ignore pci devices?
Hi,
I have a small problem: Some vendor has built a PCI board which (ab-)uses the bt848/878 chip in creative ways to do some DMA. It is *not* a video card, thus letting the bttv driver control the card isn't very useful and causes trouble. The card has no PCI Subsystem ID, so I can't identify and blacklist it easily. Thus I need some way to allow the users to tell bttv (or the kernel) to ignore that particular PCI card.
Is there already something generic for this? Some kernel parameter which makes pci_module_init() skip a given PCI device for example?
Another way to fix this I'm thinking about is to add a insmod option along the lines "ignore=<bus>,<slot>" to bttv. Comments? Better idea?
Gerd
Signature · 3 lines
You can't please everybody. And usually if you _try_ to please everybody, the end result is one big mess. -- Linus Torvalds, 2002-04-20
Quoted 7 lines · Gerd Knorr
I have a small problem: Some vendor has built a PCI board which (ab-)uses the bt848/878 chip in creative ways to do some DMA. It is *not* a video card, thus letting the bttv driver control the card isn't very useful and causes trouble. The card has no PCI Subsystem ID, so I can't identify and blacklist it easily. Thus I need some way to allow the users to tell bttv (or the kernel) to ignore that particular PCI card.
Doh.. If the vendor isnt setting subsystem ids then its not valid hardware for windows nowdays. Obvious question - what else is on that board that might let you do the idents. We already find the USB on the NSC SuperIO by peeking at the next device along and checking if its the SuperIO functions 8)
Quoted 7 lines
Quoted 3 lines
can't identify and blacklist it easily. Thus I need some way to allow the users to tell bttv (or the kernel) to ignore that particular PCI card.
Doh.. If the vendor isnt setting subsystem ids then its not valid hardware for windows nowdays. Obvious question - what else is on that board that might let you do the idents.
Well, at least nothing in PCI space. It looks just like a random, cheap bt878 card.
Quoted 2 lines
We already find the USB on the NSC SuperIO by peeking at the next device along and checking if its the SuperIO functions 8)
It is a PCI card you can plug into some slot, not a motherboard ...
Gerd
Signature · 3 lines
You can't please everybody. And usually if you _try_ to please everybody, the end result is one big mess. -- Linus Torvalds, 2002-04-20
Quoted 4 lines · Gerd Knorr
Quoted 2 lines
We already find the USB on the NSC SuperIO by peeking at the next device along and checking if its the SuperIO functions 8)
It is a PCI card you can plug into some slot, not a motherboard ...
Yes but what other PCI devices are on that card ?
Quoted 2 lines · Gerd Knorr
Well, at least nothing in PCI space. It looks just like a random, cheap bt878 card.
Arggh read the email properly this time 8)
If its a BT878 on a board then assuming we can't tell the difference I guess it doesn't work mixed with tv cards. Does it work in windows paired with a real video capture card by some cheapo cardmaker that also doesn't set the ssid ?
Quoted 10 lines · Alan Cox
Quoted 2 lines · Gerd Knorr
Well, at least nothing in PCI space. It looks just like a random, cheap bt878 card.
Arggh read the email properly this time 8)
If its a BT878 on a board then assuming we can't tell the difference I guess it doesn't work mixed with tv cards. Does it work in windows paired with a real video capture card by some cheapo cardmaker that also doesn't set the ssid ?
Don't know, Cc:'ed the owner of the card. Currently it is paired with another bt878 grabber card which has a PCI Subsystem ID.
The windows driver's I've seen so far do check the ssid, althrough some handle it better than others: The (old) Hauppauge drivers I have become greatly confused if they find a card with a unknown ssid. Maybe that is fixed by now. The Terratec Windows drivers just pick the card they know about and ignore the other one as they should. Don't know what the generic/reference/cheapo-card drivers do.
The problem bttv has here is that it can't simply ignore cards without ssid because that would break all the crap^Wcheap boards out there ...
Gerd
Signature · 3 lines
You can't please everybody. And usually if you _try_ to please everybody, the end result is one big mess. -- Linus Torvalds, 2002-04-20
Hi!
Quoted 10 lines
I have a small problem: Some vendor has built a PCI board which (ab-)uses the bt848/878 chip in creative ways to do some DMA. It is *not* a video card, thus letting the bttv driver control the card isn't very useful and causes trouble. The card has no PCI Subsystem ID, so I can't identify and blacklist it easily. Thus I need some way to allow the users to tell bttv (or the kernel) to ignore that particular PCI card.
