mpt2sas : the sanity check in base_interrupt needs to be on dword boundary
PATCH 2/10 mpt2sas : the sanity check in base_interrupt needs to be on dword boundary
The poison sanity check on the reply_post_free register needs to be by 32bit, not 64bit. The poison check is there because its possible that the driver read the 1st 32bit before the 2nd 32bit has been written to by firmware. In other words, this handles race between driver reading the 64 bit register, and it being dma'd across pci memory from controller firmware as two 32bit pci writes.
Signed-off-by Eric Moore <[email protected]>
drivers/scsi/mpt2sas/mpt2sas_base.c +12 -3
@@ -636,6 +636,14 @@ _base_unmask_interrupts(struct MPT2SAS_Astatic irqreturn_t _base_interrupt (int irq,void *bus_id) { +union reply_descriptor { +u64 word; +struct { +u32 low; +u32 high; + } u; + }; +union reply_descriptor rd;u32 post_index, post_index_next, completed_cmds;u8 request_desript_type;u16 smid;
@@ -656,7 +664,8 @@ _base_interrupt(int irq, void *bus_id) completed_cmds =0 ; do { - if (ioc->reply_post_free[post_index].Words == ~0ULL) + rd.word = ioc->reply_post_free[post_index].Words; + if (rd.u.low == UINT_MAX || rd.u.high == UINT_MAX) goto out; reply = 0; cb_idx = 0xFF;
@@ -721,7 +730,7 @@ _base_interrupt(int irq, void *bus_id)for (i =0 ; i < completed_cmds; i++) { post_index = post_index_next;/* poison the reply post descriptor */ - ioc->reply_post_free[post_index_next].Words = ~0ULL ; + ioc->reply_post_free[post_index_next].Words =ULLONG_MAX ; post_index_next = (post_index == (ioc->reply_post_queue_depth -1 )) ?0 : post_index +1 ;
@@ -3068,7 +3077,7 @@ _base_make_ioc_operational(struct MPT2SA/* initialize Reply Post Free Queue */ for (i =0 ; i < ioc->reply_post_queue_depth; i++) - ioc->reply_post_free[i].Words = ~0ULL ; + ioc->reply_post_free[i].Words =ULLONG_MAX ; r =_base_send_ioc_init (ioc,VF_ID , sleep_flag);if (r)