[PATCH] kvm tools: Fix type mismatches on GCC 4.4 on 32-bit systems

[PATCH] kvm tools: Fix type mismatches on GCC 4.4 on 32-bit systems

[Ingo Molnar]

FYI, the tools/kvm build still fails on 32-bit:

 cc1: warnings being treated as errors
 qcow.c: In function ‘qcow1_write_sector’:
 qcow.c:307: error: comparison between signed and unsigned integer expressions
 make: *** [qcow.o] Error 1
 make: *** Waiting for unfinished jobs....

using:

 gcc version 4.4.4 20100630 (Red Hat 4.4.4-10) (GCC)

The patch below addresses them but i haven't tested it beyond checking that it 
builds.

The double cast of userspace_addr is doubly sad - it highlights our 32-bitness 
problems which are visible in the guest_pfn_to_host() function as well.

Thanks,

	Ingo

---

Signed-off-by: Ingo Molnar <mingo@elte.hu>

diff --git a/tools/kvm/kvm.c b/tools/kvm/kvm.c
index c69fcc4..e3f9d02 100644
--- a/tools/kvm/kvm.c
+++ b/tools/kvm/kvm.c
@@ -162,7 +162,7 @@ static void kvm_register_mem_slot(struct kvm *kvm, u32 slot, u64 guest_phys, u64
 		.slot			= slot,
 		.guest_phys_addr	= guest_phys,
 		.memory_size		= size,
-		.userspace_addr		= (u64)userspace_addr,
+		.userspace_addr		= (u64)(long)userspace_addr,
 	};
 
 	ret = ioctl(kvm->vm_fd, KVM_SET_USER_MEMORY_REGION, &mem);
diff --git a/tools/kvm/qcow.c b/tools/kvm/qcow.c
index 8e1b70d..bb2345c 100644
--- a/tools/kvm/qcow.c
+++ b/tools/kvm/qcow.c
@@ -295,7 +295,7 @@ static int qcow1_write_sector(struct disk_image *disk, u64 sector, void *src, u3
 {
 	struct qcow *q = disk->priv;
 	struct qcow_header *header = q->header;
-	ssize_t nr_written;
+	u32 nr_written;
 	char *buf;
 	u64 offset;
 	ssize_t nr;

Re: [PATCH] kvm tools: Fix type mismatches on GCC 4.4 on 32-bit systems

[Pekka Enberg]

On Fri, May 13, 2011 at 11:19 AM, Ingo Molnar <mingo@elte.hu> wrote:
> @@ -162,7 +162,7 @@ static void kvm_register_mem_slot(struct kvm *kvm, u32 slot, u64 guest_phys, u64
>                .slot                   = slot,
>                .guest_phys_addr        = guest_phys,
>                .memory_size            = size,
> -               .userspace_addr         = (u64)userspace_addr,
> +               .userspace_addr         = (u64)(long)userspace_addr,
>        };

Isn't

+               .userspace_addr         = (unsigned long)userspace_addr,

the right thing to do here?

Re: [PATCH] kvm tools: Fix type mismatches on GCC 4.4 on 32-bit systems

[Ingo Molnar]

* Pekka Enberg <penberg@kernel.org> wrote:

> On Fri, May 13, 2011 at 11:19 AM, Ingo Molnar <mingo@elte.hu> wrote:
> > @@ -162,7 +162,7 @@ static void kvm_register_mem_slot(struct kvm *kvm, u32 slot, u64 guest_phys, u64
> >                .slot                   = slot,
> >                .guest_phys_addr        = guest_phys,
> >                .memory_size            = size,
> > -               .userspace_addr         = (u64)userspace_addr,
> > +               .userspace_addr         = (u64)(long)userspace_addr,
> >        };
> 
> Isn't
> 
> +               .userspace_addr         = (unsigned long)userspace_addr,
> 
> the right thing to do here?

Yeah, you are right - and userspace_addr will always be 32-bit on 32-bit hosts 
so this is unrelated to the guest-pfn conversion thing.

More than 1-2 GB of RAM can be supported in the future by mmap()-ing a chunk, 
passing the address to KVM and then unmapping it. In theory it is possible to 
implement more than 4GB RAM support on 32-bit hosts without having to do 
highmem alike tricks in tools/kvm/, but i doubt there's much interest in that - 
everything is so much easier on 64-bit systems ...

Thanks,

	Ingo

Re: [PATCH] kvm tools: Fix type mismatches on GCC 4.4 on 32-bit systems

[Sasha Levin]

On Fri, 2011-05-13 at 10:19 +0200, Ingo Molnar wrote:
> FYI, the tools/kvm build still fails on 32-bit:
> 
>  cc1: warnings being treated as errors
>  qcow.c: In function ‘qcow1_write_sector’:
>  qcow.c:307: error: comparison between signed and unsigned integer expressions
>  make: *** [qcow.o] Error 1
>  make: *** Waiting for unfinished jobs....
> 
> using:
> 
>  gcc version 4.4.4 20100630 (Red Hat 4.4.4-10) (GCC)
> 
> The patch below addresses them but i haven't tested it beyond checking that it 
> builds.
> 
> The double cast of userspace_addr is doubly sad - it highlights our 32-bitness 
> problems which are visible in the guest_pfn_to_host() function as well.

KVM API uses 64-bit addresses no matter the host bitness, so we can't
really get around doing these sort of casts.

-- 

Sasha.

Re: [PATCH] kvm tools: Fix type mismatches on GCC 4.4 on 32-bit systems

[Ingo Molnar]

* Sasha Levin <levinsasha928@gmail.com> wrote:

> On Fri, 2011-05-13 at 10:19 +0200, Ingo Molnar wrote:
> > FYI, the tools/kvm build still fails on 32-bit:
> > 
> >  cc1: warnings being treated as errors
> >  qcow.c: In function ‘qcow1_write_sector’:
> >  qcow.c:307: error: comparison between signed and unsigned integer expressions
> >  make: *** [qcow.o] Error 1
> >  make: *** Waiting for unfinished jobs....
> > 
> > using:
> > 
> >  gcc version 4.4.4 20100630 (Red Hat 4.4.4-10) (GCC)
> > 
> > The patch below addresses them but i haven't tested it beyond checking that it 
> > builds.
> > 
> > The double cast of userspace_addr is doubly sad - it highlights our 32-bitness 
> > problems which are visible in the guest_pfn_to_host() function as well.
> 
> KVM API uses 64-bit addresses no matter the host bitness, so we can't
> really get around doing these sort of casts.

that bit is OK - the KVM ABI has to be for the largest bit width.

Note that this kind of ABI compatibility allows (in theory) to run a 32-bit kvm 
binary on a 64-bit kernel, and still everything would work despite hypervisor 
user-space being 32-bit.

So the cast to (unsigned long) is fine and clean.

Thanks,

	Ingo