[v2,02/12] scsi: mpt3sas: Make MPI2_CONFIG_PAGE_IO_UNIT_8::Sensor[] a flexible array

This terminal 1-length variable array can be directly converted into
a C99 flexible array member.

As all users of MPI2_CONFIG_PAGE_IO_UNIT_8 (Mpi2IOUnitPage8_t) do not
use Sensor[], no further source changes are required to accommodate
its reduced sizeof():

- mpt3sas_config.c:mpt3sas_config_get_iounit_pg8() fetches a
  Mpi2IOUnitPage8_t into a caller-provided buffer, assuming
  sizeof(Mpi2IOUnitPage8_t) as the buffer size. It has one caller:

  - mpt3sas_base.c:_base_static_config_pages() passes the address of
    the Mpi2IOUnitPage8_t iounit_pg8 member of the per-adapter struct
    (struct MPT3SAS_ADAPTER *ioc) as the buffer. The assumed buffer
    size is therefore correct.

    However, the only subsequent use in mpt3sas of the thus populated
    ioc->iounit_pg8 is a little further on in the same function, and
    this use does not involve ioc->iounit_pg8.Sensor[].

    Note that iounit_pg8 occurs in the middle of the per-adapter
    struct, not at the end. The per-adapter struct is extensively
    used throughout mpt3sas even if its iounit_pg8 member isn't,
    resulting in an especially large amount of noise when comparing
    binary changes attributable to this commit.

Signed-off-by: James Seo <james@equiv.tech>
---
 drivers/scsi/mpt3sas/mpi/mpi2_cnfg.h | 10 +++-------
 1 file changed, 3 insertions(+), 7 deletions(-)

On Sun, Aug 06, 2023 at 10:05:54AM -0700, James Seo wrote:
> This terminal 1-length variable array can be directly converted into
> a C99 flexible array member.
> 
> As all users of MPI2_CONFIG_PAGE_IO_UNIT_8 (Mpi2IOUnitPage8_t) do not
> use Sensor[], no further source changes are required to accommodate
> its reduced sizeof():
> 
> - mpt3sas_config.c:mpt3sas_config_get_iounit_pg8() fetches a
>   Mpi2IOUnitPage8_t into a caller-provided buffer, assuming
>   sizeof(Mpi2IOUnitPage8_t) as the buffer size. It has one caller:
> 
>   - mpt3sas_base.c:_base_static_config_pages() passes the address of
>     the Mpi2IOUnitPage8_t iounit_pg8 member of the per-adapter struct
>     (struct MPT3SAS_ADAPTER *ioc) as the buffer. The assumed buffer
>     size is therefore correct.
> 
>     However, the only subsequent use in mpt3sas of the thus populated
>     ioc->iounit_pg8 is a little further on in the same function, and
>     this use does not involve ioc->iounit_pg8.Sensor[].
> 
>     Note that iounit_pg8 occurs in the middle of the per-adapter
>     struct, not at the end. The per-adapter struct is extensively

This is especially bad/weird. Flex arrays aren't supposed to live there,
so I think it'd be best to avoid this conversion (see below).

>     used throughout mpt3sas even if its iounit_pg8 member isn't,
>     resulting in an especially large amount of noise when comparing
>     binary changes attributable to this commit.

Since the size reduction makes it hard to validate, how about just
leaving it alone? Since nothing is using Sensor[], you could just make
it a single instance:

-     MPI2_IOUNIT8_SENSOR
-             Sensor[MPI2_IOUNITPAGE8_SENSOR_ENTRIES];/*0x10 */
+     MPI2_IOUNIT8_SENSOR     Sensor;                 /*0x10 */


or leave it as-is (i.e. drop this patch).

