Skip to contentSkip to conversation

Short question regarding generic_make_request()

DISCUSSION Steffen Persvold 3 participants 11 messages Feb 3, 2002 – Feb 4, 2002

lore
Share & tools
lore
#0 lore raw

Hi all,

Can generic_make_request() be called from interrupt level (or tasklet) ?

Regards,

Signature · 4 lines
  Steffen Persvold   | Scalable Linux Systems |   Try out the world's best
 mailto:[email protected] |  http://www.scali.com  | performing MPI implementation:
Tel: (+47) 2262 8950 |   Olaf Helsets vei 6   |      - ScaMPI 1.13.8 -
Fax: (+47) 2262 8951 |   N0621 Oslo, NORWAY   | >320MBytes/s and <4uS latency
#1 lore raw
Quoted 2 lines · Steffen Persvold

Can generic_make_request() be called from interrupt level (or tasklet) ?

no.

Ingo

#2 lore raw
Quoted 8 lines · Ingo Molnar
Quoted 2 lines · Steffen Persvold

Can generic_make_request() be called from interrupt level (or tasklet) ?

no.

OK, so are there any other way I can submit a block request from a tasklet (that is interrupt context, right ?) ?

Thanks,

Signature · 4 lines
  Steffen Persvold   | Scalable Linux Systems |   Try out the world's best
 mailto:[email protected] |  http://www.scali.com  | performing MPI implementation:
Tel: (+47) 2262 8950 |   Olaf Helsets vei 6   |      - ScaMPI 1.13.8 -
Fax: (+47) 2262 8951 |   N0621 Oslo, NORWAY   | >320MBytes/s and <4uS latency
#3 lore raw
Quoted 2 lines · Steffen Persvold

OK, so are there any other way I can submit a block request from a tasklet (that is interrupt context, right ?) ?

submitting IO is something that needs a process context currently, ie. a helper kernel thread.

Ingo

#4 lore raw
In reply to Steffen Persvold
Quoted 7 lines · Ingo Molnar
Quoted 2 lines · Steffen Persvold

Can generic_make_request() be called from interrupt level (or tasklet) ?

no.

In theory, READA from interrupt context would be ok, though. That doesn't work in real-life due to the non flag saving spin locking in __make_request.

Signature · 1 line
Jens Axboe
#5 lore raw
In reply to Steffen Persvold
Quoted 5 lines · Jens Axboe
Quoted 3 lines · Ingo Molnar

On Sun, 3 Feb 2002, Steffen Persvold wrote:

Quoted 9 lines
Quoted 4 lines
Quoted 2 lines

Can generic_make_request() be called from interrupt level (or tasklet) ?

no.

In theory, READA from interrupt context would be ok, though. That doesn't work in real-life due to the non flag saving spin locking in __make_request.

Ok, the reason I'm asking is that I receive a request from a remote machine on interrupt level (tasklet) and want to submit this to the local device. The reason I'm using a tasklet instead of a kernel thread is that somewhere between RedHat's 2.4.3-12 and 2.4.9-12 kernels the latency of waking up a kernel thread increased (using a semaphore method similar to the one used in loop.c). I don't know why this happened, but I guess that if I still could use a kernel thread there wouldn't be any problems using generic_make_request().

Regards,

Signature · 4 lines
  Steffen Persvold   | Scalable Linux Systems |   Try out the world's best
 mailto:[email protected] |  http://www.scali.com  | performing MPI implementation:
Tel: (+47) 2262 8950 |   Olaf Helsets vei 6   |      - ScaMPI 1.13.8 -
Fax: (+47) 2262 8951 |   N0621 Oslo, NORWAY   | >320MBytes/s and <4uS latency
#6 lore raw
Quoted 8 lines · Steffen Persvold

Ok, the reason I'm asking is that I receive a request from a remote machine on interrupt level (tasklet) and want to submit this to the local device. The reason I'm using a tasklet instead of a kernel thread is that somewhere between RedHat's 2.4.3-12 and 2.4.9-12 kernels the latency of waking up a kernel thread increased (using a semaphore method similar to the one used in loop.c). I don't know why this happened, but I guess that if I still could use a kernel thread there wouldn't be any problems using generic_make_request().

you really want a kernel thread for this. The wakeup latency of a kernel thread is on the order of 2-3 usecs (context switch overhead included), nothing compared to usual block IO costs.

you say that the latency of waking up a kernel thread has increased - by how much?

