct octeon_soft_command));
> + memset(resp, 0, sizeof(struct liquidio_if_cfg_resp));
> + num_iqueues =
> + CFG_GET_NUM_TXQS_NIC_IF(octeon_get_conf(octeon_dev), i);
> + num_oqueues =
> + CFG_GET_NUM_RXQS_NIC_IF(octeon_get_conf(octeon_dev), i);
> + if (num_iqueues > num_cpus)
> + num_iqueues = num_cpus;
> + if (num_oqueues > num_cpus)
> + num_oqueues = num_cpus;
> + lio_dev_dbg(octeon_dev,
> + "requesting config for interface %d, iqs %d, oqs %d\n",
> + i, num_iqueues, num_oqueues);
> + ACCESS_ONCE(resp->s.cond) = 0;
> + resp->s.octeon_id = get_octeon_device_id(octeon_dev);
> + init_waitqueue_head(&resp->s.wc);
> +
> + octeon_prepare_soft_command(octeon_dev, sc, OPCODE_NIC,
> + OPCODE_NIC_IF_CFG, i, num_iqueues,
> + num_oqueues, NULL, 0, &resp->rh,
> + (sizeof(struct liquidio_if_cfg_resp)
> + - sizeof(resp->s)));
> +
> + sc->callback = if_cfg_callback;
> + sc->callback_arg = resp;
> + sc->wait_time = 1000;
> +
> + retval = octeon_send_soft_command(octeon_dev, sc);
> + if (retval) {
> + lio_dev_err(octeon_dev,
> + "Link status instruction failed status: %x\n",
> + retval);
> + /* Soft instr is freed by driver in case of failure. */
> + goto setup_nic_dev_fail;
> + }
> +
> + /* Sleep on a wait queue till the cond flag indicates that the
> + * response arrived or timed-out.
> + */
> + sleep_cond(&resp->s.wc, &resp->s.cond);
> + retval = resp->status;
> + if (retval) {
> + lio_dev_err(octeon_dev, "Link status failed\n");
> + goto setup_nic_dev_fail;
> + }
> + octeon_swap_8B_data((uint64_t *)(&resp->cfg_info),
> + (sizeof(struct liquidio_if_cfg_info)) >> 3);
> +
> + num_iqueues = hweight64(resp->cfg_info.iqmask);
> + num_oqueues = hweight64(resp->cfg_info.oqmask);
> +
> + if (!(num_iqueues) || !(num_oqueues)) {
> + lio_dev_err(octeon_dev,
> + "Got bad iqueues (%016llx) or oqueues (%016llx) from firmware.\n",
> + resp->cfg_info.iqmask,
> + resp->cfg_info.oqmask);
> + goto setup_nic_dev_fail;
> + }
> + lio_dev_dbg(octeon_dev,
> + "interface %d, iqmask %016llx, oqmask %016llx, numiqueues %d, numoqueues %d\n",
> + i, resp->cfg_info.iqmask, resp->cfg_info.oqmask,
> + num_iqueues, num_oqueues);
> + netdev = liquidio_alloc_netdev(LIO_SIZE, num_iqueues);
> +
> + if (!netdev) {
> + lio_dev_err(octeon_dev, "Device allocation failed\n");
> + goto setup_nic_dev_fail;
> + }
> +
> + octeon_dev->props.netdev[i] = netdev;
> +
> + if (num_iqueues > 1)
> + lionetdevops.ndo_select_queue = select_q;
> +
> + /* Associate the routines that will handle different
> + * netdev tasks.
