Re: how to tell Linux *not* to share IRQs ?

Re: how to tell Linux *not* to share IRQs ?

[Per Jessen]

On Wed, 8 Aug 2001 12:13:40 -0500 (EST), Mark H. Wood wrote:

>On Wed, 8 Aug 2001, Andrew McNamara wrote:
>[snippage]
>> The problem is largely historical - each interrupt traditionally had a
>> physically line associated with it, and lines on your backplane were a
>> limited resource.
>>
>> If you were to do it again these days, you might have some sort of
>> shared serial bus, so devices could give detailed data to the cpu
>> (not only to uniquely identify the interrupting device, but also
>> identify sub-devices - say a USB peripheral).
>
>See for example "vectored interrupts" on the PDP10.  The device driver
>tells the device where the driver's ISR is, and when the device
>interrupts, it puts that address on the bus.  The interrupt logic jumps
>directly to the ISR, which "knows" it is the only driver that would be
>interested in this interrupt.  (You could set up a jump table if you
>wanted to, so that each device of the same type could identify itself
>uniquely, but that typically wasn't a big problem in '10 installations
>where multiples were most likely in a PDP11 on the other side of a DTE20,
>or Massbus devices on a single RH20.)
>
>Apparently this idea is now so old that it is new. :-)

Yeah - I believe the same was possible on the Z80 - though I'd have to 
go read the manual to be certain.


regards,
Per Jessen, Zurich

Windows 2001: "I'm sorry Dave ...  I'm afraid I can't do that."




regards,
Per Jessen, Zurich
http://www.enidan.com - home of the J1 serial console.

Windows 2001: "I'm sorry Dave ...  I'm afraid I can't do that."

Re: how to tell Linux *not* to share IRQs ?

[Chris Meadors]

On Wed, 8 Aug 2001, Per Jessen wrote:

> Yeah - I believe the same was possible on the Z80 - though I'd have to
> go read the manual to be certain.

It is.  When I was reading the description in the previous e-mail, I was
thinking, "that sounds a lot like the Z80".

I missed the beginning of this thead, what did the so old its new mean?
Is there a new chip supporting vectored interupts?  Or are they just being
simulated in software?

-Chris
-- 
Two penguins were walking on an iceberg.  The first penguin said to the
second, "you look like you are wearing a tuxedo."  The second penguin
said, "I might be..."                         --David Lynch, Twin Peaks

Re: how to tell Linux *not* to share IRQs ?

[H. Peter Anvin]

Followup to:  <Pine.LNX.4.31.0108081515300.3853-100000@rc.priv.hereintown.net>
By author:    Chris Meadors <clubneon@hereintown.net>
In newsgroup: linux.dev.kernel
>
> On Wed, 8 Aug 2001, Per Jessen wrote:
> 
> > Yeah - I believe the same was possible on the Z80 - though I'd have to
> > go read the manual to be certain.
> 
> It is.  When I was reading the description in the previous e-mail, I was
> thinking, "that sounds a lot like the Z80".
> 
> I missed the beginning of this thead, what did the so old its new mean?
> Is there a new chip supporting vectored interupts?  Or are they just being
> simulated in software?
> 

The x86 supports vectored interrupts, too (at least in the same sense
the Z80 does.)  There are 256 interrupt vectors possible, 32 of which
are reserved (although that didn't keep IBM from foolishly use them
rather than the 224 that weren't...)  However, the PIC architecture in
the PC doesn't use this -- it lets the vectors be generated by the
PIC; the standard dual PIC allows only for 15 vectors and the APIC
something like 24, and it's all controlled by which input line gets
pulled.

Sad, isn't it?

	-hpa

-- 
<hpa@transmeta.com> at work, <hpa@zytor.com> in private!
"Unix gives you enough rope to shoot yourself in the foot."
http://www.zytor.com/~hpa/puzzle.txt	<amsp@zytor.com>

Re: how to tell Linux *not* to share IRQs ?

