Re: [PATCH] lib/scatterlist: Avoid a double memset

From: Christophe JAILLET
Date: Sun Sep 20 2020 - 04:54:34 EST


Le 20/09/2020 à 10:32, Julia Lawall a écrit :


On Sun, 20 Sep 2020, Joe Perches wrote:

On Sun, 2020-09-20 at 09:15 +0200, Christophe JAILLET wrote:
'sgl' is zeroed a few lines below in 'sg_init_table()'. There is no need to
clear it twice.

Remove the redundant initialization.

I didn't look very thoroughly, but there are at least
a few more of these with kcalloc and kzalloc like

block/bsg-lib.c- size_t sz = (sizeof(struct scatterlist) * req->nr_phys_segments);
block/bsg-lib.c-
block/bsg-lib.c- BUG_ON(!req->nr_phys_segments);
block/bsg-lib.c-
block/bsg-lib.c- buf->sg_list = kzalloc(sz, GFP_KERNEL);
block/bsg-lib.c- if (!buf->sg_list)
block/bsg-lib.c- return -ENOMEM;
block/bsg-lib.c: sg_init_table(buf->sg_list, req->nr_phys_segments);
---
drivers/target/target_core_rd.c- sg = kcalloc(sg_per_table + chain_entry, sizeof(*sg),
drivers/target/target_core_rd.c- GFP_KERNEL);
drivers/target/target_core_rd.c- if (!sg)
drivers/target/target_core_rd.c- return -ENOMEM;
drivers/target/target_core_rd.c-
drivers/target/target_core_rd.c: sg_init_table(sg, sg_per_table + chain_entry);
---
net/rds/rdma.c- sg = kcalloc(nents, sizeof(*sg), GFP_KERNEL);
net/rds/rdma.c- if (!sg) {
net/rds/rdma.c- ret = -ENOMEM;
net/rds/rdma.c- goto out;
net/rds/rdma.c- }
net/rds/rdma.c- WARN_ON(!nents);
net/rds/rdma.c: sg_init_table(sg, nents);

I found 16 occurrences in the following files:

net/rds/rdma.c
drivers/infiniband/hw/efa/efa_verbs.c
drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.c
drivers/misc/mic/scif/scif_nodeqp.c
block/bsg-lib.c
drivers/dma/sh/rcar-dmac.c
drivers/spi/spi-topcliff-pch.c
net/sunrpc/xprtrdma/frwr_ops.c
drivers/dma/imx-dma.c
drivers/pci/p2pdma.c
drivers/dma/sh/shdma-base.c
drivers/target/target_core_rd.c
drivers/media/common/saa7146/saa7146_core.c
drivers/tty/serial/pch_uart.c
drivers/net/wireless/intel/iwlwifi/fw/dbg.c

julia


Also in:
sound/soc/sprd/sprd-pcm-dma.c
sound/soc/sprd/sprd-pcm-compress.c
which are a bit tricky. It also uses some, IMHO, pointless devm_ functions.

Feel free to propose patches. I'm not focused on that at the moment.

CJ