Re: [dpdk-dev] [PATCH 19.02 2/2] mem: use memfd for no-huge mode

["Burakov, Anatoly" <anatoly.burakov@intel.com>]

On 28-Nov-18 4:57 AM, Tiwei Bie wrote:
> On Tue, Nov 13, 2018 at 05:54:48PM +0000, Anatoly Burakov wrote:
>> When running in no-huge mode, we anonymously allocate our memory.
>> While this works for regular NICs and vdev's, it's not suitable
>> for memory sharing scenarios such as virtio with vhost_user
>> backend.
>>
>> To fix this, allocate no-huge memory using memfd, and register
>> it with memalloc just like any other memseg fd. This will enable
>> using rte_memseg_get_fd() API with --no-huge EAL flag.
>>
>> Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
>> ---
>>   lib/librte_eal/linuxapp/eal/eal_memory.c | 46 ++++++++++++++++++++++--
>>   1 file changed, 44 insertions(+), 2 deletions(-)
>>
>> diff --git a/lib/librte_eal/linuxapp/eal/eal_memory.c b/lib/librte_eal/linuxapp/eal/eal_memory.c
>> index 48b23ce19..8feac2c56 100644
>> --- a/lib/librte_eal/linuxapp/eal/eal_memory.c
>> +++ b/lib/librte_eal/linuxapp/eal/eal_memory.c
>> @@ -25,6 +25,7 @@
>>   #include <sys/time.h>
>>   #include <signal.h>
>>   #include <setjmp.h>
>> +#include <linux/memfd.h>
>>   #ifdef RTE_EAL_NUMA_AWARE_HUGEPAGES
>>   #include <numa.h>
>>   #include <numaif.h>
>> @@ -1345,12 +1346,15 @@ eal_legacy_hugepage_init(void)
>>   	/* hugetlbfs can be disabled */
>>   	if (internal_config.no_hugetlbfs) {
>>   		struct rte_memseg_list *msl;
>> +		int n_segs, cur_seg, fd, memfd, flags;
>>   		uint64_t page_sz;
>> -		int n_segs, cur_seg;
>>   
>>   		/* nohuge mode is legacy mode */
>>   		internal_config.legacy_mem = 1;
>>   
>> +		/* nohuge mode is single-file segments mode */
>> +		internal_config.single_file_segments = 1;
>> +
>>   		/* create a memseg list */
>>   		msl = &mcfg->memsegs[0];
>>   
>> @@ -1363,8 +1367,36 @@ eal_legacy_hugepage_init(void)
>>   			return -1;
>>   		}
>>   
>> +		/* set up parameters for anonymous mmap */
>> +		fd = -1;
>> +		flags = MAP_PRIVATE | MAP_ANONYMOUS;
>> +
>> +		/* create a memfd and store it in the segment fd table */
>> +		memfd = memfd_create("nohuge", 0);
>> +		if (memfd < 0) {
>> +			RTE_LOG(ERR, EAL, "Cannot create memfd: %s\n",
>> +					strerror(errno));
>> +			RTE_LOG(ERR, EAL, "Falling back to anonymous map\n");
>> +		} else {
>> +			/* we got an fd - now resize it */
>> +			if (ftruncate(memfd, internal_config.memory) < 0) {
>> +				RTE_LOG(ERR, EAL, "Cannot resize memfd: %s\n",
>> +						strerror(errno));
>> +				RTE_LOG(ERR, EAL, "Falling back to anonymous map\n");
>> +				close(memfd);
>> +			} else {
>> +				/* creating memfd-backed file was successful.
>> +				 * we want changes to memfd to be visible to
>> +				 * other processes (such as vhost backend), so
>> +				 * map it as shared memory.
>> +				 */
>> +				RTE_LOG(DEBUG, EAL, "Using memfd for anonymous memory\n");
>> +				fd = memfd;
>> +				flags = MAP_SHARED;
>> +			}
>> +		}
>>   		addr = mmap(NULL, internal_config.memory, PROT_READ | PROT_WRITE,
>> -				MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
>> +				flags, fd, 0);
>>   		if (addr == MAP_FAILED) {
>>   			RTE_LOG(ERR, EAL, "%s: mmap() failed: %s\n", __func__,
>>   					strerror(errno));
>> @@ -1375,6 +1407,16 @@ eal_legacy_hugepage_init(void)
>>   		msl->socket_id = 0;
>>   		msl->len = internal_config.memory;
>>   
>> +		/* we're in single-file segments mode, so only the segment list
>> +		 * fd needs to be set up.
>> +		 */
>> +		if (fd != -1) {
>> +			if (eal_memalloc_set_seg_list_fd(0, fd) < 0) {
>> +				RTE_LOG(ERR, EAL, "Cannot set up segment list fd\n");
>> +				/* not a serious error, proceed */
>> +			}
>> +		}
> 
> Hi Anatoly,
> 
> Thanks for the work!
> 
> It seems the support for getting fd offset is missing in no-huge
> mode. I got below error in virtio-user while trying this series
> with --no-huge:
> 
> update_memory_region(): Failed to get offset, ms=0x10002e000 rte_errno=19

That's weird, it should have been working. I'll look into it, thanks!

> 
> Thanks
> 
>> +
>>   		/* populate memsegs. each memseg is one page long */
>>   		for (cur_seg = 0; cur_seg < n_segs; cur_seg++) {
>>   			arr = &msl->memseg_arr;
>> -- 
>> 2.17.1
> 


-- 
Thanks,
Anatoly