[Andreas Bombe]

On Wed, Aug 08, 2001 at 09:03:14PM +0200, Per Jessen wrote:
> On Wed, 8 Aug 2001 12:13:40 -0500 (EST), Mark H. Wood wrote:
> 
> >On Wed, 8 Aug 2001, Andrew McNamara wrote:
> >[snippage]
> >> The problem is largely historical - each interrupt traditionally had a
> >> physically line associated with it, and lines on your backplane were a
> >> limited resource.
> >>
> >> If you were to do it again these days, you might have some sort of
> >> shared serial bus, so devices could give detailed data to the cpu
> >> (not only to uniquely identify the interrupting device, but also
> >> identify sub-devices - say a USB peripheral).
> >
> >See for example "vectored interrupts" on the PDP10.  The device driver
> >tells the device where the driver's ISR is, and when the device
> >interrupts, it puts that address on the bus.  The interrupt logic jumps
> >directly to the ISR, which "knows" it is the only driver that would be
> >interested in this interrupt.  (You could set up a jump table if you
> >wanted to, so that each device of the same type could identify itself
> >uniquely, but that typically wasn't a big problem in '10 installations
> >where multiples were most likely in a PDP11 on the other side of a DTE20,
> >or Massbus devices on a single RH20.)
> >
> >Apparently this idea is now so old that it is new. :-)
> 
> Yeah - I believe the same was possible on the Z80 - though I'd have to 
> go read the manual to be certain.

And with the m68k, too.  In fact, that's what it always tries to do.  It
runs an interrupt acknowledge cycle when an interrupt is raised so that
the interrupting device can respond with an 8 bit vector number which is
used to load the address of the handler from the vector table.

When the /AVEC (autovector) input is true it generates a vector number
automatically depending on the interrupt level.  So it supports both
modes of operation.

So it's not like the PDP-10 was the only to do that.  If only a better
design would have become standard instead of the ugly IBM PC
architecture...

-- 
Andreas E. Bombe <andreas.bombe@munich.netsurf.de>    DSA key 0x04880A44

how to tell Linux *not* to share IRQs ?

[Scott McDermott]

I cannot for the life of me get Linux to use all the available IRQs; it
insists on sharing IRQs even though there are enough available to give
each PCI device a separate one.

I have a system which I'm configuring as a gateway; it has three 3c509B
NICs, an AIC7892 SCSI card and an ATI 128 AGP.  I'm getting interrupts
shared between the SCSI and video card, and two of the NICs are sharing
an interrupt as well (the third takes its own).  There are IRQs
available to give each of these devices a different IRQ, but Linux won't
do it on this hardware it seems.  System is a UP Dell T800r.

I have tried using PnP OS setting in BIOS, non-PnP OS, disabling
parallel and USB, giving all manner of "pci=" kernel parameters, tried
with APIC and without (even though it's UP, it seems to still use it if
enabled), and tried lots of "pirq=" lines (although I gather that's only
relevant with SMP?).  If I move the cards around they change IRQs and
even use different IRQs which are unused with different slot
configuration, but they always share interrupts.

Is there some way to tell Linux "don't even boot if you have to share an
IRQ?"

Here's lspci presently:

   00:00.0 Host bridge: Intel Corporation 440BX/ZX - 82443BX/ZX Host bridge (rev 03)
           Flags: bus master, medium devsel, latency 64
           Memory at f4000000 (32-bit, prefetchable) [size=64M]
           Capabilities: [a0] AGP version 1.0

   00:01.0 PCI bridge: Intel Corporation 440BX/ZX - 82443BX/ZX AGP bridge (rev 03) (prog-if 00 [Normal decode])
           Flags: bus master, 66Mhz, medium devsel, latency 128
           Bus: primary=00, secondary=01, subordinate=01, sec-latency=64
           I/O behind bridge: 00009000-00009fff
           Memory behind bridge: f0100000-f01fffff
           Prefetchable memory behind bridge: f8000000-fbffffff

