Skip to contentSkip to conversation

From __cpu_raise_softirq() to net_rx_action()

DISCUSSION Joshua Stewart 3 participants 5 messages Dec 22, 2002

lore
Share & tools
lore
#0 lore raw

I'm still trying to follow a packet (or even better an sk_buff) from the NIC card to user space. I think I have a good chunk of it figured out, but I'm missing a bit from the time that the __netif_rx_schedule() routine calls __cpu_raise_softirq() until the routine net_rx_action() occurs. I read in a book on Linux TCP/IP implementation that the softirq basically leads to a call to net_rx_action(), but I don't see the connection yet. It's probably due to my lack of understanding of IRQ's (and software IRQ's).

Any help is appreciated.

Thanks,
Josh

#1 lore raw

On Sun, 22 Dec 2002, Joshua Stewart wrote:

| I'm still trying to follow a packet (or even better an sk_buff) from the
| NIC card to user space.  I think I have a good chunk of it figured out,
| but I'm missing a bit from the time that the __netif_rx_schedule()
| routine calls __cpu_raise_softirq() until the routine net_rx_action()
| occurs.  I read in a book on Linux TCP/IP implementation that the
| softirq basically leads to a call to net_rx_action(), but I don't see
| the connection yet.  It's probably due to my lack of understanding of
| IRQ's (and software IRQ's).
|
| Any help is appreciated.

What are you going to do with this good info when you have it? Something like putting it on a web page would be very good, so that other people with similar questions can have a reference to look at.

Signature · 1 line
~Randy
#2 lore raw

After I understand it all, I will not be against putting together an explanation of the whole process and posting it back to this mailing list, or trying to find a Linux website that will host it. Is this type of this suitable for the Linux Documentation Project website?

Josh

Quoted 25 lines · Randy.Dunlap

On Sun, 22 Dec 2002, Joshua Stewart wrote:

| I'm still trying to follow a packet (or even better an sk_buff) from the
| NIC card to user space.  I think I have a good chunk of it figured out,
| but I'm missing a bit from the time that the __netif_rx_schedule()
| routine calls __cpu_raise_softirq() until the routine net_rx_action()
| occurs.  I read in a book on Linux TCP/IP implementation that the
| softirq basically leads to a call to net_rx_action(), but I don't see
| the connection yet.  It's probably due to my lack of understanding of
| IRQ's (and software IRQ's).
|
| Any help is appreciated.

What are you going to do with this good info when you have it? Something like putting it on a web page would be very good, so that other people with similar questions can have a reference to look at.

Signature · 7 lines
~Randy

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/
#3 lore raw

On Sun, 22 Dec 2002, Joshua Stewart wrote:

| After I understand it all, I will not be against putting together an
| explanation of the whole process and posting it back to this mailing
| list, or trying to find a Linux website that will host it.  Is this type
| of this suitable for the Linux Documentation Project website?
|
| Josh

Yes, it should be fine for LDP.

Thanks,

Signature · 1 line
~Randy
| On Sun, 2002-12-22 at 16:15, Randy.Dunlap wrote:
| > On Sun, 22 Dec 2002, Joshua Stewart wrote:
| >
| > | I'm still trying to follow a packet (or even better an sk_buff) from the
| > | NIC card to user space.  I think I have a good chunk of it figured out,
| > | but I'm missing a bit from the time that the __netif_rx_schedule()
| > | routine calls __cpu_raise_softirq() until the routine net_rx_action()
| > | occurs.  I read in a book on Linux TCP/IP implementation that the
| > | softirq basically leads to a call to net_rx_action(), but I don't see
| > | the connection yet.  It's probably due to my lack of understanding of
| > | IRQ's (and software IRQ's).
| > |
| > | Any help is appreciated.
| >
| > What are you going to do with this good info when you have it?
| > Something like putting it on a web page would be very good, so that
| > other people with similar questions can have a reference to look at.
| >
| > --
| > ~Randy
#4 lore raw
In reply to Joshua Stewart

Em Sun, Dec 22, 2002 at 04:11:37PM -0500, Joshua Stewart escreveu:

Quoted 8 lines

I'm still trying to follow a packet (or even better an sk_buff) from the NIC card to user space. I think I have a good chunk of it figured out, but I'm missing a bit from the time that the __netif_rx_schedule() routine calls __cpu_raise_softirq() until the routine net_rx_action() occurs. I read in a book on Linux TCP/IP implementation that the softirq basically leads to a call to net_rx_action(), but I don't see the connection yet. It's probably due to my lack of understanding of IRQ's (and software IRQ's).

You need to read about softirqs, but here is a quick explanation: there are several points where softirqs are serviced, like bottom halves were in the past (softirqs are much nicer), one example:

do_IRQ -> irq_exit -> do_softirq -> net_rx_action

So basically search for places calling do_softirq, they look if there is softirqs pending and call do_softirq that will call the appropriate _action function registered at softirq creation (look at net/core/dev.c, function net_dev_init, line 2875.

- Arnaldo