Stradis driver conflicts with all other SAA7146 drivers
Hi,
[Nathan Laredo is the maintainer of the stradis driver but Jiri Slaby submitted the PCI probing change that went into 2.6.16 so I'm Cc-ing him too. I'm not a member of the video4linux mailing list so please Cc me on any responses.]
The stradis driver in the 2.6.16 kernel only looks at the SAA7146 vendor and product ID and binds to any SAA7146 based device even if it is not a stradis card. This stops all other SAA7146 drivers from working, for example my WinTV Nova-T card using the budget-ci driver doesn't work any longer. A lot of other people have also been bitten by this.
So could you please modify your driver so that it only binds to real stradis cards? If stradis cards have a subvendor and subdevice ID you could just modify the pci_device_id table to look something like this:
static struct pci_device_id stradis_pci_tbl[] = {
{
.vendor = PCI_VENDOR_ID_PHILIPS,
.device = PCI_DEVICE_ID_PHILIPS_SAA7146,
.subvendor = 0xdead,
.subdevice = 0xbeef,
},
{ 0 }
};Quoted 1 line
From the code it looks as if some boards don't have a good subsystem
vendor and device ID:
if (!pdev->subsystem_vendor)
dev_info(&pdev->dev, "%d: rev1 decoder\n", saa_num);
else
dev_info(&pdev->dev, "%d: SDM2xx found\n", saa_num);Are those rev1 boards out in the real world or are they just prototypes? If they are prototypes, can you add the subvendor ID anway and add a module parameter so that your driver only binds to unspecified vendor IDs if you use that module parameter? Something like this:
static int bind_to_anything = 0;module_param(bind_to_anything, int, 0);
int __init stradis_init(void)
{
...
if (bind_to_anything) {
stradis_pci_tbl[0].subvendor = PCI_ANY_ID;
stradis_pci_tbl[0].subdevice = PCI_ANY_ID;
}
...
pci_register_driver(&stradis_driver);
...
}For anyone submitting a new SAA7146 driver to the kernel in the future, please be aware that your card isn't the only one that uses that chip, so always add a subvendor/subdevice to your drivers.
/Christer
Signature · 4 lines
"Just how much can I get away with and still go to heaven?" Freelance consultant specializing in device driver programming for Linux Christer Weinigel <[email protected]> http://www.weinigel.se
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
Christer Weinigel napsal(a):
Quoted 57 lines
Hi,
[Nathan Laredo is the maintainer of the stradis driver but Jiri Slaby submitted the PCI probing change that went into 2.6.16 so I'm Cc-ing him too. I'm not a member of the video4linux mailing list so please Cc me on any responses.]
The stradis driver in the 2.6.16 kernel only looks at the SAA7146 vendor and product ID and binds to any SAA7146 based device even if it is not a stradis card. This stops all other SAA7146 drivers from working, for example my WinTV Nova-T card using the budget-ci driver doesn't work any longer. A lot of other people have also been bitten by this.
So could you please modify your driver so that it only binds to real stradis cards? If stradis cards have a subvendor and subdevice ID you could just modify the
pci_device_idtable to look something like this:static struct pci_device_id stradis_pci_tbl[] = { { .vendor = PCI_VENDOR_ID_PHILIPS, .device = PCI_DEVICE_ID_PHILIPS_SAA7146, .subvendor = 0xdead, .subdevice = 0xbeef, }, { 0 } };Quoted 1 line
From the code it looks as if some boards don't have a good subsystem
vendor and device ID:
if (!pdev->subsystem_vendor) dev_info(&pdev->dev, "%d: rev1 decoder\n", saa_num); else dev_info(&pdev->dev, "%d: SDM2xx found\n", saa_num);Are those rev1 boards out in the real world or are they just prototypes? If they are prototypes, can you add the subvendor ID anway and add a module parameter so that your driver only binds to unspecified vendor IDs if you use that module parameter? Something like this:
static int
bind_to_anything= 0;module_param(bind_to_anything, int, 0);int
__initstradis_init(void){ ... if (bind_to_anything) { stradis_pci_tbl[0].subvendor = PCI_ANY_ID; stradis_pci_tbl[0].subdevice = PCI_ANY_ID; } ... pci_register_driver(&stradis_driver); ... }
Quoted 3 lines
For anyone submitting a new SAA7146 driver to the kernel in the future, please be aware that your card isn't the only one that uses that chip, so always add a subvendor/subdevice to your drivers.
The only difference is in order of searching for devices. Stradis now gets control before your "real" driver. Kick stradis from your config or blacklist it. Or, why you ever load module, you don't want to use?