> + */
> + netdev->netdev_ops = &lionetdevops;
> +
> + lio = GET_LIO(netdev);
> +
> + memset(lio, 0, sizeof(struct lio));
> +
> + lio->linfo.ifidx = resp->cfg_info.ifidx;
> + lio->ifidx = resp->cfg_info.ifidx;
> +
> + lio->linfo.num_rxpciq = num_oqueues;
> + lio->linfo.num_txpciq = num_iqueues;
> + q_mask = resp->cfg_info.oqmask;
> + /* q_mask is 0-based and already verified mask is nonzero */
> + for (j = 0; j < num_oqueues; j++) {
> + q_no = __ffs64(q_mask);
> + q_mask &= (~(1UL << q_no));
> + lio->linfo.rxpciq[j] = q_no;
> + }
> + q_mask = resp->cfg_info.iqmask;
> + for (j = 0; j < num_iqueues; j++) {
> + q_no = __ffs64(q_mask);
> + q_mask &= (~(1UL << q_no));
> + lio->linfo.txpciq[j] = q_no;
> + }
> + retval = get_inittime_link_status(octeon_dev,
> + &octeon_dev->props, i);
> + if (retval) {
> + lio_dev_err(octeon_dev, "link status failed\n");
> + goto setup_nic_dev_fail;
> + }
> + lio->linfo.hw_addr = octeon_dev->props.ls->link_info.hw_addr;
> + lio->linfo.gmxport = octeon_dev->props.ls->link_info.gmxport;
> +
> + lio->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
> +
> + lio->dev_capability =
> + NETIF_F_HW_CSUM | NETIF_F_SG | NETIF_F_RXCSUM;
> + lio->dev_capability |=
> + (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_LRO);
> + netif_set_gso_max_size(netdev, GSO_MAX_SIZE - ETH_HLEN - 4);
> +
> + netdev->features = lio->dev_capability;
> + netdev->vlan_features = lio->dev_capability;
> +
> + netdev->hw_features = lio->dev_capability;
> +
> + /* Point to the properties for octeon device to which this
> + * interface belongs.
> + */
> + lio->oct_dev = get_octeon_device_ptr(octeon_id);
> + lio->octprops = &octeon_dev->props;
> + lio->netdev = netdev;
> + spin_lock_init(&lio->lock);
> +
> + lio_dev_dbg(octeon_dev, "if%d gmx: %d hw_addr: 0x%llx\n", i,
> + lio->linfo.gmxport, CVM_CAST64(lio->linfo.hw_addr));
> +
> + /* 64-bit swap required on LE machines */
> + octeon_swap_8B_data(&lio->linfo.hw_addr, 1);
> + for (j = 0; j < 6; j++)
> + macaddr[j] =
> + *((uint8_t *)(((uint8_t *)&lio->linfo.hw_addr) +
> + 2 + j));
> +
> + /* Copy MAC Address to OS network device structure */
> +
> + ether_addr_copy(netdev->dev_addr, macaddr);
> +
> + lio->linfo.link.u64 = octeon_dev->props.ls->link_info.link.u64;
> + spin_lock_init(&lio->link_update_lock);
> +
> + if (setup_io_queues(octeon_dev, netdev)) {
> + lio_dev_err(octeon_dev, "I/O queues creation failed\n");
> + goto setup_nic_dev_fail;
> + }
> +
> + ifstate_set(lio, LIO_IFSTATE_DROQ_OPS);
> +
> + /* By default all interfaces on a single Octeon uses the same
> + * tx and rx queues
> + */
> + lio->txq = lio->linfo.txpciq[0];
> + lio->rxq = lio->linfo.rxpciq[0];
> +
> + lio->tx_qsize = octeon_get_tx_qsize(octeon_id, lio->txq);
> + lio->rx_qsize = octeon_get_rx_qsize(octeon_id, lio->rxq);
> +
> + if (setup_glist(lio)) {
> + lio_dev_err(octeon_dev,
> + "Gather list allocation failed\n");
> + goto setup_nic_dev_fail;
> + }
> +
> + /* Register ethtool support */
> + liquidio_set_ethtool_ops(netdev);
> +
> + /* Register the network device with the OS */
> + if (register_netdev(netdev)) {
> + lio_dev_err(octeon_dev, "Device registration failed\n");
> + goto setup_nic_dev_fail;
> + }
> +
> + lio_dev_dbg(octeon_dev,
> + "Setup NIC ifidx:%d mac:%02x%02x%02x%02x%02x%02x\n",
> + i, macaddr[0], macaddr[1], macaddr[2], macaddr[3],
> + macaddr[4], macaddr[5]);
> + netif_carrier_off(netdev);
> +
> + if (lio->linfo.link.s.status) {
> + netif_carrier_on(netdev);
> + start_txq(netdev);
> + } else {
> + netif_carrier_off(netdev);
> + }
> +
> + ifstate_set(lio, LIO_IFSTATE_REGISTERED);
> +
> + liquidio_set_lro(netdev, OCTNET_CMD_LRO_ENABLE);
> +
> + print_link_info(netdev);
> + lio_dev_dbg(octeon_dev, "NIC ifidx:%d Setup successful\n", i);
> + }
> + kfree(sc);
> + kfree(resp);
I wonder if this is correct. AFAICS sc is the same memory that we put on
the octeon_instr_queue (see send_soft_command() ->
send_command()->add_to_nrlist()). Is that queue cleared before sc is freed?
