Skip to contentSkip to conversation

KVM: Suppress warning in __kvm_gfn_to_hva_cache_init

PATCH In review Oliver Upton 3 participants 9 messages Feb 18, 2020 – Feb 21, 2020 KVM

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 KVM: Suppress warning in __kvm_gfn_to_hva_cache_init

Particularly draconian compilers warn of a possible uninitialized use of the nr_pages_avail variable. Silence this warning by initializing it to zero.

---
 virt/kvm/kvm_main.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
virt/kvm/kvm_main.c +1 -1
index 70f03ce0e5c1..dc8a67ad082d 100644
@@ -2219,7 +2219,7 @@ static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
 	gfn_t start_gfn = gpa >> PAGE_SHIFT;
 	gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
 	gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
-	gfn_t nr_pages_avail;
+	gfn_t nr_pages_avail = 0;
 
 	/* Update ghc->generation before performing any error checks. */
 	ghc->generation = slots->generation;
Signature · 1 line
2.25.0.265.gbab2e86ba0-goog
#1 lore raw
Quoted 3 lines · Oliver Upton

Particularly draconian compilers warn of a possible uninitialized use of the nr_pages_avail variable. Silence this warning by initializing it to zero.

Can you check if the warning still exists with commit 6ad1e29fe0ab ("KVM:
Clean up __kvm_gfn_to_hva_cache_init() and its callers")? I'm guessing (hoping?) the suppression is no longer necessary.

commit 6ad1e29fe0aba843dfffc714fced0ef6a2e19502

Author: Sean Christopherson <[email protected]>
Date:   Thu Jan 9 14:58:55 2020 -0500

KVM: Clean up __kvm_gfn_to_hva_cache_init() and its callers

Barret reported a (technically benign) bug where nr_pages_avail can be accessed without being initialized if gfn_to_hva_many() fails.

virt/kvm/kvm_main.c:2193:13: warning: 'nr_pages_avail' may be used uninitialized in this function [-Wmaybe-uninitialized]

Rather than simply squashing the warning by initializing nr_pages_avail, fix the underlying issues by reworking __kvm_gfn_to_hva_cache_init() to return immediately instead of continuing on. Now that all callers check the result and/or bail immediately on a bad hva, there's no need to explicitly nullify the memslot on error.

Fixes f1b9dd5eb86c ("kvm: Disallow wraparound in kvm_gfn_to_hva_cache_init")
Cc Jim Mattson <[email protected]>
Quoted 21 lines
---
 virt/kvm/kvm_main.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
virt/kvm/kvm_main.c +1 -1
index 70f03ce0e5c1..dc8a67ad082d 100644
@@ -2219,7 +2219,7 @@ static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
 	gfn_t start_gfn = gpa >> PAGE_SHIFT;
 	gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
 	gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
-	gfn_t nr_pages_avail;
+	gfn_t nr_pages_avail = 0;
 
 	/* Update ghc->generation before performing any error checks. */
 	ghc->generation = slots->generation;
Signature · 1 line
2.25.0.265.gbab2e86ba0-goog
#2 lore raw

Hey Sean,

Quoted 8 lines · Sean Christopherson
Quoted 3 lines · Oliver Upton

Particularly draconian compilers warn of a possible uninitialized use of the nr_pages_avail variable. Silence this warning by initializing it to zero.

Can you check if the warning still exists with commit 6ad1e29fe0ab ("KVM:
Clean up __kvm_gfn_to_hva_cache_init() and its callers")? I'm guessing (hoping?) the suppression is no longer necessary.

Hmm. I rebased this patch right before sending out + it seems that it is required (at least for me) to silence the compiler warning. For good measure, I ran git branch --contains to ensure I had your change. Looks like my topic branch did in fact have your fix.

