target: Don't zero pages used for data buffers

Doing alloc_page(GFP_KERNEL | __GFP_ZERO) to get pages used for data
buffers wastes a lot of CPU clearing pages that will be quickly be
overwritten by the actual data.  However, for emulated control
commands such as INQUIRY and so on, the code does assume that the
buffer is zeroed.

To avoid this CPU overhead, skip the __GFP_ZERO for commands that are
actually moving data, ie cmds that have SCF_SCSI_DATA_SG_IO_CDB set.

Signed-off-by: Roland Dreier <roland@purestorage.com>
Signed-off-by: Nicholas Bellinger <nab@linux-iscsi.org>

diff --git a/drivers/target/target_core_transport.c b/drivers/target/target_core_transport.c
index d3ddd13619492d87f680258149f80987988bb9e4..289bc0f125f9cdd413b36818366481ada4f2b1c5 100644 (file)
--- a/drivers/target/target_core_transport.c
+++ b/drivers/target/target_core_transport.c
@@ -3516,6 +3516,7 @@ transport_generic_get_mem(struct se_cmd *cmd)
        u32 length = cmd->data_length;
        unsigned int nents;
        struct page *page;
+       gfp_t zero_flag;
        int i = 0;
 
        nents = DIV_ROUND_UP(length, PAGE_SIZE);
@@ -3526,9 +3527,11 @@ transport_generic_get_mem(struct se_cmd *cmd)
        cmd->t_data_nents = nents;
        sg_init_table(cmd->t_data_sg, nents);
 
+       zero_flag = cmd->se_cmd_flags & SCF_SCSI_DATA_SG_IO_CDB ? 0 : __GFP_ZERO;
+
        while (length) {
                u32 page_len = min_t(u32, length, PAGE_SIZE);
-               page = alloc_page(GFP_KERNEL | __GFP_ZERO);
+               page = alloc_page(GFP_KERNEL | zero_flag);
                if (!page)
                        goto out;