net/gve: Enable stats reporting for GQ format
PATCH 1/1 net/gve: Enable stats reporting for GQ format
Read from shared region to retrieve imissed statistics for GQ from device. Tested using show port xstats <port-id> in interactive mode. This metric can be triggered by using queues > cores.
--- drivers/net/gve/base/gve_adminq.h | 11 ++++ drivers/net/gve/gve_ethdev.c | 95 +++++++++++++++++++++++++++++++ drivers/net/gve/gve_ethdev.h | 6 ++ 3 files changed, 112 insertions(+)
drivers/net/gve/base/gve_adminq.h +11 -0
@@ -314,6 +314,17 @@ struct gve_stats_report {GVE_CHECK_STRUCT_LEN (8 , gve_stats_report); +/* Numbers of gve tx/rx stats in stats report. */ +#define GVE_TX_STATS_REPORT_NUM 6 +#define GVE_RX_STATS_REPORT_NUM 2 + +/* Interval to schedule a stats report update, 20000ms. */ +#define GVE_STATS_REPORT_TIMER_PERIOD 20000 + +/* Numbers of NIC tx/rx stats in stats report. */ +#define NIC_TX_STATS_REPORT_NUM 0 +#define NIC_RX_STATS_REPORT_NUM 4 +enum gve_stat_names {/* stats from gve */ TX_WAKE_CNT =1 ,
drivers/net/gve/gve_ethdev.c +95 -0
137 lines not on this page · raw message
drivers/net/gve/gve_ethdev.h +6 -0
@@ -85,6 +85,7 @@ struct gve_rx_stats {uint64_t errors;uint64_t no_mbufs;uint64_t no_mbufs_bulk; +uint64_t imissed; };struct gve_xstats_name_offset {
@@ -272,6 +273,11 @@ struct gve_priv {struct gve_tx_queue **txqs;struct gve_rx_queue **rxqs; + +uint32_t stats_report_len; +const struct rte_memzone *stats_report_mem; +uint16_t stats_start_idx;/* start index of array of stats written by NIC */ +uint16_t stats_end_idx;/* end index of array of stats written by NIC */ }; static inline bool
Signature · 1 line
2.43.0.472.g3155946c3a-goog
Quoted 4 lines · Rushil Gupta
Read from shared region to retrieve imissed statistics for GQ from device. Tested using
show port xstats <port-id>in interactive mode. This metric can be triggered by using queues > cores.
Looks good but please check following comments:
Checkpatch gives warning on the patch title, and this patch adds 'imissed' support so it can be added to the patch title, something like:
"net/gve: enable imissed stats for GQ format"
<...>
Quoted 6 lines
drivers/net/gve/gve_ethdev.c +5
-0+static int gve_alloc_stats_report (struct gve_priv *priv, +uint16_t nb_tx_queues,uint16_t nb_rx_queues) +{ +char z_name[RTE_MEMZONE_NAMESIZE ]; +int tx_stats_cnt;
drivers/net/gve/gve_ethdev.c gve_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete) · line 142
Quoted 10 lines
drivers/net/gve/gve_ethdev.c +10
-0+int rx_stats_cnt; + + tx_stats_cnt = (GVE_TX_STATS_REPORT_NUM +NIC_TX_STATS_REPORT_NUM ) * + nb_tx_queues; + rx_stats_cnt = (GVE_RX_STATS_REPORT_NUM +NIC_RX_STATS_REPORT_NUM ) * + nb_rx_queues; + priv->stats_report_len =sizeof (struct gve_stats_report ) + +sizeof (struct stats ) * (tx_stats_cnt + rx_stats_cnt); + +snprintf (z_name,sizeof (z_name),"stats_report_%s" , priv->pci_dev->device.name);
Can you please add 'gve_' prefix to the memzone name, to prevent any possible collision.