   00:07.0 ISA bridge: Intel Corporation 82371AB PIIX4 ISA (rev 02)
           Flags: bus master, medium devsel, latency 0

   00:07.1 IDE interface: Intel Corporation 82371AB PIIX4 IDE (rev 01) (prog-if 80 [Master])
           Flags: bus master, medium devsel, latency 64
           I/O ports at 1000 [size=16]

   00:07.2 USB Controller: Intel Corporation 82371AB PIIX4 USB (rev 01) (prog-if 00 [UHCI])
           Flags: bus master, medium devsel, latency 64, IRQ 9
           I/O ports at 1020 [size=32]

   00:07.3 Bridge: Intel Corporation 82371AB PIIX4 ACPI (rev 02)
           Flags: medium devsel, IRQ 9

   00:0e.0 Ethernet controller: 3Com Corporation 3c905B 100BaseTX [Cyclone]
           Subsystem: 3Com Corporation 3C905B Fast Etherlink XL 10/100
           Flags: bus master, medium devsel, latency 80, IRQ 10
           I/O ports at 1080 [size=128]
           Memory at f0000000 (32-bit, non-prefetchable) [size=128]
           Expansion ROM at <unassigned> [disabled] [size=128K]
           Capabilities: [dc] Power Management version 1

   00:0f.0 Ethernet controller: 3Com Corporation 3c905B 100BaseTX [Cyclone]
           Subsystem: 3Com Corporation 3C905B Fast Etherlink XL 10/100
           Flags: bus master, medium devsel, latency 80, IRQ 7
           I/O ports at 1400 [size=128]
           Memory at f0000400 (32-bit, non-prefetchable) [size=128]
           Expansion ROM at <unassigned> [disabled] [size=128K]
           Capabilities: [dc] Power Management version 1

   00:10.0 Ethernet controller: 3Com Corporation 3c905B 100BaseTX [Cyclone]
           Subsystem: 3Com Corporation 3C905B Fast Etherlink XL 10/100
           Flags: bus master, medium devsel, latency 80, IRQ 9
           I/O ports at 1480 [size=128]
           Memory at f0000800 (32-bit, non-prefetchable) [size=128]
           Expansion ROM at <unassigned> [disabled] [size=128K]
           Capabilities: [dc] Power Management version 1

   00:11.0 SCSI storage controller: Adaptec 7892A (rev 02)
           Subsystem: Adaptec: Unknown device e2a0
           Flags: bus master, 66Mhz, medium devsel, latency 72, IRQ 7
           BIST result: 00
           I/O ports at 1800 [disabled] [size=256]
           Memory at f0001000 (64-bit, non-prefetchable) [size=4K]
           Expansion ROM at <unassigned> [disabled] [size=128K]
           Capabilities: [dc] Power Management version 2

   01:00.0 VGA compatible controller: ATI Technologies Inc Rage 128 PF (prog-if 00 [VGA])
           Subsystem: ATI Technologies Inc: Unknown device 0404
           Flags: bus master, stepping, fast Back2Back, 66Mhz, medium devsel, latency 66, IRQ 11
           Memory at f8000000 (32-bit, prefetchable) [size=64M]
           I/O ports at 9000 [size=256]
           Memory at f0100000 (32-bit, non-prefetchable) [size=16K]
           Expansion ROM at <unassigned> [disabled] [size=128K]
           Capabilities: [50] AGP version 2.0
           Capabilities: [5c] Power Management version 2

I've -DDEBUG in pci-i386.h to get some more verbose IRQ config messages:

