The MMC IPC interrupt status and interrupt mask registers are of
little use as Ethernet statistics, but incrementing counters
based on the current interrupt and interrupt mask registers
makes them worse than useless.
For example, if the interrupt mask is set to 0x08420842,
the current code will increment by that amount each iteration,
leading to the following sequence of nonsense:
mmc_rx_ipc_intr_mask: 969816526
mmc_rx_ipc_intr_mask: 1108361744
Change the increment to a straight assignment to make the
statistics at least nominally useful.
Signed-off-by: Jesper Nilsson <[email protected]>
---
drivers/net/ethernet/stmicro/stmmac/mmc_core.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/drivers/net/ethernet/stmicro/stmmac/mmc_core.c b/drivers/net/ethernet/stmicro/stmmac/mmc_core.c
index 6a7c1d325c46..6051a22b3cec 100644
--- a/drivers/net/ethernet/stmicro/stmmac/mmc_core.c
+++ b/drivers/net/ethernet/stmicro/stmmac/mmc_core.c
@@ -280,8 +280,8 @@ static void dwmac_mmc_read(void __iomem *mmcaddr, struct stmmac_counters *mmc)
mmc->mmc_rx_vlan_frames_gb += readl(mmcaddr + MMC_RX_VLAN_FRAMES_GB);
mmc->mmc_rx_watchdog_error += readl(mmcaddr + MMC_RX_WATCHDOG_ERROR);
/* IPC */
- mmc->mmc_rx_ipc_intr_mask += readl(mmcaddr + MMC_RX_IPC_INTR_MASK);
- mmc->mmc_rx_ipc_intr += readl(mmcaddr + MMC_RX_IPC_INTR);
+ mmc->mmc_rx_ipc_intr_mask = readl(mmcaddr + MMC_RX_IPC_INTR_MASK);
+ mmc->mmc_rx_ipc_intr = readl(mmcaddr + MMC_RX_IPC_INTR);
/* IPv4 */
mmc->mmc_rx_ipv4_gd += readl(mmcaddr + MMC_RX_IPV4_GD);
mmc->mmc_rx_ipv4_hderr += readl(mmcaddr + MMC_RX_IPV4_HDERR);
---
base-commit: 0dd3ee31125508cd67f7e7172247f05b7fd1753a
change-id: 20240216-stmmac_stats-e3561d460d0e
Best regards,
--
/^JN - Jesper Nilsson
--
Jesper Nilsson -- [email protected]
On 2/16/24 07:24, Jesper Nilsson wrote:
> The MMC IPC interrupt status and interrupt mask registers are of
> little use as Ethernet statistics, but incrementing counters
> based on the current interrupt and interrupt mask registers
> makes them worse than useless.
>
> For example, if the interrupt mask is set to 0x08420842,
> the current code will increment by that amount each iteration,
> leading to the following sequence of nonsense:
>
> mmc_rx_ipc_intr_mask: 969816526
> mmc_rx_ipc_intr_mask: 1108361744
>
> Change the increment to a straight assignment to make the
> statistics at least nominally useful.
>
> Signed-off-by: Jesper Nilsson <[email protected]>
> ---
> drivers/net/ethernet/stmicro/stmmac/mmc_core.c | 4 ++--
> 1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/net/ethernet/stmicro/stmmac/mmc_core.c b/drivers/net/ethernet/stmicro/stmmac/mmc_core.c
> index 6a7c1d325c46..6051a22b3cec 100644
> --- a/drivers/net/ethernet/stmicro/stmmac/mmc_core.c
> +++ b/drivers/net/ethernet/stmicro/stmmac/mmc_core.c
> @@ -280,8 +280,8 @@ static void dwmac_mmc_read(void __iomem *mmcaddr, struct stmmac_counters *mmc)
> mmc->mmc_rx_vlan_frames_gb += readl(mmcaddr + MMC_RX_VLAN_FRAMES_GB);
> mmc->mmc_rx_watchdog_error += readl(mmcaddr + MMC_RX_WATCHDOG_ERROR);
> /* IPC */
> - mmc->mmc_rx_ipc_intr_mask += readl(mmcaddr + MMC_RX_IPC_INTR_MASK);
> - mmc->mmc_rx_ipc_intr += readl(mmcaddr + MMC_RX_IPC_INTR);
> + mmc->mmc_rx_ipc_intr_mask = readl(mmcaddr + MMC_RX_IPC_INTR_MASK);
> + mmc->mmc_rx_ipc_intr = readl(mmcaddr + MMC_RX_IPC_INTR);
So in premise I agree with the patch, that incrementing those is not the
right way to go about them. However these registers are currently
provided as part of the statistics set, but they should instead be
accessed via the register dumping method.
In either case you will get at best a snapshot of those two registers at
any given time and I suppose this can help diagnose a stuck RX
condition, but not much more than that.
--
Florian
On Fri, Feb 16, 2024 at 09:13:51AM -0800, Florian Fainelli wrote:
> On 2/16/24 07:24, Jesper Nilsson wrote:
> > The MMC IPC interrupt status and interrupt mask registers are of
> > little use as Ethernet statistics, but incrementing counters
> > based on the current interrupt and interrupt mask registers
> > makes them worse than useless.
> >
> > For example, if the interrupt mask is set to 0x08420842,
> > the current code will increment by that amount each iteration,
> > leading to the following sequence of nonsense:
> >
> > mmc_rx_ipc_intr_mask: 969816526
> > mmc_rx_ipc_intr_mask: 1108361744
> >
> > Change the increment to a straight assignment to make the
> > statistics at least nominally useful.
> >
> > Signed-off-by: Jesper Nilsson <[email protected]>
> > ---
> > drivers/net/ethernet/stmicro/stmmac/mmc_core.c | 4 ++--
> > 1 file changed, 2 insertions(+), 2 deletions(-)
> >
> > diff --git a/drivers/net/ethernet/stmicro/stmmac/mmc_core.c b/drivers/net/ethernet/stmicro/stmmac/mmc_core.c
> > index 6a7c1d325c46..6051a22b3cec 100644
> > --- a/drivers/net/ethernet/stmicro/stmmac/mmc_core.c
> > +++ b/drivers/net/ethernet/stmicro/stmmac/mmc_core.c
> > @@ -280,8 +280,8 @@ static void dwmac_mmc_read(void __iomem *mmcaddr, struct stmmac_counters *mmc)
> > mmc->mmc_rx_vlan_frames_gb += readl(mmcaddr + MMC_RX_VLAN_FRAMES_GB);
> > mmc->mmc_rx_watchdog_error += readl(mmcaddr + MMC_RX_WATCHDOG_ERROR);
> > /* IPC */
> > - mmc->mmc_rx_ipc_intr_mask += readl(mmcaddr + MMC_RX_IPC_INTR_MASK);
> > - mmc->mmc_rx_ipc_intr += readl(mmcaddr + MMC_RX_IPC_INTR);
> > + mmc->mmc_rx_ipc_intr_mask = readl(mmcaddr + MMC_RX_IPC_INTR_MASK);
> > + mmc->mmc_rx_ipc_intr = readl(mmcaddr + MMC_RX_IPC_INTR);
>
> So in premise I agree with the patch, that incrementing those is not the
> right way to go about them. However these registers are currently provided
> as part of the statistics set, but they should instead be accessed via the
> register dumping method.
>
> In either case you will get at best a snapshot of those two registers at any
> given time and I suppose this can help diagnose a stuck RX condition, but
> not much more than that.
Could you please clarify why do those CSRs state need to be exposed in
the statistics anyway? Who would need such information really?
Wouldn't that be better to just drop the
stmmac_counters::{mmc_rx_ipc_intr_mask,mmc_rx_ipc_intr}
fields? Is it because of the statistics nodes are a kind of kernel
ABI? Even in that case I don't see much reason to support something
that has been absolutely useless so far seeing the nodes currently
returning basically some random values.
-Serge(y)
> --
> Florian
>
>
On Fri, Feb 16, 2024 at 09:13:51AM -0800, Florian Fainelli wrote:
> So in premise I agree with the patch, that incrementing those is not the
> right way to go about them. However these registers are currently provided
> as part of the statistics set, but they should instead be accessed via the
> register dumping method.
You mean extending the dump register code to dump the MAC Management Counter
registers that are not counters?
From what I understand it's only the Rx and Tx interrupt and interrupt mask
registers that aren't counters. Oh, and the MMC control register itself?
To be honest, I don't think their use can justify the code churn.
> In either case you will get at best a snapshot of those two registers at any
> given time and I suppose this can help diagnose a stuck RX condition, but
> not much more than that.
Yeah, their use is very doubtful. For me, they only introduce more nonsense
data in my logs.
The Rx registers looks to have been added in the first version of the MMC
back in 2011, but the Tx registers never was.
In commit 1c901a46d57 Giuseppe mentions the MMC interrupts as something to
add later (if actually useful).
So Serge's suggestion to drop the entries completely is actually quite attractive.
> Florian
/^JN - Jesper Nilsson
--
Jesper Nilsson -- [email protected]
On 2/19/2024 2:40 AM, Jesper Nilsson wrote:
> On Fri, Feb 16, 2024 at 09:13:51AM -0800, Florian Fainelli wrote:
>> So in premise I agree with the patch, that incrementing those is not the
>> right way to go about them. However these registers are currently provided
>> as part of the statistics set, but they should instead be accessed via the
>> register dumping method.
>
> You mean extending the dump register code to dump the MAC Management Counter
> registers that are not counters?
Yes that is what I meant.
> From what I understand it's only the Rx and Tx interrupt and interrupt mask
> registers that aren't counters. Oh, and the MMC control register itself?
>
> To be honest, I don't think their use can justify the code churn.
Fair enough, it really depends whether they are actually useful in
troubleshooting or not.
>
>> In either case you will get at best a snapshot of those two registers at any
>> given time and I suppose this can help diagnose a stuck RX condition, but
>> not much more than that.
>
> Yeah, their use is very doubtful. For me, they only introduce more nonsense
> data in my logs.
>
> The Rx registers looks to have been added in the first version of the MMC
> back in 2011, but the Tx registers never was.
> In commit 1c901a46d57 Giuseppe mentions the MMC interrupts as something to
> add later (if actually useful).
>
> So Serge's suggestion to drop the entries completely is actually quite attractive.
No concerns from me to drop them, not much value in having inaccurate
information.
--
Florian
On Mon, 2024-02-19 at 11:40 +0100, Jesper Nilsson wrote:
> On Fri, Feb 16, 2024 at 09:13:51AM -0800, Florian Fainelli wrote:
> > So in premise I agree with the patch, that incrementing those is not the
> > right way to go about them. However these registers are currently provided
> > as part of the statistics set, but they should instead be accessed via the
> > register dumping method.
>
> You mean extending the dump register code to dump the MAC Management Counter
> registers that are not counters?
> > From what I understand it's only the Rx and Tx interrupt and interrupt mask
> registers that aren't counters. Oh, and the MMC control register itself?
>
> To be honest, I don't think their use can justify the code churn.
>
> > In either case you will get at best a snapshot of those two registers at any
> > given time and I suppose this can help diagnose a stuck RX condition, but
> > not much more than that.
>
> Yeah, their use is very doubtful. For me, they only introduce more nonsense
> data in my logs.
>
> The Rx registers looks to have been added in the first version of the MMC
> back in 2011, but the Tx registers never was.
> In commit 1c901a46d57 Giuseppe mentions the MMC interrupts as something to
> add later (if actually useful).
>
> So Serge's suggestion to drop the entries completely is actually quite attractive.
Please, go ahead and drop such entries.
Thank!
Paolo