--
Oliver

Quoted 46 lines

commit 6ad1e29fe0aba843dfffc714fced0ef6a2e19502

Author: Sean Christopherson <[email protected]>
Date:   Thu Jan 9 14:58:55 2020 -0500

KVM: Clean up __kvm_gfn_to_hva_cache_init() and its callers

Barret reported a (technically benign) bug where nr_pages_avail can be accessed without being initialized if gfn_to_hva_many() fails.

virt/kvm/kvm_main.c:2193:13: warning: 'nr_pages_avail' may be used uninitialized in this function [-Wmaybe-uninitialized]

Rather than simply squashing the warning by initializing nr_pages_avail, fix the underlying issues by reworking __kvm_gfn_to_hva_cache_init() to return immediately instead of continuing on. Now that all callers check the result and/or bail immediately on a bad hva, there's no need to explicitly nullify the memslot on error.

Fixes f1b9dd5eb86c ("kvm: Disallow wraparound in kvm_gfn_to_hva_cache_init")
Cc Jim Mattson <[email protected]>
Quoted 21 lines
---
 virt/kvm/kvm_main.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
virt/kvm/kvm_main.c +1 -1
index 70f03ce0e5c1..dc8a67ad082d 100644
@@ -2219,7 +2219,7 @@ static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
 	gfn_t start_gfn = gpa >> PAGE_SHIFT;
 	gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
 	gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
-	gfn_t nr_pages_avail;
+	gfn_t nr_pages_avail = 0;
 
 	/* Update ghc->generation before performing any error checks. */
 	ghc->generation = slots->generation;
Signature · 1 line
2.25.0.265.gbab2e86ba0-goog
#3 lore raw
Quoted 8 lines · Oliver Upton

Hey Sean,

Quoted 5 lines · Sean Christopherson
Quoted 3 lines · Oliver Upton

Particularly draconian compilers warn of a possible uninitialized use of the nr_pages_avail variable. Silence this warning by initializing it to zero.

Quoted 8 lines
Quoted 3 lines

Can you check if the warning still exists with commit 6ad1e29fe0ab ("KVM:
Clean up __kvm_gfn_to_hva_cache_init() and its callers")? I'm guessing (hoping?) the suppression is no longer necessary.

Hmm. I rebased this patch right before sending out + it seems that it is required (at least for me) to silence the compiler warning. For good measure, I ran git branch --contains to ensure I had your change. Looks like my topic branch did in fact have your fix.

Bummer. By "particularly draconian compilers", do you mean "clang"? Not that it really matters, but it'd explain why no one else is complaining.

Anyways, comment on the code below...

Quoted 33 lines

--
Oliver

Quoted 30 lines

commit 6ad1e29fe0aba843dfffc714fced0ef6a2e19502

Author: Sean Christopherson <[email protected]>
Date:   Thu Jan 9 14:58:55 2020 -0500

KVM: Clean up __kvm_gfn_to_hva_cache_init() and its callers

Barret reported a (technically benign) bug where nr_pages_avail can be accessed without being initialized if gfn_to_hva_many() fails.

virt/kvm/kvm_main.c:2193:13: warning: 'nr_pages_avail' may be used uninitialized in this function [-Wmaybe-uninitialized]

Rather than simply squashing the warning by initializing nr_pages_avail, fix the underlying issues by reworking __kvm_gfn_to_hva_cache_init() to return immediately instead of continuing on. Now that all callers check the result and/or bail immediately on a bad hva, there's no need to explicitly nullify the memslot on error.

