Skip to contentSkip to conversation

xen: handle paged gfn in wrmsr_hypervisor_regs

PATCH Superseded Olaf Hering 3 participants 27 messages May 2, 2013 – May 8, 2013

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 xen: handle paged gfn in wrmsr_hypervisor_regs

# HG changeset patch
# User Olaf Hering <[email protected]>
# Date 1367511750 -7200
# Node ID 19777301f89653c640de73c6319c9d386a28327f # Parent 9df019eef776d129c2abb130d1458914fe1ecac4
xen: handle paged gfn in wrmsr_hypervisor_regs

If xenpaging is started very early for a guest the gfn for the hypercall page may be paged-out already. This leads to a guest crash:

...
(XEN) HVM10: Allocated Xen hypercall page at 169ff000
(XEN) traps.c:654:d10 Bad GMFN 169ff (MFN 3e900000000) to MSR 40000000 (XEN) HVM10: Detected Xen v4.3
(XEN) io.c:201:d10 MMIO emulation failed @ 0008:c2c2c2c2: 18 7c 55 6d 03 83 ff ff 10 7c (XEN) hvm.c:1253:d10 Triple fault on VCPU0 - invoking HVM shutdown action 1. (XEN) HVM11: HVM Loader
...

Handle paged gfns in wrmsr_hypervisor_regs and update callers to deal with the new return code -EAGAIN.

Also update the gdprintk to handle a page value of NULL to avoid printing a bogus MFN value.

xen/arch/x86/hvm/svm/svm.c +4 -3
diff -r 9df019eef776 -r 19777301f896 xen/arch/x86/hvm/svm/svm.c
@@ -1569,7 +1569,7 @@ static int svm_msr_read_intercept(unsign
 
 static int svm_msr_write_intercept(unsigned int msr, uint64_t msr_content)
 {
-    int ret;
+    int ret, retry = 0;
     struct vcpu *v = current;
     struct vmcb_struct *vmcb = v->arch.hvm_svm.vmcb;
     int sync = 0;
@@ -1682,14 +1682,15 @@ static int svm_msr_write_intercept(unsig
         if ( wrmsr_viridian_regs(msr, msr_content) )
             break;
 
-        wrmsr_hypervisor_regs(msr, msr_content);
+        ret = wrmsr_hypervisor_regs(msr, msr_content);
+        retry = ret == -EAGAIN;
         break;
     }
 
     if ( sync )
         svm_vmload(vmcb);
 
-    return X86EMUL_OKAY;
+    return retry ? X86EMUL_RETRY : X86EMUL_OKAY;
 
  gpf:
     hvm_inject_hw_exception(TRAP_gp_fault, 0);
xen/arch/x86/hvm/vmx/vmx.c +4 -1
diff -r 9df019eef776 -r 19777301f896 xen/arch/x86/hvm/vmx/vmx.c
@@ -2020,6 +2020,7 @@ void vmx_vlapic_msr_changed(struct vcpu 
 
 static int vmx_msr_write_intercept(unsigned int msr, uint64_t msr_content)
 {
+    int ret = 0;
     struct vcpu *v = current;
 
     HVM_DBG_LOG(DBG_LEVEL_1, "ecx=%#x, msr_value=%#"PRIx64, msr, msr_content);
@@ -2088,7 +2089,9 @@ static int vmx_msr_write_intercept(unsig
             case HNDL_unhandled:
                 if ( (vmx_write_guest_msr(msr, msr_content) != 0) &&
                      !is_last_branch_msr(msr) )
-                    wrmsr_hypervisor_regs(msr, msr_content);
+                    ret = wrmsr_hypervisor_regs(msr, msr_content);
+                    if ( ret == -EAGAIN )
+                        return X86EMUL_RETRY;
                 break;
             case HNDL_exception_raised:
                 return X86EMUL_EXCEPTION;
xen/arch/x86/traps.c +9 -2
diff -r 9df019eef776 -r 19777301f896 xen/arch/x86/traps.c
@@ -634,6 +634,7 @@ int wrmsr_hypervisor_regs(uint32_t idx, 
         unsigned long gmfn = val >> 12;
         unsigned int idx  = val & 0xfff;
         struct page_info *page;
+        p2m_type_t t;
 
         if ( idx > 0 )
         {
@@ -643,15 +644,21 @@ int wrmsr_hypervisor_regs(uint32_t idx, 
             return 0;
         }
 
-        page = get_page_from_gfn(d, gmfn, NULL, P2M_ALLOC);
+        page = get_page_from_gfn(d, gmfn, &t, P2M_ALLOC);
 
         if ( !page || !get_page_type(page, PGT_writable_page) )
         {
             if ( page )
                 put_page(page);
+            if ( p2m_is_paging(t) )
+            {
+                p2m_mem_paging_populate(d, gmfn);
+                return -EAGAIN;
+            }
+
             gdprintk(XENLOG_WARNING,
                      "Bad GMFN %lx (MFN %lx) to MSR %08x\n",
-                     gmfn, page_to_mfn(page), base + idx);
+                     gmfn, page ? page_to_mfn(page) : -1UL, base + idx);
             return 0;
         }
#1 lore raw

xen/arch/x86/hvm/svm/svm.c static int svm_msr_write_intercept(unsig · line 1686

Quoted 8 lines
Quoted 1 line
Quoted 1 line · Olaf Hering
xen/arch/x86/hvm/svm/svm.c +2 -1
@@ -1682,14 +1682,15 @@ static int svm_msr_write_intercept(unsig
         if ( wrmsr_viridian_regs(msr, msr_content) )
             break;
 
-        wrmsr_hypervisor_regs(msr, msr_content);
+        ret = wrmsr_hypervisor_regs(msr, msr_content);
+        retry = ret == -EAGAIN;

If you add error handling, don't constrain this to a single error code please. For the case here, the easiest would appear to be a switch converting to X86EMUL_OKAY, X86EMUL_RETRY, or X86EMUL_UNHANDLEABLE. If the function had ways to fail before, it would have been a bug anyway to not check the return value.

xen/arch/x86/hvm/vmx/vmx.c static int vmx_msr_write_intercept(unsig · line 2094

Quoted 8 lines
xen/arch/x86/hvm/vmx/vmx.c +3 -1
@@ -2088,7 +2089,9 @@ static int vmx_msr_write_intercept(unsig
             case HNDL_unhandled:
                 if ( (vmx_write_guest_msr(msr, msr_content) != 0) &&
                      !is_last_branch_msr(msr) )
-                    wrmsr_hypervisor_regs(msr, msr_content);
+                    ret = wrmsr_hypervisor_regs(msr, msr_content);
+                    if ( ret == -EAGAIN )
+                        return X86EMUL_RETRY;

Similarly here, obviously.

