All of lore.kernel.org
 help / color / mirror / Atom feed
From: "Nicholas Piggin" <npiggin@gmail.com>
To: "Akihiko Odaki" <akihiko.odaki@daynix.com>
Cc: <qemu-devel@nongnu.org>,
	"Dmitry Fleytman" <dmitry.fleytman@gmail.com>,
	"Jason Wang" <jasowang@redhat.com>,
	"Sriram Yagnaraman" <sriram.yagnaraman@ericsson.com>,
	"Fabiano Rosas" <farosas@suse.de>,
	"Laurent Vivier" <lvivier@redhat.com>,
	"Paolo Bonzini" <pbonzini@redhat.com>
Subject: Re: [PATCH v3 11/12] net/e1000e|igb: Fix interrupt throttling rearming
Date: Mon, 05 May 2025 16:49:57 +1000	[thread overview]
Message-ID: <D9O0ZI8EP5J8.2RYFER9KJ3IA5@gmail.com> (raw)
In-Reply-To: <57c9b65a-bfdf-47e6-a438-6559f791f7dd@daynix.com>

On Mon May 5, 2025 at 4:03 PM AEST, Akihiko Odaki wrote:
> On 2025/05/02 12:17, Nicholas Piggin wrote:
>> Timer expiry that results in an interrupt does not rearm the timer so
>> an interrupt can appear immediately after the interrupt generated by
>> timer expiry.
>> 
>> Fix this by rearming the throttle timer when a delayed interrupt is
>> processed. e1000e gets this by reusing the e1000e_msix_notify()
>> logic, igb calls igb_intrmgr_rearm_timer() directly.
>> 
>> Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
>> ---
>>   hw/net/e1000e_core.c |  5 ++--
>>   hw/net/igb_core.c    | 55 ++++++++++++++++++++++++++------------------
>>   2 files changed, 35 insertions(+), 25 deletions(-)
>> 
>> diff --git a/hw/net/e1000e_core.c b/hw/net/e1000e_core.c
>> index d53f70065ef..2932122c04b 100644
>> --- a/hw/net/e1000e_core.c
>> +++ b/hw/net/e1000e_core.c
>> @@ -218,7 +218,7 @@ static uint32_t find_msix_causes(E1000ECore *core, int vec)
>>   }
>>   
>>   static void
>> -e1000e_msix_auto_clear_mask(E1000ECore *core, uint32_t cause);
>> +e1000e_msix_notify(E1000ECore *core, uint32_t causes);
>>   
>>   static void
>>   e1000e_intrmgr_on_msix_throttling_timer(void *opaque)
>> @@ -233,8 +233,7 @@ e1000e_intrmgr_on_msix_throttling_timer(void *opaque)
>>       causes = find_msix_causes(core, idx) & core->mac[IMS] & core->mac[ICR];
>>       if (causes) {
>>           trace_e1000e_irq_msix_notify_postponed_vec(idx);
>> -        msix_notify(core->owner, causes);
>> -        e1000e_msix_auto_clear_mask(core, causes);
>> +        e1000e_msix_notify(core, causes);
>>       }
>>   }
>>   
>> diff --git a/hw/net/igb_core.c b/hw/net/igb_core.c
>> index 035637f81f8..cc25a1d5baa 100644
>> --- a/hw/net/igb_core.c
>> +++ b/hw/net/igb_core.c
>> @@ -152,11 +152,14 @@ igb_intrmgr_arm_timer(IGBIntrDelayTimer *timer, int64_t delay_ns)
>>   static inline void
>>   igb_intrmgr_rearm_timer(IGBIntrDelayTimer *timer)
>>   {
>> -    uint32_t interval = (timer->core->mac[timer->delay_reg] &
>> -                         E1000_EITR_INTERVAL) >> 2;
>> -    int64_t delay_ns = (int64_t)interval * timer->delay_resolution_ns;
>> +    uint32_t eitr = timer->core->mac[timer->delay_reg];
>>   
>> -    igb_intrmgr_arm_timer(timer, delay_ns);
>> +    if (eitr != 0) {
>> +        uint32_t interval = (eitr & E1000_EITR_INTERVAL) >> 2;
>> +        int64_t delay_ns = (int64_t)interval * timer->delay_resolution_ns;
>> +
>> +        igb_intrmgr_arm_timer(timer, delay_ns);
>> +    }
>>   }
>>   
>>   static void
>> @@ -168,21 +171,7 @@ igb_intmgr_timer_resume(IGBIntrDelayTimer *timer)
>>   }
>>   
>>   static void
>> -igb_intrmgr_on_msix_throttling_timer(void *opaque)
>> -{
>> -    IGBIntrDelayTimer *timer = opaque;
>> -    IGBCore *core = timer->core;
>> -    int vector = timer - &core->eitr[0];
>> -    uint32_t causes;
>> -
>> -    timer->running = false;
>> -
>> -    causes = core->mac[EICR] & core->mac[EIMS];
>> -    if (causes & BIT(vector)) {
>> -        trace_e1000e_irq_msix_notify_postponed_vec(vector);
>> -        igb_msix_notify(core, vector);
>> -    }
>> -}
>> +igb_intrmgr_on_msix_throttling_timer(void *opaque);
>>   
>>   static void
>>   igb_intrmgr_initialize_all_timers(IGBCore *core, bool create)
>> @@ -2258,9 +2247,7 @@ igb_postpone_interrupt(IGBIntrDelayTimer *timer)
>>           return true;
>>       }
>>   
>> -    if (timer->core->mac[timer->delay_reg] != 0) {
>> -        igb_intrmgr_rearm_timer(timer);
>> -    }
>> +    igb_intrmgr_rearm_timer(timer);
>>   
>>       return false;
>>   }
>> @@ -2284,6 +2271,30 @@ static void igb_send_msix(IGBCore *core, uint32_t causes)
>>       }
>>   }
>>   
>> +static void
>> +igb_intrmgr_on_msix_throttling_timer(void *opaque)
>> +{
>> +    IGBIntrDelayTimer *timer = opaque;
>> +    IGBCore *core = timer->core;
>> +    int vector = timer - &core->eitr[0];
>> +    uint32_t causes;
>> +
>> +    timer->running = false;
>> +
>> +    causes = core->mac[EICR] & core->mac[EIMS];
>> +    if (causes & BIT(vector)) {
>> +        /*
>> +         * The moderation counter is loaded with interval value whenever the
>> +         * interrupt is signaled. This includes when the interrupt is signaled
>> +         * by the counter reaching 0.
>> +         */
>> +        igb_intrmgr_rearm_timer(timer);
>> +
>> +        trace_e1000e_irq_msix_notify_postponed_vec(vector);
>> +        igb_msix_notify(core, vector);
>> +    }
>> +}
>> +
>
> I wonder why the definition is moved. This patch adds a 
> igb_intrmgr_rearm_timer() call but it's already placed earlier than this 
> function.