> 
> Signed-off-by: James Seo <james@equiv.tech>
> ---
>  drivers/scsi/mpt3sas/mpi/mpi2_cnfg.h | 10 +++-------
>  1 file changed, 3 insertions(+), 7 deletions(-)
> 
> diff --git a/drivers/scsi/mpt3sas/mpi/mpi2_cnfg.h b/drivers/scsi/mpt3sas/mpi/mpi2_cnfg.h
> index 42d820159c44..12b656bd883d 100644
> --- a/drivers/scsi/mpt3sas/mpi/mpi2_cnfg.h
> +++ b/drivers/scsi/mpt3sas/mpi/mpi2_cnfg.h
> @@ -1200,12 +1200,9 @@ typedef struct _MPI2_IOUNIT8_SENSOR {
>  #define MPI2_IOUNIT8_SENSOR_FLAGS_T0_ENABLE         (0x0001)
>  
>  /*
> - *Host code (drivers, BIOS, utilities, etc.) should leave this define set to
> - *one and check the value returned for NumSensors at runtime.
> + *Host code (drivers, BIOS, utilities, etc.) should check the value returned
> + *for NumSensors at runtime before using Sensor[].
>   */
> -#ifndef MPI2_IOUNITPAGE8_SENSOR_ENTRIES
> -#define MPI2_IOUNITPAGE8_SENSOR_ENTRIES     (1)
> -#endif
>  
>  typedef struct _MPI2_CONFIG_PAGE_IO_UNIT_8 {
>  	MPI2_CONFIG_PAGE_HEADER Header;                 /*0x00 */
> @@ -1214,8 +1211,7 @@ typedef struct _MPI2_CONFIG_PAGE_IO_UNIT_8 {
>  	U8                      NumSensors;             /*0x0C */
>  	U8                      PollingInterval;        /*0x0D */
>  	U16                     Reserved3;              /*0x0E */
> -	MPI2_IOUNIT8_SENSOR
> -		Sensor[MPI2_IOUNITPAGE8_SENSOR_ENTRIES];/*0x10 */
> +	MPI2_IOUNIT8_SENSOR     Sensor[];               /*0x10 */
>  } MPI2_CONFIG_PAGE_IO_UNIT_8,
>  	*PTR_MPI2_CONFIG_PAGE_IO_UNIT_8,
>  	Mpi2IOUnitPage8_t, *pMpi2IOUnitPage8_t;
> -- 
> 2.39.2
>

On Fri, Aug 25, 2023 at 01:37:09PM -0700, Kees Cook wrote:
> On Sun, Aug 06, 2023 at 10:05:54AM -0700, James Seo wrote:
>>     Note that iounit_pg8 occurs in the middle of the per-adapter
>>     struct, not at the end. The per-adapter struct is extensively
> 
> This is especially bad/weird. Flex arrays aren't supposed to live there,
> so I think it'd be best to avoid this conversion (see below).
>
>>     used throughout mpt3sas even if its iounit_pg8 member isn't,
>>     resulting in an especially large amount of noise when comparing
>>     binary changes attributable to this commit.
> 
> Since the size reduction makes it hard to validate, how about just
> leaving it alone? Since nothing is using Sensor[], you could just make
> it a single instance:
> 
> -     MPI2_IOUNIT8_SENSOR
> -             Sensor[MPI2_IOUNITPAGE8_SENSOR_ENTRIES];/*0x10 */
> +     MPI2_IOUNIT8_SENSOR     Sensor;                 /*0x10 */
> 
> 
> or leave it as-is (i.e. drop this patch).
> 

I'd prefer not to paper it over by just up and pretending it's not a
flex array at all, but leaving things as-is feels like a waste, and I
understand the need to be conservative with storage drivers.

How do you feel about removing the struct containing the flex array
from the middle of the per-adapter struct, as per patch 8 in this
series? Moving that patch before this one in the ordering would
cleanly fix the misplaced flex array, but I imagine you'd be
especially keen on seeing Broadcom's approval for that one.

In any case, I'm fine with turning this into a single instance if it
comes down to it.