On 2/27/2012 10:28 AM, Clark, Steven wrote:
Resending with signed-off line and libfc in the subject. (Sorry, new to this).
Signed-off-by: Steven Clark<sclark at crossbeam.com>
[Bhanu Prakash Gollapudi wrote:]
On 2/23/2012 10:08 AM, Clark, Steven wrote:
I am seeing a failure in fcoe_transport_create on a 64-bit single-CPU
environment (I'm testing in a VM with one CPU
assigned).
The size in the call to __alloc_percpu() is greater than the max of 32K because
it allocates 4K pointers + header across all
CPUs.
Reducing FCOE_MAX_ID from 0x0FFF to 0x0F7F fixes the problem at the expense of
0x80 fewer allowed open
exchanges.
I can prepare a patch for that, but maybe someone knowledgeable can suggest a
better fix.
One way to fix is to use smaller XID range based on nr_cpu_ids.
Looks like FCP should be seeing this same issue since it also allocates 4K
exchanges:
#define FCPIO_HOST_EXCH_RANGE_START 0x1000
#define FCPIO_HOST_EXCH_RANGE_END 0x1fff
So the right way to fix is what you suggest.
fc_exch_mgr_alloc is already adjusting the allocation based on the number of
cpus so can do this
as a special case. I'll try that out and prepare a patch.
My colleague Vlad Senkov suggested a more general way to solve the
issue than making single CPU a special case.
I verified that this gives the original allocation on multi-CPU and reduces the
allocation from 4096 to 4088 on the problem single CPU case.
--- orig/drivers/scsi/libfc/fc_exch.c 2012-02-24 11:22:32.242157149 -0500
+++ fcoe-next/drivers/scsi/libfc/fc_exch.c 2012-02-24 15:14:35.033962311
-0500
@@ -2284,8 +2284,19 @@
mp->class = class;
/* adjust em exch xid range for offload */
mp->min_xid = min_xid;
- mp->max_xid = max_xid;
+ /* reduce range so per cpu pool fits into PCPU_MIN_UNIT_SIZE pool */
+ pool_exch_range = (PCPU_MIN_UNIT_SIZE - sizeof(*pool)) /
+ sizeof(struct fc_exch *);
+ if ((max_xid - min_xid + 1) / (fc_cpu_mask + 1)> pool_exch_range) {
+ mp->max_xid = pool_exch_range * (fc_cpu_mask + 1) +
+ min_xid - 1;
+ } else {
+ mp->max_xid = max_xid;
+ pool_exch_range = (mp->max_xid - mp->min_xid + 1) /
+ (fc_cpu_mask + 1);
+ }
+
mp->ep_pool = mempool_create_slab_pool(2, fc_em_cachep);
if (!mp->ep_pool)
goto free_mp;
@@ -2295,7 +2306,6 @@
* divided across all cpus. The exch pointers array memory is
* allocated for exch range per pool.
*/
- pool_exch_range = (mp->max_xid - mp->min_xid + 1) / (fc_cpu_mask + 1);
mp->pool_max_index = pool_exch_range - 1;
/*
Looks good to me. You may want to resend it after removing the email
conversation and please add proper patch description.
Thanks,
Bhanu
_______________________________________________
devel mailing list
devel@open-fcoe.org
https://lists.open-fcoe.org/mailman/listinfo/devel
_______________________________________________
devel mailing list
devel@open-fcoe.org
https://lists.open-fcoe.org/mailman/listinfo/devel
