Skip to contentSkip to conversation

compat ioctl32 for /dev/snapshot?

DISCUSSION Michael Tokarev 5 participants 19 messages May 4, 2009 – Jul 14, 2009

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

Is there any reason why 32-bit uswsusp &Friends does not work on 64bits kernel?

For one, 32bits s2disk produces the following when trying to suspend:

ioctl32(s2disk:4134): Unknown cmd fd(4) cmd(400c330d){t:'3';sz:12} arg(ff853554) on /dev/snapshot ioctl32(s2disk:4134): Unknown cmd fd(4) cmd(4004330a){t:'3';sz:4} arg(00000805) on /dev/snapshot

this is coming from:

    error = ioctl(dev, SNAPSHOT_SET_SWAP_AREA, &swap);
    if (error && !offset)
            error = ioctl(dev, SNAPSHOT_SET_SWAP_FILE, blkdev);

but I guess (just guess!) that other SNAPSHOT_* operations will also fail the same way.

Is there a reason why those are not in compat_ioctl?

Thanks.

/mjt

#1 lore raw
Quoted 11 lines · Michael Tokarev

Is there any reason why 32-bit uswsusp &Friends does not work on 64bits kernel?

For one, 32bits s2disk produces the following when trying to suspend:

ioctl32(s2disk:4134): Unknown cmd fd(4) cmd(400c330d){t:'3';sz:12} arg(ff853554) on /dev/snapshot ioctl32(s2disk:4134): Unknown cmd fd(4) cmd(4004330a){t:'3';sz:4} arg(00000805) on /dev/snapshot

this is coming from:

Quoted 8 lines
    error = ioctl(dev, SNAPSHOT_SET_SWAP_AREA, &swap);
    if (error && !offset)
            error = ioctl(dev, SNAPSHOT_SET_SWAP_FILE, blkdev);

but I guess (just guess!) that other SNAPSHOT_* operations will also fail the same way.

Is there a reason why those are not in compat_ioctl?

It's probably just that nobody has written the code yet. In general all missing compat_ioctls are bugs.

-Andi

Signature · 1 line
[email protected] -- Speaking for myself only.
#2 lore raw
Quoted 11 lines · Andi Kleen
Quoted 9 lines · Michael Tokarev

Is there any reason why 32-bit uswsusp &Friends does not work on 64bits kernel?

For one, 32bits s2disk produces the following when trying to suspend:

ioctl32(s2disk:4134): Unknown cmd fd(4) cmd(400c330d){t:'3';sz:12} arg(ff853554) on /dev/snapshot ioctl32(s2disk:4134): Unknown cmd fd(4) cmd(4004330a){t:'3';sz:4} arg(00000805) on /dev/snapshot

[]

Quoted 2 lines

It's probably just that nobody has written the code yet. In general all missing compat_ioctls are bugs.

Oh well.

Is the following patch ok? I just pulled all the SNAPSHOT_* stuff from include/linux/suspend_ioctls.h and added them into fs/compat_ioctl.c. The ioctls are:
o argument-less (most of them are)
o have single loff_t argument (other ioctls with the same argument are marked as COMPAT_IOCTL
o have single int argument - they's also marked as COMPAT_IOCTL,
o and one othem, SNAPSHOT_SET_SWAP_AREA, has argument pointing to the following structure (include/linux/suspend_ioctls.h):

    struct resume_swap_area {
        loff_t offset;
        u_int32_t dev;
    } __attribute__((packed));

so I think it also does not need any translation layer.

I can't test it so far, because uswsusp tools are broken in mixed 32/64bit case in other places. But at least it compiles fine and does not complain anymore.

I never touched this area before so I may be wrong... but if it's ok...

Thanks!