   Linux version 2.4.7 (root@asmodeus) (gcc version 2.96 20000731 (Red Hat Linux 7.1 2.96-85)) #7 Wed Jul 25 12:54:14 EDT 2001
   BIOS-provided physical RAM map:
    BIOS-e820: 0000000000000000 - 000000000009f400 (usable)
    BIOS-e820: 000000000009f400 - 00000000000a0000 (reserved)
    BIOS-e820: 00000000000e6c00 - 0000000000100000 (reserved)
    BIOS-e820: 0000000000100000 - 00000000040fd800 (usable)
    BIOS-e820: 00000000040fd800 - 00000000040ff800 (ACPI data)
    BIOS-e820: 00000000040ff800 - 00000000040ffc00 (ACPI NVS)
    BIOS-e820: 00000000040ffc00 - 0000000018000000 (usable)
    BIOS-e820: 00000000fff80000 - 0000000100000000 (reserved)
   Scan SMP from c0000000 for 1024 bytes.
   Scan SMP from c009fc00 for 1024 bytes.
   Scan SMP from c00f0000 for 65536 bytes.
   Scan SMP from c009f400 for 4096 bytes.
   On node 0 totalpages: 98304
   zone(0): 4096 pages.
   zone(1): 94208 pages.
   zone(2): 0 pages.
   mapped APIC to ffffe000 (01664000)
   Kernel command line: auto BOOT_IMAGE=production ro root=802 BOOT_FILE=/boot/bzImage video=aty128fb:1024x768-16@76
   Initializing CPU#0
   Detected 798.177 MHz processor.
   Console: colour VGA+ 80x25
   Calibrating delay loop... 1592.52 BogoMIPS
   Memory: 384508k/393216k available (956k kernel code, 8308k reserved, 362k data, 204k init, 0k highmem)
   Dentry-cache hash table entries: 65536 (order: 7, 524288 bytes)
   Inode-cache hash table entries: 32768 (order: 6, 262144 bytes)
   Mount-cache hash table entries: 8192 (order: 4, 65536 bytes)
   Buffer-cache hash table entries: 32768 (order: 5, 131072 bytes)
   Page-cache hash table entries: 131072 (order: 7, 524288 bytes)
   CPU: Before vendor init, caps: 0383f9ff 00000000 00000000, vendor = 0
   CPU: L1 I cache: 16K, L1 D cache: 16K
   CPU: L2 cache: 256K
   Intel machine check architecture supported.
   Intel machine check reporting enabled on CPU#0.
   CPU: After vendor init, caps: 0383f9ff 00000000 00000000 00000000
   CPU:     After generic, caps: 0383f9ff 00000000 00000000 00000000
   CPU:             Common caps: 0383f9ff 00000000 00000000 00000000
   CPU: Intel Pentium III (Coppermine) stepping 03
   Enabling fast FPU save and restore... done.
   Enabling unmasked SIMD FPU exception support... done.
   Checking 'hlt' instruction... OK.
   POSIX conformance testing by UNIFIX
   mtrr: v1.40 (20010327) Richard Gooch (rgooch@atnf.csiro.au)
   mtrr: detected mtrr type: Intel
   PCI: BIOS32 Service Directory structure at 0xc00f6be0
   PCI: BIOS32 Service Directory entry at 0xfd790
   PCI: BIOS probe returned s=00 hw=01 ver=02.10 l=01
   PCI: PCI BIOS revision 2.10 entry at 0xfd993, last bus=1
   PCI: Using configuration type 1
   PCI: Probing PCI hardware
   PCI: IDE base address fixup for 00:07.1
   PCI: Scanning for ghost devices on bus 0
   PCI: Scanning for ghost devices on bus 1
   PCI: IRQ init
   PCI: Interrupt Routing Table found at 0xc00fdf20
   00:0d slot=01 0:60/deb8 1:61/deb8 2:62/deb8 3:63/deb8
