diff mbox series

examples/vhost_crypto: fix zero copy

Message ID 20181030144852.43339-1-roy.fan.zhang@intel.com (mailing list archive)
State Superseded, archived
Delegated to: Maxime Coquelin
Headers
Series examples/vhost_crypto: fix zero copy |

Checks

Context Check Description
ci/checkpatch success coding style OK
ci/Intel-compilation success Compilation OK

Commit Message

Fan Zhang Oct. 30, 2018, 2:48 p.m. UTC 
  This patch fixes the zero copy enable problem for vhost crypto
sample application.

For some Crypto PMDs such as AESNI-MB and AESNI-GCM the data to
be processed will be made a copy in the same buffer but next to the
data. For example, to encrypt 64 bytes data the PMD will copy this
data from offset 64 to offset 123. This requires the application
provides the buffer with at least double of the data size.

However there is no way for VMs to know this limitation. When
zero-copy is enabled in Vhost the PMD may overwrite the buffer
next to the VM data to be processed, and further cause problems
such as Segmentation Fault or even worse, crashes the VM.

To fix the problem the user should avoid enabling the zero copy
for these Crypto PMDs. This patch adds the checking of the PMD
names to see if zero copy can be applied.

Fixes: 709521f4c2cd ("examples/vhost_crypto: support multi-core")

Signed-off-by: Fan Zhang <roy.fan.zhang@intel.com>
---
 examples/vhost_crypto/main.c | 23 +++++++++++++++++++++--
 1 file changed, 21 insertions(+), 2 deletions(-)

Comments

Mattias Rönnblom Oct. 30, 2018, 7:38 p.m. UTC | #1 
On 2018-10-30 15:48, Fan Zhang wrote:
> This patch fixes the zero copy enable problem for vhost crypto
> sample application.
> 
> For some Crypto PMDs such as AESNI-MB and AESNI-GCM the data to
> be processed will be made a copy in the same buffer but next to the
> data. For example, to encrypt 64 bytes data the PMD will copy this
> data from offset 64 to offset 123. This requires the application
> provides the buffer with at least double of the data size.
> 
> However there is no way for VMs to know this limitation. When
> zero-copy is enabled in Vhost the PMD may overwrite the buffer
> next to the VM data to be processed, and further cause problems
> such as Segmentation Fault or even worse, crashes the VM.
> 
> To fix the problem the user should avoid enabling the zero copy
> for these Crypto PMDs. This patch adds the checking of the PMD
> names to see if zero copy can be applied.
> 
> Fixes: 709521f4c2cd ("examples/vhost_crypto: support multi-core")
> 
> Signed-off-by: Fan Zhang <roy.fan.zhang@intel.com>
> ---
>   examples/vhost_crypto/main.c | 23 +++++++++++++++++++++--
>   1 file changed, 21 insertions(+), 2 deletions(-)
> 
> diff --git a/examples/vhost_crypto/main.c b/examples/vhost_crypto/main.c
> index cbb5e49d2..887e3eb6f 100644
> --- a/examples/vhost_crypto/main.c
> +++ b/examples/vhost_crypto/main.c
> @@ -4,6 +4,7 @@
>   
>   #include <stdio.h>
>   #include <stdlib.h>
> +#include <string.h>
>   #include <unistd.h>
>   #include <stdbool.h>
>   #include <assert.h>
> @@ -442,8 +443,13 @@ free_resource(void)
>   		struct lcore_option *lo = &options.los[i];
>   		struct vhost_crypto_info *info = options.infos[i];
>   
> -		rte_mempool_free(info->cop_pool);
> -		rte_mempool_free(info->sess_pool);
> +		if (!info)
> +			continue;
> +
> +		if (info->cop_pool)
> +			rte_mempool_free(info->cop_pool);
> +		if (info->sess_pool)
> +			rte_mempool_free(info->sess_pool);
>   

rte_mempool_free() already does a NULL-check (as per libc free() 
convention), and if you are to do a NULL-check it should be an explicit 
one ("!= NULL").

>   		for (j = 0; j < lo->nb_sockets; j++) {
>   			rte_vhost_driver_unregister(lo->socket_files[i]);
> @@ -493,6 +499,19 @@ main(int argc, char *argv[])
>   		info->nb_vids = lo->nb_sockets;
>   
>   		rte_cryptodev_info_get(info->cid, &dev_info);
> +		if (options.zero_copy == RTE_VHOST_CRYPTO_ZERO_COPY_ENABLE) {
> +#define VHOST_CRYPTO_CDEV_NAME_AESNI_MB_PMD	crypto_aesni_mb
> +#define VHOST_CRYPTO_CDEV_NAME_AESNI_GCM_PMD	crypto_aesni_gcm

What's the purpose of these defines?