Jan

#2 lore raw
Quoted 5 lines · Jan Beulich
Quoted 5 lines
Quoted 1 line
Quoted 1 line · Olaf Hering
xen/arch/x86/hvm/svm/svm.c +0 -0
@@ -1682,14 +1682,15 @@ static int svm_msr_write_intercept(unsig
         if ( wrmsr_viridian_regs(msr, msr_content) )
             break;
 
Quoted 9 lines

xen/arch/x86/hvm/svm/svm.c static int svm_msr_write_intercept(unsig · line 1686

Quoted 3 lines
xen/arch/x86/hvm/svm/svm.c +2 -1
-        wrmsr_hypervisor_regs(msr, msr_content);
+        ret = wrmsr_hypervisor_regs(msr, msr_content);
+        retry = ret == -EAGAIN;

If you add error handling, don't constrain this to a single error code please. For the case here, the easiest would appear to be a switch converting to X86EMUL_OKAY, X86EMUL_RETRY, or X86EMUL_UNHANDLEABLE. If the function had ways to fail before, it would have been a bug anyway to not check the return value.

I just sent v2 of this patch.

Olaf

#3 lore raw
Quoted 17 lines · Olaf Hering
Quoted 5 lines · Jan Beulich
Quoted 5 lines
Quoted 1 line
Quoted 1 line · Olaf Hering
xen/arch/x86/hvm/svm/svm.c +0 -0
@@ -1682,14 +1682,15 @@ static int svm_msr_write_intercept(unsig
         if ( wrmsr_viridian_regs(msr, msr_content) )
             break;
 
Quoted 9 lines

xen/arch/x86/hvm/svm/svm.c static int svm_msr_write_intercept(unsig · line 1686

Quoted 3 lines
xen/arch/x86/hvm/svm/svm.c +2 -1
-        wrmsr_hypervisor_regs(msr, msr_content);
+        ret = wrmsr_hypervisor_regs(msr, msr_content);
+        retry = ret == -EAGAIN;

If you add error handling, don't constrain this to a single error code please. For the case here, the easiest would appear to be a switch converting to X86EMUL_OKAY, X86EMUL_RETRY, or X86EMUL_UNHANDLEABLE. If the function had ways to fail before, it would have been a bug anyway to not check the return value.

Quoted 1 line

I just sent v2 of this patch.

My change v2 causes a boot failure, the return value 0 is not handled correctly.
Did you really mean to translate ret == 0 to X86EMUL_UNHANDLEABLE?

Olaf

+++ b/xen/arch/x86/hvm/vmx/vmx.c
@@ -2020,6 +2020,7 @@ void vmx_vlapic_msr_changed(struct vcpu

static int vmx_msr_write_intercept(unsigned int msr, uint64_t msr_content) {

+    int ret = 0;
     struct vcpu *v = current;

     HVM_DBG_LOG(DBG_LEVEL_1, "ecx=%#x, msr_value=%#"PRIx64, msr, msr_content);
@@ -2088,7 +2089,17 @@ static int vmx_msr_write_intercept(unsig
             case HNDL_unhandled:
                 if ( (vmx_write_guest_msr(msr, msr_content) != 0) &&
                      !is_last_branch_msr(msr) )
-                    wrmsr_hypervisor_regs(msr, msr_content);
+                    ret = wrmsr_hypervisor_regs(msr, msr_content);
+                    printk("%s(%u) msr %x mc %lx r %d\n", __func__, __LINE__, msr, msr_content, ret);
+                    switch ( ret )
+                    {

+ case -EAGAIN:
+ return X86EMUL_RETRY;
+ case 0:
+ return X86EMUL_UNHANDLEABLE;
+ default:
+ break;
+ }

                 break;
             case HNDL_exception_raised:
                 return X86EMUL_EXCEPTION;

satriani login: (XEN) HVM2: HVM Loader
(XEN) vmx_msr_write_intercept(2093) msr 40000000 mc 80000 r 1 (XEN) HVM2: Detected Xen v4.3.26961-20130503
(XEN) HVM2: Xenbus rings @0xfeffc000, event channel 6 (XEN) HVM2: System requested SeaBIOS
(XEN) HVM2: CPU speed is 2926 MHz
(XEN) irq.c:270: Dom2 PCI link 0 changed 0 -> 5 (XEN) HVM2: PCI-ISA link 0 routed to IRQ5
(XEN) irq.c:270: Dom2 PCI link 1 changed 0 -> 10 (XEN) HVM2: PCI-ISA link 1 routed to IRQ10
(XEN) irq.c:270: Dom2 PCI link 2 changed 0 -> 11 (XEN) HVM2: PCI-ISA link 2 routed to IRQ11
(XEN) irq.c:270: Dom2 PCI link 3 changed 0 -> 5 (XEN) HVM2: PCI-ISA link 3 routed to IRQ5
(XEN) HVM2: pci dev 01:2 INTD->IRQ5
(XEN) HVM2: pci dev 01:3 INTA->IRQ10
(XEN) HVM2: pci dev 03:0 INTA->IRQ5
(XEN) HVM2: pci dev 02:0 bar 10 size lx: 02000000 (XEN) HVM2: pci dev 03:0 bar 14 size lx: 01000000 (XEN) HVM2: pci dev 02:0 bar 30 size lx: 00010000 (XEN) HVM2: pci dev 02:0 bar 14 size lx: 00001000 (XEN) HVM2: pci dev 03:0 bar 10 size lx: 00000100 (XEN) HVM2: pci dev 01:2 bar 20 size lx: 00000020 (XEN) HVM2: pci dev 01:1 bar 20 size lx: 00000010 (XEN) HVM2: Multiprocessor initialisation:

(XEN) HVM2:  - CPU0 ... 40-bit phys ... fixed MTRRs ... var MTRRs [2/8] ... done.
(XEN) HVM2:  - CPU1 ... 40-bit phys ... fixed MTRRs ... var MTRRs [2/8] ... done.
(XEN) HVM2:  - CPU2 ... 40-bit phys ... fixed MTRRs ... var MTRRs [2/8] ... done.
(XEN) HVM2:  - CPU3 ... 40-bit phys ... fixed MTRRs ... var MTRRs [2/8] ... done.

