Hi,
Thank you for your reply. Annexed to this you will find the final part of the
source code of function fxp_start with the changes I have made. I hope the
comments in the code are clear enough. The idea is to read and modify the
mbufs right before they go to the wire, although I haven't yet written this
code. So far I am just trying to read the contents of the mbufs (first six
bytes of m_data).
Kind Regards,
Marcus.
Bosko Milekic wrote:
> Hi,
>
> Please show us how you're manipulating the mbufs.
> You should not be getting page faults unless you're doing
> something wrong.
> A good idea would also be to check the fault virtual
> address and enable debugging support in your kernel in order
> to be able to analyze the situation as much as you can.
>
> Regards,
> Bosko.
>
> ----- Original Message -----
> From: "Marcus Vinícius Midena Ramos" <[EMAIL PROTECTED]>
> To: <[EMAIL PROTECTED]>
> Sent: Thursday, December 28, 2000 8:58 AM
> Subject: Intel ethernet driver source code
>
> > Hello,
> >
> > I am developing an application to measure the performance of an IP
> > communications network, accounting for packet loss rate and transmission
> > delay values between any two sites on a net. This work resembles that of
> > www.advanced.org and is also based on IETF's IPPM initiative.
> >
> > One of the goals is to insert a GPS timestamp on the packets to be
> > transmitted as late as possible before they are put on the wire. This
> > means I cannot do this job on the application level. Instead, this has
> > to be done on the ethernet driver level in order to achieve the maximum
> > accuracy, specially when the measurement machine is under heavy load.
> >
> > I've studied the "if_fxp.c" (Intel ethernet driver) source code plus
> > Stevens' TCP Illustrated vol. 2, which helped me get a good
> > understanding of the data structures manipulated by this code. My aim
> > now is to insert the GPS timestamps on the packets to be transmitted
> > right before they go to the interface.
> >
> > To do this, I am trying to modify the driver source code to achieve the
> > following objectives: (1) read the contents of the mbufs in order to
> > identify the ones that hold the packets to be transmitted, and (2)
> > insert the timestamps on those ellegible for so.
> >
> > However, any access to mbufs from inside the driver causes an error
> > message "page fault while in kernel mode" to be displayed the new kernel
> > is booted. It then enters a loop, trying to reboot forever.
> >
> > I guess the mbufs are in an area somehow protected by the kernel, which
> > prevents even the driver to have access to it. May question is: how can
> > I go around this problem and be able to read and change the contents of
> > the mbufs from inside the driver ?
> >
> > Thank you very much.
> > Marcus.
> >
> >
> >
> > To Unsubscribe: send mail to [EMAIL PROTECTED]
> > with "unsubscribe freebsd-net" in the body of the message
> >
>
> To Unsubscribe: send mail to [EMAIL PROTECTED]
> with "unsubscribe freebsd-net" in the body of the message
/* The code below is a duplication of the final part of function fxp_start,
* from file /usr/src/sys/pci/if_fxp.c, with the added code to display the
* strucuture and partial contents of the mbufs.
* The message "page fault while in kernel mode" appears when the m_copydata
* function call is added to this code by removing the comment delimiters.
*/
/*
* We're finished. If we added to the list, issue a RESUME to get DMA
* going again if suspended.
*/
if (txp != NULL) {
/* BEGIN OF ADDED CODE */
int i;
char b [20];
struct fxp_cb_tx *p;
printf ("\nTX_QUEUED: %d", sc->tx_queued);
for (i = 0, p = sc->cbl_first; i < sc->tx_queued; i++, p = p->next)
{
int j;
u_int32_t size, addr;
printf ("\n\tQUEUE [%d] * TBD_NUMBER: %d", i,
p->tbd_number);
/*
m_copydata (p->mb_head, 0, 6, b);
printf ("\n\t.%d.%d.%d.%d.%d.%d.", b[0], b[1], b[2], b[3],
b[4], b[5]);
*/
for (j = 0; j < p->tbd_number; j++)
{
size = p->tbd [j].tb_size;
addr = p->tbd [j].tb_addr;
printf ("\n\t\t TBD [%d] * SIZE %d & ADDRESS %d", j,
size, addr);
}
}
/* END OF ADDED CODE */
fxp_scb_wait(sc);
CSR_WRITE_1(sc, FXP_CSR_SCB_COMMAND, FXP_SCB_COMMAND_CU_RESUME);
}
}
