]> git.itanic.dy.fi Git - linux-stable/commit
scsi: zfcp: fix GCC compiler warning emitted with -Wmaybe-uninitialized
authorBenjamin Block <bblock@linux.ibm.com>
Tue, 2 Jul 2019 21:02:02 +0000 (23:02 +0200)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Tue, 6 Aug 2019 16:28:26 +0000 (18:28 +0200)
commitad96d13da411e184a335ca11a97d025fc80365b8
treecfd9d99fa1fcee3ae9634fcf852286d07a4f4c70
parent4d8583e7fbc60a5f421b07fd384ffc7f9b5efca8
scsi: zfcp: fix GCC compiler warning emitted with -Wmaybe-uninitialized

[ Upstream commit 484647088826f2f651acbda6bcf9536b8a466703 ]

GCC v9 emits this warning:
      CC      drivers/s390/scsi/zfcp_erp.o
    drivers/s390/scsi/zfcp_erp.c: In function 'zfcp_erp_action_enqueue':
    drivers/s390/scsi/zfcp_erp.c:217:26: warning: 'erp_action' may be used uninitialized in this function [-Wmaybe-uninitialized]
      217 |  struct zfcp_erp_action *erp_action;
          |                          ^~~~~~~~~~

This is a possible false positive case, as also documented in the GCC
documentations:
    https://gcc.gnu.org/onlinedocs/gcc/Warning-Options.html#index-Wmaybe-uninitialized

The actual code-sequence is like this:
    Various callers can invoke the function below with the argument "want"
    being one of:
    ZFCP_ERP_ACTION_REOPEN_ADAPTER,
    ZFCP_ERP_ACTION_REOPEN_PORT_FORCED,
    ZFCP_ERP_ACTION_REOPEN_PORT, or
    ZFCP_ERP_ACTION_REOPEN_LUN.

    zfcp_erp_action_enqueue(want, ...)
        ...
        need = zfcp_erp_required_act(want, ...)
            need = want
            ...
            maybe: need = ZFCP_ERP_ACTION_REOPEN_PORT
            maybe: need = ZFCP_ERP_ACTION_REOPEN_ADAPTER
            ...
            return need
        ...
        zfcp_erp_setup_act(need, ...)
            struct zfcp_erp_action *erp_action; // <== line 217
            ...
            switch(need) {
            case ZFCP_ERP_ACTION_REOPEN_LUN:
                    ...
                    erp_action = &zfcp_sdev->erp_action;
                    WARN_ON_ONCE(erp_action->port != port); // <== access
                    ...
                    break;
            case ZFCP_ERP_ACTION_REOPEN_PORT:
            case ZFCP_ERP_ACTION_REOPEN_PORT_FORCED:
                    ...
                    erp_action = &port->erp_action;
                    WARN_ON_ONCE(erp_action->port != port); // <== access
                    ...
                    break;
            case ZFCP_ERP_ACTION_REOPEN_ADAPTER:
                    ...
                    erp_action = &adapter->erp_action;
                    WARN_ON_ONCE(erp_action->port != NULL); // <== access
                    ...
                    break;
            }
            ...
            WARN_ON_ONCE(erp_action->adapter != adapter); // <== access

When zfcp_erp_setup_act() is called, 'need' will never be anything else
than one of the 4 possible enumeration-names that are used in the
switch-case, and 'erp_action' is initialized for every one of them, before
it is used. Thus the warning is a false positive, as documented.

We introduce the extra if{} in the beginning to create an extra code-flow,
so the compiler can be convinced that the switch-case will never see any
other value.

BUG_ON()/BUG() is intentionally not used to not crash anything, should
this ever happen anyway - right now it's impossible, as argued above; and
it doesn't introduce a 'default:' switch-case to retain warnings should
'enum zfcp_erp_act_type' ever be extended and no explicit case be
introduced. See also v5.0 commit 399b6c8bc9f7 ("scsi: zfcp: drop old
default switch case which might paper over missing case").

Signed-off-by: Benjamin Block <bblock@linux.ibm.com>
Reviewed-by: Jens Remus <jremus@linux.ibm.com>
Reviewed-by: Steffen Maier <maier@linux.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
drivers/s390/scsi/zfcp_erp.c