<...>
drivers/net/gve/gve_ethdev.c gve_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete) · line 164
Quoted 5 lines
drivers/net/gve/gve_ethdev.c +4
-0+static void gve_free_stats_report (struct rte_eth_dev *dev) +{ +struct gve_priv *priv = dev->data->dev_private; +rte_memzone_free (priv->stats_report_mem);
What will happen if user asks stats/xstats after port stopped?
<...>
drivers/net/gve/gve_ethdev.c static int · line 447
Quoted 6 lines
gve_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats){drivers/net/gve/gve_ethdev.c +2
-0uint16_t i; +if (gve_is_gqi (dev->data->dev_private)) +gve_get_imissed_from_nic (dev);
This updates imissed in RxQ struct for all queues for basic stats, but what if user only calls xstats, I guess in that case stat won't be updated.
Quoted 32 lines · Ferruh Yigit
Quoted 1 line · Rushil Gupta
Read from shared region to retrieve imissed statistics for GQ from
device.
Quoted 3 lines
Tested using
show port xstats <port-id>in interactive mode. This metric can be triggered by using queues > cores.Looks good but please check following comments:
Checkpatch gives warning on the patch title, and this patch adds 'imissed' support so it can be added to the patch title, something like:
"net/gve: enable imissed stats for GQ format"<...>
drivers/net/gve/gve_ethdev.c gve_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
Quoted 8 lines
drivers/net/gve/gve_ethdev.c +8
-0+static int gve_alloc_stats_report (struct gve_priv *priv, +uint16_t nb_tx_queues,uint16_t nb_rx_queues) +{ +char z_name[RTE_MEMZONE_NAMESIZE ]; +int tx_stats_cnt; +int rx_stats_cnt; + + tx_stats_cnt = (GVE_TX_STATS_REPORT_NUM +NIC_TX_STATS_REPORT_NUM )*
Review
Quoted 2 lines
(quoted diff) +2
-0+ nb_tx_queues; + rx_stats_cnt = (GVE_RX_STATS_REPORT_NUM + NIC_RX_STATS_REPORT_NUM)*
drivers/net/gve/gve_ethdev.c gve_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
Quoted 5 lines
drivers/net/gve/gve_ethdev.c +5
-0+ nb_rx_queues; + priv->stats_report_len =sizeof (struct gve_stats_report ) + +sizeof (struct stats ) * (tx_stats_cnt + rx_stats_cnt); + + snprintf(z_name, sizeof(z_name), "stats_report_%s", priv->pci_dev->
Quoted 6 lines
device.name);
Quoted 1 line
Can you please add '
gve_' prefix to the memzone name, to prevent any possible collision.
Done.
Quoted 11 lines
<...>
drivers/net/gve/gve_ethdev.c gve_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete) · line 164
Quoted 5 lines
drivers/net/gve/gve_ethdev.c +4
-0+static void gve_free_stats_report (struct rte_eth_dev *dev) +{ +struct gve_priv *priv = dev->data->dev_private; +rte_memzone_free (priv->stats_report_mem);What will happen if user asks stats/xstats after port stopped?
Good catch. I have added a null check so that the driver doesn't try to read stats from memory region that doesn't exist.
Quoted 10 lines
<...>
drivers/net/gve/gve_ethdev.c static int · line 447
Quoted 6 lines
gve_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats){ uint16_t i;drivers/net/gve/gve_ethdev.c +2
-0+if (gve_is_gqi (dev->data->dev_private)) +gve_get_imissed_from_nic (dev);
Quoted 3 lines
This updates imissed in RxQ struct for all queues for basic stats, but what if user only calls xstats, I guess in that case stat won't be updated.
Yes; that is expected. Since imissed is a member of rte_eth_stats; calling gve_dev_stats_get is the right way to get this stat.