(XEN) HVM2: Testing HVM environment:
(XEN) HVM2: - REP INSB across page boundaries ... passed (XEN) HVM2: - GS base MSRs and SWAPGS ... passed (XEN) HVM2: Passed 2 of 2 tests
(XEN) HVM2: Writing SMBIOS tables ...
(XEN) HVM2: Loading SeaBIOS ...
(XEN) HVM2: Creating MP tables ...
(XEN) HVM2: Loading ACPI ...
(XEN) HVM2: vm86 TSS at fc00a100
(XEN) HVM2: BIOS map:
(XEN) HVM2: 10000-100d3: Scratch space
(XEN) HVM2: e0000-fffff: Main BIOS
(XEN) HVM2: E820 table:

(XEN) HVM2:  [00]: 00000000:00000000 - 00000000:000a0000: RAM
(XEN) HVM2:  HOLE: 00000000:000a0000 - 00000000:000e0000
(XEN) HVM2:  [01]: 00000000:000e0000 - 00000000:00100000: RESERVED
(XEN) HVM2:  [02]: 00000000:00100000 - 00000000:16a00000: RAM
(XEN) HVM2:  HOLE: 00000000:16a00000 - 00000000:fc000000
(XEN) HVM2:  [03]: 00000000:fc000000 - 00000001:00000000: RESERVED

(XEN) HVM2: Invoking SeaBIOS ...
(XEN) HVM2: SeaBIOS (version ?-20130503_132523-bax)
(XEN) HVM2:
(XEN) HVM2: Found Xen hypervisor signature at 40000000
(XEN) HVM2: xen: copy e820...
(XEN) HVM2: Ram Size=0x16a00000 (0x0000000000000000 high)
(XEN) HVM2: Relocating low data from 0x000e2490 to 0x000ef790 (size 2156) (XEN) HVM2: Relocating init from 0x000e2cfc to 0x169e20f0 (size 56804) (XEN) HVM2: CPU Mhz=2926
(XEN) HVM2: Found 7 PCI devices (max PCI bus is 00)
(XEN) HVM2: Allocated Xen hypercall page at 169ff000
(XEN) vmx_msr_write_intercept(2093) msr 40000000 mc 169ff000 r 1
(XEN) HVM2: Detected Xen v4.3.26961-20130503
(XEN) HVM2: Found 4 cpu(s) max supported 4 cpu(s)
(XEN) HVM2: xen: copy BIOS tables...
(XEN) HVM2: Copying SMBIOS entry point from 0x00010010 to 0x000fdb10
(XEN) HVM2: Copying MPTABLE from 0xfc0011c0/fc0011d0 to 0x000fd9f0
(XEN) HVM2: Copying PIR from 0x00010030 to 0x000fd970
(XEN) HVM2: Copying ACPI RSDP from 0x000100b0 to 0x000fd940
(XEN) HVM2: Scan for VGA option rom
(XEN) HVM2: Running option rom at c000:0003
(XEN) stdvga.c:147:d2 entering stdvga and caching modes
(XEN) HVM2: Turning on vga text mode console
(XEN) HVM2: SeaBIOS (version ?-20130503_132523-bax)
(XEN) HVM2:
(XEN) HVM2: UHCI init on dev 00:01.2 (io=c100)
(XEN) HVM2: Found 1 lpt ports
(XEN) HVM2: Found 1 serial ports
(XEN) HVM2: ATA controller 1 at 1f0/3f4/c120 (irq 14 dev 9)
(XEN) HVM2: ATA controller 2 at 170/374/c128 (irq 15 dev 9)
(XEN) HVM2: ata0-0: QEMU HARDDISK ATA-7 Hard-Disk (10240 MiBytes)
(XEN) HVM2: Searching bootorder for: /pci@i0cf8/*@1,1/drive@0/disk@0
(XEN) HVM2: DVD/CD [ata1-0: QEMU DVD-ROM ATAPI-4 DVD/CD]
(XEN) HVM2: Searching bootorder for: /pci@i0cf8/*@1,1/drive@1/disk@0
(XEN) HVM2: PS2 keyboard initialized
(XEN) HVM2: All threads complete.
(XEN) HVM2: Scan for option roms
(XEN) HVM2: Press F12 for boot menu.
(XEN) HVM2:
(XEN) HVM2: drive 0x000fd8f0: PCHS=16383/16/63 translation=lba LCHS=1024/255/63 s=20971520 (XEN) HVM2:
(XEN) HVM2: Space available for UMB: 000c9000-000ee800
(XEN) HVM2: Returned 61440 bytes of ZoneHigh
(XEN) HVM2: e820 map has 6 items:

(XEN) HVM2:   0: 0000000000000000 - 000000000009fc00 = 1 RAM
(XEN) HVM2:   1: 000000000009fc00 - 00000000000a0000 = 2 RESERVED
(XEN) HVM2:   2: 00000000000f0000 - 0000000000100000 = 2 RESERVED
(XEN) HVM2:   3: 0000000000100000 - 00000000169ff000 = 1 RAM
(XEN) HVM2:   4: 00000000169ff000 - 0000000016a00000 = 2 RESERVED
(XEN) HVM2:   5: 00000000fc000000 - 0000000100000000 = 2 RESERVED

(XEN) HVM2: enter handle_19:
(XEN) HVM2: NULL
(XEN) HVM2: Booting from DVD/CD...
(XEN) HVM2: Booting from 0000:7c00
(XEN) stdvga.c:151:d2 leaving stdvga
(XEN) stdvga.c:147:d2 entering stdvga and caching modes (XEN) stdvga.c:151:d2 leaving stdvga
(XEN) vmx_msr_write_intercept(2093) msr 8b mc 0 r 0 (XEN) vmx_msr_write_intercept(2093) msr 8b mc 0 r 0 (XEN) vmx_msr_write_intercept(2093) msr 8b mc 0 r 0 (XEN) vmx_msr_write_intercept(2093) msr 8b mc 0 r 0 (XEN) vmx_msr_write_intercept(2093) msr 8b mc 0 r 0 (XEN) vmx_msr_write_intercept(2093) msr 8b mc 0 r 0 (XEN) vmx_msr_write_intercept(2093) msr 8b mc 0 r 0 (XEN) vmx_msr_write_intercept(2093) msr 8b mc 0 r 0 (XEN) vmx_msr_write_intercept(2093) msr 8b mc 0 r 0 (XEN) vmx_msr_write_intercept(2093) msr 8b mc 0 r 0 (XEN) vmx_msr_write_intercept(2093) msr 8b mc 0 r 0 (XEN) vmx_msr_write_intercept(2093) msr 8b mc 0 r 0 (XEN) vmx_msr_write_intercept(2093) msr 8b mc 0 r 0 (XEN) vmx_msr_write_intercept(2093) msr 8b mc 0 r 0 (XEN) vmx_msr_write_intercept(2093) msr 8b mc 0 r 0 (XEN) vmx_msr_write_intercept(2093) msr 8b mc 0 r 0
...

