Julia pointed out beiscsi_boot_get_sinfo does not unlock mbox_lock on
nonemb_cmd memory allocation failure.
Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
alloc_wrb_handle already does memset zero of iscsi_wrb descriptor so
remove redundant memset in WRB submission paths.
Add pwrb_handle NULL check before memsett'ing pwrb.
Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
We know that 'ret' is not an error code because it has been tested a few
lines above. So, if one of these function fails, 0 will be returned
instead of an error code. Return -ENOMEM instead.
Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
'ret' needs to be set with error code if hba_setup_cid_tbls fails.
Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Current EQ delay is set to 0 to receive very high max interrupt per sec.
Set EQ delay to 32 - reducing max interrupt rate from 65K to 20K per sec.
Set TCP connection window size to 64K with scale shift count 2.
Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
In very rare scenario, connection gets killed after throwing this error:
scsi host0: BM_2312 : Event CXN_KILLED_BAD_WRB_INDEX_ERROR[15]... CID : 4
connection1:0: detected conn error (1011)
memset ISCSI_WRB descriptor to zero for all allocations of WRB handle.
Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
BUG: unable to handle kernel NULL pointer dereference at 000000000000015e
IP: [<ffffffffa0081700>]
hwi_get_async_handle.isra.23.constprop.39+0x90/0x1d0 [be2iscsi]
PGD 0
Oops: 0000 [#1] SMP
...
Call Trace:
<IRQ>
[<ffffffffa00818bc>] hwi_process_default_pdu_ring+0x7c/0x280 [be2iscsi]
[<ffffffffa0088f51>] beiscsi_process_cq+0x321/0xb90 [be2iscsi]
[<ffffffff810af028>] ? __wake_up_common+0x58/0x90
[<ffffffff810b0d84>] ? __wake_up+0x44/0x50
[<ffffffffa0089a2d>] be_iopoll+0x1d/0xb0 [be2iscsi]
[<ffffffff812d1f61>] blk_iopoll_softirq+0xc1/0x100
[<ffffffff81084b0f>] __do_softirq+0xef/0x280
The symptom observed is multiple async handles get queued for same index
thus causing leak in buffers posted to FW.
The root cause is:
- async handle is continued to be used even if it does not match the
completion.
- list_move operation done on already filled index.
1. Remove use of writables, host_write_ptr and ep_read_ptr.
2. Remove consumed logic to update writables. Instead, use only
free_entries to do the accounting of handles to be posted back.
3. Remove busy_list, instead use simple slot to index handles.
4. Added check no data, header less and overflow to make sure
all async_handles are flushed in error cases.
5. Added code to verify gathering of handles to form PDU by
checking final bit before forwarding PDU.
6. Added code to catch mismatch with CQE and handle gracefully.
7. Use AMAP, traverse cri_wait_queue list to post buffers, log
"async PDU" related errors.
8. Rearranged few data structures and added comments in init &
processing path.
9. Added WARN_ONs to catch any HD ring corruption.
Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Driver unload should call COMMON_FUNCTION_RESET. For TPE feature, this
ensures that FW has knowledge about driver getting unloaded and can
reset its bit vector.
Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Sessions are no longer valid, so schedule sess_work to fail the sessions
immediately when error is detected. This is done to avoid iSCSI transport
layer to keep sending NOP-Out which driver any ways fail.
Schedule sess_work immediately in case of HBA error. Old sessions are gone
for good and need to be re-established.
iscsi_session_failure needs process context hence this work.
Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
After UE is detected, check for recoverable error by reading
SLIPORT SEMAPHORE register. If transient parity error i.e. 0xExxx
then schedule recovery work on driver wq.
FLag this error to prevent any transactions for the duration of ue2rp to
restart polling. After that, if FW becomes ready then recover port.
Wake up processes in wq before going offline.
Wait for process to execute before cleaning up.
Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
mgmt_epfw_cleanup does not implement v1 of OPCODE_COMMON_ISCSI_CLEANUP
IOCTL for SkyHawk.
Replace use of MCCQ with BMBX for issuing the IOCTL.
Remove be_mcc_compl_poll which is no longer needed.
Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
SLIPORT FUNCTION_RESET does not reset the chip.
So POST status needs to be checked before issuing FUNCTION_RESET.
The completion of FUNCTION_RESET is indicated in BMBX Rdy bit.
be_cmd_fw_initialize too needs to be done before issuing any cmd to FW.
be_cmd_fw_initialize is renamed as beiscsi_cmd_special_wrb.
Rearrange and rename few functions in init and cleanup path.
Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
beiscsi_fail_session is defined in be_cmds.c: move it to be_iscsi.c
Move card configuration commands to be_cmds.c.
Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
BE3 and SH cards can recover from transient parity errors treated
earlier as unrecoverable errors.
Add IOCTL to query FW support for this feature.
Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
UE detection in health check is done in a work scheduled in global wq.
UE caused due to transient parity errors are recoverable and reported
within 1s. If this check for TPE gets delayed, PF0 might initiate
soft-reset and then status of UE recoverable is lost.
Handle UE detection in timer routine. Move out EQ delay update work
from health check. Make the IOCTL for EQ delay update non-blocking
as the completion status is ignored.
Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Boot work involves:
1. Find and fetch configured boot session and its handle.
2. Attempt to open the session if its not.
3. Get the session details for boot kset creation.
4. Logout of that session owned by FW.
5. Create boot kset for session details.
All these actions were done in blocking call with retries in global wq.
Other works in wq suffered if the IOCTLs stalled or timed out.
This change moves all the boot work to make it non-blocking.
The work queued in global wq just issues the IOCTL depending on the action
to be taken and mcc wq schedules work depending on status of the IOCTL.
Initial boot_work is started on link and ASYNC event.
The other code changes move all boot related functions in one place and
follow naming conventions.
Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Save ue_detected and fw_timeout errors in state field of beiscsi_hba.
BEISCSI_HBA_RUNNING
BEISCSI_HBA_LINK_UP
BEISCSI_HBA_BOOT_FOUND
BEISCSI_HBA_PCI_ERR
BEISCSI_HBA_FW_TIMEOUT
BEISCSI_HBA_IN_UE
Make sure no PCI transaction happens once in error state.
Add checks in IO path to detect HBA in error.
Skip hwi_purge_eq step which can't be done in error state.
Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
todo_mcc_cq is not needed as only MCC work is queued.
todo_cq is not used at all.
Rename functions to be consistent.
Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
alloc_mcc_tag was replaced with alloc_mcc_wrb and is no more used.
beiscsi_pci_soft_reset is not used at all and won't be needed.
Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Redefine FW IP types.
Before issuing IOCTL to clear IP, check if IP is all zeroes.
All zeroes IP implies IP is not set in FW so FW fails that IOCTL.
Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Wrong settings displayed for iface:
iface.header_digest = 192.168.197.22
iface.data_digest = 255.255.255.0
iface.immediate_data = 192.168.197.1
Process ISCSI_NET_PARAM only in beiscsi_iface_get_param.
Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Rename mgmt_get_if_info to be consistent with APIs name.
Rename create/destroy APIs to indicate IFACE operations.
Remove legacy be2iscsi and use beiscsi.
Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Move mgmt_get_all_if_id before any set param operation.
Rename mgmt_get_all_if_id to beiscsi_if_get_handle.
Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
VLAN tag is L2 construct, move VLAN code out from configuring IP.
Rearrange and rename the APIs to make it consistent.
Replace ENOSYS with EPERM.
Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
If BOOTPROTO is changed to static, the DHCP IP address should be released.
All cases are being handled mgmt_set_ip and mgmt_static_ip_modify.
Rearrange IFACE APIs to:
beiscsi_if_clr_ip
beiscsi_if_set_ip
beiscsi_if_en_static
beiscsi_if_en_dhcp
This simplifies release of DHCP IP when BOOTPROTO is set to static.
Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Gateway APIs assume IP type as IPv4. Modify it to be generic to allow
clearing of IPv6 gateway set using BIOS.
Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
ipv4_iface and ipv6_iface fields need to be set to NULL when destroyed.
Before creation these are checked. Use these to report correct states.
Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Driver takes significant time to load 1m:20s and unload 40s.
Checkpatch script threw warning:
WARNING: msleep < 20ms can sleep for up to 20ms; see
Documentation/timers/timers-howto.txt
To eliminate this warning msleep(1) was replaced with msleep(20) before
submitting.
msleep(20) in init and uninit path for creation and destroying of number
of WRBQs, CQs, and EQs is adding to load/unload time.
Replace msleep with schedule_timeout_uninterruptible of 1ms as its
enough in most cases.
Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
This got unnecessarily introduced with other changes in previous
commits. mcc_lock is taken only in process contexts.
Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Following configuration is created with what driver exports:
iface.vlan_id = 65535
iface.vlan_priority = 255
iface.vlan_state = <empty>
vlan_state is empty as iscsiadm doesn't process "Disabled".