   00:0e slot=02 0:61/deb8 1:62/deb8 2:63/deb8 3:60/deb8
   00:0f slot=03 0:62/deb8 1:63/deb8 2:60/deb8 3:61/deb8
   00:10 slot=04 0:63/deb8 1:60/deb8 2:61/deb8 3:62/deb8
   00:11 slot=05 0:62/deb8 1:63/deb8 2:60/deb8 3:61/deb8
   00:0c slot=00 0:63/deb8 1:00/def8 2:00/def8 3:00/def8
   00:00 slot=00 0:60/def8 1:61/def8 2:62/def8 3:63/def8
   00:07 slot=00 0:60/def8 1:61/def8 2:62/def8 3:63/def8
   00:01 slot=00 0:60/def8 1:61/def8 2:62/def8 3:63/def8
   01:00 slot=00 0:60/def8 1:61/def8 2:00/def8 3:00/def8
   PCI: Attempting to find IRQ router for 8086:122e
   PCI: Using IRQ router PIIX [8086/7110] at 00:07.0
   PCI: IRQ fixup
   PCI: Allocating resources
   PCI: Resource f4000000-f7ffffff (f=1208, d=0, p=0)
   PCI: Resource 00001000-0000100f (f=101, d=0, p=0)
   PCI: Resource 00001020-0000103f (f=101, d=0, p=0)
   PCI: Resource 00001800-000018ff (f=101, d=0, p=0)
   PCI: Resource f0001000-f0001fff (f=204, d=0, p=0)
   PCI: Resource f8000000-fbffffff (f=1208, d=0, p=0)
   PCI: Resource 00009000-000090ff (f=101, d=0, p=0)
   PCI: Resource f0100000-f0103fff (f=200, d=0, p=0)
   PCI: Resource 00001080-000010ff (f=101, d=1, p=1)
   PCI: Resource f0000000-f000007f (f=200, d=1, p=1)
   PCI: Resource 00001400-0000147f (f=101, d=1, p=1)
   PCI: Resource f0000400-f000047f (f=200, d=1, p=1)
   PCI: Resource 00001480-000014ff (f=101, d=1, p=1)
   PCI: Resource f0000800-f000087f (f=200, d=1, p=1)
   PCI: Sorting device list...
   Limiting direct PCI/PCI transfers.
   Linux NET4.0 for Linux 2.4
   Based upon Swansea University Computer Society NET3.039
   Initializing RT netlink socket
   Starting kswapd v1.8
   IRQ for 01:00.0:0 -> PIRQ 60, mask def8, excl 0000 -> newirq=11 -> got IRQ 11
   PCI: Found IRQ 11 for device 01:00.0
   aty128fb: Rage128 BIOS located at segment C00C0000
   aty128fb: Rage128 Pro PF (AGP) [chip rev 0x1] 16M 128-bit SDR SGRAM (1:1)
   Console: switching to colour frame buffer device 128x48
   fb0: ATY Rage128 frame buffer device on PCI
   aty128fb: Rage128 MTRR set to ON
   Detected PS/2 Mouse Port.
   pty: 256 Unix98 ptys configured
   Serial driver version 5.05c (2001-07-08) with MANY_PORTS SHARE_IRQ SERIAL_PCI enabled
   ttyS00 at 0x03f8 (irq = 4) is a 16550A
   block: queued sectors max/low 255200kB/124128kB, 768 slots per queue
   NET4: Frame Diverter 0.46
   Linux agpgart interface v0.99 (c) Jeff Hartmann
   agpgart: Maximum main memory to use for agp memory: 322M
   agpgart: Detected Intel 440BX chipset
   agpgart: AGP aperture is 64M @ 0xf4000000
   SCSI subsystem driver Revision: 1.00
   IRQ for 00:11.0:0 -> PIRQ 62, mask deb8, excl 0000 -> newirq=7 -> got IRQ 7
   PCI: Found IRQ 7 for device 00:11.0
   PCI: Sharing IRQ 7 with 00:0f.0
   scsi0 : Adaptec AIC7XXX EISA/VLB/PCI SCSI HBA DRIVER, Rev 6.1.13
           <Adaptec 29160 Ultra160 SCSI adapter>
           aic7892: Ultra160 Wide Channel A, SCSI Id=7, 32/255 SCBs
     Vendor: QUANTUM   Model: ATLAS 10K 18WLS   Rev: UCH0
     Type:   Direct-Access                      ANSI SCSI revision: 03