Is there already something generic for this? Some kernel parameter which makes
pci_module_init()skip a given PCI device for example?
What about writing a "driver" which will just bind to a given PCI device, so that the other drivers will see it's already handled?
Have a nice fortnight
Signature · 3 lines
Martin `MJ' Mares <[email protected]> http://atrey.karlin.mff.cuni.cz/~mj/ Faculty of Math and Physics, Charles University, Prague, Czech Rep., Earth For every complex problem, there's a solution that is simple, neat and wrong.
Quoted 5 lines · Martin Mares
Quoted 2 lines
Is there already something generic for this? Some kernel parameter which makes
pci_module_init()skip a given PCI device for example?What about writing a "driver" which will just bind to a given PCI device, so that the other drivers will see it's already handled?
pci_driver has no implicit ordering.
Quoted 1 line
pci_driverhas no implicit ordering.
Agreed, but I meant inserting it as a module before the other modules.
Have a nice fortnight
Signature · 3 lines
Martin `MJ' Mares <[email protected]> http://atrey.karlin.mff.cuni.cz/~mj/ Faculty of Math and Physics, Charles University, Prague, Czech Rep., Earth Current root password is "p3s5vwF50". Keep secret.
Quoted 4 lines · Martin Mares
Quoted 1 line
pci_driverhas no implicit ordering.Agreed, but I meant inserting it as a module before the other modules.
Which breaks the moment it meets a hotplug system
Quoted 7 lines · Alan Cox
Quoted 4 lines · Martin Mares
Quoted 1 line
pci_driverhas no implicit ordering.Agreed, but I meant inserting it as a module before the other modules.
Which breaks the moment it meets a hotplug system
Neverless this works fine in the non-hotplugging modular driver case, i.e. it can be used to fix that particular bt878 card issue. Might also be useful for testing ressource allocation error paths of PCI drivers.
Gerd
==============================[ cut here ]============================== #include <linux/module.h>
#include <linux/errno.h>
#include <linux/pci.h>
#include <linux/init.h>
#include <linux/kernel.h>
MODULE_DESCRIPTION("reserve pci card ressources to make device drivers *not* touch the card"); MODULE_AUTHOR("Gerd Knorr");MODULE_LICENSE("GPL");
static char *ignore = "none";MODULE_PARM(ignore,"s");
static int bus = -1, slot = -1,function = -1;
static void remove(struct pci_dev *dev)
{
int i;
printk("reserve: remove device %02x:%02x.%x\n",
dev->bus->number,PCI_SLOT(dev->devfn),PCI_FUNC(dev->devfn));
for (i = 0;; i++) {
if (0 == pci_resource_start(dev,i))
break;
if (pci_resource_start(dev,i) < 0x10000) {
release_region(pci_resource_start(dev,i),
pci_resource_len(dev,i));
} else {
release_mem_region(pci_resource_start(dev,i),
pci_resource_len(dev,i));
}
}
return;
}static int __devinit probe(struct pci_dev *dev,
const struct pci_device_id *id)
{
struct resource *res;
int i;
if (bus != dev->bus->number ||
slot != PCI_SLOT(dev->devfn) ||
function != PCI_FUNC(dev->devfn))
return -EBUSY;
printk("reserve: probe device %02x:%02x.%x\n",
dev->bus->number,PCI_SLOT(dev->devfn),PCI_FUNC(dev->devfn));
for (i = 0;; i++) {
if (0 == pci_resource_start(dev,i))
break;
if (pci_resource_start(dev,i) < 0x10000) {
res = request_region(pci_resource_start(dev,i),
pci_resource_len(dev,i),
"reserve");
