Re: [PATCH 13/13] net/ionic: optimize device start operation

[Boyer, Andrew]

> On Feb 2, 2024, at 11:28 PM, Stephen Hemminger <step...@networkplumber.org> 
> wrote:
> 
> Caution: This message originated from an External Source. Use proper caution 
> when opening attachments, clicking links, or responding.
> 
> 
> On Fri, 2 Feb 2024 11:32:38 -0800
> Andrew Boyer <andrew.bo...@amd.com> wrote:
> 
>> +     memset(ctx, 0, sizeof(*ctx));
>> +     ctx->pending_work = true;
>> +     ctx->cmd.q_init.opcode = IONIC_CMD_Q_INIT;
>> +     ctx->cmd.q_init.type = q->type;
>> +     ctx->cmd.q_init.ver = lif->qtype_info[q->type].version;
>> +     ctx->cmd.q_init.index = rte_cpu_to_le_32(q->index);
>> +     ctx->cmd.q_init.flags = rte_cpu_to_le_16(IONIC_QINIT_F_ENA);
>> +     ctx->cmd.q_init.intr_index = rte_cpu_to_le_16(IONIC_INTR_NONE);
>> +     ctx->cmd.q_init.ring_size = rte_log2_u32(q->num_descs);
>> +     ctx->cmd.q_init.cq_ring_base = rte_cpu_to_le_64(cq->base_pa);
>> +     ctx->cmd.q_init.sg_ring_base = rte_cpu_to_le_64(q->sg_base_pa);
>> +
> 
> memset followed by assignment is technically slower than structure
> initialization because it requires two writes to the data.
> But the optimizer may in some cases figure that out.

Thanks for looking. Since DPDK requires C11 now, and I see other users of 
compound literals in lib/, I'll respin this to use that instead.
(The internet says unset fields are implicitly initialized to zero, hopefully 
that's correct.)

It makes the diff a lot cleaner, too.

Thanks,
Andrew