#4 lore raw
Quoted 20 lines
Quoted 1 line
Quoted 1 line · Olaf Hering
Quoted 17 lines · Olaf Hering
Quoted 14 lines · Jan Beulich

xen/arch/x86/hvm/svm/svm.c static int svm_msr_write_intercept(unsig · line 1686

Quoted 8 lines
Quoted 1 line
Quoted 1 line · Olaf Hering
xen/arch/x86/hvm/svm/svm.c +2 -1
@@ -1682,14 +1682,15 @@ static int svm_msr_write_intercept(unsig
         if ( wrmsr_viridian_regs(msr, msr_content) )
             break;
 
-        wrmsr_hypervisor_regs(msr, msr_content);
+        ret = wrmsr_hypervisor_regs(msr, msr_content);
+        retry = ret == -EAGAIN;

If you add error handling, don't constrain this to a single error code please. For the case here, the easiest would appear to be a switch converting to X86EMUL_OKAY, X86EMUL_RETRY, or X86EMUL_UNHANDLEABLE. If the function had ways to fail before, it would have been a bug anyway to not check the return value.

Quoted 5 lines
Quoted 1 line

I just sent v2 of this patch.

My change v2 causes a boot failure, the return value 0 is not handled correctly.
Did you really mean to translate ret == 0 to X86EMUL_UNHANDLEABLE?

Of course I didn't - I merely enumerated the values that I'd expect to result, with no meaning implied from their ordering.

Jan

#5 lore raw

PATCH v2 1/1 xen: handle paged gfn in wrmsr_hypervisor_regs

In reply to Olaf Hering

# HG changeset patch
# User Olaf Hering <[email protected]>
# Date 1367585741 -7200
# Node ID 219e97e92394eb8aea443144ae93e48d9cce9d87 # Parent 9df019eef776d129c2abb130d1458914fe1ecac4
xen: handle paged gfn in wrmsr_hypervisor_regs

If xenpaging is started very early for a guest the gfn for the hypercall page may be paged-out already. This leads to a guest crash:

...
(XEN) HVM10: Allocated Xen hypercall page at 169ff000
(XEN) traps.c:654:d10 Bad GMFN 169ff (MFN 3e900000000) to MSR 40000000 (XEN) HVM10: Detected Xen v4.3
(XEN) io.c:201:d10 MMIO emulation failed @ 0008:c2c2c2c2: 18 7c 55 6d 03 83 ff ff 10 7c (XEN) hvm.c:1253:d10 Triple fault on VCPU0 - invoking HVM shutdown action 1. (XEN) HVM11: HVM Loader
...

Handle paged gfns in wrmsr_hypervisor_regs and update callers to deal with the new return code -EAGAIN.

Also update the gdprintk to handle a page value of NULL to avoid printing a bogus MFN value.

xen/arch/x86/hvm/svm/svm.c +14 -3
diff -r 9df019eef776 -r 219e97e92394 xen/arch/x86/hvm/svm/svm.c
@@ -1569,7 +1569,7 @@ static int svm_msr_read_intercept(unsign
 
 static int svm_msr_write_intercept(unsigned int msr, uint64_t msr_content)
 {
-    int ret;
+    int ret, result = X86EMUL_OKAY;
     struct vcpu *v = current;
     struct vmcb_struct *vmcb = v->arch.hvm_svm.vmcb;
     int sync = 0;
@@ -1682,14 +1682,25 @@ static int svm_msr_write_intercept(unsig
         if ( wrmsr_viridian_regs(msr, msr_content) )
             break;
 
-        wrmsr_hypervisor_regs(msr, msr_content);
+        ret = wrmsr_hypervisor_regs(msr, msr_content);
+        switch ( ret )
+        {
+            case -EAGAIN:
+                result = X86EMUL_RETRY;
+                break;
+            case 0:
+                result = X86EMUL_UNHANDLEABLE;
+                break;
+            default:
+                break;
+        }
         break;
     }
 
     if ( sync )
         svm_vmload(vmcb);
 
-    return X86EMUL_OKAY;
+    return result;
 
  gpf:
     hvm_inject_hw_exception(TRAP_gp_fault, 0);
xen/arch/x86/hvm/vmx/vmx.c +11 -1
diff -r 9df019eef776 -r 219e97e92394 xen/arch/x86/hvm/vmx/vmx.c
@@ -2020,6 +2020,7 @@ void vmx_vlapic_msr_changed(struct vcpu 
 
 static int vmx_msr_write_intercept(unsigned int msr, uint64_t msr_content)
 {
+    int ret = 0;
     struct vcpu *v = current;
 
     HVM_DBG_LOG(DBG_LEVEL_1, "ecx=%#x, msr_value=%#"PRIx64, msr, msr_content);
@@ -2088,7 +2089,16 @@ static int vmx_msr_write_intercept(unsig
             case HNDL_unhandled:
                 if ( (vmx_write_guest_msr(msr, msr_content) != 0) &&
                      !is_last_branch_msr(msr) )
-                    wrmsr_hypervisor_regs(msr, msr_content);
+                    ret = wrmsr_hypervisor_regs(msr, msr_content);
+                    switch ( ret )
+                    {
+                        case -EAGAIN:
+                            return X86EMUL_RETRY;
+                        case 0:
+                            return X86EMUL_UNHANDLEABLE;
+                        default:
+                            break;
+                    }
                 break;
             case HNDL_exception_raised:
                 return X86EMUL_EXCEPTION;
xen/arch/x86/traps.c +10 -2
diff -r 9df019eef776 -r 219e97e92394 xen/arch/x86/traps.c
@@ -634,6 +634,7 @@ int wrmsr_hypervisor_regs(uint32_t idx, 
         unsigned long gmfn = val >> 12;
         unsigned int idx  = val & 0xfff;
         struct page_info *page;
+        p2m_type_t t;
 
         if ( idx > 0 )
         {
@@ -643,15 +644,22 @@ int wrmsr_hypervisor_regs(uint32_t idx, 
             return 0;
         }
 