When applying this configuration, iscsiadm checks for if vlan_state is
"disable" if not it enables with value in vlan_id. 65535 not being valid
value, 0 is applied.
Use "enable" or "disable" for ISCSI_NET_PARAM.
Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
In beiscsi_setup_boot_info(), the boot_kset pointer should be set to
NULL in case of failure otherwise an invalid pointer dereference may
occur later.
Cc: <stable@vger.kernel.org>
Signed-off-by: Maurizio Lombardi <mlombard@redhat.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Calling synchronize_irq() right before free_irq() is quite useless. On one
hand the IRQ can easily fire again before free_irq() is entered, on the
other hand free_irq() itself calls synchronize_irq() internally (in a race
condition free way), before any state associated with the IRQ is freed.
Patch was generated using the following semantic patch:
// <smpl>
@@
expression irq;
@@
-synchronize_irq(irq);
free_irq(irq, ...);
// </smpl>
Signed-off-by: Lars-Peter Clausen <lars@metafoo.de>
Reviewed-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
This adds the missing error check and path for if the call to the
function hwi_init_controller fails as this error path was clearly missed
when writing beiscsi_eeh_resume and thus we must add it now in order to
be able to handle this nonrecoverable failing function call gracefully
in beiscsi_eeh_resume.
Signed-off-by: Nicholas Krause <xerofoify@gmail.com>
Reviewed-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
In case of error, the memory allocated for phwi_ctrlr was not freed.
Signed-off-by: Maurizio Lombardi <mlombard@redhat.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
FW got into UE after running IO stress test
With kernel change to split session lock in frwd_lock and back_lock for
tx and rx path correspondingly, in the IO path, common resource used in
driver such as WRB was left unprotected.
Add wrb_lock spinlock to protect allocation and freeing of WRB.
Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Processing of mgmt and IO tasks are done in process context and
softirqs.
Allocation and freeing of sgl_handles needs to be done under
spin_lock_bh/spin_unlock_bh and move the locks to the routines.
Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Connection resets observed from some targets when NOP-Out with wrong
ExpStatSn is sent.
FW keeps track of StatSn and fills up ExpStatSn accordingly. The header
filled up by the stack needs to be modified by driver to clear
ExpStatSn. If the field is not cleared, FW recalculates ExpStatSn and
wrong offset'ed ExpStatSn is seen in the wire trace.
Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
WARN_ON(atomic_read(&mccq->used) >= mccq->len) seen when FW gets into
UE.
MCCQ overflow is happening because driver discards any new request and
frees up the tag. The tag allocation controls the number of MCC WRB
posted. It is being replenished but WRBs are not hence the WARN_ON.
Allocation and freeing of WRB and tags for MCC is now done in one place.
This helps to achieve proper accounting of WRB indices and MCC tags.
Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
In open with IP of unknown address family, only tag is freed and error
returned. MCC WRB allocated for the operation is not freed.
Added check for supported family of IP in the beginning before
allocating the tag and WRB.
Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Remove confusingly named be_mcc_compl_is_new and be_mcc_compl_use
functions in processing of BMBX. Rearrange beiscsi_process_mbox_compl
function.
Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
be_mcc_compl_poll waits till 'used' count of MCC WRBQ is zero. This is
to determine the completion of an MCC sent.
Change function to poll for the tag of MCC sent, instead, and wait till
its tag_state is cleared.
Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
be_mbox_notify_wait does exactly same thing as be_mbox_notify.
Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
be_mcc_compl_process_isr is removed. MCC CQ processing is done only in
beiscsi_process_mcc_cq and MCC CQE processing is done only in
beiscsi_process_mcc_compl.
Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Rename mcc_numtag to mcc_tag_status. MCC CQE status is processed using
macros already defined in be_cmds.h.
Add MCC_Q_WRB_ and MCC_Q_CMD_TAG_MASK macros to map to already defined
CQE_STATUS_ macros to be consistent when posting MCC.
Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
mcc_cq_lock spin_lock is used only in beiscsi_process_mcc which is
called only when all interrupts are disabled from mgmt_epfw_cleanup
during unloading of driver. There is no other context where there can be
contention for the processing of CQ.
Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
hwi_write_buffer() does not check if mapping dma memory succeed. The
patch adds the check and failure handling.
Found by Linux Driver Verification project (linuxtesting.org).
Signed-off-by: Alexey Khoroshilov <khoroshilov@ispras.ru>
Reviewed-by: Jitendra Bhivare <jitendra.bhivare@avagotech.com>
Acked-by: Jitendra Bhivare <jitendra.bhivare@avagotech.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
