On Wed, Jun 26, 2019 at 2:03 PM David Marchand <david.march...@redhat.com>
wrote:
>
>
> On Tue, Jun 25, 2019 at 5:51 PM Anatoly Burakov <anatoly.bura...@intel.com>
> wrote:
>
>> Linux EAL will attach the shared config at an arbitrary address,
>> find out where the shared config is mapped in the primary, and
>> then will reattach it at that exact address.
>>
>> FreeBSD version doesn't seem to go for that extra reattach step,
>> which makes one wonder how did it ever work in the first place.
>>
>> Fix the FreeBSD init to also reattach shared config to the exact
>> same place the primary process has it.
>>
>> Fixes: 764bf26873b9 ("add FreeBSD support")
>> Cc: bruce.richard...@intel.com
>> Cc: sta...@dpdk.org
>>
>> Signed-off-by: Anatoly Burakov <anatoly.bura...@intel.com>
>> ---
>> lib/librte_eal/freebsd/eal/eal.c | 36 ++++++++++++++++++++++++++++++++
>> 1 file changed, 36 insertions(+)
>>
>> diff --git a/lib/librte_eal/freebsd/eal/eal.c
>> b/lib/librte_eal/freebsd/eal/eal.c
>> index 8c399c799..ce7a5f91d 100644
>> --- a/lib/librte_eal/freebsd/eal/eal.c
>> +++ b/lib/librte_eal/freebsd/eal/eal.c
>> @@ -280,6 +280,41 @@ rte_eal_config_attach(void)
>> rte_config.mem_config = rte_mem_cfg_addr;
>> }
>>
>> +/* reattach the shared config at exact memory location primary process
>> has it */
>> +static void
>> +rte_eal_config_reattach(void)
>> +{
>> + struct rte_mem_config *mem_config;
>> + void *rte_mem_cfg_addr;
>> +
>> + if (internal_config.no_shconf)
>> + return;
>> +
>> + /* save the address primary process has mapped shared config to */
>> + rte_mem_cfg_addr =
>> + (void
>> *)(uintptr_t)rte_config.mem_config->mem_cfg_addr;
>>
>
> It should be within the 80 columns limit on a single line.
>
> +
>> + /* unmap original config */
>> + munmap(rte_config.mem_config, sizeof(struct rte_mem_config));
>>
>
> Hum, the previous mapping is PROT_WRITE while unneeded, right?
>
>
> +
>> + /* remap the config at proper address */
>> + mem_config = (struct rte_mem_config *) mmap(rte_mem_cfg_addr,
>> + sizeof(*mem_config), PROT_READ | PROT_WRITE,
>> MAP_SHARED,
>> + mem_cfg_fd, 0);
>>
>
> mem_cfg_fd should have been closed in rte_eal_config_attach().
> So it should fail here.
>
>
> + if (mem_config == MAP_FAILED || mem_config != rte_mem_cfg_addr) {
>> + if (mem_config != MAP_FAILED)
>> + /* errno is stale, don't use */
>> + rte_panic("Cannot mmap memory for rte_config at
>> [%p], got [%p]\n",
>> + rte_mem_cfg_addr, mem_config);
>> + else
>> + rte_panic("Cannot mmap memory for rte_config!
>> error %i (%s)\n",
>> + errno, strerror(errno));
>> + }
>> + close(mem_cfg_fd);
>> +
>> + rte_config.mem_config = mem_config;
>> +}
>> +
>> /* Detect if we are a primary or a secondary process */
>> enum rte_proc_type_t
>> eal_proc_type_detect(void)
>> @@ -318,6 +353,7 @@ rte_config_init(void)
>> case RTE_PROC_SECONDARY:
>> rte_eal_config_attach();
>> rte_eal_mcfg_wait_complete(rte_config.mem_config);
>> + rte_eal_config_reattach();
>> break;
>> case RTE_PROC_AUTO:
>> case RTE_PROC_INVALID:
>> --
>> 2.17.1
>>
>
>
> Is there a reason why FreeBSD EAL does not support the virtaddr hint like
> for Linux EAL?
>
Not sure you noticed, but Arnon also had sent a patch touching the config
mapping parts.
http://patchwork.dpdk.org/patch/54583/
--
David Marchand
