Re: [RFC net-next 03/17] net/smc: Remove error handling of unregister_dmb()

From
Alexandra Winter <wintera@linux.ibm.com>
Date
2025-08-11 11:28:52
Also in
linux-rdma, linux-s390
Reply to
Dust Li, 2025-08-10

On 10.08.25 13:03, Dust Li wrote:
On 2025-08-06 17:41:08, Alexandra Winter wrote:
quoted
smcd_buf_free() calls smc_ism_unregister_dmb(lgr->smcd, buf_desc) and
then unconditionally frees buf_desc.

Remove the cleaning up of fields of buf_desc in
smc_ism_unregister_dmb(), because it is not helpful.

This removes the only usage of ISM_ERROR from the smc module. So move it
to drivers/s390/net/ism.h.

Signed-off-by: Alexandra Winter <wintera@linux.ibm.com>
Reviewed-by: Mahanta Jambigi <mjambigi@linux.ibm.com>
---
drivers/s390/net/ism.h |  1 +
include/net/smc.h      |  2 --
net/smc/smc_ism.c      | 14 +++++---------
net/smc/smc_ism.h      |  3 ++-
4 files changed, 8 insertions(+), 12 deletions(-)
diff --git a/drivers/s390/net/ism.h b/drivers/s390/net/ism.h
index 047fa6101555..b5b03db52fce 100644
--- a/drivers/s390/net/ism.h
+++ b/drivers/s390/net/ism.h
@@ -10,6 +10,7 @@
#include <asm/pci_insn.h>

#define UTIL_STR_LEN	16
+#define ISM_ERROR	0xFFFF

/*
 * Do not use the first word of the DMB bits to ensure 8 byte aligned access.
diff --git a/include/net/smc.h b/include/net/smc.h
index db84e4e35080..a9c023dd1380 100644
--- a/include/net/smc.h
+++ b/include/net/smc.h
@@ -44,8 +44,6 @@ struct smcd_dmb {
#define ISM_RESERVED_VLANID	0x1FFF

-#define ISM_ERROR	0xFFFF
-
struct smcd_dev;

struct smcd_gid {
diff --git a/net/smc/smc_ism.c b/net/smc/smc_ism.c
index 84f98e18c7db..a94e1450d095 100644
--- a/net/smc/smc_ism.c
+++ b/net/smc/smc_ism.c
@@ -205,13 +205,13 @@ int smc_ism_put_vlan(struct smcd_dev *smcd, unsigned short vlanid)
	return rc;
}

-int smc_ism_unregister_dmb(struct smcd_dev *smcd, struct smc_buf_desc *dmb_desc)
+void smc_ism_unregister_dmb(struct smcd_dev *smcd,
+			    struct smc_buf_desc *dmb_desc)
{
	struct smcd_dmb dmb;
-	int rc = 0;

	if (!dmb_desc->dma_addr)
-		return rc;
+		return;

	memset(&dmb, 0, sizeof(dmb));
	dmb.dmb_tok = dmb_desc->token;
@@ -219,13 +219,9 @@ int smc_ism_unregister_dmb(struct smcd_dev *smcd, struct smc_buf_desc *dmb_desc)
	dmb.cpu_addr = dmb_desc->cpu_addr;
	dmb.dma_addr = dmb_desc->dma_addr;
	dmb.dmb_len = dmb_desc->len;
-	rc = smcd->ops->unregister_dmb(smcd, &dmb);
-	if (!rc || rc == ISM_ERROR) {
-		dmb_desc->cpu_addr = NULL;
-		dmb_desc->dma_addr = 0;
-	}
+	smcd->ops->unregister_dmb(smcd, &dmb);
Hmm, I think the old way of handling error here is certainly not good.
But completely ignoring error handling here would make bugs harder
to detect.

What about adding a WARN_ON_ONCE(rc) ?

Also, I think we can just remove the rc == ISM_ERROR to remove
the dependency of ISM_ERROR in smc.

Best regards,
Dust
As I wrote in the commit message, I removed rc, because it is ignored by the caller anyhow today.
If you want to I can add it back into this function and then you can think about how SMC should
handle such an error.

My thoughts on this are:
There is not really much smc can do about a problem in unregister_dmb.

I think it is Linux strategy to report and handle errors at the lowest level possible. I have some
patches on my harddisk to improve error handling and error reporting of the ism device diver. And
we are already discussing internally which errors should do a WARN_ON_ONCE in the ism device driver.

So I don't think there is much to do in the smc layer at the moment.
How does that sound to you?







Thread (48 messages) 48 messages, 4 authors, last reply 2025-09-01
Keyboard shortcuts
hback out one level
jnext message in thread / next series-diff position
kprevious message in thread / previous series-diff position
ldrill in
Escclose help / fold thread tree
?toggle this help