^ permalink raw reply
* Re: [PATCH net-next v3] Add support of Cavium Liquidio ethernet adapters
From: Lino Sanfilippo @ 2014-12-19 22:04 UTC (permalink / raw)
To: Raghu Vatsavayi, davem
Cc: netdev, Derek Chickles, Satanand Burla, Felix Manlunas,
Raghu Vatsavayi, Stephen Hemminger
In-Reply-To: <1418959519-31681-1-git-send-email-rvatsavayi@caviumnetworks.com>
On 19.12.2014 04:25, Raghu Vatsavayi wrote:
> +static void cn6xxx_disable_interrupt(void *chip)
> +{
> + struct octeon_cn6xxx *cn6xxx = (struct octeon_cn6xxx *)chip;
> +
> + /* Disable Interrupts */
> + writeq(0, cn6xxx->intr_enb_reg64);
> +}
> +
This could also be a good candidate for forced write posting. The code
assumes that interrupts are actually deactivated after that.
Regards,
Lino
^ permalink raw reply
* Re: [PATCH net-next v3] Add support of Cavium Liquidio ethernet adapters
From: Lino Sanfilippo @ 2014-12-19 22:24 UTC (permalink / raw)
To: Raghu Vatsavayi, davem
Cc: netdev, Derek Chickles, Satanand Burla, Felix Manlunas,
Raghu Vatsavayi, Stephen Hemminger
In-Reply-To: <1418959519-31681-1-git-send-email-rvatsavayi@caviumnetworks.com>
There seem to be various places where GFP_KERNEL could be used instead
of GFP_ATOMIC, e.g.:
> +static int setup_nic_devices(struct octeon_device *octeon_dev)
> +{
> +
> + sc = kmalloc(sizeof(*sc), GFP_ATOMIC);
> + if (!sc)
> + return -ENOMEM;
> +
and
> +static int liquidio_init_nic_module(int octeon_id, void *octeon_dev)
> +{
> + /* Allocate a soft command to be used to send link status requests
> + * to the core app.
> + */
> + sc = kmalloc(sizeof(*sc), GFP_ATOMIC);
> + if (!sc) {
> + kfree(ls);
> + return -ENOMEM;
also
> +/* Configure interrupt moderation parameters */
> +static int octnet_set_intrmod_cfg(void *oct, struct oct_intrmod_cfg *intr_cfg)
> +{
> + struct octeon_soft_command *sc;
> + struct oct_intrmod_cmd *cmd;
> + int retval;
> + struct octeon_device *oct_dev = (struct octeon_device *)oct;
> + unsigned char *cfg;
> +
> + /* Alloc soft command */
> + sc = (struct octeon_soft_command *)
> + kmalloc(sizeof(struct octeon_soft_command), GFP_ATOMIC);
> + if (!sc)
> + return -ENOMEM;
seems only to be called from process context.
Regards,
Lino
^ permalink raw reply
page: next (older) | prev (newer) | latest
- recent:[subjects (threaded)|topics (new)|topics (active)]
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox