diff mbox series

[RFC,net-next,v6,3/3] r8169: Implement dynamic ASPM mechanism

Message ID 20211007161552.272771-4-kai.heng.feng@canonical.com (mailing list archive)
State Superseded
Delegated to: Netdev Maintainers
Headers show
Series r8169: Implement dynamic ASPM mechanism for recent 1.0/2.5Gbps Realtek NICs | expand

Checks

Context Check Description
netdev/cover_letter success Series has a cover letter
netdev/fixes_present success Fixes tag not required for -next series
netdev/patch_count success Link
netdev/tree_selection success Clearly marked for net-next
netdev/subject_prefix success Link
netdev/cc_maintainers success CCed 5 of 5 maintainers
netdev/source_inline success Was 0 now: 0
netdev/verify_signedoff success Signed-off-by tag matches author and committer
netdev/module_param success Was 0 now: 0
netdev/build_32bit success Errors and warnings before: 0 this patch: 0
netdev/kdoc success Errors and warnings before: 0 this patch: 0
netdev/verify_fixes success No Fixes tag
netdev/checkpatch warning CHECK: struct mutex definition without comment WARNING: line length of 88 exceeds 80 columns
netdev/build_allmodconfig_warn success Errors and warnings before: 0 this patch: 0
netdev/header_inline success No static functions without inline keyword in header files

Commit Message

Kai-Heng Feng Oct. 7, 2021, 4:15 p.m. UTC
r8169 NICs on some platforms have abysmal speed when ASPM is enabled.
Same issue can be observed with older vendor drivers.

The issue is however solved by the latest vendor driver. There's a new
mechanism, which disables r8169's internal ASPM when the NIC traffic has
more than 10 packets per second, and vice versa. The possible reason for
this is likely because the buffer on the chip is too small for its ASPM
exit latency.

Realtek confirmed that all their PCIe LAN NICs, r8106, r8168 and r8125
use dynamic ASPM under Windows. So implement the same mechanism here to
resolve the issue.

Also introduce a lock to prevent race on accessing config registers.

Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=214307
Signed-off-by: Kai-Heng Feng <kai.heng.feng@canonical.com>
---
v6:
 - Wording change.
 - Add bugzilla link.

v5:
 - Split out aspm_manageable replacement as another patch.
 - Introduce a lock for lock_config_regs() and unlock_config_regs().

v4:
 - Squash two patches
 - Remove aspm_manageable and use pcie_aspm_capable()
   pcie_aspm_enabled() accordingly

v3:
 - Use msecs_to_jiffies() for delay time
 - Use atomic_t instead of mutex for bh
 - Mention the buffer size and ASPM exit latency in commit message

v2: 
 - Use delayed_work instead of timer_list to avoid interrupt context
 - Use mutex to serialize packet counter read/write
 - Wording change
 drivers/net/ethernet/realtek/r8169_main.c | 58 +++++++++++++++++++++--
 1 file changed, 53 insertions(+), 5 deletions(-)

Comments

Bjorn Helgaas Oct. 7, 2021, 7:11 p.m. UTC | #1
On Fri, Oct 08, 2021 at 12:15:52AM +0800, Kai-Heng Feng wrote:
> r8169 NICs on some platforms have abysmal speed when ASPM is enabled.
> Same issue can be observed with older vendor drivers.
> 
> The issue is however solved by the latest vendor driver. There's a new
> mechanism, which disables r8169's internal ASPM when the NIC traffic has
> more than 10 packets per second, and vice versa. The possible reason for
> this is likely because the buffer on the chip is too small for its ASPM
> exit latency.

Because the NIC works fine on some platforms with ASPM fully enabled,
I would describe this as a "workaround" for a bug where we don't know
the root cause, not a "solution".

> Realtek confirmed that all their PCIe LAN NICs, r8106, r8168 and r8125
> use dynamic ASPM under Windows. So implement the same mechanism here to
> resolve the issue.
> 
> Also introduce a lock to prevent race on accessing config registers.

Strictly speaking, the addition of the lock should be a separate patch
since it's not directly related to the ASPM change.

A little more below...

> Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=214307
> Signed-off-by: Kai-Heng Feng <kai.heng.feng@canonical.com>
> ---
> v6:
>  - Wording change.
>  - Add bugzilla link.
> 
> v5:
>  - Split out aspm_manageable replacement as another patch.
>  - Introduce a lock for lock_config_regs() and unlock_config_regs().
> 
> v4:
>  - Squash two patches
>  - Remove aspm_manageable and use pcie_aspm_capable()
>    pcie_aspm_enabled() accordingly
> 
> v3:
>  - Use msecs_to_jiffies() for delay time
>  - Use atomic_t instead of mutex for bh
>  - Mention the buffer size and ASPM exit latency in commit message
> 
> v2: 
>  - Use delayed_work instead of timer_list to avoid interrupt context
>  - Use mutex to serialize packet counter read/write
>  - Wording change
>  drivers/net/ethernet/realtek/r8169_main.c | 58 +++++++++++++++++++++--
>  1 file changed, 53 insertions(+), 5 deletions(-)
> 
> diff --git a/drivers/net/ethernet/realtek/r8169_main.c b/drivers/net/ethernet/realtek/r8169_main.c
> index 53936ebb3b3a6..9c10a908c08fb 100644
> --- a/drivers/net/ethernet/realtek/r8169_main.c
> +++ b/drivers/net/ethernet/realtek/r8169_main.c
> @@ -622,6 +622,11 @@ struct rtl8169_private {
>  	} wk;
>  
>  	unsigned supports_gmii:1;
> +	unsigned rtl_aspm_enabled:1;
> +	struct delayed_work aspm_toggle;
> +	atomic_t aspm_packet_count;
> +	struct mutex config_lock;
> +
>  	dma_addr_t counters_phys_addr;
>  	struct rtl8169_counters *counters;
>  	struct rtl8169_tc_offsets tc_offset;
> @@ -670,12 +675,14 @@ static inline struct device *tp_to_dev(struct rtl8169_private *tp)
>  
>  static void rtl_lock_config_regs(struct rtl8169_private *tp)
>  {
> +	mutex_lock(&tp->config_lock);
>  	RTL_W8(tp, Cfg9346, Cfg9346_Lock);
>  }
>  
>  static void rtl_unlock_config_regs(struct rtl8169_private *tp)
>  {
>  	RTL_W8(tp, Cfg9346, Cfg9346_Unlock);
> +	mutex_unlock(&tp->config_lock);
>  }
>  
>  static void rtl_pci_commit(struct rtl8169_private *tp)
> @@ -2669,6 +2676,8 @@ static void rtl_hw_aspm_clkreq_enable(struct rtl8169_private *tp, bool enable)
>  	if (!pcie_aspm_support_enabled() || !pcie_aspm_capable(pdev))
>  		return;
>  
> +	tp->rtl_aspm_enabled = enable;
> +
>  	if (enable) {
>  		RTL_W8(tp, Config5, RTL_R8(tp, Config5) | ASPM_en);
>  		RTL_W8(tp, Config2, RTL_R8(tp, Config2) | ClkReqEn);
> @@ -4407,6 +4416,7 @@ static void rtl_tx(struct net_device *dev, struct rtl8169_private *tp,
>  
>  	dirty_tx = tp->dirty_tx;
>  
> +	atomic_add(tp->cur_tx - dirty_tx, &tp->aspm_packet_count);
>  	while (READ_ONCE(tp->cur_tx) != dirty_tx) {
>  		unsigned int entry = dirty_tx % NUM_TX_DESC;
>  		u32 status;
> @@ -4551,6 +4561,8 @@ static int rtl_rx(struct net_device *dev, struct rtl8169_private *tp, int budget
>  		rtl8169_mark_to_asic(desc);
>  	}
>  
> +	atomic_add(count, &tp->aspm_packet_count);
> +
>  	return count;
>  }
>  
> @@ -4658,8 +4670,39 @@ static int r8169_phy_connect(struct rtl8169_private *tp)
>  	return 0;
>  }
>  
> +#define ASPM_PACKET_THRESHOLD 10
> +#define ASPM_TOGGLE_INTERVAL 1000
> +
> +static void rtl8169_aspm_toggle(struct work_struct *work)
> +{
> +	struct rtl8169_private *tp = container_of(work, struct rtl8169_private,
> +						  aspm_toggle.work);
> +	int packet_count;
> +	bool enable;
> +
> +	packet_count = atomic_xchg(&tp->aspm_packet_count, 0);
> +
> +	if (pcie_aspm_enabled(tp->pci_dev)) {
> +		enable = packet_count <= ASPM_PACKET_THRESHOLD;
> +
> +		if (tp->rtl_aspm_enabled != enable) {
> +			rtl_unlock_config_regs(tp);
> +			rtl_hw_aspm_clkreq_enable(tp, enable);
> +			rtl_lock_config_regs(tp);
> +		}
> +	} else if (tp->rtl_aspm_enabled) {
> +		rtl_unlock_config_regs(tp);
> +		rtl_hw_aspm_clkreq_enable(tp, false);
> +		rtl_lock_config_regs(tp);
> +	}

IIUC the way the "dynamic ASPM" works is that rtl8169_aspm_toggle()
runs every second (1000ms).  If the NIC has sent or received fewer
than 10 packets in the last second, you make sure ASPM is enabled.  If
it has sent or received more than 10 packets, you disable ASPM.

Since the disable is done in rtl_hw_aspm_clkreq_enable() with
chip-specific registers, I suppose lspci and the like still show ASPM
as being enabled.  Not really a problem, I guess.

It looks like this disables ASPM completely, even though the NIC
apparently works correctly with L0s and L1.1 enabled, right?

I suppose that on the Intel system, if we enable ASPM, the link goes
to L1.2, and the NIC immediately receives 1000 packets in that second
before we can disable ASPM again, we probably drop a few packets?

Whereas on the AMD system, we probably *never* drop any packets even
with L1.2 enabled all the time?

And if we actually knew the root cause and could set the correct LTR
values or whatever is wrong on the Intel system, we probably wouldn't
need this dynamic scheme?

> +	schedule_delayed_work(&tp->aspm_toggle, msecs_to_jiffies(ASPM_TOGGLE_INTERVAL));
> +}
> +
>  static void rtl8169_down(struct rtl8169_private *tp)
>  {
> +	cancel_delayed_work_sync(&tp->aspm_toggle);
> +
>  	/* Clear all task flags */
>  	bitmap_zero(tp->wk.flags, RTL_FLAG_MAX);
>  
> @@ -4686,6 +4729,10 @@ static void rtl8169_up(struct rtl8169_private *tp)
>  	rtl_reset_work(tp);
>  
>  	phy_start(tp->phydev);
> +
> +	/* pcie_aspm_capable may change after system resume */
> +	if (pcie_aspm_support_enabled() && pcie_aspm_capable(tp->pci_dev))
> +		schedule_delayed_work(&tp->aspm_toggle, 0);
>  }
>  
>  static int rtl8169_close(struct net_device *dev)
> @@ -5273,11 +5320,6 @@ static int rtl_init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
>  	if (rc)
>  		return rc;
>  
> -	/* Disable ASPM L1 as that cause random device stop working
> -	 * problems as well as full system hangs for some PCIe devices users.
> -	 */
> -	pci_disable_link_state(pdev, PCIE_LINK_STATE_L1);
> -
>  	/* enable device (incl. PCI PM wakeup and hotplug setup) */
>  	rc = pcim_enable_device(pdev);
>  	if (rc < 0) {
> @@ -5307,6 +5349,8 @@ static int rtl_init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
>  		return rc;
>  	}
>  
> +	mutex_init(&tp->config_lock);
> +
>  	tp->mmio_addr = pcim_iomap_table(pdev)[region];
>  
>  	xid = (RTL_R32(tp, TxConfig) >> 20) & 0xfcf;
> @@ -5344,6 +5388,10 @@ static int rtl_init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
>  
>  	INIT_WORK(&tp->wk.work, rtl_task);
>  
> +	INIT_DELAYED_WORK(&tp->aspm_toggle, rtl8169_aspm_toggle);
> +
> +	atomic_set(&tp->aspm_packet_count, 0);
> +
>  	rtl_init_mac_address(tp);
>  
>  	dev->ethtool_ops = &rtl8169_ethtool_ops;
> -- 
> 2.32.0
>
Kai-Heng Feng Oct. 8, 2021, 6:18 a.m. UTC | #2
On Fri, Oct 8, 2021 at 3:11 AM Bjorn Helgaas <helgaas@kernel.org> wrote:
>
> On Fri, Oct 08, 2021 at 12:15:52AM +0800, Kai-Heng Feng wrote:
> > r8169 NICs on some platforms have abysmal speed when ASPM is enabled.
> > Same issue can be observed with older vendor drivers.
> >
> > The issue is however solved by the latest vendor driver. There's a new
> > mechanism, which disables r8169's internal ASPM when the NIC traffic has
> > more than 10 packets per second, and vice versa. The possible reason for
> > this is likely because the buffer on the chip is too small for its ASPM
> > exit latency.
>
> Because the NIC works fine on some platforms with ASPM fully enabled,
> I would describe this as a "workaround" for a bug where we don't know
> the root cause, not a "solution".

OK, will change the wording.

>
> > Realtek confirmed that all their PCIe LAN NICs, r8106, r8168 and r8125
> > use dynamic ASPM under Windows. So implement the same mechanism here to
> > resolve the issue.
> >
> > Also introduce a lock to prevent race on accessing config registers.
>
> Strictly speaking, the addition of the lock should be a separate patch
> since it's not directly related to the ASPM change.

Will separate it to another patch.

>
> A little more below...
>
> > Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=214307
> > Signed-off-by: Kai-Heng Feng <kai.heng.feng@canonical.com>
> > ---
> > v6:
> >  - Wording change.
> >  - Add bugzilla link.
> >
> > v5:
> >  - Split out aspm_manageable replacement as another patch.
> >  - Introduce a lock for lock_config_regs() and unlock_config_regs().
> >
> > v4:
> >  - Squash two patches
> >  - Remove aspm_manageable and use pcie_aspm_capable()
> >    pcie_aspm_enabled() accordingly
> >
> > v3:
> >  - Use msecs_to_jiffies() for delay time
> >  - Use atomic_t instead of mutex for bh
> >  - Mention the buffer size and ASPM exit latency in commit message
> >
> > v2:
> >  - Use delayed_work instead of timer_list to avoid interrupt context
> >  - Use mutex to serialize packet counter read/write
> >  - Wording change
> >  drivers/net/ethernet/realtek/r8169_main.c | 58 +++++++++++++++++++++--
> >  1 file changed, 53 insertions(+), 5 deletions(-)
> >
> > diff --git a/drivers/net/ethernet/realtek/r8169_main.c b/drivers/net/ethernet/realtek/r8169_main.c
> > index 53936ebb3b3a6..9c10a908c08fb 100644
> > --- a/drivers/net/ethernet/realtek/r8169_main.c
> > +++ b/drivers/net/ethernet/realtek/r8169_main.c
> > @@ -622,6 +622,11 @@ struct rtl8169_private {
> >       } wk;
> >
> >       unsigned supports_gmii:1;
> > +     unsigned rtl_aspm_enabled:1;
> > +     struct delayed_work aspm_toggle;
> > +     atomic_t aspm_packet_count;
> > +     struct mutex config_lock;
> > +
> >       dma_addr_t counters_phys_addr;
> >       struct rtl8169_counters *counters;
> >       struct rtl8169_tc_offsets tc_offset;
> > @@ -670,12 +675,14 @@ static inline struct device *tp_to_dev(struct rtl8169_private *tp)
> >
> >  static void rtl_lock_config_regs(struct rtl8169_private *tp)
> >  {
> > +     mutex_lock(&tp->config_lock);
> >       RTL_W8(tp, Cfg9346, Cfg9346_Lock);
> >  }
> >
> >  static void rtl_unlock_config_regs(struct rtl8169_private *tp)
> >  {
> >       RTL_W8(tp, Cfg9346, Cfg9346_Unlock);
> > +     mutex_unlock(&tp->config_lock);
> >  }
> >
> >  static void rtl_pci_commit(struct rtl8169_private *tp)
> > @@ -2669,6 +2676,8 @@ static void rtl_hw_aspm_clkreq_enable(struct rtl8169_private *tp, bool enable)
> >       if (!pcie_aspm_support_enabled() || !pcie_aspm_capable(pdev))
> >               return;
> >
> > +     tp->rtl_aspm_enabled = enable;
> > +
> >       if (enable) {
> >               RTL_W8(tp, Config5, RTL_R8(tp, Config5) | ASPM_en);
> >               RTL_W8(tp, Config2, RTL_R8(tp, Config2) | ClkReqEn);
> > @@ -4407,6 +4416,7 @@ static void rtl_tx(struct net_device *dev, struct rtl8169_private *tp,
> >
> >       dirty_tx = tp->dirty_tx;
> >
> > +     atomic_add(tp->cur_tx - dirty_tx, &tp->aspm_packet_count);
> >       while (READ_ONCE(tp->cur_tx) != dirty_tx) {
> >               unsigned int entry = dirty_tx % NUM_TX_DESC;
> >               u32 status;
> > @@ -4551,6 +4561,8 @@ static int rtl_rx(struct net_device *dev, struct rtl8169_private *tp, int budget
> >               rtl8169_mark_to_asic(desc);
> >       }
> >
> > +     atomic_add(count, &tp->aspm_packet_count);
> > +
> >       return count;
> >  }
> >
> > @@ -4658,8 +4670,39 @@ static int r8169_phy_connect(struct rtl8169_private *tp)
> >       return 0;
> >  }
> >
> > +#define ASPM_PACKET_THRESHOLD 10
> > +#define ASPM_TOGGLE_INTERVAL 1000
> > +
> > +static void rtl8169_aspm_toggle(struct work_struct *work)
> > +{
> > +     struct rtl8169_private *tp = container_of(work, struct rtl8169_private,
> > +                                               aspm_toggle.work);
> > +     int packet_count;
> > +     bool enable;
> > +
> > +     packet_count = atomic_xchg(&tp->aspm_packet_count, 0);
> > +
> > +     if (pcie_aspm_enabled(tp->pci_dev)) {
> > +             enable = packet_count <= ASPM_PACKET_THRESHOLD;
> > +
> > +             if (tp->rtl_aspm_enabled != enable) {
> > +                     rtl_unlock_config_regs(tp);
> > +                     rtl_hw_aspm_clkreq_enable(tp, enable);
> > +                     rtl_lock_config_regs(tp);
> > +             }
> > +     } else if (tp->rtl_aspm_enabled) {
> > +             rtl_unlock_config_regs(tp);
> > +             rtl_hw_aspm_clkreq_enable(tp, false);
> > +             rtl_lock_config_regs(tp);
> > +     }
>
> IIUC the way the "dynamic ASPM" works is that rtl8169_aspm_toggle()
> runs every second (1000ms).  If the NIC has sent or received fewer
> than 10 packets in the last second, you make sure ASPM is enabled.  If
> it has sent or received more than 10 packets, you disable ASPM.

Yes, this is what this patch does.

>
> Since the disable is done in rtl_hw_aspm_clkreq_enable() with
> chip-specific registers, I suppose lspci and the like still show ASPM
> as being enabled.  Not really a problem, I guess.
>
> It looks like this disables ASPM completely, even though the NIC
> apparently works correctly with L0s and L1.1 enabled, right?

I've seen bug reports that ASPM L0s and L1.1 caused the NIC stops to working.
So dynamic ASPM strikes the right

>
> I suppose that on the Intel system, if we enable ASPM, the link goes
> to L1.2, and the NIC immediately receives 1000 packets in that second
> before we can disable ASPM again, we probably drop a few packets?
>
> Whereas on the AMD system, we probably *never* drop any packets even
> with L1.2 enabled all the time?

Yes and yes.

>
> And if we actually knew the root cause and could set the correct LTR
> values or whatever is wrong on the Intel system, we probably wouldn't
> need this dynamic scheme?

Because Realtek already implemented the dynamic ASPM workaround in
their Windows and Linux driver, they never bother to find the root
cause.
So we'll never know what really happens here.

Kai-Heng

>
> > +     schedule_delayed_work(&tp->aspm_toggle, msecs_to_jiffies(ASPM_TOGGLE_INTERVAL));
> > +}
> > +
> >  static void rtl8169_down(struct rtl8169_private *tp)
> >  {
> > +     cancel_delayed_work_sync(&tp->aspm_toggle);
> > +
> >       /* Clear all task flags */
> >       bitmap_zero(tp->wk.flags, RTL_FLAG_MAX);
> >
> > @@ -4686,6 +4729,10 @@ static void rtl8169_up(struct rtl8169_private *tp)
> >       rtl_reset_work(tp);
> >
> >       phy_start(tp->phydev);
> > +
> > +     /* pcie_aspm_capable may change after system resume */
> > +     if (pcie_aspm_support_enabled() && pcie_aspm_capable(tp->pci_dev))
> > +             schedule_delayed_work(&tp->aspm_toggle, 0);
> >  }
> >
> >  static int rtl8169_close(struct net_device *dev)
> > @@ -5273,11 +5320,6 @@ static int rtl_init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
> >       if (rc)
> >               return rc;
> >
> > -     /* Disable ASPM L1 as that cause random device stop working
> > -      * problems as well as full system hangs for some PCIe devices users.
> > -      */
> > -     pci_disable_link_state(pdev, PCIE_LINK_STATE_L1);
> > -
> >       /* enable device (incl. PCI PM wakeup and hotplug setup) */
> >       rc = pcim_enable_device(pdev);
> >       if (rc < 0) {
> > @@ -5307,6 +5349,8 @@ static int rtl_init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
> >               return rc;
> >       }
> >
> > +     mutex_init(&tp->config_lock);
> > +
> >       tp->mmio_addr = pcim_iomap_table(pdev)[region];
> >
> >       xid = (RTL_R32(tp, TxConfig) >> 20) & 0xfcf;
> > @@ -5344,6 +5388,10 @@ static int rtl_init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
> >
> >       INIT_WORK(&tp->wk.work, rtl_task);
> >
> > +     INIT_DELAYED_WORK(&tp->aspm_toggle, rtl8169_aspm_toggle);
> > +
> > +     atomic_set(&tp->aspm_packet_count, 0);
> > +
> >       rtl_init_mac_address(tp);
> >
> >       dev->ethtool_ops = &rtl8169_ethtool_ops;
> > --
> > 2.32.0
> >
Bjorn Helgaas Oct. 8, 2021, 1:58 p.m. UTC | #3
On Fri, Oct 08, 2021 at 02:18:55PM +0800, Kai-Heng Feng wrote:
> On Fri, Oct 8, 2021 at 3:11 AM Bjorn Helgaas <helgaas@kernel.org> wrote:
> > On Fri, Oct 08, 2021 at 12:15:52AM +0800, Kai-Heng Feng wrote:
> > > r8169 NICs on some platforms have abysmal speed when ASPM is enabled.
> > > Same issue can be observed with older vendor drivers.
> > >
> > > The issue is however solved by the latest vendor driver. There's a new
> > > mechanism, which disables r8169's internal ASPM when the NIC traffic has
> > > more than 10 packets per second, and vice versa. The possible reason for
> > > this is likely because the buffer on the chip is too small for its ASPM
> > > exit latency.
> > > ...

> > I suppose that on the Intel system, if we enable ASPM, the link goes
> > to L1.2, and the NIC immediately receives 1000 packets in that second
> > before we can disable ASPM again, we probably drop a few packets?
> >
> > Whereas on the AMD system, we probably *never* drop any packets even
> > with L1.2 enabled all the time?
> 
> Yes and yes.

The fact that we drop some packets with dynamic ASPM on the Intel
system means we must be giving up some performance.

And I guess that on the AMD system, we should get full performance but
we must be using a little more power (probably unmeasurable) because
ASPM *could* be always enabled but dynamic ASPM disables it some of
the time.

> > And if we actually knew the root cause and could set the correct LTR
> > values or whatever is wrong on the Intel system, we probably wouldn't
> > need this dynamic scheme?
> 
> Because Realtek already implemented the dynamic ASPM workaround in
> their Windows and Linux driver, they never bother to find the root
> cause.
> So we'll never know what really happens here.

Looks like it.  Somebody with a PCIe analyzer could probably make
progress, but I agree, that doesn't seem likely.

Realtek no doubt has the equipment to do this, but apparently they
don't think it's worthwhile.  In their defense, the Linux ASPM code is
pretty impenetrable and there could be a problem there that causes or
contributes to this.

Bjorn
Kai-Heng Feng Oct. 15, 2021, 4:11 a.m. UTC | #4
On Fri, Oct 8, 2021 at 9:58 PM Bjorn Helgaas <helgaas@kernel.org> wrote:
>
> On Fri, Oct 08, 2021 at 02:18:55PM +0800, Kai-Heng Feng wrote:
> > On Fri, Oct 8, 2021 at 3:11 AM Bjorn Helgaas <helgaas@kernel.org> wrote:
> > > On Fri, Oct 08, 2021 at 12:15:52AM +0800, Kai-Heng Feng wrote:
> > > > r8169 NICs on some platforms have abysmal speed when ASPM is enabled.
> > > > Same issue can be observed with older vendor drivers.
> > > >
> > > > The issue is however solved by the latest vendor driver. There's a new
> > > > mechanism, which disables r8169's internal ASPM when the NIC traffic has
> > > > more than 10 packets per second, and vice versa. The possible reason for
> > > > this is likely because the buffer on the chip is too small for its ASPM
> > > > exit latency.
> > > > ...
>
> > > I suppose that on the Intel system, if we enable ASPM, the link goes
> > > to L1.2, and the NIC immediately receives 1000 packets in that second
> > > before we can disable ASPM again, we probably drop a few packets?
> > >
> > > Whereas on the AMD system, we probably *never* drop any packets even
> > > with L1.2 enabled all the time?
> >
> > Yes and yes.
>
> The fact that we drop some packets with dynamic ASPM on the Intel
> system means we must be giving up some performance.
>
> And I guess that on the AMD system, we should get full performance but
> we must be using a little more power (probably unmeasurable) because
> ASPM *could* be always enabled but dynamic ASPM disables it some of
> the time.

Yes that's the case here.

>
> > > And if we actually knew the root cause and could set the correct LTR
> > > values or whatever is wrong on the Intel system, we probably wouldn't
> > > need this dynamic scheme?
> >
> > Because Realtek already implemented the dynamic ASPM workaround in
> > their Windows and Linux driver, they never bother to find the root
> > cause.
> > So we'll never know what really happens here.
>
> Looks like it.  Somebody with a PCIe analyzer could probably make
> progress, but I agree, that doesn't seem likely.
>
> Realtek no doubt has the equipment to do this, but apparently they
> don't think it's worthwhile.  In their defense, the Linux ASPM code is
> pretty impenetrable and there could be a problem there that causes or
> contributes to this.

I do hope they can put more effort on their ethernet driver like what
they do on their wireless drivers.

Kai-Heng

>
> Bjorn
diff mbox series

Patch

diff --git a/drivers/net/ethernet/realtek/r8169_main.c b/drivers/net/ethernet/realtek/r8169_main.c
index 53936ebb3b3a6..9c10a908c08fb 100644
--- a/drivers/net/ethernet/realtek/r8169_main.c
+++ b/drivers/net/ethernet/realtek/r8169_main.c
@@ -622,6 +622,11 @@  struct rtl8169_private {
 	} wk;
 
 	unsigned supports_gmii:1;
+	unsigned rtl_aspm_enabled:1;
+	struct delayed_work aspm_toggle;
+	atomic_t aspm_packet_count;
+	struct mutex config_lock;
+
 	dma_addr_t counters_phys_addr;
 	struct rtl8169_counters *counters;
 	struct rtl8169_tc_offsets tc_offset;
@@ -670,12 +675,14 @@  static inline struct device *tp_to_dev(struct rtl8169_private *tp)
 
 static void rtl_lock_config_regs(struct rtl8169_private *tp)
 {
+	mutex_lock(&tp->config_lock);
 	RTL_W8(tp, Cfg9346, Cfg9346_Lock);
 }
 
 static void rtl_unlock_config_regs(struct rtl8169_private *tp)
 {
 	RTL_W8(tp, Cfg9346, Cfg9346_Unlock);
+	mutex_unlock(&tp->config_lock);
 }
 
 static void rtl_pci_commit(struct rtl8169_private *tp)
@@ -2669,6 +2676,8 @@  static void rtl_hw_aspm_clkreq_enable(struct rtl8169_private *tp, bool enable)
 	if (!pcie_aspm_support_enabled() || !pcie_aspm_capable(pdev))
 		return;
 
