scsi: pm80xx: Avoid -Wflex-array-member-not-at-end warning

PATCH Superseded Gustavo A. R. Silva 5 participants 6 messages Sep 3, 2025 – Sep 4, 2025

Subsystems (2)

SCSI PMC sierra PM8001 driver

lore

Share & tools

lore

Gustavo A. R. Silva

author

2025-09-03 18:44 UTC #0 lore raw

PATCH 1/1 scsi: pm80xx: Avoid -Wflex-array-member-not-at-end warning

-Wflex-array-member-not-at-end was introduced in GCC-14, and we are getting ready to enable it, globally.

Move the conflicting declarations to the end of the corresponding structures. Notice that struct ssp_response_iu is a flexible structure, this is a structure that contains a flexible-array member.

With these changes fix the following warnings:

drivers/scsi/pm8001/pm8001_hwi.h:342:33: warning: structure containing a flexible array member is not at the end of another structure [-Wflex-array-member-not-at-end] drivers/scsi/pm8001/pm80xx_hwi.h:561:32: warning: structure containing a flexible array member is not at the end of another structure [-Wflex-array-member-not-at-end]

Signed-off-by Gustavo A. R. Silva <[email protected]>

---
 drivers/scsi/pm8001/pm8001_hwi.h | 4 +++-
 drivers/scsi/pm8001/pm80xx_hwi.h | 4 +++-
 2 files changed, 6 insertions(+), 2 deletions(-)

drivers/scsi/pm8001/pm8001_hwi.h +3 -1

index fc2127dcb58d..7dc7870a8f86 100644
@@ -339,8 +339,10 @@ struct ssp_completion_resp {
 	__le32	status;
 	__le32	param;
 	__le32	ssptag_rescv_rescpad;
-	struct ssp_response_iu  ssp_resp_iu;
 	__le32	residual_count;
+
+	/* Must be last --ends in a flexible-array member. */
+	struct ssp_response_iu  ssp_resp_iu;
 } __attribute__((packed, aligned(4)));

drivers/scsi/pm8001/pm80xx_hwi.h +3 -1

index eb8fd37b2066..21afc28d9875 100644
@@ -558,8 +558,10 @@ struct ssp_completion_resp {
 	__le32	status;
 	__le32	param;
 	__le32	ssptag_rescv_rescpad;
-	struct ssp_response_iu ssp_resp_iu;
 	__le32	residual_count;
+
+	/* Must be last --ends in a flexible-array member. */
+	struct ssp_response_iu ssp_resp_iu;
 } __attribute__((packed, aligned(4)));
 
 #define SSP_RESCV_BIT	0x00010000

Signature · 1 line

2.43.0

Jinpu Wang

Acked-by on v1 1/1

2025-09-04 05:27 UTC #1 lore raw

Quoted 8 lines · Gustavo A. R. Silva

-Wflex-array-member-not-at-end was introduced in GCC-14, and we are getting ready to enable it, globally.
Move the conflicting declarations to the end of the corresponding structures. Notice that struct ssp_response_iu is a flexible structure, this is a structure that contains a flexible-array member.

Quoted 6 lines

With these changes fix the following warnings:
drivers/scsi/pm8001/pm8001_hwi.h:342:33: warning: structure containing a flexible array member is not at the end of another structure [-Wflex-array-member-not-at-end] drivers/scsi/pm8001/pm80xx_hwi.h:561:32: warning: structure containing a flexible array member is not at the end of another structure [-Wflex-array-member-not-at-end]
Signed-off-by Gustavo A. R. Silva <[email protected]>

Acked-by Jack Wang <[email protected]>

Quoted 40 lines

---
 drivers/scsi/pm8001/pm8001_hwi.h | 4 +++-
 drivers/scsi/pm8001/pm80xx_hwi.h | 4 +++-
 2 files changed, 6 insertions(+), 2 deletions(-)

drivers/scsi/pm8001/pm8001_hwi.h +3 -1

index fc2127dcb58d..7dc7870a8f86 100644
@@ -339,8 +339,10 @@ struct ssp_completion_resp {
        __le32  status;
        __le32  param;
        __le32  ssptag_rescv_rescpad;
-       struct ssp_response_iu  ssp_resp_iu;
        __le32  residual_count;
+
+       /* Must be last --ends in a flexible-array member. */
+       struct ssp_response_iu  ssp_resp_iu;
 } __attribute__((packed, aligned(4)));

drivers/scsi/pm8001/pm80xx_hwi.h +3 -2

index eb8fd37b2066..21afc28d9875 100644
@@ -558,8 +558,10 @@ struct ssp_completion_resp {
        __le32  status;
        __le32  param;
        __le32  ssptag_rescv_rescpad;
-       struct ssp_response_iu ssp_resp_iu;
        __le32  residual_count;
+
+       /* Must be last --ends in a flexible-array member. */
+       struct ssp_response_iu ssp_resp_iu;
 } __attribute__((packed, aligned(4)));
 