-        page = get_page_from_gfn(d, gmfn, NULL, P2M_ALLOC);
+        page = get_page_from_gfn(d, gmfn, &t, P2M_ALLOC);
 
         if ( !page || !get_page_type(page, PGT_writable_page) )
         {
             if ( page )
                 put_page(page);
+
+            if ( p2m_is_paging(t) )
+            {
+                p2m_mem_paging_populate(d, gmfn);
+                return -EAGAIN;
+            }
+
             gdprintk(XENLOG_WARNING,
                      "Bad GMFN %lx (MFN %lx) to MSR %08x\n",
-                     gmfn, page_to_mfn(page), base + idx);
+                     gmfn, page ? page_to_mfn(page) : -1UL, base + idx);
             return 0;
         }
#6 lore raw
Quoted 7 lines
Quoted 1 line
Quoted 1 line · Olaf Hering
xen/arch/x86/hvm/svm/svm.c +1 -1
@@ -1682,14 +1682,25 @@ static int svm_msr_write_intercept(unsig
         if ( wrmsr_viridian_regs(msr, msr_content) )
             break;
 
-        wrmsr_hypervisor_regs(msr, msr_content);
+        ret = wrmsr_hypervisor_regs(msr, msr_content);

xen/arch/x86/hvm/svm/svm.c static int svm_msr_write_intercept(unsig · line 1695

Quoted 10 lines
xen/arch/x86/hvm/svm/svm.c +10 -0
+        switch ( ret )
+        {
+            case -EAGAIN:
+                result = X86EMUL_RETRY;
+                break;
+            case 0:
+                result = X86EMUL_UNHANDLEABLE;
+                break;
+            default:
+                break;

As you had already noticed the hard way - case 0 and default of course need to be switched (0 -> okay, anything else -> unhandleable).

Jan

#7 lore raw
Quoted 5 lines · Jan Beulich
Quoted 5 lines
Quoted 1 line
Quoted 1 line · Olaf Hering
xen/arch/x86/hvm/svm/svm.c +0 -0
@@ -1682,14 +1682,25 @@ static int svm_msr_write_intercept(unsig
         if ( wrmsr_viridian_regs(msr, msr_content) )
             break;
 
Quoted 16 lines

xen/arch/x86/hvm/svm/svm.c static int svm_msr_write_intercept(unsig · line 1695

Quoted 12 lines
xen/arch/x86/hvm/svm/svm.c +11 -1
-        wrmsr_hypervisor_regs(msr, msr_content);
+        ret = wrmsr_hypervisor_regs(msr, msr_content);
+        switch ( ret )
+        {
+            case -EAGAIN:
+                result = X86EMUL_RETRY;
+                break;
+            case 0:
+                result = X86EMUL_UNHANDLEABLE;
+                break;
+            default:
+                break;

As you had already noticed the hard way - case 0 and default of course need to be switched (0 -> okay, anything else -> unhandleable).

I dont follow.
ret == 1 looks like success to me, ret == 0 some sort of failure.

Olaf

#8 lore raw
Quoted 25 lines
Quoted 1 line
Quoted 1 line · Olaf Hering
Quoted 5 lines · Jan Beulich
Quoted 5 lines
Quoted 1 line
Quoted 1 line · Olaf Hering
xen/arch/x86/hvm/svm/svm.c +0 -0
@@ -1682,14 +1682,25 @@ static int svm_msr_write_intercept(unsig
         if ( wrmsr_viridian_regs(msr, msr_content) )
             break;
 
Quoted 16 lines

xen/arch/x86/hvm/svm/svm.c static int svm_msr_write_intercept(unsig · line 1695

Quoted 12 lines
xen/arch/x86/hvm/svm/svm.c +11 -1
-        wrmsr_hypervisor_regs(msr, msr_content);
+        ret = wrmsr_hypervisor_regs(msr, msr_content);
+        switch ( ret )
+        {
+            case -EAGAIN:
+                result = X86EMUL_RETRY;
+                break;
+            case 0:
+                result = X86EMUL_UNHANDLEABLE;
+                break;
+            default:
+                break;

As you had already noticed the hard way - case 0 and default of course need to be switched (0 -> okay, anything else -> unhandleable).

Quoted 2 lines

I dont follow.
ret == 1 looks like success to me, ret == 0 some sort of failure.

Let me check again... Positive values (or 1 in particular) mean "handled", 0 means not handled, negative values are errors.

Jan

#9 lore raw
In reply to Jan Beulich
Quoted 5 lines · Jan Beulich
Quoted 5 lines
Quoted 1 line
Quoted 1 line · Olaf Hering
xen/arch/x86/hvm/svm/svm.c +0 -0
@@ -1682,14 +1682,25 @@ static int svm_msr_write_intercept(unsig
         if ( wrmsr_viridian_regs(msr, msr_content) )
             break;
 
Quoted 16 lines

xen/arch/x86/hvm/svm/svm.c static int svm_msr_write_intercept(unsig · line 1695

Quoted 12 lines
xen/arch/x86/hvm/svm/svm.c +11 -1
-        wrmsr_hypervisor_regs(msr, msr_content);
+        ret = wrmsr_hypervisor_regs(msr, msr_content);
+        switch ( ret )
+        {
+            case -EAGAIN:
+                result = X86EMUL_RETRY;
+                break;
+            case 0:
+                result = X86EMUL_UNHANDLEABLE;
+                break;
+            default:
+                break;

As you had already noticed the hard way - case 0 and default of course need to be switched (0 -> okay, anything else -> unhandleable).

No!

Actually anything other than -EAGAIN should be handled here as 'okay'. In fact the return codes from wrmsr_hypervisor_regs are going to be a bit of a mess if we're not careful.

I suggest the following return codes:
0: not handled
1: handled
-EINVAL: error during handling
-EAGAIN: retry

The HVM callers should then handle as follows:
-EAGAIN: rc = X86EMUL_RETRY
-EINVAL: goto gp_fault
0: try other msr handlers (if any)
1: we're done, return X86EMUL_OKAY

Does that make sense?