Quoted 11 lines
Quoted 11 lines
<...>
Quoted 6 lines
gve_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats){ uint16_t i;drivers/net/gve/gve_ethdev.c +2
-0+if (gve_is_gqi (dev->data->dev_private)) +gve_get_imissed_from_nic (dev);
Quoted 8 lines
Quoted 4 lines
This updates imissed in RxQ struct for all queues for basic stats, but what if user only calls xstats, I guess in that case stat won't be updated.
Yes; that is expected. Since imissed is a member of
rte_eth_stats; callinggve_dev_stats_getis the right way to get this stat.
I actually think that this should be a counter backed by an xstat as well. As far as I can tell xstats is meant to be a more flexible version of stats, and ultimately should be able to handle a superset of what basic stats can handle, for the purposes of querying, etc.
Quoted 76 lines · Rushil Gupta
Quoted 32 lines · Ferruh Yigit
Quoted 1 line · Rushil Gupta
Read from shared region to retrieve imissed statistics for GQ from
device.
Quoted 3 lines
Tested using
show port xstats <port-id>in interactive mode. This metric can be triggered by using queues > cores.Looks good but please check following comments:
Checkpatch gives warning on the patch title, and this patch adds 'imissed' support so it can be added to the patch title, something like:
"net/gve: enable imissed stats for GQ format"<...>
drivers/net/gve/gve_ethdev.c gve_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
Quoted 8 lines
drivers/net/gve/gve_ethdev.c +8
-0+static int gve_alloc_stats_report (struct gve_priv *priv, +uint16_t nb_tx_queues,uint16_t nb_rx_queues) +{ +char z_name[RTE_MEMZONE_NAMESIZE ]; +int tx_stats_cnt; +int rx_stats_cnt; + + tx_stats_cnt = (GVE_TX_STATS_REPORT_NUM +
NIC_TX_STATS_REPORT_NUM) *Review
Quoted 2 lines
(quoted diff) +2
-0+ nb_tx_queues; + rx_stats_cnt = (GVE_RX_STATS_REPORT_NUM +
NIC_RX_STATS_REPORT_NUM) *drivers/net/gve/gve_ethdev.c gve_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
Quoted 5 lines
drivers/net/gve/gve_ethdev.c +5
-0+ nb_rx_queues; + priv->stats_report_len =sizeof (struct gve_stats_report ) + +sizeof (struct stats ) * (tx_stats_cnt + rx_stats_cnt); + + snprintf(z_name, sizeof(z_name), "stats_report_%s",Quoted 6 lines
priv->pci_dev->device.name);Quoted 1 line
Can you please add '
gve_' prefix to the memzone name, to prevent any possible collision.Done.
Quoted 11 lines
<...>
drivers/net/gve/gve_ethdev.c gve_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete) · line 164
Quoted 5 lines
drivers/net/gve/gve_ethdev.c +4
-0+static void gve_free_stats_report (struct rte_eth_dev *dev) +{ +struct gve_priv *priv = dev->data->dev_private; +rte_memzone_free (priv->stats_report_mem);What will happen if user asks stats/xstats after port stopped?
Good catch. I have added a null check so that the driver doesn't try to read stats from memory region that doesn't exist.
Quoted 10 lines
<...>
drivers/net/gve/gve_ethdev.c static int · line 447
Quoted 6 lines
gve_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats){ uint16_t i;drivers/net/gve/gve_ethdev.c +2
-0+if (gve_is_gqi (dev->data->dev_private)) +gve_get_imissed_from_nic (dev);Quoted 4 lines
This updates imissed in RxQ struct for all queues for basic stats, but what if user only calls xstats, I guess in that case stat won't be updated.
Yes; that is expected. Since imissed is a member of
rte_eth_stats; callinggve_dev_stats_getis the right way to get this stat.