 #define SSP_RESCV_BIT  0x00010000
--

2.43.0

John Garry

2025-09-04 06:53 UTC #2 lore raw

In reply to Gustavo A. R. Silva

Quoted 23 lines · Gustavo A. R. Silva

-Wflex-array-member-not-at-end was introduced in GCC-14, and we are getting ready to enable it, globally.
Move the conflicting declarations to the end of the corresponding structures. Notice that struct ssp_response_iu is a flexible structure, this is a structure that contains a flexible-array member.
With these changes fix the following warnings:
drivers/scsi/pm8001/pm8001_hwi.h:342:33: warning: structure containing a flexible array member is not at the end of another structure [-Wflex-array-member-not-at-end] drivers/scsi/pm8001/pm80xx_hwi.h:561:32: warning: structure containing a flexible array member is not at the end of another structure [-Wflex-array-member-not-at-end]
Signed-off-by Gustavo A. R. Silva <[email protected]>
---
  drivers/scsi/pm8001/pm8001_hwi.h | 4 +++-
  drivers/scsi/pm8001/pm80xx_hwi.h | 4 +++-
  2 files changed, 6 insertions(+), 2 deletions(-)
drivers/scsi/pm8001/pm8001_hwi.h +0 -0
index fc2127dcb58d..7dc7870a8f86 100644

drivers/scsi/pm8001/pm8001_hwi.h struct ssp_completion_resp { · line 345

Quoted 8 lines

drivers/scsi/pm8001/pm8001_hwi.h +3 -1

@@ -339,8 +339,10 @@ struct ssp_completion_resp {
  	__le32	status;
  	__le32	param;
  	__le32	ssptag_rescv_rescpad;
-	struct ssp_response_iu  ssp_resp_iu;
  	__le32	residual_count;
+
+	/* Must be last --ends in a flexible-array member. */
+	struct ssp_response_iu  ssp_resp_iu;

this is a HW structure, right? I did not think that it is ok to simply re-order them...

Quoted 19 lines

drivers/scsi/pm8001/pm8001_hwi.h +0 -0

  } __attribute__((packed, aligned(4)));

drivers/scsi/pm8001/pm80xx_hwi.h +3 -1

index eb8fd37b2066..21afc28d9875 100644
@@ -558,8 +558,10 @@ struct ssp_completion_resp {
  	__le32	status;
  	__le32	param;
  	__le32	ssptag_rescv_rescpad;
-	struct ssp_response_iu ssp_resp_iu;
  	__le32	residual_count;
+
+	/* Must be last --ends in a flexible-array member. */
+	struct ssp_response_iu ssp_resp_iu;
  } __attribute__((packed, aligned(4)));
 
  #define SSP_RESCV_BIT	0x00010000

James Bottomley

maintainer

2025-09-04 12:39 UTC #3 lore raw

Quoted 10 lines · John Garry

Quoted 9 lines · Gustavo A. R. Silva

drivers/scsi/pm8001/pm8001_hwi.h +0 -0

index fc2127dcb58d..7dc7870a8f86 100644
@@ -339,8 +339,10 @@ struct ssp_completion_resp {
   	__le32	status;
   	__le32	param;
   	__le32	ssptag_rescv_rescpad;

Quoted 8 lines

drivers/scsi/pm8001/pm8001_hwi.h struct ssp_completion_resp { · line 345
Quoted 5 lines
drivers/scsi/pm8001/pm8001_hwi.h +3 -1
-	struct ssp_response_iu  ssp_resp_iu;
   	__le32	residual_count;
+
+	/* Must be last --ends in a flexible-array member. */
+	struct ssp_response_iu  ssp_resp_iu;
this is a HW structure, right? I did not think that it is ok to simply re-order them...

Agreed, this is a standards defined information unit corresponding to an on the wire data structure. The patch is clearly wrong.

That being said, the three things the flexible member can contain are no data, response data or sense data. None of them has a residual count at the beginning and, indeed, this field is never referred to in the driver, so it looks like it can simply be deleted to fix the warning.

That being said, this pattern of adding fields after flexible members to represent data that's common to all content types of the union is not unknown in SCSI so if you want to enable this warning, what are we supposed to do when we encounter a genuine use case?