-- Keir

Quoted 7 lines

Jan

_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel

#10 lore raw
Quoted 37 lines
Quoted 1 line
Quoted 1 line · Keir Fraser
Quoted 5 lines · Jan Beulich
Quoted 5 lines
Quoted 1 line
Quoted 1 line · Olaf Hering
xen/arch/x86/hvm/svm/svm.c +0 -0
@@ -1682,14 +1682,25 @@ static int svm_msr_write_intercept(unsig
         if ( wrmsr_viridian_regs(msr, msr_content) )
             break;
 
Quoted 16 lines

xen/arch/x86/hvm/svm/svm.c static int svm_msr_write_intercept(unsig · line 1695

Quoted 12 lines
xen/arch/x86/hvm/svm/svm.c +11 -1
-        wrmsr_hypervisor_regs(msr, msr_content);
+        ret = wrmsr_hypervisor_regs(msr, msr_content);
+        switch ( ret )
+        {
+            case -EAGAIN:
+                result = X86EMUL_RETRY;
+                break;
+            case 0:
+                result = X86EMUL_UNHANDLEABLE;
+                break;
+            default:
+                break;

As you had already noticed the hard way - case 0 and default of course need to be switched (0 -> okay, anything else -> unhandleable).

No!

Actually anything other than -EAGAIN should be handled here as 'okay'. In fact the return codes from wrmsr_hypervisor_regs are going to be a bit of a mess if we're not careful.

I suggest the following return codes:
0: not handled
1: handled
-EINVAL: error during handling
-EAGAIN: retry

Quoted 7 lines

The HVM callers should then handle as follows:
-EAGAIN: rc = X86EMUL_RETRY
-EINVAL: goto gp_fault
0: try other msr handlers (if any)
1: we're done, return X86EMUL_OKAY

Does that make sense?

Sure - you may have seen my later reply where I also notide this mistake in my first response to Olaf.

The only thing I'd like to ensure is to not constrain the code to specific error codes - any negative value other than -EAGAIN should result in #GP (or whatever a suitable action is) imo.

Jan

#11 lore raw
Quoted 13 lines · Jan Beulich
Quoted 13 lines
Quoted 1 line
Quoted 1 line · Keir Fraser

I suggest the following return codes:
0: not handled
1: handled
-EINVAL: error during handling
-EAGAIN: retry

The HVM callers should then handle as follows:
-EAGAIN: rc = X86EMUL_RETRY
-EINVAL: goto gp_fault
0: try other msr handlers (if any)
1: we're done, return X86EMUL_OKAY

Quoted 4 lines
Quoted 1 line

Does that make sense?

Sure - you may have seen my later reply where I also notide this mistake in my first response to Olaf.

I sent v3 of the patch. I had to guess which 'return 0' is supposed to be an error.

Olaf

#12 lore raw
Quoted 15 lines · Olaf Hering
Quoted 13 lines · Jan Beulich
Quoted 13 lines
Quoted 1 line
Quoted 1 line · Keir Fraser

I suggest the following return codes:
0: not handled
1: handled
-EINVAL: error during handling
-EAGAIN: retry

The HVM callers should then handle as follows:
-EAGAIN: rc = X86EMUL_RETRY
-EINVAL: goto gp_fault
0: try other msr handlers (if any)
1: we're done, return X86EMUL_OKAY

Quoted 7 lines
Quoted 4 lines
Quoted 1 line

Does that make sense?

Sure - you may have seen my later reply where I also notide this mistake in my first response to Olaf.

I sent v3 of the patch. I had to guess which 'return 0' is supposed to be an error.

You guessed correctly ;)

Quoted 1 line

Olaf

#13 lore raw

PATCH v3 1/1 xen: handle paged gfn in wrmsr_hypervisor_regs

In reply to Olaf Hering

# HG changeset patch
# User Olaf Hering <[email protected]>
# Date 1367593457 -7200
# Node ID b8af60cf8282bfddb13cc10e4ffaf0c396a15104 # Parent 9df019eef776d129c2abb130d1458914fe1ecac4
xen: handle paged gfn in wrmsr_hypervisor_regs

If xenpaging is started very early for a guest the gfn for the hypercall page may be paged-out already. This leads to a guest crash:

...
(XEN) HVM10: Allocated Xen hypercall page at 169ff000
(XEN) traps.c:654:d10 Bad GMFN 169ff (MFN 3e900000000) to MSR 40000000 (XEN) HVM10: Detected Xen v4.3
(XEN) io.c:201:d10 MMIO emulation failed @ 0008:c2c2c2c2: 18 7c 55 6d 03 83 ff ff 10 7c (XEN) hvm.c:1253:d10 Triple fault on VCPU0 - invoking HVM shutdown action 1. (XEN) HVM11: HVM Loader
...

Update return codes of wrmsr_hypervisor_regs, update callers to deal with the new return codes:
0: not handled
1: handled
-EINVAL: error during handling
-EAGAIN: retry

Also update the gdprintk to handle a page value of NULL to avoid printing a bogus MFN value. Update also computing of MSR value in gdprintk, the idx was always zero.

xen/arch/x86/hvm/svm/svm.c +13 -3
diff -r 9df019eef776 -r b8af60cf8282 xen/arch/x86/hvm/svm/svm.c
@@ -1569,7 +1569,7 @@ static int svm_msr_read_intercept(unsign
 
 static int svm_msr_write_intercept(unsigned int msr, uint64_t msr_content)
 {
-    int ret;
+    int ret, result = X86EMUL_OKAY;
     struct vcpu *v = current;
     struct vmcb_struct *vmcb = v->arch.hvm_svm.vmcb;
     int sync = 0;
@@ -1682,14 +1682,24 @@ static int svm_msr_write_intercept(unsig
         if ( wrmsr_viridian_regs(msr, msr_content) )
             break;
 
-        wrmsr_hypervisor_regs(msr, msr_content);
+        switch ( wrmsr_hypervisor_regs(msr, msr_content) )
+        {
+            case -EAGAIN:
+                result = X86EMUL_RETRY;
+                break;
+            case 0:
+            case 1:
+                break;
+            default:
+	        goto gpf;
+        }
         break;
     }
 
     if ( sync )
         svm_vmload(vmcb);
 