There is no change in searching devices, it didn't check for subvendors before not even now. If Nathan knows, there are some subvendor/subdevices ids, which we should compare to, then yes, we can change the behaviour, otherwise, I am afraid, we can't. It's vendors' problem, that they don't use this pci registers (and it's evil) -- i think, that stradis cards have that two zeroed.
regards,
- --
Jiri Slaby www.fi.muni.cz/~xslaby
\_.-^-._ [email protected] _.-^-._/
B67499670407CE62ACC8 22A032CC55C339D47A7E -----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.3 (GNU/Linux)
Comment: Using GnuPG with Fedora - http://enigmail.mozdev.org
iD8DBQFEeZ0kMsxVwznUen4RAoh5AJ4lmYmYd6PYUZsnKOEw6nBnd+BIyACgim5k PWpWmQPZc5PDpJdntNQ3eic=
=O7It
-----END PGP SIGNATURE-----
Em Dom, 2006-05-28 às 14:52 +0159, Jiri Slaby escreveu:
Quoted 7 lines
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1Quoted 3 lines
For anyone submitting a new SAA7146 driver to the kernel in the future, please be aware that your card isn't the only one that uses that chip, so always add a subvendor/subdevice to your drivers.
The better is to use the existent driver instead of creating newer ones.
Quoted 3 lines
The only difference is in order of searching for devices. Stradis now gets control before your "real" driver. Kick stradis from your config or blacklist it.
This sucks. Distros should compile all drivers to support as much hardware as possible. For now, it is better to implement Christer suggestions.
Quoted 1 line
Or, why you ever load module, you don't want to use?
There's no safe way to make a driver to get control after the others, if both are modules, so the option is to use the saa7146 driver instead of stradis.
Quoted 5 lines
There is no change in searching devices, it didn't check for subvendors before not even now. If Nathan knows, there are some subvendor/subdevices ids, which we should compare to, then yes, we can change the behaviour, otherwise, I am afraid, we can't. It's vendors' problem, that they don't use this pci registers (and it's evil) -- i think, that stradis cards have that two zeroed.
This is, in fact, a trouble on several video capture boards. A real merge work should be done by migrating stradis to use the generic support for SAA7146. Then, having just one board, those conflicts won't happen.
Quoted 4 lines
regards,
- --
Jiri Slaby www.fi.muni.cz/~xslaby
Cheers,
Mauro.
I agree that the real fix is to unify the stradis driver so that it uses the existing saa7146 driver (and extending the saa7146 driver if
it doesn't have all the support necessary yet). Keep in mind that at
the time the driver was written, there was no other saa7146-based card on the market (mid 1999).
Until the pci change there was never a single complaint.
Unfortunately it was ill timed to happen when I was busiest at work, and even now I'm on my way to New Zealand for a month where I'll be out of touc as well, so the "right" fix is not likely to happen soon.
I didn't even know that other saa7146 cards had been developed until the bug reports about the conflicts started pouring in. I don't run bleeding edge kernels anymore due to work having a rhel3 requirement because the lame cad tool vendors are so far behind the curve.
- Nathan Laredo
[email protected]
Quoted 38 lines · Mauro Carvalho Chehab
Em Dom, 2006-05-28 às 14:52 +0159, Jiri Slaby escreveu:
Quoted 7 lines
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1Quoted 3 lines
For anyone submitting a new SAA7146 driver to the kernel in the future, please be aware that your card isn't the only one that uses that chip, so always add a subvendor/subdevice to your drivers.
The better is to use the existent driver instead of creating newer ones.
Quoted 3 lines
The only difference is in order of searching for devices. Stradis now gets control before your "real" driver. Kick stradis from your config or blacklist it.
This sucks. Distros should compile all drivers to support as much hardware as possible. For now, it is better to implement Christer suggestions.
Quoted 1 line
Or, why you ever load module, you don't want to use?
There's no safe way to make a driver to get control after the others, if both are modules, so the option is to use the saa7146 driver instead of stradis.
Quoted 5 lines
There is no change in searching devices, it didn't check for subvendors before not even now. If Nathan knows, there are some subvendor/subdevices ids, which we should compare to, then yes, we can change the behaviour, otherwise, I am afraid, we can't. It's vendors' problem, that they don't use this pci registers (and it's evil) -- i think, that stradis cards have that two zeroed.
This is, in fact, a trouble on several video capture boards. A real merge work should be done by migrating stradis to use the generic support for SAA7146. Then, having just one board, those conflicts won't happen.
