[ofw] RE: IA64 vstate cmd execution induced system crash
Tzachi Dar
tzachid at mellanox.co.il
Thu Sep 20 16:05:27 PDT 2007
Seems fine, please commit.
Thanks
Tzachi
> -----Original Message-----
> From: Smith, Stan [mailto:stan.smith at intel.com]
> Sent: Friday, September 21, 2007 12:03 AM
> To: Tzachi Dar
> Cc: ofw at lists.openfabrics.org; arlin.r.davis at intel.com;
> Woodruff, Robert J
> Subject: IA64 vstate cmd execution induced system crash
>
> Hello,
> After poking around with windbg using a minidump file, I
> found what the problem is.....drum roll, surprise, surprise,
> an IA64 misaligned access.
>
> The problem is an unaligned (0xe0000165`3721def9) local MAD
> is passed into mthca_process_mad(): 'mthca_mad.c'. The first
> few local MAD header references are to u8 fields, when the
> attr_id field(be16)is accessed @ line 229 it's unaligned - oops!
>
> After receiving mentoring from a well qualified
> openib-windows SW coach, the problem is best fixed in
> gen1\trunk\inc\iba\ib_al_ioctl.h @ line #2486, the typedef
> union _ual_local_mad_ioctl definition; see enclosed files as
> the email tabs/spaces don't really lineup well.
>
> Testing on IA64 reveals no patch side effects not to mention
> vstat works as expected.
> Testing on x64 & x86 reveals no changes in system behavior as
> well as working vstat commands.
>
> Please advise if this patch is acceptible and I will commit.
>
> Stan.
>
>
> Index: ib_al_ioctl.h
> ===================================================================
> --- ib_al_ioctl.h (revision 825)
> +++ ib_al_ioctl.h (working copy)
> @@ -2298,7 +2298,7 @@
> uint64_t h_av;
> ib_mad_element_t* __ptr64 p_mad_element;
> uint32_t size;
> - void* __ptr64* __ptr64 ph_proxy;
> + void* __ptr64* __ptr64 ph_proxy;
>
> } in;
> struct _ual_send_mad_ioctl_out
> @@ -2488,14 +2488,15 @@
> struct _ual_local_mad_ioctl_in
> {
> uint64_t h_ca;
> + __declspec(align(8)) uint8_t mad_in[MAD_BLOCK_SIZE];
> uint8_t port_num;
> - uint8_t mad_in[MAD_BLOCK_SIZE];
>
> } in;
> struct _ual_local_mad_ioctl_out
> {
> ib_api_status_t status;
> - uint8_t mad_out[MAD_BLOCK_SIZE];
> + uint32_t _pad; /* 8-byte
> alignment needed for ia64 */
> + __declspec(align(8)) uint8_t mad_out[MAD_BLOCK_SIZE];
>
> } out;
>
>
> Offending call chain.
>
> nt!KeBugCheck2+0x170
> nt!KiSystemServiceHandler+0x190
> nt!RtlpExecuteEmHandlerForException+0x50
> nt!RtlDispatchException+0x580
> nt!KiDispatchException+0x470
> nt!KiExceptionDispatch+0x190
> nt!KiGenericExceptionHandler+0x330
> mthca!mthca_process_mad(struct ib_device * ibdev =
> 0xe0000165`3862f110, int mad_flags = 0, unsigned char
> port_num = 0x01 '', struct _ib_wc * in_wc =
> 0x00000000`00000000, struct _ib_grh * in_grh =
> 0x00000000`00000000, struct ib_mad * in_mad =
> 0xe0000165`3721def9, struct ib_mad * out_mad =
> 0xe0000165`3721def4)+0x710
> [d:\openib-windows-svn\769\gen1\trunk\hw\mthca\kernel\mthca_ma
> d.c @ 229]
>
> mthca!mlnx_local_mad(struct _ib_ca * h_ca =
> 0xe0000165`389dfd70, unsigned char port_num = 0x01 '', struct
> _ib_av_attr * p_av_attr = 0x00000000`00000000, struct _ib_mad
> * p_mad_in = 0xe0000165`3721def9, struct _ib_mad * p_mad_out
> = 0xe0000165`3721def4)+0x760
> [d:\openib-windows-svn\769\gen1\trunk\hw\mthca\kernel\hca_verb
> s.c @ 1541]
>
> ibbus!al_local_mad(struct _ib_ca * h_ca =
> 0xe0000165`373ecf40, unsigned char port_num = 0x01 '', struct
> _ib_av_attr * p_src_av_attr = 0x00000000`00000000, void *
> p_mad_in = 0xe0000165`3721def9, void * p_mad_out =
> 0xe0000165`3721def4)+0x16c0
> [d:\openib-windows-svn\769\gen1\trunk\core\al\al_mad.c @
> 3229] ibbus!ib_local_mad(struct _ib_ca * h_ca =
> 0xe0000165`373ecf40, unsigned char port_num = 0x01 '', void *
> p_mad_in = 0xe0000165`3721def9, void * p_mad_out =
> 0xe0000165`3721def4)+0xf80
> [d:\openib-windows-svn\769\gen1\trunk\core\al\al_mad.c @ 3188]
>
> ibbus!proxy_local_mad(void * p_open_context =
> 0xe0000165`38811140, struct _IRP * h_ioctl =
> 0xe0000165`373c0cc0, unsigned int64 * p_ret_bytes =
> 0xe0000165`24ef73a8)+0xbb0
> [d:\openib-windows-svn\769\gen1\trunk\core\al\kernel\al_proxy_
> subnet.c @ 1077]
>
> ibbus!subnet_ioctl(struct _IRP * h_ioctl =
> 0xe0000165`373c0cc0, unsigned
> int64 * p_ret_bytes = 0xe0000165`24ef73a8)+0xc50
> [d:\openib-windows-svn\769\gen1\trunk\core\al\kernel\al_proxy_
> subnet.c @ 1150]
>
> ibbus!al_dev_ioctl(struct _IRP * h_ioctl =
> 0xe0000165`373c0cc0)+0xcd0
> [d:\openib-windows-svn\769\gen1\trunk\core\al\kernel\al_dev.c @ 460]
>
> ibbus!bus_drv_ioctl(struct _DEVICE_OBJECT * p_dev_obj =
> 0xe0000165`386d38a0, struct _IRP * p_irp =
> 0xe0000165`373c0cc0)+0x8b0
> [d:\openib-windows-svn\769\gen1\trunk\core\bus\kernel\bus_driv
> er.c @ 402]
>
> nt!IofCallDriver+0x120
> nt!IopSynchronousServiceTail+0x230
> nt!IopXxxControlFile+0x1140
> nt!NtDeviceIoControlFile+0x80
> nt!KiSystemServiceExit
>
More information about the ofw
mailing list