/mjt

fs/compat_ioctl.c renamed from fs/compat_ioctl.c.orig +17 -0
@@ -112,2 +112,4 @@
 
+#include <linux/suspend_ioctls.h>
+
 #ifdef CONFIG_SPARC
@@ -2301,2 +2303,17 @@ COMPATIBLE_IOCTL(SOUND_MIXER_SETLEVELS)
 COMPATIBLE_IOCTL(OSS_GETVERSION)
+/* SNAPSHOT */
+COMPATIBLE_IOCTL(SNAPSHOT_FREEZE)
+COMPATIBLE_IOCTL(SNAPSHOT_UNFREEZE)
+COMPATIBLE_IOCTL(SNAPSHOT_ATOMIC_RESTORE)
+COMPATIBLE_IOCTL(SNAPSHOT_FREE)
+COMPATIBLE_IOCTL(SNAPSHOT_FREE_SWAP_PAGES)
+COMPATIBLE_IOCTL(SNAPSHOT_S2RAM)
+COMPATIBLE_IOCTL(SNAPSHOT_SET_SWAP_AREA)	/* struct resume_swap_area */
+COMPATIBLE_IOCTL(SNAPSHOT_GET_IMAGE_SIZE)
+COMPATIBLE_IOCTL(SNAPSHOT_PLATFORM_SUPPORT)
+COMPATIBLE_IOCTL(SNAPSHOT_POWER_OFF)
+COMPATIBLE_IOCTL(SNAPSHOT_CREATE_IMAGE)
+COMPATIBLE_IOCTL(SNAPSHOT_PREF_IMAGE_SIZE)
+COMPATIBLE_IOCTL(SNAPSHOT_AVAIL_SWAP_SIZE)
+COMPATIBLE_IOCTL(SNAPSHOT_ALLOC_SWAP_PAGE)
 /* AUTOFS */
#3 lore raw
Quoted 14 lines · Michael Tokarev
Quoted 5 lines · Andi Kleen
Quoted 3 lines · Michael Tokarev

Is there any reason why 32-bit uswsusp &Friends does not work on 64bits kernel?

Quoted 8 lines
Quoted 8 lines

For one, 32bits s2disk produces the following when trying to suspend:

ioctl32(s2disk:4134): Unknown cmd fd(4) cmd(400c330d){t:'3';sz:12} arg(ff853554) on /dev/snapshot
ioctl32(s2disk:4134): Unknown cmd fd(4) cmd(4004330a){t:'3';sz:4} arg(00000805) on /dev/snapshot

Quoted 7 lines

[]

Quoted 2 lines

It's probably just that nobody has written the code yet. In general all missing compat_ioctls are bugs.

Oh well.

Is the following patch ok? I just pulled all the SNAPSHOT_* stuff from

You should ask Rafael (cc'ed) who maintains that code.

Quoted 4 lines

I can't test it so far, because uswsusp tools are broken in mixed 32/64bit case in other places. But at least it compiles fine and does not complain anymore.

It would be good to test before merging.

-Andi

Signature · 1 line
[email protected] -- Speaking for myself only.
#4 lore raw
Quoted 16 lines · Andi Kleen
Quoted 15 lines · Michael Tokarev
Quoted 5 lines · Andi Kleen
Quoted 3 lines · Michael Tokarev

Is there any reason why 32-bit uswsusp &Friends does not work on 64bits kernel?

Quoted 8 lines
Quoted 8 lines

For one, 32bits s2disk produces the following when trying to suspend:

ioctl32(s2disk:4134): Unknown cmd fd(4) cmd(400c330d){t:'3';sz:12} arg(ff853554) on /dev/snapshot
ioctl32(s2disk:4134): Unknown cmd fd(4) cmd(4004330a){t:'3';sz:4} arg(00000805) on /dev/snapshot

[]

Quoted 8 lines
Quoted 6 lines
Quoted 2 lines

It's probably just that nobody has written the code yet. In general all missing compat_ioctls are bugs.

Oh well.

Is the following patch ok? I just pulled all the SNAPSHOT_* stuff from

You should ask Rafael (cc'ed) who maintains that code.

Thanks for CCing me. :-)

In fact I don't think the 32-bit user space will work with 64-bit kernels as is in this particular case, because of the different pointer size.

