Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Subject: Re: [PATCH v3 4/4] objtool: orc_gen: Move orc_entry out of
 instruction structure
To:     peterz@infradead.org
Cc:     linux-kernel@vger.kernel.org, jpoimboe@redhat.com,
        mhelsley@vmware.com, mbenes@suse.cz
References: <20200730094143.27494-1-jthierry@redhat.com>
 <20200730094143.27494-5-jthierry@redhat.com>
 <20200730100304.GI2655@hirez.programming.kicks-ass.net>
From:   Julien Thierry <jthierry@redhat.com>
Message-ID: <e4e239ad-120e-bd8f-4128-6976146c8512@redhat.com>
Date:   Thu, 30 Jul 2020 13:40:48 +0100
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:68.0) Gecko/20100101
 Thunderbird/68.5.0
MIME-Version: 1.0
In-Reply-To: <20200730100304.GI2655@hirez.programming.kicks-ass.net>
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Language: en-US
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk


On 7/30/20 11:03 AM, peterz@infradead.org wrote:
> On Thu, Jul 30, 2020 at 10:41:43AM +0100, Julien Thierry wrote:
>> One orc_entry is associated with each instruction in the object file,
>> but having the orc_entry contained by the instruction structure forces
>> architectures not implementing the orc subcommands to provide a dummy
>> definition of the orc_entry.
>>
>> Avoid that by having orc_entries in a separate list, part of the
>> objtool_file.
>>
> 
>> diff --git a/tools/objtool/orc_gen.c b/tools/objtool/orc_gen.c
>> index 66fd56c33303..00f1efd05653 100644
>> --- a/tools/objtool/orc_gen.c
>> +++ b/tools/objtool/orc_gen.c
>> @@ -9,18 +9,33 @@
>>   #include "check.h"
>>   #include "warn.h"
>>   
>> +struct orc_data {
>> +	struct list_head list;
>> +	struct instruction *insn;
>> +	struct orc_entry orc;
>> +};
>> +
>>   int create_orc(struct objtool_file *file)
>>   {
>>   	struct instruction *insn;
>>   
>>   	for_each_insn(file, insn) {
>> -		struct orc_entry *orc = &insn->orc;
>>   		struct cfi_reg *cfa = &insn->cfi.cfa;
>>   		struct cfi_reg *bp = &insn->cfi.regs[CFI_BP];
>> +		struct orc_entry *orc;
>> +		struct orc_data *od;
>>   
>>   		if (!insn->sec->text)
>>   			continue;
>>   
>> +		od = calloc(1, sizeof(*od));
>> +		if (!od)
>> +			return -1;
>> +		od->insn = insn;
>> +		list_add_tail(&od->list, &file->orc_data_list);
>> +
>> +		orc = &od->orc;
>> +
>>   		orc->end = insn->cfi.end;
>>   
>>   		if (cfa->base == CFI_UNDEFINED) {
> 
> This will dramatically increase the amount of allocation calls, what, if
> anything, does this do for the performance of objtool?
> 

I guess I forgot about the usecase of running objtool on vmlinux...

On a kernel build for x86_64 defconfig, the difference in time seems to 
be withing the noise.

But I agree the proposed code is not ideal and on the other we've tried 
avoiding #ifdef in the code. Ideally I'd have an empty orc_entry 
definition when SUBCMD_ORC is not implemented.

Would you have a suggested approach to do that?

Thanks,

-- 
Julien Thierry