Fixes f1b9dd5eb86c ("kvm: Disallow wraparound in kvm_gfn_to_hva_cache_init")
Cc Jim Mattson <[email protected]>
Quoted 5 lines
---
 virt/kvm/kvm_main.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
Quoted 10 lines
Quoted 10 lines

virt/kvm/kvm_main.c static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots, · line 2222

Quoted 10 lines
virt/kvm/kvm_main.c +1 -1
index 70f03ce0e5c1..dc8a67ad082d 100644
@@ -2219,7 +2219,7 @@ static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
 	gfn_t start_gfn = gpa >> PAGE_SHIFT;
 	gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
 	gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
-	gfn_t nr_pages_avail;
+	gfn_t nr_pages_avail = 0;

Since the warning is technically bogus, uninitialized_var() can be used, which will hopefully make it more obvious (well, less misleading), that the loop's post-condition isn't broken.

	gfn uninitialized_var(nr_pages_avail);
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
virt/kvm/kvm_main.c +0 -0
 	/* Update ghc->generation before performing any error checks. */
 	ghc->generation = slots->generation;
Signature · 1 line
2.25.0.265.gbab2e86ba0-goog
maintainer
#4 lore raw
In reply to Sean Christopherson
Quoted 7 lines · Sean Christopherson
Quoted 3 lines · Oliver Upton

Particularly draconian compilers warn of a possible uninitialized use of the nr_pages_avail variable. Silence this warning by initializing it to zero.

Can you check if the warning still exists with commit 6ad1e29fe0ab ("KVM:
Clean up __kvm_gfn_to_hva_cache_init() and its callers")? I'm guessing (hoping?) the suppression is no longer necessary.

What if __gfn_to_hva_many and gfn_to_hva_many are marked __always_inline?

Thanks,

Paolo

#5 lore raw
Quoted 10 lines · Paolo Bonzini
Quoted 7 lines · Sean Christopherson
Quoted 3 lines · Oliver Upton

Particularly draconian compilers warn of a possible uninitialized use of the nr_pages_avail variable. Silence this warning by initializing it to zero.

Can you check if the warning still exists with commit 6ad1e29fe0ab ("KVM:
Clean up __kvm_gfn_to_hva_cache_init() and its callers")? I'm guessing (hoping?) the suppression is no longer necessary.

Quoted 6 lines

What if __gfn_to_hva_many and gfn_to_hva_many are marked __always_inline?

Thanks,

Paolo

Even with this suggestion the compiler is ill-convinced :/

in re to Sean: what do I mean by "draconian compiler"

Well, the public answer is that both Barret and I use the same compiler. Nothing particularly interesting about it, but idk what our toolchain folks' stance is on divulging details.

I'll instead use Sean's suggested fix (which reads much better) and resend.

--
Thanks,
Oliver

maintainer
#6 lore raw
Quoted 20 lines · Oliver Upton
Quoted 10 lines · Paolo Bonzini
Quoted 7 lines · Sean Christopherson
Quoted 3 lines · Oliver Upton

Particularly draconian compilers warn of a possible uninitialized use of the nr_pages_avail variable. Silence this warning by initializing it to zero.

Can you check if the warning still exists with commit 6ad1e29fe0ab ("KVM:
Clean up __kvm_gfn_to_hva_cache_init() and its callers")? I'm guessing (hoping?) the suppression is no longer necessary.

Quoted 6 lines

What if __gfn_to_hva_many and gfn_to_hva_many are marked __always_inline?

Thanks,

Paolo

Even with this suggestion the compiler is ill-convinced :/

Quoted 7 lines

in re to Sean: what do I mean by "draconian compiler"

Well, the public answer is that both Barret and I use the same compiler. Nothing particularly interesting about it, but idk what our toolchain folks' stance is on divulging details.

I'll instead use Sean's suggested fix (which reads much better) and resend.

Can you instead look into fixing that compiler? After inlining, it is trivial to realize that the first two returns imply kvm_is_error_hva(ghc->hva). I'm asking this because even for GCC -Wuninitialized *used to be* total crap, but these days it's quite reliable and even basic data flow should be able to thread through the jumps.

I'm more than willing to help the compiler with __always_inline hints, but an "uninitialized_var()" adds load on the code and I'm not sure it makes sense to do that for the sake of some proprietary, or at least unnamed, software.