Quoted 4 lines
regards,
- --
Jiri Slaby www.fi.muni.cz/~xslabyCheers,
Mauro.
Quoted 15 lines · Jiri Slaby
Christer Weinigel napsal(a):
Quoted 7 lines
Hi,
[Nathan Laredo is the maintainer of the stradis driver but Jiri Slaby submitted the PCI probing change that went into 2.6.16 so I'm Cc-ing him too. I'm not a member of the video4linux mailing list so please Cc me on any responses.]
Quoted 6 lines
The stradis driver in the 2.6.16 kernel only looks at the SAA7146 vendor and product ID and binds to any SAA7146 based device even if it is not a stradis card. This stops all other SAA7146 drivers from working, for example my WinTV Nova-T card using the budget-ci driver doesn't work any longer. A lot of other people have also been bitten by this.
Quoted 8 lines
The only difference is in order of searching for devices. Stradis now gets control before your "real" driver. Kick stradis from your config or blacklist it. Or, why you ever load module, you don't want to use?
There is no change in searching devices, it didn't check for subvendors before not even now. If Nathan knows, there are some subvendor/subdevices ids, which we should compare to, then yes, we can change the behaviour, otherwise, I am afraid, we can't. It's vendors' problem, that they don't use this pci registers (and it's evil) -- i think, that stradis cards have that two zeroed.
I'm running the stock Fedora Core 5 kernels, and for some reason the stradis driver is loaded. I suppose there's some magic in the FC5 hotplug scripts that tries to load all device drivers that claim to support a certain PCI device.
I have blacklisted the stradis driver on my system, which fixes it for me, but it does feels as a workaround for a problem that ought to be fixed in the driver. If the card doesn't have a subvendor/subdevice, is there some way of doing a sanity check on the board to see if it actually is a stradis card and then release the board if it isn't?
If the driver isn't fixed I'll file a bug report on the Fedora bugzilla asking them to blacklist or just not compile that driver.
/Christer
Signature · 4 lines
"Just how much can I get away with and still go to heaven?" Freelance consultant specializing in device driver programming for Linux Christer Weinigel <[email protected]> http://www.weinigel.se
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
Nathan Laredo napsal(a):
Quoted 3 lines
I agree that the real fix is to unify the stradis driver so that it uses the existing saa7146 driver (and extending the saa7146 driver if it doesn't have all the support necessary yet).
Yup, the best way to solve it.
Quoted 2 lines
Until the pci change there was never a single complaint.
There was no pci_device_id table before (and no MODULE_DEVICE_TABLE), so userspace didn't load the driver at all.
cheers,
- --
Jiri Slaby www.fi.muni.cz/~xslaby
\_.-^-._ [email protected] _.-^-._/
B67499670407CE62ACC8 22A032CC55C339D47A7E -----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.3 (GNU/Linux)
Comment: Using GnuPG with Fedora - http://enigmail.mozdev.org
iD8DBQFEeczoMsxVwznUen4RAm7QAKCUavRhZmwp4KzxdbSioaSpGNIlrACeLRU/ WU2a51USyUrMG55XTdHHV78=
=5cCd
-----END PGP SIGNATURE-----
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
Christer Weinigel napsal(a):
Quoted 4 lines
I'm running the stock Fedora Core 5 kernels, and for some reason the stradis driver is loaded. I suppose there's some magic in the FC5 hotplug scripts that tries to load all device drivers that claim to support a certain PCI device.
Userspace now knows, that stradis can take control of this hardware, so US loads it, because MODULE_DEVICE_TABLE is present in the driver (you may compare old and new /lib/modules/uname -r/modules.pcimap). It shows us the next way, how to solve this -- delete this TABLE line from the driver, to not advertise "I can take control of it".
Quoted 6 lines
I have blacklisted the stradis driver on my system, which fixes it for me, but it does feels as a workaround for a problem that ought to be fixed in the driver. If the card doesn't have a subvendor/subdevice, is there some way of doing a sanity check on the board to see if it actually is a stradis card and then release the board if it isn't?
Unfortunately not.
Quoted 3 lines
If the driver isn't fixed I'll file a bug report on the Fedora bugzilla asking them to blacklist or just not compile that driver.
It is the best short-term solution, I think.