Signature · 1 line
Joshua Washington | Software Engineer | [email protected] | (414) 366-4423
Quoted 74 lines · Rushil Gupta
On Fri, Jan 12, 2024 at 8:36 PM Ferruh Yigit <[email protected] <mailto:[email protected]>> wrote:
On 12/22/2023 3:39 PM, Rushil Gupta wrote: > Read from shared region to retrieve imissed statistics for GQ from device. > Tested using `show port xstats <port-id>` in interactive mode. > This metric can be triggered by using queues > cores. > Looks good but please check following comments: Checkpatch gives warning on the patch title, and this patch adds 'imissed' support so it can be added to the patch title, something like: "net/gve: enable imissed stats for GQ format" <...> > +static int gve_alloc_stats_report(struct gve_priv *priv, > + uint16_t nb_tx_queues, uint16_t nb_rx_queues) > +{ > + char z_name[RTE_MEMZONE_NAMESIZE]; > + int tx_stats_cnt; > + int rx_stats_cnt; > + > + tx_stats_cnt = (GVE_TX_STATS_REPORT_NUM + NIC_TX_STATS_REPORT_NUM) * > + nb_tx_queues; > + rx_stats_cnt = (GVE_RX_STATS_REPORT_NUM + NIC_RX_STATS_REPORT_NUM) * > + nb_rx_queues; > + priv->stats_report_len = sizeof(struct gve_stats_report) + > + sizeof(struct stats) * (tx_stats_cnt + rx_stats_cnt); > + > + snprintf(z_name, sizeof(z_name), "stats_report_%s", priv->pci_dev->device.name <http://device.name>); > Can you please add 'gve_' prefix to the memzone name, to prevent any possible collision.Done.
<...> > +static void gve_free_stats_report(struct rte_eth_dev *dev) > +{ > + struct gve_priv *priv = dev->data->dev_private; > + rte_memzone_free(priv->stats_report_mem); > What will happen if user asks stats/xstats after port stopped?Good catch. I have added a null check so that the driver doesn't try to read stats from memory region that doesn't exist.
<...> > gve_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats) > { > uint16_t i; > + if (gve_is_gqi(dev->data->dev_private)) > + gve_get_imissed_from_nic(dev); > This updates imissed in RxQ struct for all queues for basic stats, but what if user only calls xstats, I guess in that case stat won't be updated.
Quoted 4 lines
Yes; that is expected. Since imissed is a member ofrte_eth_stats; callinggve_dev_stats_getis the right way to get this stat.
I don't think it is expected.
xstats contains the basic stats too, if users calls xstats API, expectation is to get correct values.
Those are fair points. I'll fix this by simply
calling gve_get_imissed_from_nic from gve_xstats_get in the v3 patch.
Quoted 87 lines · Ferruh Yigit
Quoted 12 lines · Rushil Gupta
On Fri, Jan 12, 2024 at 8:36 PM Ferruh Yigit <[email protected] <mailto:[email protected]>> wrote:
On 12/22/2023 3:39 PM, Rushil Gupta wrote:
> Read from shared region to retrieve imissed statistics for GQ from device.
> Tested usingshow port xstats <port-id>in interactive mode. > This metric can be triggered by using queues > cores. >Quoted 4 lines
Looks good but please check following comments:
Checkpatch gives warning on the patch title, and this patch adds 'imissed' support so it can be added to the patch title, something
like:
Quoted 44 lines
"net/gve: enable imissed stats for GQ format" <...> > +static int gve_alloc_stats_report(struct gve_priv *priv, > + uint16_t nb_tx_queues, uint16_t nb_rx_queues) > +{ > + char z_name[RTE_MEMZONE_NAMESIZE]; > + int tx_stats_cnt; > + int rx_stats_cnt; > + > + tx_stats_cnt = (GVE_TX_STATS_REPORT_NUM + NIC_TX_STATS_REPORT_NUM) * > + nb_tx_queues; > + rx_stats_cnt = (GVE_RX_STATS_REPORT_NUM + NIC_RX_STATS_REPORT_NUM) * > + nb_rx_queues; > + priv->stats_report_len = sizeof(struct gve_stats_report) + > + sizeof(struct stats) * (tx_stats_cnt + rx_stats_cnt); > + > + snprintf(z_name, sizeof(z_name), "stats_report_%s", priv->pci_dev->device.name <http://device.name>); > Can you please add 'gve_' prefix to the memzone name, to prevent any possible collision.Done.