> +			if (strstr(dev_info.driver_name,
> +				RTE_STR(VHOST_CRYPTO_CDEV_NAME_AESNI_MB_PMD)) ||
> +				strstr(dev_info.driver_name,
> +				RTE_STR(VHOST_CRYPTO_CDEV_NAME_AESNI_GCM_PMD)))
> +			RTE_LOG(ERR, USER1, "Cannot enable Zero Copy to %s\n",
> +					dev_info.driver_name);

"Zero Copy to" should probably be "zero-copy in" or "Zero-copy in".
Maxime Coquelin Nov. 9, 2018, 11:40 a.m. UTC | #2 
Hi Fan,

Could you please have a look at Mattias comments and reply?

Thanks in advance,
Maxime

On 10/30/18 8:38 PM, Mattias Rönnblom wrote:
> On 2018-10-30 15:48, Fan Zhang wrote:
>> This patch fixes the zero copy enable problem for vhost crypto
>> sample application.
>>
>> For some Crypto PMDs such as AESNI-MB and AESNI-GCM the data to
>> be processed will be made a copy in the same buffer but next to the
>> data. For example, to encrypt 64 bytes data the PMD will copy this
>> data from offset 64 to offset 123. This requires the application
>> provides the buffer with at least double of the data size.
>>
>> However there is no way for VMs to know this limitation. When
>> zero-copy is enabled in Vhost the PMD may overwrite the buffer
>> next to the VM data to be processed, and further cause problems
>> such as Segmentation Fault or even worse, crashes the VM.
>>
>> To fix the problem the user should avoid enabling the zero copy
>> for these Crypto PMDs. This patch adds the checking of the PMD
>> names to see if zero copy can be applied.
>>
>> Fixes: 709521f4c2cd ("examples/vhost_crypto: support multi-core")
>>
>> Signed-off-by: Fan Zhang <roy.fan.zhang@intel.com>
>> ---
>>   examples/vhost_crypto/main.c | 23 +++++++++++++++++++++--
>>   1 file changed, 21 insertions(+), 2 deletions(-)
>>
>> diff --git a/examples/vhost_crypto/main.c b/examples/vhost_crypto/main.c
>> index cbb5e49d2..887e3eb6f 100644
>> --- a/examples/vhost_crypto/main.c
>> +++ b/examples/vhost_crypto/main.c
>> @@ -4,6 +4,7 @@
>>   #include <stdio.h>
>>   #include <stdlib.h>
>> +#include <string.h>
>>   #include <unistd.h>
>>   #include <stdbool.h>
>>   #include <assert.h>
>> @@ -442,8 +443,13 @@ free_resource(void)
>>           struct lcore_option *lo = &options.los[i];
>>           struct vhost_crypto_info *info = options.infos[i];
>> -        rte_mempool_free(info->cop_pool);
>> -        rte_mempool_free(info->sess_pool);
>> +        if (!info)
>> +            continue;
>> +
>> +        if (info->cop_pool)
>> +            rte_mempool_free(info->cop_pool);
>> +        if (info->sess_pool)
>> +            rte_mempool_free(info->sess_pool);
> 
> rte_mempool_free() already does a NULL-check (as per libc free() 
> convention), and if you are to do a NULL-check it should be an explicit 
> one ("!= NULL").
> 
>>           for (j = 0; j < lo->nb_sockets; j++) {
>>               rte_vhost_driver_unregister(lo->socket_files[i]);
>> @@ -493,6 +499,19 @@ main(int argc, char *argv[])
>>           info->nb_vids = lo->nb_sockets;
>>           rte_cryptodev_info_get(info->cid, &dev_info);
>> +        if (options.zero_copy == RTE_VHOST_CRYPTO_ZERO_COPY_ENABLE) {
>> +#define VHOST_CRYPTO_CDEV_NAME_AESNI_MB_PMD    crypto_aesni_mb
>> +#define VHOST_CRYPTO_CDEV_NAME_AESNI_GCM_PMD    crypto_aesni_gcm
> 
> What's the purpose of these defines?
> 
>> +            if (strstr(dev_info.driver_name,
>> +                RTE_STR(VHOST_CRYPTO_CDEV_NAME_AESNI_MB_PMD)) ||
>> +                strstr(dev_info.driver_name,
>> +                RTE_STR(VHOST_CRYPTO_CDEV_NAME_AESNI_GCM_PMD)))
>> +            RTE_LOG(ERR, USER1, "Cannot enable Zero Copy to %s\n",
>> +                    dev_info.driver_name);
> 
> "Zero Copy to" should probably be "zero-copy in" or "Zero-copy in".
Fan Zhang Nov. 14, 2018, 8:46 a.m. UTC | #3 
Hi Mattias,

Sorry for the late reply. Comments inline.

