Re: [PATCH iproute2-next 2/4] rdma: Add resource FRMR pools show command

[Michael Gur <michaelgur@nvidia.com>]

On 3/7/2026 3:45 AM, David Ahern wrote:
> External email: Use caution opening links or attachments
>
>
> On 3/2/26 8:51 AM, Chiara Meiohas wrote:
>> diff --git a/rdma/res-frmr-pools.c b/rdma/res-frmr-pools.c
>> new file mode 100644
>> index 00000000..97d59705
>> --- /dev/null
>> +++ b/rdma/res-frmr-pools.c
>> @@ -0,0 +1,190 @@
>> +// SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB
>> +/*
>> + * res-frmr-pools.c  RDMA tool
>> + * Authors:    Michael Guralnik <michaelgur@nvidia.com>
>> + */
>> +
>> +#include "res.h"
>> +#include <inttypes.h>
>> +
>> +#define FRMR_POOL_KEY_SIZE 21
>> +#define FRMR_POOL_KEY_HEX_SIZE (FRMR_POOL_KEY_SIZE * 2)
>> +union frmr_pool_key {
>> +     struct {
>> +             uint8_t ats;
>> +             uint32_t access_flags;
>> +             uint64_t vendor_key;
>> +             uint64_t num_dma_blocks;
>> +     } __attribute__((packed)) fields;
> why is packed needed on this struct? why can't the fields be re-arranged
> to not require it and just let the extra 3B be at the end unused?

The reasoning was to keep fields that are more likely to be zeroes at 
MSB to allow shortening the pool hex key in input and output.

Giving this another thought, I now think we're better dropping the logic 
of variable length in the pool key encoding/decoding and keep the hex 
pool key at fixed length.
I'll rearrange fields to drop the packed attribute and have the last 3B 
reserved.

>> +     uint8_t raw[FRMR_POOL_KEY_SIZE];
>> +};
>> +