Regards,

James

John Garry

2025-09-04 13:49 UTC #4 lore raw

Quoted 18 lines · James Bottomley

Quoted 17 lines · John Garry
Quoted 6 lines · Gustavo A. R. Silva
drivers/scsi/pm8001/pm8001_hwi.h +0 -0
index fc2127dcb58d..7dc7870a8f86 100644
drivers/scsi/pm8001/pm8001_hwi.h struct ssp_completion_resp { · line 345
Quoted 8 lines
drivers/scsi/pm8001/pm8001_hwi.h +3 -1
@@ -339,8 +339,10 @@ struct ssp_completion_resp {
   	__le32	status;
   	__le32	param;
   	__le32	ssptag_rescv_rescpad;
-	struct ssp_response_iu  ssp_resp_iu;
   	__le32	residual_count;
+
+	/* Must be last --ends in a flexible-array member. */
+	struct ssp_response_iu  ssp_resp_iu;
this is a HW structure, right? I did not think that it is ok to simply re-order them...

Quoted 8 lines

Agreed, this is a standards defined information unit corresponding to an on the wire data structure. The patch is clearly wrong.
That being said, the three things the flexible member can contain are no data, response data or sense data. None of them has a residual count at the beginning and, indeed, this field is never referred to in the driver, so it looks like it can simply be deleted to fix the warning.

Seems reasonable. I don't see how the sizeof(struct ssp_completion_resp) is relevant, as the size of the memory to hold this structure (and other response types) will be defined elsewhere.

Quoted 5 lines

That being said, this pattern of adding fields after flexible members to represent data that's common to all content types of the union is not unknown in SCSI so if you want to enable this warning, what are we supposed to do when we encounter a genuine use case?

Such a problem was solved in commit cd6856d38881, but I can't say that it is a good example as we simply dropped the flex array usage.

Thanks,
John

Gustavo A. R. Silva

2025-09-04 15:58 UTC #5 lore raw

In reply to James Bottomley

Quoted 20 lines · James Bottomley

Quoted 10 lines · John Garry
Quoted 9 lines · Gustavo A. R. Silva
drivers/scsi/pm8001/pm8001_hwi.h +0 -0
index fc2127dcb58d..7dc7870a8f86 100644
@@ -339,8 +339,10 @@ struct ssp_completion_resp {
   	__le32	status;
   	__le32	param;
   	__le32	ssptag_rescv_rescpad;
Quoted 8 lines
drivers/scsi/pm8001/pm8001_hwi.h struct ssp_completion_resp { · line 345
Quoted 5 lines
drivers/scsi/pm8001/pm8001_hwi.h +3 -1
-	struct ssp_response_iu  ssp_resp_iu;
   	__le32	residual_count;
+
+	/* Must be last --ends in a flexible-array member. */
+	struct ssp_response_iu  ssp_resp_iu;
this is a HW structure, right? I did not think that it is ok to simply re-order them...

Quoted 8 lines

Agreed, this is a standards defined information unit corresponding to an on the wire data structure. The patch is clearly wrong.
That being said, the three things the flexible member can contain are no data, response data or sense data. None of them has a residual count at the beginning and, indeed, this field is never referred to in the driver, so it looks like it can simply be deleted to fix the warning.

I see. I just submitted v2. Thanks!

Quoted 5 lines

That being said, this pattern of adding fields after flexible members to represent data that's common to all content types of the union is not unknown in SCSI so if you want to enable this warning, what are we supposed to do when we encounter a genuine use case?

It depends on the situation. Some people prefer to have a separate struct with only the header part of the flexible structure --this is excluding the flexible-array member (FAM), and then use an object of that header type embedded anywhere in any other struct. This of course (sometimes) implies having to do many other changes to accommodate the rest of the code accordingly, as in this[1] case.

To address the situation described above without necessarily having to create a separate header struct and change a lot of code, the struct_group() helper can be used [2].

In other cases, when for some reason the FAM has to be accessed through a composite struct, both struct_group() and container_of() (this to retrieve a pointer to the flexible structure and then access the FAM) can be used [3].

We have the new TRAILING_OVERLAP() helper that in many cases can be superior to the struct_group()/container_of() approach. For instance this[4] could've been fixed with the following shorter and simpler patch:

struct nfsacl_simple_acl {

-       struct posix_acl acl;
-       struct posix_acl_entry ace[4];
+       TRAILING_OVERLAP(struct posix_acl, acl, a_entries,

+ struct posix_acl_entry ace[4];
+ );
};

However, at the time we didn't have TRAILING_OVERLAP(). Also, this new macro is helping us to detect alignment issues and correct them [5]. These[6][7] are a couple more example of when and how to use this helper.

We also have the DEFINE_FLEX()/DEFINE_RAW_FLEX() helpers [8][9].

So, again, we can do different things depending on the situation and maintainer's preferences.

Ideally, FAMs-in-the-middle should be avoided, because it's so easy for them to open the door to memory corruption bugs[10] (to mention some).

Thanks
-Gustavo