printk("reserve: ... io at %lx size %lx [%s]\n",
pci_resource_start(dev,i),
pci_resource_len(dev,i),
res ? "ok" : "failed");
if (!res)
goto busy;
} else {
res = request_mem_region(pci_resource_start(dev,i),
pci_resource_len(dev,i),
"reserve");
printk("reserve: ... mem at %lx size %lx [%s]\n",
pci_resource_start(dev,i),
pci_resource_len(dev,i),
res ? "ok" : "failed");
if (!res)
goto busy;
}
}
return 0;busy:
i--;
while (i >= 0) {
if (pci_resource_start(dev,i) < 0x10000) {
release_region(pci_resource_start(dev,i),
pci_resource_len(dev,i));
} else {
release_mem_region(pci_resource_start(dev,i),
pci_resource_len(dev,i));
}
}
return -EBUSY;
}static struct pci_device_id pci_tbl[] __devinitdata = {
{
.vendor = PCI_ANY_ID,
.device = PCI_ANY_ID,
.subvendor = PCI_ANY_ID, .subdevice = PCI_ANY_ID,
},
{ /* end of list */}
};
MODULE_DEVICE_TABLE(pci, pci_tbl);
static struct pci_driver pci_driver = {
name: "reserve",
id_table: pci_tbl,
probe: probe,
remove: remove,
};
static int do_init(void)
{
if (3 == sscanf(ignore,"%x:%x.%x",&bus,&slot,&function)) {
/* nothing */;
} else if (2 == sscanf(ignore,"%x.%x",&slot,&function)) {
bus = 0;
} else if (2 == sscanf(ignore,"%x:%x",&bus,&slot)) {
function = 0;
} else if (1 == sscanf(ignore,"%x",&slot)) {
bus = 0;
function = 0;
} else {
printk("reserve: can't parse ignore=<device> insmod option\n");
return -EINVAL;
}
printk("reserve: ignore=\"%s\" -> using device %02x:%02x.%x\n",
ignore,bus,slot,function);
return pci_module_init(&pci_driver);
}static void do_fini(void)
{
pci_unregister_driver(&pci_driver);
}module_init(do_init);module_exit(do_fini);
/*
* Local variables:
* c-basic-offset: 8
* End:
*/
Quoted 3 lines · Gerd Knorr
Neverless this works fine in the non-hotplugging modular driver case, i.e. it can be used to fix that particular bt878 card issue. Might also be useful for testing ressource allocation error paths of PCI drivers.
Up until 2.4.20pre6 they are broken in the PCI core. IDE found that one out.
The reserve module is an interesting approach though. I had been pondering doing a rename module for some similar purposes but completely missed that one.
BTW we have pci_request_resource stuff that would shrink your module a fair bit
Quoted 10 lines · Alan Cox
Up until 2.4.20pre6 they are broken in the PCI core. IDE found that one out.
The reserve module is an interesting approach though. I had been pondering doing a rename module for some similar purposes but completely missed that one.
BTW we have
pci_request_resourcestuff that would shrink your module a fair bit
guess you mean pci_assign_resource()? Played with that one, now my /proc/iomem file looks *ahem* intresting. Is this the bug mentioned above?
Gerd
==============================[ cut here ]============================== Script started on Wed Sep 11 17:33:33 2002
bogomips kraxel ~/kernel/bttv-0.7.97/driver# lspci -vs a.0 00:0a.0 Multimedia video controller: Brooktree Corporation Bt878 (rev 11)
Subsystem: TERRATEC Electronic GmbH: Unknown device 1135
Flags: bus master, medium devsel, latency 32, IRQ 5 Memory at ce000000 (32-bit, prefetchable) [size=4K]
Capabilities: <available only to root>
bogomips kraxel ~/kernel/bttv-0.7.97/driver# insmod ./pci-reserve.o ignore=0a.0 bogomips kraxel ~/kernel/bttv-0.7.97/driver# dmesg | tail
lp0: using parport0 (polling).