Ingo

#7 lore raw
Quoted 12 lines · Ingo Molnar
Quoted 8 lines · Steffen Persvold

Ok, the reason I'm asking is that I receive a request from a remote machine on interrupt level (tasklet) and want to submit this to the local device. The reason I'm using a tasklet instead of a kernel thread is that somewhere between RedHat's 2.4.3-12 and 2.4.9-12 kernels the latency of waking up a kernel thread increased (using a semaphore method similar to the one used in loop.c). I don't know why this happened, but I guess that if I still could use a kernel thread there wouldn't be any problems using generic_make_request().

Quoted 7 lines

you really want a kernel thread for this. The wakeup latency of a kernel thread is on the order of 2-3 usecs (context switch overhead included), nothing compared to usual block IO costs.

you say that the latency of waking up a kernel thread has increased - by how much?

Well, I might be analyzing it wrong but, the same driver that I'm gonna use for this shared disk stuff can also be enabled for ethernet emulation. The reason I'm saying that the latency of waking up a kernel thread increased somewhere between RedHat's 2.4.3 and 2.4.9 is that with the packet receive handler running in a tasklet I get these ping-pong numbers (measured with /bin/ping) :

[root@damd1 root]# ping sci4
PING sci4 (192.168.4.4) from 192.168.4.3 : 56(84) bytes of data. 64 bytes from sci4 (192.168.4.4): icmp_seq=0 ttl=255 time=238 usec 64 bytes from sci4 (192.168.4.4): icmp_seq=1 ttl=255 time=176 usec 64 bytes from sci4 (192.168.4.4): icmp_seq=2 ttl=255 time=200 usec 64 bytes from sci4 (192.168.4.4): icmp_seq=3 ttl=255 time=177 usec 64 bytes from sci4 (192.168.4.4): icmp_seq=4 ttl=255 time=172 usec 64 bytes from sci4 (192.168.4.4): icmp_seq=5 ttl=255 time=156 usec 64 bytes from sci4 (192.168.4.4): icmp_seq=6 ttl=255 time=160 usec 64 bytes from sci4 (192.168.4.4): icmp_seq=7 ttl=255 time=177 usec 64 bytes from sci4 (192.168.4.4): icmp_seq=8 ttl=255 time=173 usec

For simplicity I'll say ~200usec. When I change the receive handler from a tasklet to a kernel thread, the numbers look like this :

[root@damd1 root]# ping sci4
PING sci4 (192.168.4.4) from 192.168.4.3 : 56(84) bytes of data. 64 bytes from sci4 (192.168.4.4): icmp_seq=0 ttl=255 time=4.215 msec 64 bytes from sci4 (192.168.4.4): icmp_seq=1 ttl=255 time=5.728 msec 64 bytes from sci4 (192.168.4.4): icmp_seq=2 ttl=255 time=3.825 msec 64 bytes from sci4 (192.168.4.4): icmp_seq=3 ttl=255 time=6.521 msec 64 bytes from sci4 (192.168.4.4): icmp_seq=4 ttl=255 time=5.700 msec 64 bytes from sci4 (192.168.4.4): icmp_seq=5 ttl=255 time=5.666 msec 64 bytes from sci4 (192.168.4.4): icmp_seq=6 ttl=255 time=6.495 msec 64 bytes from sci4 (192.168.4.4): icmp_seq=7 ttl=255 time=5.631 msec 64 bytes from sci4 (192.168.4.4): icmp_seq=8 ttl=255 time=6.480 msec

A bit up and down, but all of them in the msec range wich is means that the ping-pong latency has increased atleast 100 times.

This might of course be related to the network stack (i.e it doesn't like to netif_rx() from user context, or just a high turnaround time when trying to send the response) but on 2.4.3-12 it didn't behave like this. The strange thing is that when I run a network benchmark such as netperf, I get nice numbers on UDP bandwith (one-way traffic) :

(with kernel thread):
UDP UNIDIRECTIONAL SEND TEST to sci4

Socket  Message  Elapsed      Messages                
Size    Size     Time         Okay Errors   Throughput
bytes   bytes    secs            #      #   MBytes/sec
262144   32768   10.00       55958      0     174.92
262144           10.00       55925            174.82

(with tasklet):
UDP UNIDIRECTIONAL SEND TEST to sci4

Socket  Message  Elapsed      Messages                
Size    Size     Time         Okay Errors   Throughput
bytes   bytes    secs            #      #   MBytes/sec
262144   32768   10.01       53648      0     167.55
262144           10.01       53648            167.55

The TCP stack seems to like it better with tasklets :

(with kernel thread):
TCP STREAM TEST to sci4

Recv   Send    Send                          
Socket Socket  Message  Elapsed              
Size   Size    Size     Time     Throughput  
bytes  bytes   bytes    secs.    MBytes/sec  