Having a quick look at the code (I don't remember the details right now) I think most probably it could be modified to handle this case too, but I'm not really sure.

Best,
Rafael

#5 lore raw
Quoted 6 lines · Rafael J. Wysocki

In fact I don't think the 32-bit user space will work with 64-bit kernels as is in this particular case, because of the different pointer size.

Having a quick look at the code (I don't remember the details right now) I think most probably it could be modified to handle this case too, but I'm not really sure.

I don't see anything in the snapshot code that passes pointers to the kernel, so why should the pointer size matter?

	Arnd <><
#6 lore raw
Quoted 10 lines · Arnd Bergmann
Quoted 6 lines · Rafael J. Wysocki

In fact I don't think the 32-bit user space will work with 64-bit kernels as is in this particular case, because of the different pointer size.

Having a quick look at the code (I don't remember the details right now) I think most probably it could be modified to handle this case too, but I'm not really sure.

I don't see anything in the snapshot code that passes pointers to the kernel, so why should the pointer size matter?

It's the userspace part (uswsusp). In particular it parses swap on-disk data structures.

But in any case, without compat_ioctl32 in kernel userspace part can not be fixed.

/mjt

#7 lore raw
Quoted 9 lines · Michael Tokarev
Quoted 8 lines · Arnd Bergmann
Quoted 6 lines · Rafael J. Wysocki

In fact I don't think the 32-bit user space will work with 64-bit kernels as is in this particular case, because of the different pointer size.

Having a quick look at the code (I don't remember the details right now) I think most probably it could be modified to handle this case too, but I'm not really sure.

Quoted 8 lines
Quoted 2 lines

I don't see anything in the snapshot code that passes pointers to the kernel, so why should the pointer size matter?

It's the userspace part (uswsusp). In particular it parses swap on-disk data structures.

But in any case, without compat_ioctl32 in kernel userspace part can not be fixed.

Well, agreed. :-)

Rafael

#8 lore raw
In reply to Michael Tokarev
Quoted 4 lines · Michael Tokarev

Is the following patch ok?  I just pulled all the SNAPSHOT_* stuff from include/linux/suspend_ioctls.h and added them into fs/compat_ioctl.c. The ioctls are:
  o argument-less (most of them are)

Some of the ones that do not take a pointer argument actually do take an integer argument, in particular SNAPSHOT_PREF_IMAGE_SIZE and SNAPSHOT_PLATFORM_SUPPORT, which you need to list as ULONG_IOCTL rather than COMPAT_IOCTL if you want to add them to compat_ioctl.c.

Quoted 3 lines

  o have single loff_t argument (other ioctls with the same argument are marked as COMPAT_IOCTL
  o have single int argument - they's also marked as COMPAT_IOCTL,

right.

Quoted 7 lines

  o and one othem, SNAPSHOT_SET_SWAP_AREA, has argument pointing to the following structure (include/linux/suspend_ioctls.h):
     struct resume_swap_area {
         loff_t offset;
         u_int32_t dev;
     } __attribute__((packed));
    so I think it also does not need any translation layer.

this one would be compat_ioctl as well, because the __attribute__((packed)) turns it into a well-defined size of 12 bytes on any architecture (without the attribute, it would be 16 bytes on most architectures, but not i386. The recommended way to do this for new architectures would be explicit padding rather than packed attribute.).

You are however missing support for deprecated ioctls in your
code: SNAPSHOT_SET_SWAP_FILE and SNAPSHOT_PMOPS are trivial (COMPATIBLE_IOCTL), but you also need to add support for these

#define SNAPSHOT_ATOMIC_SNAPSHOT32 _IOW(SNAPSHOT_IOC_MAGIC, 3, compat_uptr_t) #define SNAPSHOT_SET_IMAGE_SIZE32 _IOW(SNAPSHOT_IOC_MAGIC, 6, compat_ulong) #define SNAPSHOT_AVAIL_SWAP32 _IOR(SNAPSHOT_IOC_MAGIC, 7, compat_uptr_t) #define SNAPSHOT_GET_SWAP_PAGE32 _IOR(SNAPSHOT_IOC_MAGIC, 8, compat_uptr_t)

Quoted 3 lines

I can't test it so far, because uswsusp tools are broken in mixed 32/64bit case in other places.  But at least it compiles fine and does not complain anymore.

