Message ID | 6b31f5f523af93d47cac37509caf8036e183e136.camel@gmail.com (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
Series | e1000e: Fix tx/rx counters | expand |
在 2023/4/10 23:33, Akihiko Odaki 写道: > On 2023/04/11 0:27, timothee.cocault@gmail.com wrote: >> The bytes and packets counter registers are cleared on read. >> >> Copying the "total counter" registers to the "good counter" registers >> has >> side effects. >> If the "total" register is never read by the OS, it only gets >> incremented. >> This leads to exponential growth of the "good" register. >> >> This commit increments the counters individually to avoid this. >> >> Signed-off-by: Timothée Cocault <timothee.cocault@gmail.com> > > Reviewed-by: Akihiko Odaki <akihiko.odaki@daynix.com> Queued. Thanks > >> --- >> hw/net/e1000.c | 5 ++--- >> hw/net/e1000e_core.c | 5 ++--- >> hw/net/e1000x_common.c | 5 ++--- >> hw/net/igb_core.c | 5 ++--- >> 4 files changed, 8 insertions(+), 12 deletions(-) >> >> diff --git a/hw/net/e1000.c b/hw/net/e1000.c >> index 23d660619f..59bacb5d3b 100644 >> --- a/hw/net/e1000.c >> +++ b/hw/net/e1000.c >> @@ -637,9 +637,8 @@ xmit_seg(E1000State *s) >> e1000x_inc_reg_if_not_full(s->mac_reg, TPT); >> e1000x_grow_8reg_if_not_full(s->mac_reg, TOTL, s->tx.size + 4); >> - s->mac_reg[GPTC] = s->mac_reg[TPT]; >> - s->mac_reg[GOTCL] = s->mac_reg[TOTL]; >> - s->mac_reg[GOTCH] = s->mac_reg[TOTH]; >> + e1000x_inc_reg_if_not_full(s->mac_reg, GPTC); >> + e1000x_grow_8reg_if_not_full(s->mac_reg, GOTCL, s->tx.size + 4); >> } >> static void >> diff --git a/hw/net/e1000e_core.c b/hw/net/e1000e_core.c >> index c0c09b6965..cfa3f55e96 100644 >> --- a/hw/net/e1000e_core.c >> +++ b/hw/net/e1000e_core.c >> @@ -711,9 +711,8 @@ e1000e_on_tx_done_update_stats(E1000ECore *core, >> struct NetTxPkt *tx_pkt) >> g_assert_not_reached(); >> } >> - core->mac[GPTC] = core->mac[TPT]; >> - core->mac[GOTCL] = core->mac[TOTL]; >> - core->mac[GOTCH] = core->mac[TOTH]; >> + e1000x_inc_reg_if_not_full(core->mac, GPTC); >> + e1000x_grow_8reg_if_not_full(core->mac, GOTCL, tot_len); >> } >> static void >> diff --git a/hw/net/e1000x_common.c b/hw/net/e1000x_common.c >> index b844af590a..4c8e7dcf70 100644 >> --- a/hw/net/e1000x_common.c >> +++ b/hw/net/e1000x_common.c >> @@ -220,15 +220,14 @@ e1000x_update_rx_total_stats(uint32_t *mac, >> e1000x_increase_size_stats(mac, PRCregs, data_fcs_size); >> e1000x_inc_reg_if_not_full(mac, TPR); >> - mac[GPRC] = mac[TPR]; >> + e1000x_inc_reg_if_not_full(mac, GPRC); >> /* TOR - Total Octets Received: >> * This register includes bytes received in a packet from the >> <Destination >> * Address> field through the <CRC> field, inclusively. >> * Always include FCS length (4) in size. >> */ >> e1000x_grow_8reg_if_not_full(mac, TORL, data_size + 4); >> - mac[GORCL] = mac[TORL]; >> - mac[GORCH] = mac[TORH]; >> + e1000x_grow_8reg_if_not_full(mac, GORCL, data_size + 4); >> } >> void >> diff --git a/hw/net/igb_core.c b/hw/net/igb_core.c >> index d733fed6cf..826e7a6cf1 100644 >> --- a/hw/net/igb_core.c >> +++ b/hw/net/igb_core.c >> @@ -538,9 +538,8 @@ igb_on_tx_done_update_stats(IGBCore *core, struct >> NetTxPkt *tx_pkt, int qn) >> g_assert_not_reached(); >> } >> - core->mac[GPTC] = core->mac[TPT]; >> - core->mac[GOTCL] = core->mac[TOTL]; >> - core->mac[GOTCH] = core->mac[TOTH]; >> + e1000x_inc_reg_if_not_full(core->mac, GPTC); >> + e1000x_grow_8reg_if_not_full(core->mac, GOTCL, tot_len); >> if (core->mac[MRQC] & 1) { >> uint16_t pool = qn % IGB_NUM_VM_POOLS; >
10.04.2023 18:27, timothee.cocault@gmail.com wrote: > The bytes and packets counter registers are cleared on read. > > Copying the "total counter" registers to the "good counter" registers has > side effects. > If the "total" register is never read by the OS, it only gets incremented. > This leads to exponential growth of the "good" register. > > This commit increments the counters individually to avoid this. Smells like a -stable material, is it not? Thanks, /mjt