So, what to do now, Mauro, Nathan? Fix it in US (i.e. blacklist) or KS (i.e. delete TABLE entry -- hope this helps) for the time until the driver will be integrated?
regards,
- --
Jiri Slaby www.fi.muni.cz/~xslaby
\_.-^-._ [email protected] _.-^-._/
B67499670407CE62ACC8 22A032CC55C339D47A7E -----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.3 (GNU/Linux)
Comment: Using GnuPG with Fedora - http://enigmail.mozdev.org
iD8DBQFEedFnMsxVwznUen4RAnv3AKCthKpF67t+R+iAu0cs31icG1KH2ACgnwXw UKrSzuXdr/sBAr+rDmAdPIk=
=2R9i
-----END PGP SIGNATURE-----
Quoted 5 lines · Jiri Slaby
Christer Weinigel napsal(a):
Quoted 3 lines
fixed in the driver. If the card doesn't have a subvendor/subdevice, is there some way of doing a sanity check on the board to see if it actually is a stradis card and then release the board if it isn't?
Unfortunately not.
Why not? There's an I2C bus with a bunch of devices on it. Isn't it possible to do an I2C scan and if it doesn't match what's supposed to be on the card fail the probe and release the PCI resources?
If there is no FPGA or the FPGA fails to respond, that should also be a fairly good indicator that it is not a stradis board.
And it seems that at least some of the cards have a value in the subvendor field since the driver says "SDM2xx found" if pdev->subsystem_vendor is nonzero. So how common are the rev1 boards (which I assume have zeroes in the subvendor/subdevice fields) and what does "lspci -vn" for the SDM2xx boards say?
/Christer
Signature · 4 lines
"Just how much can I get away with and still go to heaven?" Freelance consultant specializing in device driver programming for Linux Christer Weinigel <[email protected]> http://www.weinigel.se
Em Dom, 2006-05-28 às 09:01 -0700, Nathan Laredo escreveu:
Quoted 6 lines
I agree that the real fix is to unify the stradis driver so that it uses the existing saa7146 driver (and extending the saa7146 driver if
it doesn't have all the support necessary yet). Keep in mind that at
the time the driver was written, there was no other saa7146-based card on the market (mid 1999).
Until the pci change there was never a single complaint.
It seems that both drivers were developed independently, and both went to kernel in the past. Only some time after Jiri we realized that PCI IDs were conflicting.
Quoted 4 lines
Unfortunately it was ill timed to happen when I was busiest at work, and even now I'm on my way to New Zealand for a month where I'll be out of touc as well, so the "right" fix is not likely to happen soon.
So, we need a quick fix. Maybe the better for now is to do a quick workaround to 2.6.17, backported also to 2.6.16. We can work for a definitive solution to 2.6.18 or 2.6.19.
Quoted 5 lines
I didn't even know that other saa7146 cards had been developed until the bug reports about the conflicts started pouring in. I don't run bleeding edge kernels anymore due to work having a rhel3 requirement because the lame cad tool vendors are so far behind the curve.
There are some new boards from a very well known vendor in Europe (Hauppauge) that uses also saa7146.
Currently, there are some drivers probing for saa7146 PCI IDs:
probing all PCI IDs:
dpc7146.c, hexium_orion.c, mxb.c, stradis.c
probing specific IDs:
hexium_gemini.c
I can imagine some possible solutions for this right now:
1) to include an insmod flag to allow PCI generic detection for each driver. So, the default behavior would be the one at the previous kernels;
2) We may create one saa7146_driver that probes the required PCI IDs and then load the corresponding driver of the above listed. For those boards whose don't have its own ID, an insmod parameter (or an autodetection procedure) may specify if it is a dpc7146, hexium_orion, mxb or stradis board.
(2) would be an interim solution until we integrate all stuff into just one driver.
Quoted 42 lines
- Nathan Laredo
[email protected]Quoted 38 lines · Mauro Carvalho Chehab
Em Dom, 2006-05-28 às 14:52 +0159, Jiri Slaby escreveu:
Quoted 7 lines
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1Quoted 3 lines
For anyone submitting a new SAA7146 driver to the kernel in the future, please be aware that your card isn't the only one that uses that chip, so always add a subvendor/subdevice to your drivers.
The better is to use the existent driver instead of creating newer ones.
Quoted 3 lines
The only difference is in order of searching for devices. Stradis now gets control before your "real" driver. Kick stradis from your config or blacklist it.
This sucks. Distros should compile all drivers to support as much hardware as possible. For now, it is better to implement Christer suggestions.
Quoted 1 line
Or, why you ever load module, you don't want to use?
There's no safe way to make a driver to get control after the others, if both are modules, so the option is to use the saa7146 driver instead of stradis.