I try to reduce the size of compat_ioctl.c. A better implementation would be to add a ->compat_ioctl() operation to the file_operations and list the compatible ioctl numbers as well.

Please try this patch instead:

suspend: Add compat_ioctl for snapshot device

All of the ioctl numbers in here are compatible, but for some of the deprecated numbers, we need to add aliases.

kernel/power/user.c +19 -0
@@ -51,6 +51,10 @@
 #define SNAPSHOT_SET_IMAGE_SIZE		_IOW(SNAPSHOT_IOC_MAGIC, 6, unsigned long)
 #define SNAPSHOT_AVAIL_SWAP		_IOR(SNAPSHOT_IOC_MAGIC, 7, void *)
 #define SNAPSHOT_GET_SWAP_PAGE		_IOR(SNAPSHOT_IOC_MAGIC, 8, void *)
+#define SNAPSHOT_ATOMIC_SNAPSHOT32	_IOW(SNAPSHOT_IOC_MAGIC, 3, compat_uptr_t)
+#define SNAPSHOT_SET_IMAGE_SIZE32	_IOW(SNAPSHOT_IOC_MAGIC, 6, compat_ulong)
+#define SNAPSHOT_AVAIL_SWAP32		_IOR(SNAPSHOT_IOC_MAGIC, 7, compat_uptr_t)
+#define SNAPSHOT_GET_SWAP_PAGE32	_IOR(SNAPSHOT_IOC_MAGIC, 8, compat_uptr_t)
 
 
 #define SNAPSHOT_MINOR	231
@@ -249,6 +253,9 @@ static long snapshot_ioctl(struct file *filp, unsigned int cmd,
 
 	case SNAPSHOT_CREATE_IMAGE:
 	case SNAPSHOT_ATOMIC_SNAPSHOT:
+#ifdef CONFIG_COMPAT
+	case SNAPSHOT_ATOMIC_SNAPSHOT32:
+#endif
 		if (data->mode != O_RDONLY || !data->frozen  || data->ready) {
 			error = -EPERM;
 			break;
@@ -278,6 +285,9 @@ static long snapshot_ioctl(struct file *filp, unsigned int cmd,
 
 	case SNAPSHOT_PREF_IMAGE_SIZE:
 	case SNAPSHOT_SET_IMAGE_SIZE:
+#ifdef CONFIG_COMPAT
+	case SNAPSHOT_SET_IMAGE_SIZE32:
+#endif
 		image_size = arg;
 		break;
 
@@ -293,6 +303,9 @@ static long snapshot_ioctl(struct file *filp, unsigned int cmd,
 
 	case SNAPSHOT_AVAIL_SWAP_SIZE:
 	case SNAPSHOT_AVAIL_SWAP:
+#ifdef CONFIG_COMPAT
+	case SNAPSHOT_AVAIL_SWAP32:
+#endif
 		size = count_swap_pages(data->swap, 1);
 		size <<= PAGE_SHIFT;
 		error = put_user(size, (loff_t __user *)arg);
@@ -300,6 +313,9 @@ static long snapshot_ioctl(struct file *filp, unsigned int cmd,
 
 	case SNAPSHOT_ALLOC_SWAP_PAGE:
 	case SNAPSHOT_GET_SWAP_PAGE:
+#ifdef CONFIG_COMPAT
+	case SNAPSHOT_GET_SWAP_PAGE32:
+#endif
 		if (data->swap < 0 || data->swap >= MAX_SWAPFILES) {
 			error = -ENODEV;
 			break;
@@ -436,6 +452,9 @@ static const struct file_operations snapshot_fops = {
 	.write = snapshot_write,
 	.llseek = no_llseek,
 	.unlocked_ioctl = snapshot_ioctl,
+#ifdef CONFIG_COMPAT
+	.compat_ioctl = snapshot_ioctl,
+#endif
 };
 
 static struct miscdevice snapshot_device = {
#9 lore raw
Quoted 3 lines · Arnd Bergmann
Quoted 2 lines · Michael Tokarev

Is the following patch ok? I just pulled all the SNAPSHOT_* stuff from include/linux/suspend_ioctls.h and added them into fs/compat_ioctl.c.