-    return X86EMUL_OKAY;
+    return result;
 
  gpf:
     hvm_inject_hw_exception(TRAP_gp_fault, 0);
xen/arch/x86/hvm/vmx/vmx.c +10 -1
diff -r 9df019eef776 -r b8af60cf8282 xen/arch/x86/hvm/vmx/vmx.c
@@ -2088,7 +2088,16 @@ static int vmx_msr_write_intercept(unsig
             case HNDL_unhandled:
                 if ( (vmx_write_guest_msr(msr, msr_content) != 0) &&
                      !is_last_branch_msr(msr) )
-                    wrmsr_hypervisor_regs(msr, msr_content);
+                    switch ( wrmsr_hypervisor_regs(msr, msr_content) )
+                    {
+                        case -EAGAIN:
+                            return X86EMUL_RETRY;
+                        case 0:
+                        case 1:
+                            break;
+                        default:
+                            goto gp_fault;
+                    }
                 break;
             case HNDL_exception_raised:
                 return X86EMUL_EXCEPTION;
xen/arch/x86/traps.c +13 -5
diff -r 9df019eef776 -r b8af60cf8282 xen/arch/x86/traps.c
@@ -634,25 +634,33 @@ int wrmsr_hypervisor_regs(uint32_t idx, 
         unsigned long gmfn = val >> 12;
         unsigned int idx  = val & 0xfff;
         struct page_info *page;
+        p2m_type_t t;
 
         if ( idx > 0 )
         {
             gdprintk(XENLOG_WARNING,
                      "Out of range index %u to MSR %08x\n",
                      idx, 0x40000000);
-            return 0;
+            return -EINVAL;
         }
 
-        page = get_page_from_gfn(d, gmfn, NULL, P2M_ALLOC);
+        page = get_page_from_gfn(d, gmfn, &t, P2M_ALLOC);
 
         if ( !page || !get_page_type(page, PGT_writable_page) )
         {
             if ( page )
                 put_page(page);
+
+            if ( p2m_is_paging(t) )
+            {
+                p2m_mem_paging_populate(d, gmfn);
+                return -EAGAIN;
+            }
+
             gdprintk(XENLOG_WARNING,
                      "Bad GMFN %lx (MFN %lx) to MSR %08x\n",
-                     gmfn, page_to_mfn(page), base + idx);
-            return 0;
+                     gmfn, page ? page_to_mfn(page) : -1UL, base);
+            return -EINVAL;
         }
 
         hypercall_page = __map_domain_page(page);
@@ -2490,7 +2498,7 @@ static int emulate_privileged_op(struct 
                 goto fail;
             break;
         default:
-            if ( wrmsr_hypervisor_regs(regs->ecx, msr_content) )
+            if ( wrmsr_hypervisor_regs(regs->ecx, msr_content) == 1 )
                 break;
 
             rc = vmce_wrmsr(regs->ecx, msr_content);
#14 lore raw
Quoted 49 lines
Quoted 1 line
Quoted 1 line · Olaf Hering
xen/arch/x86/hvm/svm/svm.c +13 -3
@@ -1569,7 +1569,7 @@ static int svm_msr_read_intercept(unsign
 
 static int svm_msr_write_intercept(unsigned int msr, uint64_t msr_content)
 {
-    int ret;
+    int ret, result = X86EMUL_OKAY;
     struct vcpu *v = current;
     struct vmcb_struct *vmcb = v->arch.hvm_svm.vmcb;
     int sync = 0;
@@ -1682,14 +1682,24 @@ static int svm_msr_write_intercept(unsig
         if ( wrmsr_viridian_regs(msr, msr_content) )
             break;
 
-        wrmsr_hypervisor_regs(msr, msr_content);
+        switch ( wrmsr_hypervisor_regs(msr, msr_content) )
+        {
+            case -EAGAIN:
+                result = X86EMUL_RETRY;
+                break;
+            case 0:
+            case 1:
+                break;
+            default:
+	        goto gpf;
+        }
         break;
     }
 
     if ( sync )
         svm_vmload(vmcb);
 
-    return X86EMUL_OKAY;
+    return result;
 
  gpf:
     hvm_inject_hw_exception(TRAP_gp_fault, 0);
xen/arch/x86/hvm/vmx/vmx.c +3 -1
@@ -2088,7 +2088,16 @@ static int vmx_msr_write_intercept(unsig
             case HNDL_unhandled:
                 if ( (vmx_write_guest_msr(msr, msr_content) != 0) &&
                      !is_last_branch_msr(msr) )
-                    wrmsr_hypervisor_regs(msr, msr_content);
+                    switch ( wrmsr_hypervisor_regs(msr, msr_content) )
+                    {
+                        case -EAGAIN:

xen/arch/x86/hvm/vmx/vmx.c static int vmx_msr_write_intercept(unsig · line 2103

Quoted 10 lines
xen/arch/x86/hvm/vmx/vmx.c +7 -0
+                            return X86EMUL_RETRY;
+                        case 0:
+                        case 1:
+                            break;
+                        default:
+                            goto gp_fault;
+                    }
                 break;
             case HNDL_exception_raised:
                 return X86EMUL_EXCEPTION;

Apart from formatting things look okay up to here.

Quoted 5 lines
xen/arch/x86/traps.c +0 -0
@@ -634,25 +634,33 @@ int wrmsr_hypervisor_regs(uint32_t idx, 
         unsigned long gmfn = val >> 12;
         unsigned int idx  = val & 0xfff;

xen/arch/x86/traps.c int wrmsr_hypervisor_regs(uint32_t idx, · line 644

Quoted 10 lines
xen/arch/x86/traps.c +2 -1
         struct page_info *page;
+        p2m_type_t t;
 
         if ( idx > 0 )
         {
             gdprintk(XENLOG_WARNING,
                      "Out of range index %u to MSR %08x\n",
                      idx, 0x40000000);
-            return 0;
+            return -EINVAL;

But I'd stay away from converting to actual errors both here ...

Quoted 12 lines
xen/arch/x86/traps.c +4 -1
         }
 
-        page = get_page_from_gfn(d, gmfn, NULL, P2M_ALLOC);
+        page = get_page_from_gfn(d, gmfn, &t, P2M_ALLOC);
 
         if ( !page || !get_page_type(page, PGT_writable_page) )
         {
             if ( page )
                 put_page(page);
+
+            if ( p2m_is_paging(t) )
+            {

xen/arch/x86/traps.c int wrmsr_hypervisor_regs(uint32_t idx, · line 663

Quoted 10 lines
xen/arch/x86/traps.c +6 -2
+                p2m_mem_paging_populate(d, gmfn);
+                return -EAGAIN;
+            }
+
             gdprintk(XENLOG_WARNING,
                      "Bad GMFN %lx (MFN %lx) to MSR %08x\n",
-                     gmfn, page_to_mfn(page), base + idx);
-            return 0;
+                     gmfn, page ? page_to_mfn(page) : -1UL, base);
+            return -EINVAL;

... and here. If at all these ought to go into a separate patch (which we'd likely postpone until after 4.3).