<...> > +static void gve_free_stats_report(struct rte_eth_dev *dev) > +{ > + struct gve_priv *priv = dev->data->dev_private; > + rte_memzone_free(priv->stats_report_mem); > What will happen if user asks stats/xstats after port stopped?Good catch. I have added a null check so that the driver doesn't try to read stats from memory region that doesn't exist.
Quoted 11 lines
<...> > gve_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats) > { > uint16_t i; > + if (gve_is_gqi(dev->data->dev_private)) > + gve_get_imissed_from_nic(dev); > This updates imissed in RxQ struct for all queues for basic stats,but
Quoted 7 lines
what if user only calls xstats, I guess in that case stat won't be updated.
Yes; that is expected. Since imissed is a member of
rte_eth_stats; callinggve_dev_stats_getis the right way to get this stat.I don't think it is expected.
xstats contains the basic stats too, if users calls xstats API, expectation is to get correct values.
PATCH v2 1/1 net/gve: enable imissed stats for GQ format
In reply to Rushil GuptaRead from shared region to retrieve imissed statistics for GQ from device. Tested using show port xstats <port-id> in interactive mode.
--- drivers/net/gve/base/gve_adminq.h | 11 ++++ drivers/net/gve/gve_ethdev.c | 100 ++++++++++++++++++++++++++++++ drivers/net/gve/gve_ethdev.h | 6 ++ 3 files changed, 117 insertions(+)
drivers/net/gve/base/gve_adminq.h +11 -0
@@ -314,6 +314,17 @@ struct gve_stats_report {GVE_CHECK_STRUCT_LEN (8 , gve_stats_report); +/* Numbers of gve tx/rx stats in stats report. */ +#define GVE_TX_STATS_REPORT_NUM 6 +#define GVE_RX_STATS_REPORT_NUM 2 + +/* Interval to schedule a stats report update, 20000ms. */ +#define GVE_STATS_REPORT_TIMER_PERIOD 20000 + +/* Numbers of NIC tx/rx stats in stats report. */ +#define NIC_TX_STATS_REPORT_NUM 0 +#define NIC_RX_STATS_REPORT_NUM 4 +enum gve_stat_names {/* stats from gve */ TX_WAKE_CNT =1 ,
drivers/net/gve/gve_ethdev.c +100 -0
142 lines not on this page · raw message
drivers/net/gve/gve_ethdev.h +6 -0
@@ -85,6 +85,7 @@ struct gve_rx_stats {uint64_t errors;uint64_t no_mbufs;uint64_t no_mbufs_bulk; +uint64_t imissed; };struct gve_xstats_name_offset {
@@ -272,6 +273,11 @@ struct gve_priv {struct gve_tx_queue **txqs;struct gve_rx_queue **rxqs; + +uint32_t stats_report_len; +const struct rte_memzone *gve_stats_report_mem; +uint16_t stats_start_idx;/* start index of array of stats written by NIC */ +uint16_t stats_end_idx;/* end index of array of stats written by NIC */ }; static inline bool
Signature · 1 line
2.34.1
PATCH v3 1/1 net/gve: enable imissed stats for GQ format
In reply to Rushil GuptaRead from shared region to retrieve imissed statistics for GQ from device.