[]

Quoted 4 lines

You are however missing support for deprecated ioctls in your
code: SNAPSHOT_SET_SWAP_FILE and SNAPSHOT_PMOPS are trivial (COMPATIBLE_IOCTL), but you also need to add support for these

Well, as comments in that file (kernel/power/user.c) states, those ioctls are obsolete and will be removed and are only preserved for compatibility etc. Since no one complained so far (well, no one complained about whole /dev/snapshot thing at all ;), maybe there's no need to define those and the following ones too?

Quoted 5 lines

#define SNAPSHOT_ATOMIC_SNAPSHOT32 _IOW(SNAPSHOT_IOC_MAGIC, 3, compat_uptr_t) #define SNAPSHOT_SET_IMAGE_SIZE32 _IOW(SNAPSHOT_IOC_MAGIC, 6, compat_ulong) #define SNAPSHOT_AVAIL_SWAP32 _IOR(SNAPSHOT_IOC_MAGIC, 7, compat_uptr_t) #define SNAPSHOT_GET_SWAP_PAGE32 _IOR(SNAPSHOT_IOC_MAGIC, 8, compat_uptr_t)

Quoted 7 lines
Quoted 3 lines

I can't test it so far, because uswsusp tools are broken in mixed 32/64bit case in other places. But at least it compiles fine and does not complain anymore.

I try to reduce the size of compat_ioctl.c. A better implementation would be to add a ->compat_ioctl() operation to the file_operations and list the compatible ioctl numbers as well.

Well, your way is definitely a lot cleaner. But I didn't know about either of them anyway :)

Quoted 1 line

Please try this patch instead:

Ok. Your patch was garbled by your MUA (inserting =3D =20 and breaking lines), but that's ok. After de-garbling it now complains about undefined compat_uptr_t. For that, I've added

#ifdef CONFIG_COMPAT
#include <linux/compat.h>
#endif

at the end of the file. Also there was a typo (duplicated above), compat_ulong should be compat_ulong_t. After all that it finally compiles. As of "works" -- the above my comment still applies, uswsusp for one is badly broken on 32/64 bits. I'm trying to create a simpler test-case, as time permits.

Thanks!

/mjt

#10 lore raw
Quoted 11 lines · Michael Tokarev
Quoted 3 lines · Arnd Bergmann
Quoted 2 lines · Michael Tokarev

Is the following patch ok? I just pulled all the SNAPSHOT_* stuff from include/linux/suspend_ioctls.h and added them into fs/compat_ioctl.c.

[]

Quoted 4 lines

You are however missing support for deprecated ioctls in your
code: SNAPSHOT_SET_SWAP_FILE and SNAPSHOT_PMOPS are trivial (COMPATIBLE_IOCTL), but you also need to add support for these

Quoted 10 lines

Well, as comments in that file (kernel/power/user.c) states, those ioctls are obsolete and will be removed and are only preserved for compatibility etc. Since no one complained so far (well, no one complained about whole /dev/snapshot thing at all ;), maybe there's no need to define those and the following ones too?