Hmm, I did require moving it at one point, but I must have reworked it
enough to avoid it. That makes the patch smaller and nicer. Another
good catch.

Thanks,
Nick


  reply	other threads:[~2025-05-05  6:50 UTC|newest]

Thread overview: 24+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2025-05-02  3:16 [PATCH v3 00/12] hw/e1000e|igb: interrupts and qtests fixes Nicholas Piggin
2025-05-02  3:16 ` [PATCH v3 01/12] qtest/e1000e|igb: Clear interrupt-cause and msix pending bits after irq Nicholas Piggin
2025-05-19 15:06   ` Fabiano Rosas
2025-05-02  3:16 ` [PATCH v3 02/12] net/e1000e: Permit disabling interrupt throttling Nicholas Piggin
2025-05-05  5:41   ` Akihiko Odaki
2025-05-05  6:36     ` Nicholas Piggin
2025-05-02  3:16 ` [PATCH v3 03/12] hw/net/e1000e|igb: Remove xitr_guest_value logic Nicholas Piggin
2025-05-05  5:45   ` Akihiko Odaki
2025-05-05  6:38     ` Nicholas Piggin
2025-05-02  3:16 ` [PATCH v3 04/12] qtest/e1000e|igb: assert irqs are clear before triggering an irq Nicholas Piggin
2025-05-19 15:07   ` Fabiano Rosas
2025-05-02  3:16 ` [PATCH v3 05/12] net/igb: Fix interrupt throttling interval calculation Nicholas Piggin
2025-05-02  3:16 ` [PATCH v3 06/12] net/igb: Implement EITR Moderation Counter Nicholas Piggin
2025-05-02  3:16 ` [PATCH v3 07/12] igb: Add a note about re-loading timers breaking deterministic replay Nicholas Piggin
2025-05-02  3:17 ` [PATCH v3 08/12] hw/net/e1000e: Postponed msix interrupt processing should auto-clear cause Nicholas Piggin
2025-05-02  3:17 ` [PATCH v3 09/12] hw/net/e1000e: Do not auto-clear cause on postponed msix interrupt Nicholas Piggin
2025-05-02  3:17 ` [PATCH v3 10/12] net/e1000e|igb: Only send delayed msix interrupts that have a cause Nicholas Piggin
2025-05-05  5:51   ` Akihiko Odaki
2025-05-05  6:48     ` Nicholas Piggin
2025-05-02  3:17 ` [PATCH v3 11/12] net/e1000e|igb: Fix interrupt throttling rearming Nicholas Piggin
2025-05-05  6:03   ` Akihiko Odaki
2025-05-05  6:49     ` Nicholas Piggin [this message]
2025-05-02  3:17 ` [PATCH v3 12/12] qtest/e1000e|igb: Test interrupt throttling in multiple_transfers test Nicholas Piggin
2025-05-19 15:08   ` Fabiano Rosas

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=D9O0ZI8EP5J8.2RYFER9KJ3IA5@gmail.com \
    --to=npiggin@gmail.com \
    --cc=akihiko.odaki@daynix.com \
    --cc=dmitry.fleytman@gmail.com \
    --cc=farosas@suse.de \
    --cc=jasowang@redhat.com \
    --cc=lvivier@redhat.com \
    --cc=pbonzini@redhat.com \
    --cc=qemu-devel@nongnu.org \
    --cc=sriram.yagnaraman@ericsson.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.