--- drivers/net/gve/base/gve_adminq.h | 11 ++++ drivers/net/gve/gve_ethdev.c | 103 ++++++++++++++++++++++++++++++ drivers/net/gve/gve_ethdev.h | 6 ++ 3 files changed, 120 insertions(+)
drivers/net/gve/base/gve_adminq.h +11 -0
@@ -314,6 +314,17 @@ struct gve_stats_report {GVE_CHECK_STRUCT_LEN (8 , gve_stats_report); +/* Numbers of gve tx/rx stats in stats report. */ +#define GVE_TX_STATS_REPORT_NUM 6 +#define GVE_RX_STATS_REPORT_NUM 2 + +/* Interval to schedule a stats report update, 20000ms. */ +#define GVE_STATS_REPORT_TIMER_PERIOD 20000 + +/* Numbers of NIC tx/rx stats in stats report. */ +#define NIC_TX_STATS_REPORT_NUM 0 +#define NIC_RX_STATS_REPORT_NUM 4 +enum gve_stat_names {/* stats from gve */ TX_WAKE_CNT =1 ,
drivers/net/gve/gve_ethdev.c +103 -0
151 lines not on this page · raw message
drivers/net/gve/gve_ethdev.h +6 -0
@@ -85,6 +85,7 @@ struct gve_rx_stats {uint64_t errors;uint64_t no_mbufs;uint64_t no_mbufs_bulk; +uint64_t imissed; };struct gve_xstats_name_offset {
@@ -272,6 +273,11 @@ struct gve_priv {struct gve_tx_queue **txqs;struct gve_rx_queue **rxqs; + +uint32_t stats_report_len; +const struct rte_memzone *gve_stats_report_mem; +uint16_t stats_start_idx;/* start index of array of stats written by NIC */ +uint16_t stats_end_idx;/* end index of array of stats written by NIC */ }; static inline bool
Signature · 1 line
2.34.1
Re: [v3] net/gve: enable imissed stats for GQ format
Quoted 5 lines · Rushil Gupta
Read from shared region to retrieve imissed statistics for GQ from device.
Signed-off-by Rushil Gupta <[email protected]>Reviewed-by Joshua Washington <[email protected]>
<...>
Quoted 11 lines
drivers/net/gve/gve_ethdev.c +10
-0+static int +gve_alloc_stats_report (struct gve_priv *priv, +uint16_t nb_tx_queues,uint16_t nb_rx_queues) +{ +char z_name[RTE_MEMZONE_NAMESIZE ]; +int tx_stats_cnt; +int rx_stats_cnt; + + tx_stats_cnt = (GVE_TX_STATS_REPORT_NUM +NIC_TX_STATS_REPORT_NUM ) * + nb_tx_queues;
drivers/net/gve/gve_ethdev.c gve_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete) · line 147
Quoted 10 lines
drivers/net/gve/gve_ethdev.c +10
-0+ rx_stats_cnt = (GVE_RX_STATS_REPORT_NUM +NIC_RX_STATS_REPORT_NUM ) * + nb_rx_queues; + priv->stats_report_len =sizeof (struct gve_stats_report ) + +sizeof (struct stats ) * (tx_stats_cnt + rx_stats_cnt); + +snprintf (z_name,sizeof (z_name),"stats_report_%s" , priv->pci_dev->device.name); + priv->gve_stats_report_mem =rte_memzone_reserve_aligned (z_name, + priv->stats_report_len, +rte_socket_id (), +RTE_MEMZONE_IOVA_CONTIG ,PAGE_SIZE );
Adding 'gve_' prefix to memzone name comment seems missed.
Re: [v3] net/gve: enable imissed stats for GQ format
I misinterpreted your comment earlier and prefixed memzone variable name with "gve" instead of memzone name. Will fix it in v4. Thanks!
Quoted 38 lines · Ferruh Yigit
Quoted 1 line · Rushil Gupta
Read from shared region to retrieve imissed statistics for GQ from
device.