Quoted 4 lines

#define SNAPSHOT_ATOMIC_SNAPSHOT32 _IOW(SNAPSHOT_IOC_MAGIC, 3, compat_uptr_t) #define SNAPSHOT_SET_IMAGE_SIZE32 _IOW(SNAPSHOT_IOC_MAGIC, 6, compat_ulong) #define SNAPSHOT_AVAIL_SWAP32 _IOR(SNAPSHOT_IOC_MAGIC, 7, compat_uptr_t) #define SNAPSHOT_GET_SWAP_PAGE32 _IOR(SNAPSHOT_IOC_MAGIC, 8, compat_uptr_t)

The point is that they are meant for compatibility with existing binaries, which is also the reason for having the compat_ioctl in the first place. The kernel should behave in identical ways for both 32 and 64 bits, so either we add all of the ioctl numbers in compat mode, or we remove support for the obsolete calls in native mode as well.

Quoted 4 lines
Quoted 1 line

Please try this patch instead:

Ok. Your patch was garbled by your MUA (inserting =3D =20 and breaking lines), but that's ok.

sorry about that, I'll try to find out how that happened.

Quoted 6 lines

After de-garbling it now complains about
undefined compat_uptr_t. For that, I've added

#ifdef CONFIG_COMPAT
#include <linux/compat.h>
#endif

This one does not need to be enclosed in #ifdef CONFIG_COMPAT, the file can always be included. The reason for the #ifdef inside of the switch() statement is that you otherwise get warnings about duplicate case statements on 32 bit kernels.

	Arnd <><
#11 lore raw
In reply to Michael Tokarev
Quoted 11 lines · Michael Tokarev

Is there any reason why 32-bit uswsusp &Friends does not work on 64bits kernel?

For one, 32bits s2disk produces the following when trying to suspend:

ioctl32(s2disk:4134): Unknown cmd fd(4) cmd(400c330d){t:'3';sz:12} arg(ff853554) on /dev/snapshot ioctl32(s2disk:4134): Unknown cmd fd(4) cmd(4004330a){t:'3';sz:4} arg(00000805) on /dev/snapshot

this is coming from:

Quoted 8 lines
     error = ioctl(dev, SNAPSHOT_SET_SWAP_AREA, &swap);
     if (error && !offset)
             error = ioctl(dev, SNAPSHOT_SET_SWAP_FILE, blkdev);

but I guess (just guess!) that other SNAPSHOT_* operations will also fail the same way.

Is there a reason why those are not in compat_ioctl?

The user land hasn't been designed with that in mind. You'd most likely have to modify it too.

Thanks,
Rafael

#12 lore raw
In reply to Michael Tokarev
Quoted 11 lines · Michael Tokarev

Is there any reason why 32-bit uswsusp &Friends does not work on 64bits kernel?

For one, 32bits s2disk produces the following when trying to suspend:

ioctl32(s2disk:4134): Unknown cmd fd(4) cmd(400c330d){t:'3';sz:12} arg(ff853554) on /dev/snapshot ioctl32(s2disk:4134): Unknown cmd fd(4) cmd(4004330a){t:'3';sz:4} arg(00000805) on /dev/snapshot

this is coming from:

Quoted 8 lines
    error = ioctl(dev, SNAPSHOT_SET_SWAP_AREA, &swap);
    if (error && !offset)
            error = ioctl(dev, SNAPSHOT_SET_SWAP_FILE, blkdev);

but I guess (just guess!) that other SNAPSHOT_* operations will also fail the same way.

Is there a reason why those are not in compat_ioctl?

Lazyness...?

Patch would be welcome. On 4GB machine, running 64bit kernel (but staying with 32bit userland) makes sense...

Pavel

Signature · 2 lines
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html
#13 lore raw
Quoted 16 lines · Pavel Machek
Quoted 15 lines · Michael Tokarev

Is there any reason why 32-bit uswsusp &Friends does not work on 64bits kernel?

For one, 32bits s2disk produces the following when trying to suspend:

ioctl32(s2disk:4134): Unknown cmd fd(4) cmd(400c330d){t:'3';sz:12} arg(ff853554) on /dev/snapshot ioctl32(s2disk:4134): Unknown cmd fd(4) cmd(4004330a){t:'3';sz:4} arg(00000805) on /dev/snapshot

this is coming from:

    error = ioctl(dev, SNAPSHOT_SET_SWAP_AREA, &swap);
    if (error && !offset)
            error = ioctl(dev, SNAPSHOT_SET_SWAP_FILE, blkdev);
Quoted 8 lines
Quoted 4 lines

but I guess (just guess!) that other SNAPSHOT_* operations will also fail the same way.

