]> git.proxmox.com Git - mirror_ubuntu-disco-kernel.git/commitdiff
cxgb4: allow large buffer size to have page size
authorThadeu Lima de Souza Cascardo <cascardo@linux.vnet.ibm.com>
Mon, 23 Dec 2013 17:34:29 +0000 (15:34 -0200)
committerDavid S. Miller <davem@davemloft.net>
Sat, 4 Jan 2014 00:27:57 +0000 (19:27 -0500)
Since commit 52367a763d8046190754ab43743e42638564a2d1
("cxgb4/cxgb4vf: Code cleanup to enable T4 Configuration File support"),
we have failures like this during cxgb4 probe:

cxgb4 0000:01:00.4: bad SGE FL page buffer sizes [65536, 65536]
cxgb4: probe of 0000:01:00.4 failed with error -22

This happens whenever software parameters are used, without a
configuration file. That happens when the hardware was already
initialized (after kexec, or after csiostor is loaded).

It happens that these values are acceptable, rendering fl_pg_order equal
to 0, which is the case of a hard init when the page size is equal or
larger than 65536.

Accepting fl_large_pg equal to fl_small_pg solves the issue, and
shouldn't cause any trouble besides a possible performance reduction
when smaller pages are used. And that can be fixed by a configuration
file.

Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/ethernet/chelsio/cxgb4/sge.c

index cc380c36e1a8687cd1c7f3b4c3dbf2d0b8c1bf47..cc3511a5cd0c0f9ea54e0d186878821f3dad8096 100644 (file)
@@ -2581,7 +2581,7 @@ static int t4_sge_init_soft(struct adapter *adap)
        #undef READ_FL_BUF
 
        if (fl_small_pg != PAGE_SIZE ||
-           (fl_large_pg != 0 && (fl_large_pg <= fl_small_pg ||
+           (fl_large_pg != 0 && (fl_large_pg < fl_small_pg ||
                                  (fl_large_pg & (fl_large_pg-1)) != 0))) {
                dev_err(adap->pdev_dev, "bad SGE FL page buffer sizes [%d, %d]\n",
                        fl_small_pg, fl_large_pg);