***SPAM*** Re: ***SPAM*** Re: [ofa-general] ib_reg_phys_mr( ) results in crash

[neutron]

It might be related to new ConnectX card (with mlx4_ib module).

Now I tried the same program on a machine with only "mthca" card, it
succeeds without any problems.

thanks.


I remember one guy in this list also reported a similar issue:
ib_phys_reg_mr( )  fails with mlx4 module.



On Fri, Feb 20, 2009 at 3:21 AM, Liang Zhen <Zhen.Liang at sun.com> wrote:
> Hmm, I didn't see any problem in your code. Have you installed
> ofa_kernel_devel (kernel headers of  OFED) after installation of
> ofa_kernel_1_3_1?
>
> Regards
> Liang
>
> neutron:
>>
>> I'm using Mellanox HCA 'mthca0' type: MT25208, kernel version:
>> 2.6.18-53.1.14.el5,  ofed 1.3.1.
>>
>> The failed function call is like:
>>
>> {
>>
>> ctx->send_buf = dma_alloc_coherent(ctx->ib_dev->dma_device, MAX_SIZE,
>>                &dma_addr, GFP_KERNEL);
>>
>> ctx->phy_buf[0].addr = dma_addr;
>> ctx->phy_buf[0].size = MAX_SIZE;
>> ctx->iovstart = (u64) ctx->send_buf;
>>
>> printk("pd=%p, phy_buf[0].addr=%p,size=%d, iovstart=%llx\n",
>>       ctx->pd, ctx->phy_buf[0].addr, ctx->phy_buf[0].size, ctx->iovstart
>> );
>>
>> send_mr = ib_reg_phys_mr( ctx->pd, &ctx->phy_buf[0], 1,
>>                        IB_ACCESS_REMOTE_WRITE | IB_ACCESS_REMOTE_READ
>>                         | IB_ACCESS_LOCAL_WRITE, &(ctx->iovstart));
>> }
>>
>> The phy_buf[0] is a "ib_phys_buf" corresponding to "ctx->send_buf".
>>
>> Below is /var/log/messages output around the crash.
>> ----------------
>> Feb 19 12:50:22 wci30 kernel:  pd=ffff8101da3ddce0,
>> phy_buf[0].addr=00000001bbe4b000,size=1024, iovstart=ffff8101bbe4b000
>>
>> Feb 19 12:50:22 wci30 kernel: Unable to handle kernel NULL pointer
>> dereference at 0000000000000000
>>  RIP:
>> Feb 19 12:50:22 wci30 kernel:  [<0000000000000000>]
>> _stext+0x7ffff000/0x1000
>> Feb 19 12:50:22 wci30 kernel: PGD 1c06d5067 PUD 1c9dcd067 PMD 0
>> Feb 19 12:50:22 wci30 kernel: Oops: 0010 [1] SMP
>> Feb 19 12:50:22 wci30 kernel: last sysfs file: /module/libata/version
>> Feb 19 12:50:22 wci30 kernel: CPU 0
>> Feb 19 12:54:05 wci30 syslogd 1.4.1: restart.
>> Feb 19 12:54:05 wci30 kernel: klogd 1.4.1, log source = /proc/kmsg
>> started.
>> Feb 19 12:54:05 wci30 kernel: Linux version 2.6.18-53.1.14.el5
>> (brewbuilder at hs20-bc2-3.build.redha
>> t.com) (gcc version 4.1.2 20070626 (Red Hat 4.1.2-14)) #1 SMP Tue Feb
>> 19 07:18:46 EST 2008
>> Feb 19 12:54:05 wci30 kernel: Command line: ro root=LABEL=/ rhgb quiet
>>
>> ====================
>> It's strange that the kernel doesn't print out the function call stack
>> before crashing.
>>
>> Any hints?  Thanks a lot!
>>
>> On Wed, Feb 18, 2009 at 7:40 PM, Roland Dreier <rdreier at cisco.com> wrote:
>>
>>>
>>>  > Before calling ib_reg_phys_mr,  printk() shows that all its arguments
>>>  > are valid.  But the system always crashes immediately after entering
>>>  > the function ib_reg_phys_mr( ).    Any possible reasons ?  Thanks!!
>>>
>>> What do you mean by "immediately after entering ib_reg_phys_mr()"?  Do
>>> you get an oops message?  If so that would be very important info for
>>> debugging this.
>>>
>>> - R.
>>>
>>>
>>
>> _______________________________________________
>> general mailing list
>> general at lists.openfabrics.org
>> http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
>>
>> To unsubscribe, please visit
>> http://openib.org/mailman/listinfo/openib-general
>>
>
>