Paolo

#7 lore raw
Quoted 28 lines · Paolo Bonzini
Quoted 26 lines · Oliver Upton
Quoted 10 lines · Paolo Bonzini
Quoted 7 lines · Sean Christopherson
Quoted 3 lines · Oliver Upton

Particularly draconian compilers warn of a possible uninitialized use of the nr_pages_avail variable. Silence this warning by initializing it to zero.

Can you check if the warning still exists with commit 6ad1e29fe0ab ("KVM:
Clean up __kvm_gfn_to_hva_cache_init() and its callers")? I'm guessing (hoping?) the suppression is no longer necessary.

Quoted 6 lines

What if __gfn_to_hva_many and gfn_to_hva_many are marked __always_inline?

Thanks,

Paolo

Even with this suggestion the compiler is ill-convinced :/

in re to Sean: what do I mean by "draconian compiler"

Well, the public answer is that both Barret and I use the same compiler. Nothing particularly interesting about it, but idk what our toolchain folks' stance is on divulging details.

Quoted 8 lines
Quoted 1 line

I'll instead use Sean's suggested fix (which reads much better) and resend.

Can you instead look into fixing that compiler? After inlining, it is trivial to realize that the first two returns imply kvm_is_error_hva(ghc->hva). I'm asking this because even for GCC -Wuninitialized *used to be* total crap, but these days it's quite reliable and even basic data flow should be able to thread through the jumps.

Yeah, at this point I completely agree. I'll move onto a mainline toolchain going forward. This was just a glaring change when I rebased some work on top of the -Werror change (as it absolutely should).

Quoted 5 lines

I'm more than willing to help the compiler with __always_inline hints, but an "uninitialized_var()" adds load on the code and I'm not sure it makes sense to do that for the sake of some proprietary, or at least unnamed, software.

Absolutely. I thought it sensible to send out the fix in case of other toolchains out in the wild. But if nobody else other than us has complained it's quite obvious where the problem lies.

Thanks!

maintainer
#8 lore raw
Quoted 3 lines · Oliver Upton

Absolutely. I thought it sensible to send out the fix in case of other toolchains out in the wild. But if nobody else other than us has complained it's quite obvious where the problem lies.

Here is another plausible (and untested) way to fix it, in case it's the alias analysis that is throwing off the compiler (plus possibly __always_inline).

virt/kvm/kvm_main.c +14 -9
index 027259af883e..63c7dcd7c57f 100644
@@ -2218,6 +2218,8 @@ static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
 	gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
 	gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
 	gfn_t nr_pages_avail;
+	unsigned long hva;
+	struct kvm_memslot *memslot;
 
 	/* Update ghc->generation before performing any error checks. */
 	ghc->generation = slots->generation;
@@ -2231,19 +2233,22 @@ static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
 	 * If the requested region crosses two memslots, we still
 	 * verify that the entire region is valid here.
 	 */
-	for ( ; start_gfn <= end_gfn; start_gfn += nr_pages_avail) {
-		ghc->memslot = __gfn_to_memslot(slots, start_gfn);
-		ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
-					   &nr_pages_avail);
-		if (kvm_is_error_hva(ghc->hva))
+	do {
+		memslot = __gfn_to_memslot(slots, start_gfn);
+		hva = gfn_to_hva_many(memslot, start_gfn, &nr_pages_avail);
+		if (kvm_is_error_hva(hva))
 			return -EFAULT;
-	}
+		start_gfn += nr_pages_avail;
+	} while (start_gfn <= end_gfn);
 
 	/* Use the slow path for cross page reads and writes. */
-	if (nr_pages_needed == 1)
-		ghc->hva += offset;
-	else
+	if (nr_pages_needed == 1) {
+		ghc->hva = hva + offset;
+		ghc->memslot = memslot;
+	} else {
+		ghc->hva = 0;
 		ghc->memslot = NULL;
+	}
 
 	ghc->gpa = gpa;
 	ghc->len = len;

Paolo