Quoted 5 lines
There is no change in searching devices, it didn't check for subvendors before not even now. If Nathan knows, there are some subvendor/subdevices ids, which we should compare to, then yes, we can change the behaviour, otherwise, I am afraid, we can't. It's vendors' problem, that they don't use this pci registers (and it's evil) -- i think, that stradis cards have that two zeroed.
This is, in fact, a trouble on several video capture boards. A real merge work should be done by migrating stradis to use the generic support for SAA7146. Then, having just one board, those conflicts won't happen.
Quoted 4 lines
regards,
- --
Jiri Slaby www.fi.muni.cz/~xslabyCheers,
Mauro.
Cheers,
Mauro.
Christer Weinigel napsal(a):
Quoted 11 lines
Quoted 5 lines · Jiri Slaby
Christer Weinigel napsal(a):
Quoted 3 lines
fixed in the driver. If the card doesn't have a subvendor/subdevice, is there some way of doing a sanity check on the board to see if it actually is a stradis card and then release the board if it isn't?
Unfortunately not.
Why not? There's an I2C bus with a bunch of devices on it. Isn't it possible to do an I2C scan and if it doesn't match what's supposed to be on the card fail the probe and release the PCI resources?
This is an older method not used for device drivers, but only for searching for busses or i2c et al, of which drivers stands aside and controls the device.
Quoted 3 lines
If there is no FPGA or the FPGA fails to respond, that should also be a fairly good indicator that it is not a stradis board.
Yup, but pci probing doesn't have such mechanism.
Quoted 6 lines
And it seems that at least some of the cards have a value in the subvendor field since the driver says "SDM2xx found" if
pdev->subsystem_vendoris nonzero. So how common are the rev1 boards (which I assume have zeroes in the subvendor/subdevice fields) and what does "lspci -vn" for the SDM2xx boards say?
If it happens that way, it's easy to add sub*ids into pci device id table too and the problem will go away.
regards,
Signature · 3 lines
Jiri Slaby www.fi.muni.cz/~xslaby \_.-^-._ [email protected] _.-^-._/ B67499670407CE62ACC8 22A032CC55C339D47A7E
Quoted 17 lines · Jiri Slaby
Christer Weinigel napsal(a):
Quoted 9 lines
Quoted 6 lines · Jiri Slaby
Christer Weinigel napsal(a):
Quoted 3 lines
fixed in the driver. If the card doesn't have a subvendor/subdevice, is there some way of doing a sanity check on the board to see if it actually is a stradis card and then release the board if it isn't?
Quoted 5 lines
Quoted 1 line
Unfortunately not.
Why not? There's an I2C bus with a bunch of devices on it. Isn't it possible to do an I2C scan and if it doesn't match what's supposed to be on the card fail the probe and release the PCI resources?
Quoted 7 lines
This is an older method not used for device drivers, but only for searching for busses or i2c et al, of which drivers stands aside and controls the device.
Quoted 2 lines
If there is no FPGA or the FPGA fails to respond, that should also be a fairly good indicator that it is not a stradis board.
Yup, but pci probing doesn't have such mechanism.
Hum ?
The driver have to return an error (negative value) in the probbing function if it detect that the card fails to respond correctly.
Same happen for i2c.
If the driver didn't manage to find what it expect on the i2c bus, the driver won't be usefull for the device, so the driver should release the device.
Matthieu
Quoted 22 lines · Mauro Carvalho Chehab
Em Dom, 2006-05-28 às 09:01 -0700, Nathan Laredo escreveu:
Quoted 6 lines
I agree that the real fix is to unify the stradis driver so that it uses the existing saa7146 driver (and extending the saa7146 driver if
it doesn't have all the support necessary yet). Keep in mind that at
the time the driver was written, there was no other saa7146-based card on the market (mid 1999).
Until the pci change there was never a single complaint.It seems that both drivers were developed independently, and both went to kernel in the past. Only some time after Jiri we realized that PCI IDs were conflicting.
Quoted 4 lines
Unfortunately it was ill timed to happen when I was busiest at work, and even now I'm on my way to New Zealand for a month where I'll be out of touc as well, so the "right" fix is not likely to happen soon.
So, we need a quick fix. Maybe the better for now is to do a quick workaround to 2.6.17, backported also to 2.6.16. We can work for a definitive solution to 2.6.18 or 2.6.19.
Quoted 5 lines
I didn't even know that other saa7146 cards had been developed until the bug reports about the conflicts started pouring in. I don't run bleeding edge kernels anymore due to work having a rhel3 requirement because the lame cad tool vendors are so far behind the curve.
Quoted 8 lines
There are some new boards from a very well known vendor in Europe (Hauppauge) that uses also saa7146.
