Skip to contentSkip to conversation

nested vmx: fix I/O port bitmap indexing arithmetic

PATCH In review Matthew Daley 4 participants 4 messages Nov 29, 2013 – Dec 2, 2013

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 nested vmx: fix I/O port bitmap indexing arithmetic

The I/O port bitmap holds 8 ports per element, and hence the port number used when indexing into it should be shifted right by 3 bits, not 4.

---
Jan: I'm not sure if this also needs a fixup similar to what you did in SVM code with commit b1e87805bf.

 xen/arch/x86/hvm/vmx/vvmx.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
xen/arch/x86/hvm/vmx/vvmx.c +1 -1
index 248e975..7fa110e 100644
@@ -2225,7 +2225,7 @@ int nvmx_n2_vmexit_handler(struct cpu_user_regs *regs,
             __vmread(EXIT_QUALIFICATION, &qual);
             port = qual >> 16;
             bitmap = nvmx->iobitmap[port >> 15];
-            if ( bitmap[(port & 0x7fff) >> 4] & (1 << (port & 0x7)) )
+            if ( bitmap[(port & 0x7fff) >> 3] & (1 << (port & 0x7)) )
                 nvcpu->nv_vmexit_pending = 1;
             if ( !nvcpu->nv_vmexit_pending )
                gdprintk(XENLOG_WARNING, "L0 PIO %x.\n", port);
Signature · 1 line
1.7.10.4
Reviewed-by on v1 1/1
#1 lore raw

Matthew Daley wrote on 2013-11-30:

Quoted 23 lines

The I/O port bitmap holds 8 ports per element, and hence the port number used when indexing into it should be shifted right by 3 bits, not 4.

---
Jan: I'm not sure if this also needs a fixup similar to what you did in SVM code with commit b1e87805bf.

 xen/arch/x86/hvm/vmx/vvmx.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
xen/arch/x86/hvm/vmx/vvmx.c +1 -1
index 248e975..7fa110e 100644
@@ -2225,7 +2225,7 @@ int nvmx_n2_vmexit_handler(struct cpu_user_regs
 *regs,
             __vmread(EXIT_QUALIFICATION, &qual);
             port = qual >> 16;
             bitmap = nvmx->iobitmap[port >> 15];
-            if ( bitmap[(port & 0x7fff) >> 4] & (1 << (port & 0x7)) )
+            if ( bitmap[(port & 0x7fff) >> 3] & (1 << (port & 0x7)) )
                 nvcpu->nv_vmexit_pending = 1;
             if ( !nvcpu->nv_vmexit_pending )
Quoted 1 line
                gdprintk(XENLOG_WARNING, "L0 PIO %x.\n", port);

Best regards,
Yang

#2 lore raw
In reply to Matthew Daley

Acked-by Eddie Dong <[email protected]>

-----Original Message-----
From: Matthew Daley [mailto:[email protected]]
Sent: Saturday, November 30, 2013 6:39 AM
To: [email protected]
Cc: Matthew Daley; Nakajima, Jun; Dong, Eddie; Keir Fraser; Jan Beulich
Subject: [PATCH] nested vmx: fix I/O port bitmap indexing arithmetic

The I/O port bitmap holds 8 ports per element, and hence the port number used when indexing into it should be shifted right by 3 bits, not 4.

---
Jan: I'm not sure if this also needs a fixup similar to what you did in SVM code with commit b1e87805bf.

 xen/arch/x86/hvm/vmx/vvmx.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
xen/arch/x86/hvm/vmx/vvmx.c +1 -1
@@ -2225,7 +2225,7 @@ int nvmx_n2_vmexit_handler(struct cpu_user_regs *regs,
             __vmread(EXIT_QUALIFICATION, &qual);
             port = qual >> 16;
             bitmap = nvmx->iobitmap[port >> 15];
-            if ( bitmap[(port & 0x7fff) >> 4] & (1 << (port & 0x7)) )
+            if ( bitmap[(port & 0x7fff) >> 3] & (1 << (port & 0x7)) )
                 nvcpu->nv_vmexit_pending = 1;
             if ( !nvcpu->nv_vmexit_pending )
                gdprintk(XENLOG_WARNING, "L0 PIO %x.\n", port);

--
1.7.10.4

#3 lore raw
In reply to Matthew Daley
Quoted 8 lines
Quoted 1 line
Quoted 1 line · Matthew Daley

The I/O port bitmap holds 8 ports per element, and hence the port number used when indexing into it should be shifted right by 3 bits, not 4.

---
Jan: I'm not sure if this also needs a fixup similar to what you did in SVM code with commit b1e87805bf.

Yes, definitely. The wrapping behavior is different for VMX though. In fact, I'd prefer the whole fix to be done in one go (unless you tell me that you're not going to find time to do so any time soon).

Jan