   (scsi0:A:0): 160.000MB/s transfers (80.000MHz DT, offset 31, 16bit)
   scsi0:0:0:0: Tagged Queuing enabled.  Depth 253
   Attached scsi disk sda at scsi0, channel 0, id 0, lun 0
   SCSI device sda: 35877972 512-byte hdwr sectors (18370 MB)
   Partition check:
    sda: sda1 sda2 sda3 sda4 < sda5 sda6 sda7 sda8 sda9 sda10 sda11 >
   md: raid1 personality registered
   md: md driver 0.90.0 MAX_MD_DEVS=256, MD_SB_DISKS=27
   md: Autodetecting RAID arrays.
   md: autorun ...
   md: ... autorun DONE.
   NET4: Linux TCP/IP 1.0 for NET4.0
   IP Protocols: ICMP, UDP, TCP, IGMP
   IP: routing cache hash table of 4096 buckets, 32Kbytes
   TCP: Hash tables configured (established 32768 bind 32768)
   Linux IP multicast router 0.06 plus PIM-SM
   NET4: Unix domain sockets 1.0/SMP for Linux NET4.0.
   VFS: Mounted root (ext2 filesystem) readonly.
   Freeing unused kernel memory: 204k freed
   Adding Swap: 1574360k swap-space (priority -1)
   PCI: Enabling device 00:0e.0 (0114 -> 0117)
   IRQ for 00:0e.0:0 -> PIRQ 61, mask deb8, excl 0000 -> newirq=10 -> got IRQ 10
   PCI: Found IRQ 10 for device 00:0e.0
   3c59x.c:LK1.1.15 6 June 2001  Donald Becker and others. http://www.scyld.com/network/vortex.html
   See Documentation/networking/vortex.txt
   00:0e.0: 3Com PCI 3c905B Cyclone 100baseTx at 0x1080,  00:10:4b:63:97:50, IRQ 10
     product code 4e43 rev 00.0 date 02-21-98
     8K byte-wide RAM 5:3 Rx:Tx split, autoselect/Autonegotiate interface.
     MII transceiver found at address 24, status 7809.
     Enabling bus-master transmits and whole-frame receives.
   00:0e.0: scatter/gather enabled. h/w checksums enabled
   divert: allocating divert_blk for eth0
   PCI: Enabling device 00:0f.0 (0114 -> 0117)
   IRQ for 00:0f.0:0 -> PIRQ 62, mask deb8, excl 0000 -> newirq=7 -> got IRQ 7
   PCI: Found IRQ 7 for device 00:0f.0
   PCI: Sharing IRQ 7 with 00:11.0
   00:0f.0: 3Com PCI 3c905B Cyclone 100baseTx at 0x1400,  00:10:4b:63:97:26, IRQ 7
     product code 4e43 rev 00.0 date 02-21-98
     8K byte-wide RAM 5:3 Rx:Tx split, autoselect/Autonegotiate interface.
     MII transceiver found at address 24, status 7809.
     Enabling bus-master transmits and whole-frame receives.
   00:0f.0: scatter/gather enabled. h/w checksums enabled
   divert: allocating divert_blk for eth1
   PCI: Enabling device 00:10.0 (0114 -> 0117)
   IRQ for 00:10.0:0 -> PIRQ 63, mask deb8, excl 0000 -> newirq=9 -> got IRQ 9
   PCI: Found IRQ 9 for device 00:10.0
   PCI: Sharing IRQ 9 with 00:07.2
   00:10.0: 3Com PCI 3c905B Cyclone 100baseTx at 0x1480,  00:10:4b:63:95:9f, IRQ 9
     product code 4e43 rev 00.0 date 02-21-98
     8K byte-wide RAM 5:3 Rx:Tx split, autoselect/Autonegotiate interface.
     MII transceiver found at address 24, status 7809.
     Enabling bus-master transmits and whole-frame receives.
   00:10.0: scatter/gather enabled. h/w checksums enabled
   divert: allocating divert_blk for eth2
   eth0: using NWAY device table, not 8