Currently, there are some drivers probing for saa7146 PCI IDs:
probing all PCI IDs:dpc7146.c, hexium_orion.c, mxb.c, stradis.cprobing specific IDs:
hexium_gemini.c
dpc7146, hexium_orion and mxb don't match all PCI IDs, they only match boards with zero as a board ID. So they won't conflict with non-broken boards that have valid subvendor IDs. But they will conflict with each other.
How may of these boards are broken and have zeroes in the subvendor/subdevice fields? Apparently some of the dpc7146f, hexium_orion, mxb, and stradis boards are broken. How many of the boards supported by the generic saa7146 driver are broken the same way?
Can't the stradis driver do the same thing as the other drivers and explicitly match the broken zero subvendor id and the non-broken subvendor IDs?
static struct pci_device_id pci_tbl[] = {
{
.vendor = PCI_VENDOR_ID_PHILIPS,
.device = PCI_DEVICE_ID_PHILIPS_SAA7146,
.subvendor = 0x0000,
.subdevice = 0x0000,
},
{
.vendor = PCI_VENDOR_ID_PHILIPS,
.device = PCI_DEVICE_ID_PHILIPS_SAA7146,
.subvendor = STRADIS_SUBVENDOR_ID,
.subdevice = STRADIS_SUBDEVICE_ID,
},
{ }
}That would fix the problem for me and make it SEP. :-)
Can somebody with a SDM2xx stradis board mail me the output from "lspci -vn" and I'll cobble together a patch that does this?
This still needs solving properly, but at least it makes it less of a problem for people with non-broken hardware.
/Christer
Signature · 4 lines
"Just how much can I get away with and still go to heaven?" Freelance consultant specializing in device driver programming for Linux Christer Weinigel <[email protected]> http://www.weinigel.se
Em Dom, 2006-05-28 às 19:58 +0200, Christer Weinigel escreveu:
Quoted 6 lines
Mauro Carvalho Chehab <[email protected]> writes:
dpc7146,
hexium_orionand mxb don't match all PCI IDs, they only match boards with zero as a board ID. So they won't conflict with non-broken boards that have valid subvendor IDs. But they will conflict with each other.
Yes. it is this I was trying to say :)
Quoted 6 lines
How may of these boards are broken and have zeroes in the subvendor/subdevice fields? Apparently some of the dpc7146f,
hexium_orion, mxb, and stradis boards are broken. How many of the boards supported by the generic saa7146 driver are broken the same way?
I don't have any saa7146 board, but, on bttv, most boards don't have its own PCI ID.
Quoted 7 lines
Can't the stradis driver do the same thing as the other drivers and explicitly match the broken zero subvendor id and the non-broken subvendor IDs?
That would fix the problem for me and make it SEP. :-)
This seems to be an interesting idea: to make stradis just probe the recognized IDs and the generic one. It won't solve, however, the probing intersection for dpc7146f, hexium_orion, mxb, and stradis when no subvendor ID is specified on the board.
Quoted 9 lines
Can somebody with a SDM2xx stradis board mail me the output from "lspci -vn" and I'll cobble together a patch that does this?
This still needs solving properly, but at least it makes it less of a problem for people with non-broken hardware.
/Christer
Cheers,
Mauro.
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
matthieu castet napsal(a):
Quoted 11 lines
Quoted 10 lines · Jiri Slaby
Christer Weinigel napsal(a):
Quoted 3 lines
Why not? There's an I2C bus with a bunch of devices on it. Isn't it possible to do an I2C scan and if it doesn't match what's supposed to be on the card fail the probe and release the PCI resources?
This is an older method not used for device drivers, but only for searching for
busses or i2c et al, of which drivers stands aside and controls the device.
Quoted 7 lines
Quoted 4 lines
Quoted 2 lines
If there is no FPGA or the FPGA fails to respond, that should also be a fairly good indicator that it is not a stradis board.
Yup, but pci probing doesn't have such mechanism.
Hum ?
The driver have to return an error (negative value) in the probbing function if it detect that the card fails to respond correctly.