Quoted 4 lines
Signed-off-by Rushil Gupta <[email protected]>Reviewed-by Joshua Washington <[email protected]><...>
drivers/net/gve/gve_ethdev.c gve_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
Quoted 9 lines
drivers/net/gve/gve_ethdev.c +9
-0+static int +gve_alloc_stats_report (struct gve_priv *priv, +uint16_t nb_tx_queues,uint16_t nb_rx_queues) +{ +char z_name[RTE_MEMZONE_NAMESIZE ]; +int tx_stats_cnt; +int rx_stats_cnt; + + tx_stats_cnt = (GVE_TX_STATS_REPORT_NUM +NIC_TX_STATS_REPORT_NUM )*
Review
Quoted 2 lines
(quoted diff) +2
-0+ nb_tx_queues; + rx_stats_cnt = (GVE_RX_STATS_REPORT_NUM + NIC_RX_STATS_REPORT_NUM)*
drivers/net/gve/gve_ethdev.c gve_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
Quoted 5 lines
drivers/net/gve/gve_ethdev.c +5
-0+ nb_rx_queues; + priv->stats_report_len =sizeof (struct gve_stats_report ) + +sizeof (struct stats ) * (tx_stats_cnt + rx_stats_cnt); + + snprintf(z_name, sizeof(z_name), "stats_report_%s", priv->pci_dev->device.name);
drivers/net/gve/gve_ethdev.c gve_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete) · line 147
Quoted 5 lines
drivers/net/gve/gve_ethdev.c +4
-0+ priv->gve_stats_report_mem =rte_memzone_reserve_aligned (z_name, + priv->stats_report_len, +rte_socket_id (), +RTE_MEMZONE_IOVA_CONTIG ,PAGE_SIZE );Adding '
gve_' prefix to memzone name comment seems missed.
PATCH v4 1/1 net/gve: enable imissed stats for GQ format
In reply to Rushil GuptaRead from shared region to retrieve imissed statistics for GQ from device.
--- drivers/net/gve/base/gve_adminq.h | 11 ++++ drivers/net/gve/gve_ethdev.c | 104 ++++++++++++++++++++++++++++++ drivers/net/gve/gve_ethdev.h | 6 ++ 3 files changed, 121 insertions(+)
drivers/net/gve/base/gve_adminq.h +11 -0
@@ -314,6 +314,17 @@ struct gve_stats_report {GVE_CHECK_STRUCT_LEN (8 , gve_stats_report); +/* Numbers of gve tx/rx stats in stats report. */ +#define GVE_TX_STATS_REPORT_NUM 6 +#define GVE_RX_STATS_REPORT_NUM 2 + +/* Interval to schedule a stats report update, 20000ms. */ +#define GVE_STATS_REPORT_TIMER_PERIOD 20000 + +/* Numbers of NIC tx/rx stats in stats report. */ +#define NIC_TX_STATS_REPORT_NUM 0 +#define NIC_RX_STATS_REPORT_NUM 4 +enum gve_stat_names {/* stats from gve */ TX_WAKE_CNT =1 ,
drivers/net/gve/gve_ethdev.c +104 -0
152 lines not on this page · raw message
drivers/net/gve/gve_ethdev.h +6 -0
@@ -85,6 +85,7 @@ struct gve_rx_stats {uint64_t errors;uint64_t no_mbufs;uint64_t no_mbufs_bulk; +uint64_t imissed; };struct gve_xstats_name_offset {
@@ -272,6 +273,11 @@ struct gve_priv {struct gve_tx_queue **txqs;struct gve_rx_queue **rxqs; + +uint32_t stats_report_len; +const struct rte_memzone *stats_report_mem; +uint16_t stats_start_idx;/* start index of array of stats written by NIC */ +uint16_t stats_end_idx;/* end index of array of stats written by NIC */ }; static inline bool
Signature · 1 line
2.34.1
Re: [v4] net/gve: enable imissed stats for GQ format
Quoted 5 lines · Rushil Gupta
Read from shared region to retrieve imissed statistics for GQ from device.
Signed-off-by Rushil Gupta <[email protected]>Reviewed-by Joshua Washington <[email protected]>
Applied to dpdk-next-net/main, thanks.