The system is intended to be a gateway between three networks and I
expect non-trivial interrupt rates; I really don't want to share IRQs.

Any ideas how I might get these to play nice and actually use the
available IRQs?

Re: how to tell Linux *not* to share IRQs ?

[Alan Cox]

> I cannot for the life of me get Linux to use all the available IRQs; it
> insists on sharing IRQs even though there are enough available to give
> each PCI device a separate one.

The actual sharing rules for PCI interrupt lines are frequently determined
by the actual wiring on the motherboard. It is quite possible the interrupt
lines on some of your slots are physically wired together, and indeed quite
likely that this is true if you have five or more slots

> parallel and USB, giving all manner of "pci=" kernel parameters, tried
> with APIC and without (even though it's UP, it seems to still use it if

APIC can help by removing ISA/PCI sharing, but not PCI to PCI. 

Alan
--
  "Have you noticed the way people's intelligence capabilities decline
   sharply the minute they start waving guns around?"
 		-- Dr. Who

Re: how to tell Linux *not* to share IRQs ?

[Andrew McNamara]

>The actual sharing rules for PCI interrupt lines are frequently determined
>by the actual wiring on the motherboard. It is quite possible the interrupt
>lines on some of your slots are physically wired together, and indeed quite
>likely that this is true if you have five or more slots

Does this mean the ISRs for every driver sharing an interrupt have to
poll their device when an interrupt comes in (in the case of shared PCI
interrupts), or is there some additional hardware smarts so the kernel
knows which driver's ISRs need to be invoked?

 ---
Andrew McNamara (System Architect)

connect.com.au Pty Ltd
Lvl 3, 213 Miller St, North Sydney, NSW 2060, Australia
Phone: +61 2 9409 2117, Fax: +61 2 9409 2111

Re: how to tell Linux *not* to share IRQs ?

[Chris Wedgwood]

On Thu, Jul 26, 2001 at 05:06:21PM +1000, Andrew McNamara wrote:

    Does this mean the ISRs for every driver sharing an interrupt have
    to poll their device when an interrupt comes in (in the case of
    shared PCI interrupts), or is there some additional hardware
    smarts so the kernel knows which driver's ISRs need to be invoked?

Yes, drivers need to check their hardware devices and acknowledge
whether or not it's was their interrupt or not.  It sounds terrible
but even with many thousands of interrupts per second the cost doesn't
seem to be that high.




  --cw

Re: how to tell Linux *not* to share IRQs ?

[Jes Sorensen]

>>>>> "Chris" == Chris Wedgwood <cw@f00f.org> writes:

Chris> On Thu, Jul 26, 2001 at 05:06:21PM +1000, Andrew McNamara
> wrote: Does this mean the ISRs for every driver sharing an
> interrupt have to poll their device when an interrupt comes in
> (in the case of shared PCI interrupts), or is there some
> additional hardware smarts so the kernel knows which driver's
> ISRs need to be invoked?

Chris> Yes, drivers need to check their hardware devices and
Chris> acknowledge whether or not it's was their interrupt or not.  It
Chris> sounds terrible but even with many thousands of interrupts per
Chris> second the cost doesn't seem to be that high.

Not only is this the case, it's also the only sane thing to do <tm>,
any device driver should check the status of the hardware it is
serving before doing anything else.

Jes

Re: how to tell Linux *not* to share IRQs ?

[Andrew McNamara]

>Chris> Yes, drivers need to check their hardware devices and
>Chris> acknowledge whether or not it's was their interrupt or not.  It
>Chris> sounds terrible but even with many thousands of interrupts per
>Chris> second the cost doesn't seem to be that high.
>
>Not only is this the case, it's also the only sane thing to do <tm>,
>any device driver should check the status of the hardware it is
>serving before doing anything else.

That's not necessarily the case - the problem arises due the
limitations of PC interrupt routing. In an ideal world, the
interrupting device would be able to be uniquely identified, rather
than having to poll every device sharing that interrupt.