On Thu, May 11, 2023 at 2:54 AM Michael Tokarev <mjt@tls.msk.ru> wrote: > > 10.04.2023 18:27, timothee.cocault@gmail.com wrote: > > The bytes and packets counter registers are cleared on read. > > > > Copying the "total counter" registers to the "good counter" registers has > > side effects. > > If the "total" register is never read by the OS, it only gets incremented. > > This leads to exponential growth of the "good" register. > > > > This commit increments the counters individually to avoid this. > > Smells like a -stable material, is it not? Yes, tagged for stable. Thanks > > Thanks, > > /mjt >
diff --git a/hw/net/e1000.c b/hw/net/e1000.c index 23d660619f..59bacb5d3b 100644 --- a/hw/net/e1000.c +++ b/hw/net/e1000.c @@ -637,9 +637,8 @@ xmit_seg(E1000State *s) e1000x_inc_reg_if_not_full(s->mac_reg, TPT); e1000x_grow_8reg_if_not_full(s->mac_reg, TOTL, s->tx.size + 4); - s->mac_reg[GPTC] = s->mac_reg[TPT]; - s->mac_reg[GOTCL] = s->mac_reg[TOTL]; - s->mac_reg[GOTCH] = s->mac_reg[TOTH]; + e1000x_inc_reg_if_not_full(s->mac_reg, GPTC); + e1000x_grow_8reg_if_not_full(s->mac_reg, GOTCL, s->tx.size + 4); } static void diff --git a/hw/net/e1000e_core.c b/hw/net/e1000e_core.c index c0c09b6965..cfa3f55e96 100644 --- a/hw/net/e1000e_core.c +++ b/hw/net/e1000e_core.c @@ -711,9 +711,8 @@ e1000e_on_tx_done_update_stats(E1000ECore *core, struct NetTxPkt *tx_pkt) g_assert_not_reached(); } - core->mac[GPTC] = core->mac[TPT]; - core->mac[GOTCL] = core->mac[TOTL]; - core->mac[GOTCH] = core->mac[TOTH]; + e1000x_inc_reg_if_not_full(core->mac, GPTC); + e1000x_grow_8reg_if_not_full(core->mac, GOTCL, tot_len); } static void diff --git a/hw/net/e1000x_common.c b/hw/net/e1000x_common.c index b844af590a..4c8e7dcf70 100644 --- a/hw/net/e1000x_common.c +++ b/hw/net/e1000x_common.c @@ -220,15 +220,14 @@ e1000x_update_rx_total_stats(uint32_t *mac, e1000x_increase_size_stats(mac, PRCregs, data_fcs_size); e1000x_inc_reg_if_not_full(mac, TPR); - mac[GPRC] = mac[TPR]; + e1000x_inc_reg_if_not_full(mac, GPRC); /* TOR - Total Octets Received: * This register includes bytes received in a packet from the <Destination * Address> field through the <CRC> field, inclusively. * Always include FCS length (4) in size. */ e1000x_grow_8reg_if_not_full(mac, TORL, data_size + 4); - mac[GORCL] = mac[TORL]; - mac[GORCH] = mac[TORH]; + e1000x_grow_8reg_if_not_full(mac, GORCL, data_size + 4); } void diff --git a/hw/net/igb_core.c b/hw/net/igb_core.c index d733fed6cf..826e7a6cf1 100644 --- a/hw/net/igb_core.c +++ b/hw/net/igb_core.c @@ -538,9 +538,8 @@ igb_on_tx_done_update_stats(IGBCore *core, struct NetTxPkt *tx_pkt, int qn) g_assert_not_reached(); } - core->mac[GPTC] = core->mac[TPT]; - core->mac[GOTCL] = core->mac[TOTL]; - core->mac[GOTCH] = core->mac[TOTH]; + e1000x_inc_reg_if_not_full(core->mac, GPTC); + e1000x_grow_8reg_if_not_full(core->mac, GOTCL, tot_len); if (core->mac[MRQC] & 1) { uint16_t pool = qn % IGB_NUM_VM_POOLS;
The bytes and packets counter registers are cleared on read. Copying the "total counter" registers to the "good counter" registers has side effects. If the "total" register is never read by the OS, it only gets incremented. This leads to exponential growth of the "good" register. This commit increments the counters individually to avoid this. Signed-off-by: Timothée Cocault <timothee.cocault@gmail.com> --- hw/net/e1000.c | 5 ++--- hw/net/e1000e_core.c | 5 ++--- hw/net/e1000x_common.c | 5 ++--- hw/net/igb_core.c | 5 ++--- 4 files changed, 8 insertions(+), 12 deletions(-)