nfs warning: mount version older than kernel/dev/vmnet: open called by PID 1250 (vmnet-netifup)/dev/vmnet: port on hub 8 successfully opened/dev/vmnet: open called by PID 1267 (vmnet-dhcpd)/dev/vmnet: port on hub 8 successfully opened
vmnet8: no IPv6 routers present
reserve: ignore="0a.0" -> using device 00:0a.0
reserve: probe device 00:0a.0
reserve: ... ressource at 40000000 size 1000 [ok]
bogomips kraxel ~/kernel/bttv-0.7.97/driver# cat /proc/iomem 00000000-0009efff : System RAM
0009f000-0009ffff : reserved
000a0000-000bffff : Video RAM area
000c0000-000c7fff : Video ROM
000c8000-000cb7ff : Extension ROM
000f0000-000fffff : System ROM
00100000-3fffbfff : System RAM
00100000-001ecc3c : Kernel code
001ecc3d-0024aa5f : Kernel data
3fffc000-3fffefff : ACPI Tables
3ffff000-3fffffff : ACPI Non-volatile Storage
40000000-40000fff : Brooktree Corporation Bt878 Video Capture ca000000-ca0000ff : LSI Logic / Symbios Logic (formerly NCR) 53c810 ca800000-ca8003ff : Philips Semiconductors SAA7134 (#2) cb000000-cb0000ff : Lite-On Communications Inc LNE100TX
cb000000-cb0000ff : tulip
cb800000-cb8003ff : Philips Semiconductors SAA7134 cc000000-cd6fffff : PCI Bus #01
cc000000-cc7fffff : Matrox Graphics, Inc. MGA G200 AGP cc800000-cc803fff : Matrox Graphics, Inc. MGA G200 AGP
cc800000-cc803fff : matroxfb MMIO
cd800000-cd800fff : Brooktree Corporation Bt878 Audio Capture 40000000-40000fff : Brooktree Corporation Bt878 Video Capture ca000000-ca0000ff : LSI Logic / Symbios Logic (formerly NCR) 53c810 ca800000-ca8003ff : Philips Semiconductors SAA7134 (#2) cb000000-cb0000ff : Lite-On Communications Inc LNE100TX
cb000000-cb0000ff : tulip
cb800000-cb8003ff : Philips Semiconductors SAA7134 cc000000-cd6fffff : PCI Bus #01
cc000000-cc7fffff : Matrox Graphics, Inc. MGA G200 AGP cc800000-cc803fff : Matrox Graphics, Inc. MGA G200 AGP
cc800000-cc803fff : matroxfb MMIO
cd800000-cd800fff : Brooktree Corporation Bt878 Audio Capture 40000000-40000fff : Brooktree Corporation Bt878 Video Capture ca000000-ca0000ff : LSI Logic / Symbios Logic (formerly NCR) 53c810 ca800000-ca8003ff : Philips Semiconductors SAA7134 (#2) cb000000-cb0000ff : Lite-On Communications Inc LNE100TX
cb000000-cb0000ff : tulip
cb800000-cb8003ff : Philips Semiconductors SAA7134 cc000000-cd6fffff : PCI Bus #01
cc000000-cc7fffff : Matrox Graphics, Inc. MGA G200 AGP cc800000-cc803fff : Matrox Graphics, Inc. MGA G200 AGP
cc800000-cc803fff : matroxfb MMIO
cd800000-cd800fff : Brooktree Corporation Bt878 Audio Capture 40000000-40000fff : Brooktree Corporation Bt878 Video Capture ca000000-ca0000ff : LSI Logic / Symbios Logic (formerly NCR) 53c810 ca800000-ca8003ff : Philips Semiconductors SAA7134 (#2) cb000000-cb0000ff : Lite-On Communications Inc LNE100TX
cb000000-cb0000ff : tulip
cb800000-cb8003ff : Philips Semiconductors SAA7134 cc000000-cd6fffff : PCI Bus #01
cc000000-cc7fffff : Matrox Graphics, Inc. MGA G200 AGP cc800000-cc803fff : Matrox Graphics, Inc. MGA G200 AGP
cc800000-cc803fff : matroxfb MMIO
cd800000-cd800fff : Brooktree Corporation Bt878 Audio Capture 40000000-40000fff : Brooktree Corporation Bt878 Video Capture ca000000-ca0000ff : LSI Logic / Symbios Logic (formerly NCR) 53c810 ca800000-ca8003ff : Philips Semiconductors SAA7134 (#2) cb000000-cb0000ff : Lite-On Communications Inc LNE100TX
cb000000-cb0000ff : tulip
cb800000-cb8003ff : Philips Semiconductors SAA7134 cc000000-cd6fffff : PCI Bus #01
cc000000-cc7fffff : Matrox Graphics, Inc. MGA G200 AGP cc800000-cc803fff : Matrox Graphics, Inc. MGA G200 AGP
cc800000-cc803fff : matroxfb MMIO
