Re: kernel problems with smart on LSI-92xx

[Artem Bokhan <aptem@ngs.ru>]

   It appears that passing self-test command to SATA disk does not work too. 
Although nobody reports any error, self-test actually just does not start.

smartctl -d megaraid,10 -t long /dev/sg0

smartctl 5.40 2010-10-16 r3189 [x86_64-unknown-linux-gnu] (local build)
Copyright (C) 2002-10 by Bruce Allen, http://smartmontools.sourceforge.net

/dev/sg0 [megaraid_disk_10] [SAT]: Device open changed type from 'megaraid' to 'sat'
=== START OF OFFLINE IMMEDIATE AND SELF-TEST SECTION ===
Sending command: "Execute SMART Extended self-test routine immediately in 
off-line mode".
Drive command "Execute SMART Extended self-test routine immediately in off-line 
mode" successful.
Testing has begun.
Please wait 255 minutes for test to complete.
Test will complete after Sat Nov 27 01:54:44 2010

Use smartctl -X to abort test.


11.11.2010 20:06, Bjørn Mork пишет:
> Bokhan Artem<aptem@ngs.ru>  writes:
>
>> Hello.
>>
>> I have kernel problems when trying to run smart commands on LSI-92xx
>> controller.
>>
>> Running self-test  on SAS disk (smartctl /dev/sda -dmegaraid,24 -t
>> long) with smartmontools causes kernel oops (?) (and segfault). Look
>> in attachment for dmesg.
>>
>> strace of smartctl:
>>
>> mknod("/dev/megaraid_sas_ioctl_node", S_IFCHR, makedev(251, 0)) = -1 EEXIST
>> (File exists)
>> close(4)                                = 0
>> munmap(0x7f4be3a9f000, 4096)            = 0
>> open("/dev/megaraid_sas_ioctl_node", O_RDWR) = 4
>> ioctl(4, MTRRIOC_SET_ENTRY, 0x7fffa574ed30) = 0
>> ioctl(4, MTRRIOC_SET_ENTRY, 0x7fffa574eb30) = 0
>> ioctl(4, MTRRIOC_SET_ENTRY<unfinished ...>
>> +++ killed by SIGSEGV +++
>>
>>
>> Viewing  smart info is OK (smartctl /dev/sda -dmegaraid,24 -a).
>> Running self-test on SATA disk on the same system is OK.
>>
>> The problem is reproducible with 2.6.32 and 2.6.36 kernels.
> A quick look at this reveals that smartctl will happily do a
> MEGASAS_IOC_FIRMWARE ioctl with sge_count = 1 and sgl[0].iov_len = 0 if
> it is sending a command with dataLen == 0 :
>
>
> /* Issue passthrough scsi command to PERC5/6 controllers */
> bool linux_megaraid_device::megasas_cmd(int cdbLen, void *cdb,
>    int dataLen, void *data,
>    int /*senseLen*/, void * /*sense*/, int /*report*/)
> {
>    struct megasas_pthru_frame    *pthru;
>    struct megasas_iocpacket      uio;
>    struct megasas_iocpacket      uio;
>    int rc;
>
>    memset(&uio, 0, sizeof(uio));
>    pthru = (struct megasas_pthru_frame *)uio.frame.raw;
>    pthru->cmd = MFI_CMD_PD_SCSI_IO;
>    int rc;
>
>    memset(&uio, 0, sizeof(uio));
>    pthru = (struct megasas_pthru_frame *)uio.frame.raw;
>    pthru->cmd = MFI_CMD_PD_SCSI_IO;
>    pthru->cmd_status = 0xFF;
>    pthru->scsi_status = 0x0;
>    pthru->target_id = m_disknum;
>    pthru->lun = 0;
>    pthru->cdb_len = cdbLen;
>    pthru->timeout = 0;
>    pthru->flags = MFI_FRAME_DIR_READ;
>    pthru->sge_count = 1;
>    pthru->data_xfer_len = dataLen;
>    pthru->sgl.sge32[0].phys_addr = (intptr_t)data;
>    pthru->sgl.sge32[0].length = (uint32_t)dataLen;
>    memcpy(pthru->cdb, cdb, cdbLen);
>
>    uio.host_no = m_hba;
>    uio.sge_count = 1;
>    uio.sgl_off = offsetof(struct megasas_pthru_frame, sgl);
>    uio.sgl[0].iov_base = data;
>    uio.sgl[0].iov_len = dataLen;
>
>    rc = 0;
>    errno = 0;
>    rc = ioctl(m_fd, MEGASAS_IOC_FIRMWARE,&uio);
>    if (pthru->cmd_status || rc != 0) {
>      if (pthru->cmd_status == 12) {
>        return set_err(EIO, "megasas_cmd: Device %d does not exist\n", m_disknum);
>      }
>      return set_err((errno ? errno : EIO), "megasas_cmd result: %d.%d = %d/%d",
>                     m_hba, m_disknum, errno,
>                     pthru->cmd_status);
>    }
>    return true;
> }
>
>
>
> The kernel bug is that the zero valued sgl[0].iov_len is passed
> unmodified to megasas_mgmt_fw_ioctl() which again passes it on as size
> to dma_alloc_coherent():
>
>          /*
>           * For each user buffer, create a mirror buffer and copy in
>           */
>          for (i = 0; i<  ioc->sge_count; i++) {
>                  kbuff_arr[i] = dma_alloc_coherent(&instance->pdev->dev,
>                                                      ioc->sgl[i].iov_len,
>                                                      &buf_handle, GFP_KERNEL);
>
>
>
> And it looks like most (all?) of the dma_alloc_coherent()
> implementations will use get_order(size) to compute the necessary
> allocation. This will fail if size == 0.
>
> On the other hand, I may have misunderstood this entirely....
>
> But if you dare, you could try the attached patch (compile tested only
> as I don't have the hardware) and see if it helps.  Let me know how it
> goes, and I'll forward it to the megaraid manitainers if it really fixes
> your problem.
>
>
>
>
> Bjørn
>