> -----Original Message-----
> From: Mattias Rönnblom [mailto:mattias.ronnblom@ericsson.com]
> Sent: Tuesday, October 30, 2018 7:38 PM
> To: Zhang, Roy Fan <roy.fan.zhang@intel.com>; dev@dpdk.org
> Cc: maxime.coquelin@redhat.com
> Subject: Re: [dpdk-dev] [PATCH] examples/vhost_crypto: fix zero copy
> 
> On 2018-10-30 15:48, Fan Zhang wrote:
> > This patch fixes the zero copy enable problem for vhost crypto sample
> > application.
> rte_mempool_free() already does a NULL-check (as per libc free()
> convention), and if you are to do a NULL-check it should be an explicit one
> ("!= NULL").
No problem, I can change that. 
> 
> >   		for (j = 0; j < lo->nb_sockets; j++) {
> >   			rte_vhost_driver_unregister(lo->socket_files[i]);
> > @@ -493,6 +499,19 @@ main(int argc, char *argv[])
> >   		info->nb_vids = lo->nb_sockets;
> >
> >   		rte_cryptodev_info_get(info->cid, &dev_info);
> > +		if (options.zero_copy ==
> RTE_VHOST_CRYPTO_ZERO_COPY_ENABLE) {
> > +#define VHOST_CRYPTO_CDEV_NAME_AESNI_MB_PMD
> 	crypto_aesni_mb
> > +#define VHOST_CRYPTO_CDEV_NAME_AESNI_GCM_PMD
> 	crypto_aesni_gcm
> 
> What's the purpose of these defines?
These two PMD names are defined inside the PMDs and are not exposed to the vhost library.
Also these two PMDs are the ones that copy to the end of the packet for computation purposes, which will cause the problem for virtio case.
 
> 
> > +			if (strstr(dev_info.driver_name,
> > +
> 	RTE_STR(VHOST_CRYPTO_CDEV_NAME_AESNI_MB_PMD)) ||
> > +				strstr(dev_info.driver_name,
> > +
> 	RTE_STR(VHOST_CRYPTO_CDEV_NAME_AESNI_GCM_PMD)))
> > +			RTE_LOG(ERR, USER1, "Cannot enable Zero Copy
> to %s\n",
> > +					dev_info.driver_name);
> 
> "Zero Copy to" should probably be "zero-copy in" or "Zero-copy in".
Thanks I will change that.

Regards,
Fan
diff mbox series

Patch

diff --git a/examples/vhost_crypto/main.c b/examples/vhost_crypto/main.c
index cbb5e49d2..887e3eb6f 100644
--- a/examples/vhost_crypto/main.c
+++ b/examples/vhost_crypto/main.c
@@ -4,6 +4,7 @@ 
 
 #include <stdio.h>
 #include <stdlib.h>
+#include <string.h>
 #include <unistd.h>
 #include <stdbool.h>
 #include <assert.h>
@@ -442,8 +443,13 @@  free_resource(void)
 		struct lcore_option *lo = &options.los[i];
 		struct vhost_crypto_info *info = options.infos[i];
 
-		rte_mempool_free(info->cop_pool);
-		rte_mempool_free(info->sess_pool);
+		if (!info)
+			continue;
+
+		if (info->cop_pool)
+			rte_mempool_free(info->cop_pool);
+		if (info->sess_pool)
+			rte_mempool_free(info->sess_pool);
 
 		for (j = 0; j < lo->nb_sockets; j++) {
 			rte_vhost_driver_unregister(lo->socket_files[i]);
@@ -493,6 +499,19 @@  main(int argc, char *argv[])
 		info->nb_vids = lo->nb_sockets;
 
 		rte_cryptodev_info_get(info->cid, &dev_info);
+		if (options.zero_copy == RTE_VHOST_CRYPTO_ZERO_COPY_ENABLE) {
+#define VHOST_CRYPTO_CDEV_NAME_AESNI_MB_PMD	crypto_aesni_mb
+#define VHOST_CRYPTO_CDEV_NAME_AESNI_GCM_PMD	crypto_aesni_gcm
+			if (strstr(dev_info.driver_name,
+				RTE_STR(VHOST_CRYPTO_CDEV_NAME_AESNI_MB_PMD)) ||
+				strstr(dev_info.driver_name,
+				RTE_STR(VHOST_CRYPTO_CDEV_NAME_AESNI_GCM_PMD)))
+			RTE_LOG(ERR, USER1, "Cannot enable Zero Copy to %s\n",
+					dev_info.driver_name);
+			ret = -EPERM;
+			goto error_exit;
+		}
+
 		if (dev_info.max_nb_queue_pairs < info->qid + 1) {
 			RTE_LOG(ERR, USER1, "Number of queues cannot over %u",
 					dev_info.max_nb_queue_pairs);