+	tp->rtl_aspm_enabled = enable;
+
 	if (enable) {
 		RTL_W8(tp, Config5, RTL_R8(tp, Config5) | ASPM_en);
 		RTL_W8(tp, Config2, RTL_R8(tp, Config2) | ClkReqEn);
@@ -4407,6 +4416,7 @@  static void rtl_tx(struct net_device *dev, struct rtl8169_private *tp,
 
 	dirty_tx = tp->dirty_tx;
 
+	atomic_add(tp->cur_tx - dirty_tx, &tp->aspm_packet_count);
 	while (READ_ONCE(tp->cur_tx) != dirty_tx) {
 		unsigned int entry = dirty_tx % NUM_TX_DESC;
 		u32 status;
@@ -4551,6 +4561,8 @@  static int rtl_rx(struct net_device *dev, struct rtl8169_private *tp, int budget
 		rtl8169_mark_to_asic(desc);
 	}
 
+	atomic_add(count, &tp->aspm_packet_count);
+
 	return count;
 }
 
@@ -4658,8 +4670,39 @@  static int r8169_phy_connect(struct rtl8169_private *tp)
 	return 0;
 }
 
+#define ASPM_PACKET_THRESHOLD 10
+#define ASPM_TOGGLE_INTERVAL 1000
+
+static void rtl8169_aspm_toggle(struct work_struct *work)
+{
+	struct rtl8169_private *tp = container_of(work, struct rtl8169_private,
+						  aspm_toggle.work);
+	int packet_count;
+	bool enable;
+
+	packet_count = atomic_xchg(&tp->aspm_packet_count, 0);
+
+	if (pcie_aspm_enabled(tp->pci_dev)) {
+		enable = packet_count <= ASPM_PACKET_THRESHOLD;
+
+		if (tp->rtl_aspm_enabled != enable) {
+			rtl_unlock_config_regs(tp);
+			rtl_hw_aspm_clkreq_enable(tp, enable);
+			rtl_lock_config_regs(tp);
+		}
+	} else if (tp->rtl_aspm_enabled) {
+		rtl_unlock_config_regs(tp);
+		rtl_hw_aspm_clkreq_enable(tp, false);
+		rtl_lock_config_regs(tp);
+	}
+
+	schedule_delayed_work(&tp->aspm_toggle, msecs_to_jiffies(ASPM_TOGGLE_INTERVAL));
+}
+
 static void rtl8169_down(struct rtl8169_private *tp)
 {
+	cancel_delayed_work_sync(&tp->aspm_toggle);
+
 	/* Clear all task flags */
 	bitmap_zero(tp->wk.flags, RTL_FLAG_MAX);
 
@@ -4686,6 +4729,10 @@  static void rtl8169_up(struct rtl8169_private *tp)
 	rtl_reset_work(tp);
 
 	phy_start(tp->phydev);
+
+	/* pcie_aspm_capable may change after system resume */
+	if (pcie_aspm_support_enabled() && pcie_aspm_capable(tp->pci_dev))
+		schedule_delayed_work(&tp->aspm_toggle, 0);
 }
 
 static int rtl8169_close(struct net_device *dev)
@@ -5273,11 +5320,6 @@  static int rtl_init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
 	if (rc)
 		return rc;
 
-	/* Disable ASPM L1 as that cause random device stop working
-	 * problems as well as full system hangs for some PCIe devices users.
-	 */
-	pci_disable_link_state(pdev, PCIE_LINK_STATE_L1);
-
 	/* enable device (incl. PCI PM wakeup and hotplug setup) */
 	rc = pcim_enable_device(pdev);
 	if (rc < 0) {
@@ -5307,6 +5349,8 @@  static int rtl_init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
 		return rc;
 	}
 
+	mutex_init(&tp->config_lock);
+
 	tp->mmio_addr = pcim_iomap_table(pdev)[region];
 
 	xid = (RTL_R32(tp, TxConfig) >> 20) & 0xfcf;
@@ -5344,6 +5388,10 @@  static int rtl_init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
 
 	INIT_WORK(&tp->wk.work, rtl_task);
 
+	INIT_DELAYED_WORK(&tp->aspm_toggle, rtl8169_aspm_toggle);
+
+	atomic_set(&tp->aspm_packet_count, 0);
+
 	rtl_init_mac_address(tp);
 
 	dev->ethtool_ops = &rtl8169_ethtool_ops;