The problem is largely historical - each interrupt traditionally had a
physically line associated with it, and lines on your backplane were a
limited resource. 

If you were to do it again these days, you might have some sort of
shared serial bus, so devices could give detailed data to the cpu
(not only to uniquely identify the interrupting device, but also
identify sub-devices - say a USB peripheral).

 ---
Andrew McNamara (System Architect)

connect.com.au Pty Ltd
Lvl 3, 213 Miller St, North Sydney, NSW 2060, Australia
Phone: +61 2 9409 2117, Fax: +61 2 9409 2111

Re: how to tell Linux *not* to share IRQs ?

[Eric W. Biederman]

Andrew McNamara <andrewm@connect.com.au> writes:

> >Chris> Yes, drivers need to check their hardware devices and
> >Chris> acknowledge whether or not it's was their interrupt or not.  It
> >Chris> sounds terrible but even with many thousands of interrupts per
> >Chris> second the cost doesn't seem to be that high.
> >
> >Not only is this the case, it's also the only sane thing to do <tm>,
> >any device driver should check the status of the hardware it is
> >serving before doing anything else.
> 
> That's not necessarily the case - the problem arises due the
> limitations of PC interrupt routing. In an ideal world, the
> interrupting device would be able to be uniquely identified, rather
> than having to poll every device sharing that interrupt.

Interrupts are inherently race, so you still want to check your device
status, even if you get an interrupt.  Having a smaller set of devices
to poll to see if there is any work to do is of course good.
 
> The problem is largely historical - each interrupt traditionally had a
> physically line associated with it, and lines on your backplane were a
> limited resource. 
> 
> If you were to do it again these days, you might have some sort of
> shared serial bus, so devices could give detailed data to the cpu
> (not only to uniquely identify the interrupting device, but also
> identify sub-devices - say a USB peripheral).

I don't know.  The ISA bus was structured somewhat like you describe
and that is what is truly legacy at the moment.  I suspect
the one line per interrupt came of wanting to reduce latencies.

Though when you start think of things like the current ioapics where
you do get a serial bus for your interrupts anyway.  A shared serial
bus probably makes sense.

On the other side I think having drivers that can share interupts is a
very healthy thing so I'm not certain I would want to discourage that.

Eric

Re: how to tell Linux *not* to share IRQs ?

[Mark H. Wood]

On Wed, 8 Aug 2001, Andrew McNamara wrote:
[snippage]
> The problem is largely historical - each interrupt traditionally had a
> physically line associated with it, and lines on your backplane were a
> limited resource.
>
> If you were to do it again these days, you might have some sort of
> shared serial bus, so devices could give detailed data to the cpu
> (not only to uniquely identify the interrupting device, but also
> identify sub-devices - say a USB peripheral).

See for example "vectored interrupts" on the PDP10.  The device driver
tells the device where the driver's ISR is, and when the device
interrupts, it puts that address on the bus.  The interrupt logic jumps
directly to the ISR, which "knows" it is the only driver that would be
interested in this interrupt.  (You could set up a jump table if you
wanted to, so that each device of the same type could identify itself
uniquely, but that typically wasn't a big problem in '10 installations
where multiples were most likely in a PDP11 on the other side of a DTE20,
or Massbus devices on a single RH20.)

Apparently this idea is now so old that it is new. :-)

-- 
Mark H. Wood, Lead System Programmer   mwood@IUPUI.Edu
Make a good day.

Re: how to tell Linux *not* to share IRQs ?

[Francois Romieu]

Andrew McNamara <andrewm@connect.com.au> ecrit :
[...]
> Does this mean the ISRs for every driver sharing an interrupt have to
> poll their device when an interrupt comes in (in the case of shared PCI

The ISR must verify there is some pending work for him.
Thus one must poll the device or his interrupt status queue in memory if 
any. Btw one may end doing the work before the real interrupt comes.

-- 
Ueimor