Is there a reason why those are not in compat_ioctl?

Lazyness...?

Patch would be welcome.

Pavel, you really should take a look at the original thread.

As I mentioned in my first email, I'm not the right person to do the patch. But regardless, I spent about a day understanding this stuff (or trying to, anyway) - 100% useless day - and when I thought I have a patch someone else spoken up and said this way (compat_ioctl) is the wrong approach now. And sent another, also trivial patch, adding compat calls right to the proper place in kernel/power.c. Which (the patch) has been ignored too.

So "welcome" is a wrong word here, and I'm sorry about that.

Quoted 2 lines

On 4GB machine, running 64bit kernel (but
staying with 32bit userland) makes sense...

This is exactly what I'm running here by the way. But regardless, uswsusp should be fixed too before it will be useful for that. And fixing uswsusp is not trivial, unlike kernel side. But having in mind amount of useless jumping needed just for the trivial kernel part, for a 20-minute patch, -- there's not much hope really. (I wanted to fix it too, initially, -- not any more).

/mjt

#14 lore raw
Quoted 8 lines · Michael Tokarev

As I mentioned in my first email, I'm not the right person to do the patch. But regardless, I spent about a day understanding this stuff (or trying to, anyway) - 100% useless day - and when I thought I have a patch someone else spoken up and said this way (compat_ioctl) is the wrong approach now. And sent another, also trivial patch, adding compat calls right to the proper place in kernel/power.c. Which (the patch) has been ignored too.

I never got a reply from you saying whether or not the patch I sent actually worked. If you or someone else can confirm it, I'll resend with the fixes you mentioned.

	Arnd <><
#15 lore raw
Quoted 10 lines · Arnd Bergmann
Quoted 8 lines · Michael Tokarev

As I mentioned in my first email, I'm not the right person to do the patch. But regardless, I spent about a day understanding this stuff (or trying to, anyway) - 100% useless day - and when I thought I have a patch someone else spoken up and said this way (compat_ioctl) is the wrong approach now. And sent another, also trivial patch, adding compat calls right to the proper place in kernel/power.c. Which (the patch) has been ignored too.

Quoted 3 lines

I never got a reply from you saying whether or not the patch I sent actually worked. If you or someone else can confirm it, I'll resend with the fixes you mentioned.

In order to (try to) check if it works or not, another userspace component has to be fixed to support 32/64 bit mode. It's uswsusp, which currently assumes swap space structures are all 32bits. So it isn't possible to immediately check if it works or not -- just ioctl(s) aren't enough. Complete fix (kernel+user space) requires both, fixing all remaining (yet unknown) issues in old and new code on the way.

For now, I think it's best to let Pavel or Rafael to decide what to do with all this.

Thanks!

/mjt

#16 lore raw
Quoted 15 lines · Michael Tokarev
Quoted 10 lines · Arnd Bergmann
Quoted 8 lines · Michael Tokarev

As I mentioned in my first email, I'm not the right person to do the patch. But regardless, I spent about a day understanding this stuff (or trying to, anyway) - 100% useless day - and when I thought I have a patch someone else spoken up and said this way (compat_ioctl) is the wrong approach now. And sent another, also trivial patch, adding compat calls right to the proper place in kernel/power.c. Which (the patch) has been ignored too.

Quoted 3 lines

I never got a reply from you saying whether or not the patch I sent actually worked. If you or someone else can confirm it, I'll resend with the fixes you mentioned.

Quoted 7 lines

In order to (try to) check if it works or not, another userspace component has to be fixed to support 32/64 bit mode. It's uswsusp, which currently assumes swap space structures are all 32bits. So it isn't possible to immediately check if it works or not -- just ioctl(s) aren't enough. Complete fix (kernel+user space) requires both, fixing all remaining (yet unknown) issues in old and new code on the way.