cd800000-cd800fff : Brooktree Corporation Bt878 Audio Capture 40000000-40000fff : Brooktree Corporation Bt878 Video Capture ca000000-ca0000ff : LSI Logic / Symbios Logic (formerly NCR) 53c810 ca800000-ca8003ff : Philips Semiconductors SAA7134 (#2) cb000000-cb0000ff : Lite-On Communications Inc LNE100TX
cb000000-cb0000ff : tulip
cb800000-cb8003ff : Philips Semiconductors SAA7134 cc000000-cd6fffff : PCI Bus #01
cc000000-cc7fffff : Matrox Graphics, Inc. MGA G200 AGP cc800000-cc803fff : Matrox Graphics, Inc. MGA G200 AGP
cc800000-cc803fff : matroxfb MMIO
cd800000-cd800fff : Brooktree Corporation Bt878 Audio Capture 40000000-40000fff : Brooktree Corporation Bt878 Video Capture ca000000-ca0000ff : LSI Logic / Symbios Logic (formerly NCR) 53c810 ca800000-ca8003ff : Philips Semiconductors SAA7134 (#2) cb000000-cb0000ff : Lite-On Communications Inc LNE100TX
cb000000-cb0000ff : tulip
bogomips kraxel ~/kernel/bttv-0.7.97/driver#
Script done on Wed Sep 11 17:34:56 2002 ==============================[ cut here ]============================== /*
* pci-reserve.c -- reserve pci device resources to make device \
* drivers *not* touch the card
*
* (c) 2002 Gerd Knorr <[email protected]>
*
*/
#include <linux/module.h>
#include <linux/errno.h>
#include <linux/pci.h>
#include <linux/init.h>
#include <linux/kernel.h>
MODULE_DESCRIPTION("reserve pci device resources to make device drivers *not* touch the card"); MODULE_AUTHOR("Gerd Knorr");MODULE_LICENSE("GPL");
static char *ignore = "none";
static int bus = -1, slot = -1,function = -1;MODULE_PARM(ignore,"s");
#ifndef MODULE
static int __init setup(char *str) { ignore = str; return 0; } __setup("pci-reserve.ignore=", setup);
#endif
static int __devinit probe(struct pci_dev *dev,
const struct pci_device_id *id)
{
int i,rc;
if (bus != dev->bus->number ||
slot != PCI_SLOT(dev->devfn) ||
function != PCI_FUNC(dev->devfn))
return -EBUSY;
printk("reserve: probe device %02x:%02x.%x\n",
dev->bus->number,PCI_SLOT(dev->devfn),PCI_FUNC(dev->devfn));
for (i = 0;; i++) {
if (0 == pci_resource_start(dev,i))
break;
rc = pci_assign_resource(dev,i);
printk("reserve: ... ressource at %lx size %lx [%s]\n",
pci_resource_start(dev,i),
pci_resource_len(dev,i),
rc ? "failed" : "ok");
if (rc)
return -EBUSY;
}
return 0;
}static struct pci_device_id pci_tbl[] __devinitdata = {
{
.vendor = PCI_ANY_ID,
.device = PCI_ANY_ID,
.subvendor = PCI_ANY_ID, .subdevice = PCI_ANY_ID,
},
{ /* end of list */}
};
MODULE_DEVICE_TABLE(pci, pci_tbl);
static struct pci_driver pci_driver = {
.name = "reserve",
.id_table = pci_tbl,
.probe = probe,};
static int parse(char *id, int *b, int *s, int *f)
{
if (3 == sscanf(id,"%x:%x.%x",b,s,f)) {
/* nothing */;
} else if (2 == sscanf(id,"%x.%x",s,f)) {
*b = 0;
} else if (2 == sscanf(id,"%x:%x",b,s)) {
*f= 0;
} else if (1 == sscanf(id,"%x",s)) {
*b = 0; *f= 0;
} else {
printk("reserve: can't parse \"%s\"\n",id);
return -1;
}
return 0;
}static int do_init(void)
{
if (parse(ignore,&bus,&slot,&function) < 0)
return -EINVAL;
printk("reserve: ignore=\"%s\" -> using device %02x:%02x.%x\n",
ignore,bus,slot,function);
return pci_module_init(&pci_driver);
}static void do_fini(void)
{
pci_unregister_driver(&pci_driver);
}module_init(do_init);module_exit(do_fini);
/*
* Local variables:
* c-basic-offset: 8
* End:
*/
Quoted 3 lines · Gerd Knorr
guess you mean
pci_assign_resource()? Played with that one, now my/proc/iomemfile looks *ahem* intresting. Is this the bug mentioned above?
Well I've not seen it appear that way but I guess it could do - the old code on finding a clash when reserving PCI resources (and you'll tickle that anyway with pci_module_init and two candidate drivers) freed resources it never allocated.
I'd have expected lots of printks first