megaraid_sas : Round down max sge supported by controller to power of two

Resending the patch. Addressed the review comments from Tomas Henzl.

Round down the max sge to power of two.

Earlier max sge limit is 70 SGE, which will allow block layer to send 280K IO frame.
It is optimal to provide max IO size aligned to the smallest possible stripe size.
E.a
Consider that we have configured RAID Volumes which does not allow Fast Path across the stripe.
Raid volume with stripe size = 256K, will have peformance hit if we get io frame of size 280K.
Driver will not send IO frame large than stripe size to the Fast Path.
Also, FW will convert 280K frame into 256K + 24K. This is an additional overhead.

Signed-off-by: Sumit Saxena <sumit.saxena@avagotech.com>
Signed-off-by: Kashyap Desai <kashyap.desai@avagotech.com>
Reviewed-by: Tomas Henzl <thenzl@redhat.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

diff --git a/drivers/scsi/megaraid/megaraid_sas_fusion.c b/drivers/scsi/megaraid/megaraid_sas_fusion.c
index ed24a848df5be3ec39d3e0d678a1d5edefa41477..d35ac34ea6242d3f7366aae541c9025757c92538 100644 (file)
--- a/drivers/scsi/megaraid/megaraid_sas_fusion.c
+++ b/drivers/scsi/megaraid/megaraid_sas_fusion.c
@@ -1001,8 +1001,8 @@ megasas_init_adapter_fusion(struct megasas_instance *instance)
        fusion->max_sge_in_chain =
                MEGASAS_MAX_SZ_CHAIN_FRAME / sizeof(union MPI2_SGE_IO_UNION);
 
-       instance->max_num_sge = fusion->max_sge_in_main_msg +
-               fusion->max_sge_in_chain - 2;
+       instance->max_num_sge = rounddown_pow_of_two(
+               fusion->max_sge_in_main_msg + fusion->max_sge_in_chain - 2);
 
        /* Used for pass thru MFI frame (DCMD) */
        fusion->chain_offset_mfi_pthru =