Well, there seems to be single structure in s2disk; that does not seem that hard to fix.

Quoted 2 lines

For now, I think it's best to let Pavel or Rafael to decide what to do with all this.

I don't currently have easy access to 64bit machine, so I guess it is up to someone else.

Signature · 2 lines
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html
#17 lore raw

Pavel Machek wrote:
[]

Quoted 10 lines
Quoted 7 lines

In order to (try to) check if it works or not, another userspace component has to be fixed to support 32/64 bit mode. It's uswsusp, which currently assumes swap space structures are all 32bits. So it isn't possible to immediately check if it works or not -- just ioctl(s) aren't enough. Complete fix (kernel+user space) requires both, fixing all remaining (yet unknown) issues in old and new code on the way.

Well, there seems to be single structure in s2disk; that does not seem that hard to fix.

Which one do you mean?

Also, do you want to make it 32/64 bit clean in userspace too, so that an image produced by 32-bit s2disk can be read by 64bit resume and vise versa? Sure it's good thing to have, to avoid possible issues with 32bit initramfs on 64bit system for example...

Quoted 5 lines
Quoted 2 lines

For now, I think it's best to let Pavel or Rafael to decide what to do with all this.

I don't currently have easy access to 64bit machine, so I guess it is up to someone else.

[]

Ok. I applied Arnd's patch again (with two fixes -- adding #include <linux/compat.h> and s/compat_ulong/compat_ulong_t/ and tried suspend/resume cycle with unmodified uswsusp-0.8. Suspend worked (seemengly - it reported about 4G pages written, on a machine with 4G memory) but resume failed after reading all the pages and submitting them to /dev/snapshot.

I've no time right now to debug it further (and again, yet again:
I don't know the code, neither kernel nor userspace part).

So I'd merge this change for now and deal with possible bugs later.

At least there's no failed ioctl()s on /dev/snapshot anymore, neither at suspend nor at resume time.

/mjt

#18 lore raw
Quoted 10 lines · Michael Tokarev

Pavel Machek wrote:
[]

Quoted 8 lines
Quoted 7 lines

In order to (try to) check if it works or not, another userspace component has to be fixed to support 32/64 bit mode. It's uswsusp, which currently assumes swap space structures are all 32bits. So it isn't possible to immediately check if it works or not -- just ioctl(s) aren't enough. Complete fix (kernel+user space) requires both, fixing all remaining (yet unknown) issues in old and new code on the way.

Quoted 4 lines
Quoted 2 lines

Well, there seems to be single structure in s2disk; that does not seem that hard to fix.

Which one do you mean?

struct swsusp_info should be the one...

Quoted 4 lines

Also, do you want to make it 32/64 bit clean in userspace too, so that an image produced by 32-bit s2disk can be read by 64bit resume and vise versa? Sure it's good thing to have, to avoid possible issues with 32bit initramfs on 64bit system for example...

It would be cool, but...

Quoted 14 lines
Quoted 5 lines
Quoted 2 lines

For now, I think it's best to let Pavel or Rafael to decide what to do with all this.

I don't currently have easy access to 64bit machine, so I guess it is up to someone else.

[]

Ok. I applied Arnd's patch again (with two fixes -- adding #include <linux/compat.h> and s/compat_ulong/compat_ulong_t/ and tried suspend/resume cycle with unmodified uswsusp-0.8. Suspend worked (seemengly - it reported about 4G pages written, on a machine with 4G memory) but resume failed after reading all the pages and submitting them to /dev/snapshot.

Quoted 7 lines

I've no time right now to debug it further (and again, yet again:
I don't know the code, neither kernel nor userspace part).

So I'd merge this change for now and deal with possible bugs later.

At least there's no failed ioctl()s on /dev/snapshot anymore, neither at suspend nor at resume time.

ACK.
Pavel

Signature · 2 lines
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html