I meant something other. Now it's clear, ignore that sentence in my reply, please. Of course, there is a mechanism how driver can tell layers upwards it can't drive the device.
regrads,
- --
Jiri Slaby www.fi.muni.cz/~xslaby
\_.-^-._ [email protected] _.-^-._/
B67499670407CE62ACC8 22A032CC55C339D47A7E -----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.3 (GNU/Linux)
Comment: Using GnuPG with Fedora - http://enigmail.mozdev.org
iD8DBQFEeggJMsxVwznUen4RAprXAKCNwSJpVIItEPm7cPSLvtS5xeT/gwCeI6UP 4SXefweIJhM0j2DA1wxlyZ4=
=RmKM
-----END PGP SIGNATURE-----
Hello,
sorry for jumping into this thread so late.
on 28.05.2006 19:58 Christer Weinigel said the following:
Quoted 7 lines
Quoted 1 line · Mauro Carvalho Chehab
Em Dom, 2006-05-28 às 09:01 -0700, Nathan Laredo escreveu:
dpc7146,
hexium_orionand mxb don't match all PCI IDs, they only match boards with zero as a board ID. So they won't conflict with non-broken boards that have valid subvendor IDs. But they will conflict with each other.
That's unfortunately right.
Quoted 3 lines
How may of these boards are broken and have zeroes in the subvendor/subdevice fields? Apparently some of the dpc7146f,
hexium_orion, mxb, and stradis boards are broken.
I would not call them broken.
They simply don't have subvendor/subdevice informations, but when these informations are requested, the card gives back zeroes.
Quoted 3 lines
How many of the
boards supported by the generic saa7146 driver are broken the same way?
I can confirm this for the dpc7146, the mxb and the hexium_orion.
Quoted 2 lines
This still needs solving properly, but at least it makes it less of a problem for people with non-broken hardware.
Up to now, this problem did not show up so drastically. On the one hand, just a few dozen people use the MXB, a handful use the hexium_orion and the dpc7146 is really rare. All these users then simply tweaked their environment so that it works for the next boot.
In order to fix this, these drivers should not be autoloaded because there is no sane way to autodetect these cards. In theory, you could do an i2c bus scan and check if all devices are there. But since MXB and dpc7146 both use the saa7111 video decoder on address 0x11 IIRC, the dpc7146 will grab any MXB device if loaded before.
Distributions probably should keep these drivers from being autoloaded.
Quoted 1 line
/Christer
Best regards
Michael.
Hi,
on 28.05.2006 20:40 Mauro Carvalho Chehab said the following:
Quoted 2 lines
I don't have any saa7146 board, but, on bttv, most boards don't have its own PCI ID.
As I said in the other mail, dpc7146, mxb and hexium_orion don't have subvendor/subdevice ids.
Quoted 3 lines
It won't solve, however, the probing
intersection for dpc7146f,hexium_orion, mxb, and stradis when no subvendor ID is specified on the board.
Probing is probably not possible for the devices mentioned above.
Quoted 2 lines
Cheers,
Mauro.
CU
Michael.
Hi, Michael,
Em Seg, 2006-05-29 às 14:46 +0200, Michael Hunold escreveu:
Quoted 6 lines
Hi,
on 28.05.2006 20:40 Mauro Carvalho Chehab said the following:
Quoted 2 lines
I don't have any saa7146 board, but, on bttv, most boards don't have its own PCI ID.
Quoted 8 lines
As I said in the other mail, dpc7146, mxb and
hexium_oriondon't have subvendor/subdevice ids.Quoted 3 lines
It won't solve, however, the probing
intersection for dpc7146f,hexium_orion, mxb, and stradis when no subvendor ID is specified on the board.Probing is probably not possible for the devices mentioned above.
On bttv and other boards, were we have such conflicts, we have an option to specify what board is used (called card). When the driver locates a board without PCI subvendor ID, it shows a help msg at dmesg and exits (or load a generic handler). The user may then use card=xx (where xx is the number of the board). IMHO, this is the better for saa7146 boards.
Quoted 6 lines
Quoted 3 lines
Cheers,
Mauro.CU
Michael.
Cheers,
Mauro.
Hi,
on 29.05.2006 15:33 Mauro Carvalho Chehab said the following:
Quoted 5 lines
On bttv and other boards, were we have such conflicts, we have an option to specify what board is used (called card). When the driver locates a board without PCI subvendor ID, it shows a help msg at dmesg and exits (or load a generic handler). The user may then use card=xx (where xx is the number of the board). IMHO, this is the better for saa7146 boards.
bttv is a monolithic driver for all devices using Bt8x8 chipsets, whereas saa7146 and saa7146_vv only hold the core infrastructure that is common for all saa7146 cards.
saa7146 does not know anything about a card, but the so-called extension driver (like mxb, hexium_orion or dpc7146) does. It holds the PCI IDs and is responsible for telling the system what card it supports and do any probing if necessary.