262144 262144 262144 10.02 16.24

(with tasklet):
TCP STREAM TEST to sci4

Recv   Send    Send                          
Socket Socket  Message  Elapsed              
Size   Size    Size     Time     Throughput  
bytes  bytes   bytes    secs.    MBytes/sec  

262144 262144 262144 10.00 116.00

Regards,

Signature · 4 lines
  Steffen Persvold   | Scalable Linux Systems |   Try out the world's best
 mailto:[email protected] |  http://www.scali.com  | performing MPI implementation:
Tel: (+47) 2262 8950 |   Olaf Helsets vei 6   |      - ScaMPI 1.13.8 -
Fax: (+47) 2262 8951 |   N0621 Oslo, NORWAY   | >320MBytes/s and <4uS latency
#8 lore raw
Quoted 11 lines · Steffen Persvold

[root@damd1 root]# ping sci4
PING sci4 (192.168.4.4) from 192.168.4.3 : 56(84) bytes of data. 64 bytes from sci4 (192.168.4.4): icmp_seq=0 ttl=255 time=238 usec

For simplicity I'll say ~200usec. When I change the receive handler from a tasklet to a kernel thread, the numbers look like this :

[root@damd1 root]# ping sci4
PING sci4 (192.168.4.4) from 192.168.4.3 : 56(84) bytes of data. 64 bytes from sci4 (192.168.4.4): icmp_seq=0 ttl=255 time=4.215 msec 64 bytes from sci4 (192.168.4.4): icmp_seq=1 ttl=255 time=5.728 msec

this shows some sort of wakeup or softirq handling irregularity. A number of softirq latency bugs were fixed, i'd suggest to try this with any recent kernel (or recent errata kernel rpms).

Ingo

#9 lore raw
Quoted 10 lines · Ingo Molnar
Quoted 7 lines · Steffen Persvold

[root@damd1 root]# ping sci4
PING sci4 (192.168.4.4) from 192.168.4.3 : 56(84) bytes of data. 64 bytes from sci4 (192.168.4.4): icmp_seq=0 ttl=255 time=238 usec

For simplicity I'll say ~200usec. When I change the receive handler from a tasklet to a kernel thread, the numbers look like this :

Quoted 9 lines
Quoted 4 lines

[root@damd1 root]# ping sci4
PING sci4 (192.168.4.4) from 192.168.4.3 : 56(84) bytes of data. 64 bytes from sci4 (192.168.4.4): icmp_seq=0 ttl=255 time=4.215 msec 64 bytes from sci4 (192.168.4.4): icmp_seq=1 ttl=255 time=5.728 msec

this shows some sort of wakeup or softirq handling irregularity. A number of softirq latency bugs were fixed, i'd suggest to try this with any recent kernel (or recent errata kernel rpms).

You are right indeed. I tried a vanilla 2.4.17 kernel at it performs well with a kernel thread (even better than a tasklet actually) :

TCP STREAM TEST to sci9

Recv   Send    Send                          
Socket Socket  Message  Elapsed              
Size   Size    Size     Time     Throughput  
bytes  bytes   bytes    secs.    MBytes/sec  

262144 262144 262144 10.00 157.73

Any idea when this bug was fixed (haven't tried the latest RedHat errata kernel yet, but will do).

I guess now that I can use the kernel thread without loosing performance I can start using generic_make_request() in the block device server.

Thanks for all the help,

Regards,

Signature · 4 lines
  Steffen Persvold   | Scalable Linux Systems |   Try out the world's best
 mailto:[email protected] |  http://www.scali.com  | performing MPI implementation:
Tel: (+47) 2262 8950 |   Olaf Helsets vei 6   |      - ScaMPI 1.13.8 -
Fax: (+47) 2262 8951 |   N0621 Oslo, NORWAY   | >320MBytes/s and <4uS latency
#10 lore raw
In reply to Ingo Molnar
Quoted 6 lines · Ingo Molnar

this shows some sort of wakeup or softirq handling irregularity. A number of softirq latency bugs were fixed, i'd suggest to try this with any recent kernel (or recent errata kernel rpms).

I just testet the latest RedHat errata kernel (2.4.9-21) and it performs badly with a kernel thread. Is there a simple patch out there that fixes this problem on a 2.4.9 kernel (I guess it's only related to the network stack) so that I can send it to RedHat's bugzilla system ?

Regards,

Signature · 4 lines
  Steffen Persvold   | Scalable Linux Systems |   Try out the world's best
 mailto:[email protected] |  http://www.scali.com  | performing MPI implementation:
Tel: (+47) 2262 8950 |   Olaf Helsets vei 6   |      - ScaMPI 1.13.8 -
Fax: (+47) 2262 8951 |   N0621 Oslo, NORWAY   | >320MBytes/s and <4uS latency