Regression: ixgb warning on MTU change
Subsystems (2)
I just found an issue related to ixgb that seems to be a regression. After the interface is up and running (packets already transmitted), if I try to change the MTU, running "ifconfig ethX mtu 70", for example, causes the following warning[1].
Bisecting I found that the commit that caused this warning is fc2d14e36c69a8d44a2f5230835b54e95025363e. Reverting it solves the problem, ie, no more warnings.
Digging further, I found that just removing the following line added by fc2d14e36c69a8d44a2f5230835b54e95025363e's patch "buffer_info->dma = 0;", also solves the problem.
Jay Vosburgh suggested me some tests with barriers, mainly because I hit this error on a PPC, but I didn't get any luck adding barriers around "buffer_info->dma = 0" line.
[1]iommu_free: invalid entry
entry = 0x0 dma_addr = 0x0 Table = 0xc0000000be02b600 bus# = 0x0 size = 0x38000 startOff = 0x48000 index = 0x4000004 ------------[ cut here ]------------ Badness at arch/powerpc/kernel/iommu.c:258 NIP: c000000000028488 LR: c000000000028484 CTR: 0000000000000001 REGS: c0000000ba10b280 TRAP: 0700 Tainted: G W (2.6.27-rc7) MSR: 8000000000029032 <EE,ME,IR,DR> CR: 24002422 XER: 20000000
TASK = c0000000be5f4c20[23325] 'ifconfig' THREAD: c0000000ba108000 CPU: 2
GPR00: c000000000028484 c0000000ba10b500 c000000000681ae0 000000000000001a
GPR04: 0000000000000001 0000000000000001 0000000000000000 c0000000005b2f5c
GPR08: 000000000001ffff c0000000005ea4f8 c0000000007154f4 c0000000007154f0
GPR12: d00000000047f770 c0000000006f3880 0000000000000000 0000000000000000
GPR16: 0000000000000000 0000000000000000 0000000000000004 0000000000000000
GPR20: 0000000010020000 0000000010022860 0000000000000004 0000000000000001
GPR24: c0000000ba10bca0 0000000000000000 0000000000000000 0000000000000001
GPR28: fffffffffffb8000 c0000000be02b600 c00000000062b200 0000000000000000 NIP [c000000000028488] .__iommu_free+0xe8/0x150
LR [c000000000028484] .__iommu_free+0xe4/0x150
Call Trace: [c0000000ba10b500] [c000000000028484] .__iommu_free+0xe4/0x150 (unreliable) [c0000000ba10b5a0] [c000000000028548] .iommu_free+0x58/0xc0 [c0000000ba10b640] [c0000000000280cc] .dma_iommu_unmap_single+0x14/0x28 [c0000000ba10b6c0] [d000000000472998] .ixgb_clean_rx_ring+0xa8/0x1e8 [ixgb] [c0000000ba10b760] [d0000000004763fc] .ixgb_change_mtu+0xa0/0xec [ixgb] [c0000000ba10b800] [c00000000036df34] .dev_set_mtu+0x68/0xb0 [c0000000ba10b880] [c00000000036f8cc] .dev_ioctl+0x67c/0x780 [c0000000ba10b990] [c00000000035ed44] .sock_ioctl+0x2d8/0x310 [c0000000ba10ba30] [c0000000001095b4] .vfs_ioctl+0x5c/0xf0 [c0000000ba10bad0] [c000000000109a54] .do_vfs_ioctl+0x40c/0x448 [c0000000ba10bb80] [c000000000109b00] .sys_ioctl+0x70/0xb4 [c0000000ba10bc30] [c00000000013eaa4] .dev_ifsioc+0x1b0/0x3e4 [c0000000ba10bd40] [c00000000013e064] .compat_sys_ioctl+0x3d4/0x468 [c0000000ba10be30] [c0000000000086b4] syscall_exit+0x0/0x40
Instruction dump:e89d0008 e87e8058 4804aec1 60000000 e89d0010 e87e8060 4804aeb1 60000000 e87e8068 e89d0020 4804aea1 60000000 <0fe00000> 4800003c e93e8070 e9290050
ixgb: eth4: ixgb_watchdog: NIC Link is Up 10000 Mbps Full Duplex
Quoted 5 lines · Breno Leitao
I just found an issue related to ixgb that seems to be a regression. After the interface is up and running (packets already transmitted), if I try to change the MTU, running "ifconfig ethX mtu 70", for example, causes the following warning[1].
Quoted 8 lines
Bisecting I found that the commit that caused this warning is
fc2d14e36c69a8d44a2f5230835b54e95025363e. Reverting it solves the problem, ie, no more warnings.Digging further, I found that just removing the following line added by fc2d14e36c69a8d44a2f5230835b54e95025363e's patch "
buffer_info->dma= 0;", also solves the problem.
Thanks for the report, I think this patch should fix it. Please test and let us know. This was compile tested only.
----
From: Jesse Brandeburg <[email protected]>
ixgb: fix bug when freeing resources
It was pointed out by Breno Leitao <[email protected]> that ixgb would crash on PPC when an IOMMU was in use, if change_mtu was called.
It appears to be a pretty simple issue in the driver that wasn't discovered because most systems don't run with an IOMMU. The driver needs to only unmap buffers that are mapped (duh).
--- drivers/net/ixgb/ixgb_main.c | 8 +++++--- 1 files changed, 5 insertions(+), 3 deletions(-)
drivers/net/ixgb/ixgb_main.c +5 -3
@@ -977,15 +977,17 @@ ixgb_clean_rx_ring(struct ixgb_adapter *adapter)for (i =0 ; i < rx_ring->count; i++) { buffer_info = &rx_ring->buffer_info[i]; -if (buffer_info->skb) { - +if (buffer_info->dma) {pci_unmap_single (pdev, buffer_info->dma, buffer_info->length,PCI_DMA_FROMDEVICE ); + buffer_info->dma =0 ; + buffer_info->length =0 ; + } +if (buffer_info->skb) {dev_kfree_skb (buffer_info->skb); - buffer_info->skb =NULL ; } }
Hi Jesse,
Quoted 3 lines · Jesse
On Wed, 24 Sep 2008, Breno Leitao wrote:
Thanks for the report, I think this patch should fix it. Please test and let us know. This was compile tested only.
I just tested this patch and it fixed that problem.
Thanks
Quoted 47 lines
----
From: Jesse Brandeburg <[email protected]>
ixgb: fix bug when freeing resources
It was pointed out by Breno Leitao <[email protected]> that ixgb would crash on PPC when an IOMMU was in use, if
change_mtuwas called.It appears to be a pretty simple issue in the driver that wasn't discovered because most systems don't run with an IOMMU. The driver needs to only unmap buffers that are mapped (duh).
CC Breno Leitao <[email protected]>Signed-off-by Jesse Brandeburg <[email protected]>--- drivers/net/ixgb/ixgb_main.c | 8 +++++--- 1 files changed, 5 insertions(+), 3 deletions(-)drivers/net/ixgb/ixgb_main.c +5
-3@@ -977,15 +977,17 @@ ixgb_clean_rx_ring(struct ixgb_adapter *adapter)for (i =0 ; i < rx_ring->count; i++) { buffer_info = &rx_ring->buffer_info[i]; -if (buffer_info->skb) { - +if (buffer_info->dma) {pci_unmap_single (pdev, buffer_info->dma, buffer_info->length,PCI_DMA_FROMDEVICE ); + buffer_info->dma =0 ; + buffer_info->length =0 ; + } +if (buffer_info->skb) {dev_kfree_skb (buffer_info->skb); - buffer_info->skb =NULL ; } }