I don't know where to put the card=xx parameter in that case, because the hexium_orion does not know mxb nor dpc7146. 8-(
Since these cards don't have subvendor/subdevice IDs, it's impossible to find out which card is in the system.
CU
Michael.
Em Seg, 2006-05-29 às 15:43 +0200, Michael Hunold escreveu:
Quoted 18 lines
Hi,
on 29.05.2006 15:33 Mauro Carvalho Chehab said the following:
Quoted 5 lines
On bttv and other boards, were we have such conflicts, we have an option to specify what board is used (called card). When the driver locates a board without PCI subvendor ID, it shows a help msg at dmesg and exits (or load a generic handler). The user may then use card=xx (where xx is the number of the board). IMHO, this is the better for saa7146 boards.
bttv is a monolithic driver for all devices using Bt8x8 chipsets, whereas saa7146 and
saa7146_vvonly hold the core infrastructure that is common for all saa7146 cards.saa7146 does not know anything about a card, but the so-called extension driver (like mxb,
hexium_orionor dpc7146) does. It holds the PCI IDs and is responsible for telling the system what card it supports and do any probing if necessary.
Quoted 5 lines
I don't know where to put the card=xx parameter in that case, because the
hexium_oriondoes not know mxb nor dpc7146. 8-(Since these cards don't have subvendor/subdevice IDs, it's impossible to find out which card is in the system.
We have two tasks:
1) Integrate your code and Nathan one;
2) create a generic handler for all saa7146 boards, moving all PCI probe to the newer module. After detecting the card number, it should request the specific module.
Quoted 3 lines
CU
Michael.
Cheers,
Mauro.
Quoted 10 lines
Quoted 2 lines
Since these cards don't have subvendor/subdevice IDs, it's impossible to find out which card is in the system.
We have two tasks:
1) Integrate your code and Nathan one;
2) create a generic handler for all saa7146 boards, moving all PCI probe to the newer module. After detecting the card number, it should request the specific module.
Well, that doesn't solve the fundamental problem that you can't get it right without help of the user. Maybe the best solution would be that no driver is allowed to occupy a device without subsystem id unless explicitly asked by the user to do so (that would also a good idea for bttv btw). A small PCI probing module would be one way to do that, but it is certainly not the only one. You could also do something like adding a pcislot=... insmod option to the existing drivers.
just my 2 cent,
Gerd
Signature · 2 lines
Gerd Hoffmann <[email protected]> http://www.suse.de/~kraxel/julika-dora.jpeg
Quoted 10 lines · Michael Hunold
Up to now, this problem did not show up so drastically. On the one hand, just a few dozen people use the MXB, a handful use the
hexium_orionand the dpc7146 is really rare. All these users then simply tweaked their environment so that it works for the next boot.In order to fix this, these drivers should not be autoloaded because there is no sane way to autodetect these cards. In theory, you could do an i2c bus scan and check if all devices are there. But since MXB and dpc7146 both use the saa7111 video decoder on address
0x11IIRC, the dpc7146 will grab any MXB device if loaded before.
Ouch. Oh well, that means that those cards will need some kind of
manual intervention.
Quoted 1 line
Distributions probably should keep these drivers from being autoloaded.
I'm not too sure of how the module autoloading works, but would it help just remove the line with:
MODULE_DEVICE_TABLE(pci, pci_tbl);
from those drivers? If I understand correctly, the contens of MODULE_DEVICE_TABLE is what ends up in modules.pcimap. That should fix it for most distributions and the owners of those cards have to add a manual modprobe command to their rc.local.
/Christer
Signature · 4 lines
"Just how much can I get away with and still go to heaven?" Freelance consultant specializing in device driver programming for Linux Christer Weinigel <[email protected]> http://www.weinigel.se
Quoted 5 lines · Mauro Carvalho Chehab
1) Integrate your code and Nathan one;
2) create a generic handler for all saa7146 boards, moving all PCI probe to the newer module. After detecting the card number, it should request the specific module.
#2 breaks some existing setups that build modules to load based on the PCI tables or built initrds this way. You can have two drivers for the same PCI identifier providing that they both know how to bail out for the wrong type of card
Quoted 9 lines · Alan Cox
Quoted 5 lines · Mauro Carvalho Chehab
1) Integrate your code and Nathan one;
2) create a generic handler for all saa7146 boards, moving all PCI probe to the newer module. After detecting the card number, it should request the specific module.
#2 breaks some existing setups that build modules to load based on the PCI tables or built initrds this way.
not if you make this meta module depend on the other two, that way all this infrastructure will just pull the right stuff in.. at the price of some wasted memory for the one you don't need