Jan

Quoted 12 lines
xen/arch/x86/traps.c +1 -1
         }
 
         hypercall_page = __map_domain_page(page);
@@ -2490,7 +2498,7 @@ static int emulate_privileged_op(struct 
                 goto fail;
             break;
         default:
-            if ( wrmsr_hypervisor_regs(regs->ecx, msr_content) )
+            if ( wrmsr_hypervisor_regs(regs->ecx, msr_content) == 1 )
                 break;
 
             rc = vmce_wrmsr(regs->ecx, msr_content);
#15 lore raw
Quoted 1 line · Jan Beulich

Apart from formatting things look okay up to here.

Did you mean just the single tab in svm_msr_write_intercept?

Quoted 8 lines
Quoted 8 lines
xen/arch/x86/traps.c +1 -0
@@ -634,25 +634,33 @@ int wrmsr_hypervisor_regs(uint32_t idx, 
         unsigned long gmfn = val >> 12;
         unsigned int idx  = val & 0xfff;
         struct page_info *page;
+        p2m_type_t t;
 
Quoted 9 lines

xen/arch/x86/traps.c int wrmsr_hypervisor_regs(uint32_t idx, · line 644

Quoted 7 lines
xen/arch/x86/traps.c +1 -1
         if ( idx > 0 )
         {
             gdprintk(XENLOG_WARNING,
                      "Out of range index %u to MSR %08x\n",
                      idx, 0x40000000);
-            return 0;
+            return -EINVAL;

But I'd stay away from converting to actual errors both here ...

I sent v4 with the -EINVAL removed.

Olaf

#16 lore raw
Quoted 5 lines
Quoted 1 line
Quoted 1 line · Olaf Hering
Quoted 1 line · Jan Beulich

Apart from formatting things look okay up to here.

Did you mean just the single tab in svm_msr_write_intercept?

That and the too deep indentation of the case labels within the new switch blocks.

Jan

#17 lore raw
In reply to Jan Beulich
Quoted 27 lines · Jan Beulich

Review

Quoted 7 lines
         if ( idx > 0 )
         {
             gdprintk(XENLOG_WARNING,
                      "Out of range index %u to MSR %08x\n",
                      idx, 0x40000000);
(quoted diff) +1 -1
-            return 0;
+            return -EINVAL;

But I'd stay away from converting to actual errors both here ...

Quoted 17 lines
         }
xen/arch/x86/traps.c +8 -1
-        page = get_page_from_gfn(d, gmfn, NULL, P2M_ALLOC);
+        page = get_page_from_gfn(d, gmfn, &t, P2M_ALLOC);
 
         if ( !page || !get_page_type(page, PGT_writable_page) )
         {
             if ( page )
                 put_page(page);
+
+            if ( p2m_is_paging(t) )
+            {
+                p2m_mem_paging_populate(d, gmfn);
+                return -EAGAIN;
+            }
+
             gdprintk(XENLOG_WARNING,
Quoted 8 lines

xen/arch/x86/traps.c int wrmsr_hypervisor_regs(uint32_t idx, · line 663

Quoted 5 lines
                      "Bad GMFN %lx (MFN %lx) to MSR %08x\n",
xen/arch/x86/traps.c +2 -2
-                     gmfn, page_to_mfn(page), base + idx);
-            return 0;
+                     gmfn, page ? page_to_mfn(page) : -1UL, base);
+            return -EINVAL;

... and here. If at all these ought to go into a separate patch (which we'd likely postpone until after 4.3).

But they are errors?

Could agree with postponing to post-4.3 though.

-- Keir

#18 lore raw
Quoted 39 lines
Quoted 1 line
Quoted 1 line · Keir Fraser
Quoted 27 lines · Jan Beulich

Review

Quoted 7 lines
         if ( idx > 0 )
         {
             gdprintk(XENLOG_WARNING,
                      "Out of range index %u to MSR %08x\n",
                      idx, 0x40000000);
(quoted diff) +1 -1
-            return 0;
+            return -EINVAL;

But I'd stay away from converting to actual errors both here ...

Quoted 17 lines
         }
xen/arch/x86/traps.c +8 -1
-        page = get_page_from_gfn(d, gmfn, NULL, P2M_ALLOC);
+        page = get_page_from_gfn(d, gmfn, &t, P2M_ALLOC);
 
         if ( !page || !get_page_type(page, PGT_writable_page) )
         {
             if ( page )
                 put_page(page);
+
+            if ( p2m_is_paging(t) )
+            {
+                p2m_mem_paging_populate(d, gmfn);
+                return -EAGAIN;
+            }
+
             gdprintk(XENLOG_WARNING,
Quoted 8 lines

xen/arch/x86/traps.c int wrmsr_hypervisor_regs(uint32_t idx, · line 663

Quoted 5 lines
                      "Bad GMFN %lx (MFN %lx) to MSR %08x\n",
xen/arch/x86/traps.c +2 -2
-                     gmfn, page_to_mfn(page), base + idx);
-            return 0;
+                     gmfn, page ? page_to_mfn(page) : -1UL, base);
+            return -EINVAL;

... and here. If at all these ought to go into a separate patch (which we'd likely postpone until after 4.3).

Quoted 1 line

But they are errors?

Right. But they weren't handled as such so far, and doing the conversion isn't the purpose of the patch.

Quoted 1 line

Could agree with postponing to post-4.3 though.

Hence the request to split off that part of the patch. I'd be happy to apply that after the trees got branched.

Jan

#19 lore raw
Quoted 7 lines · Jan Beulich
Quoted 4 lines
Quoted 2 lines

... and here. If at all these ought to go into a separate patch (which we'd likely postpone until after 4.3).

But they are errors?

Right. But they weren't handled as such so far, and doing the conversion isn't the purpose of the patch.

Okay, I wasn't sure how dismissive your "if at all" was. :)

Quoted 4 lines
Quoted 1 line

Could agree with postponing to post-4.3 though.

Hence the request to split off that part of the patch. I'd be happy to apply that after the trees got branched.

Fine.

-- Keir

Acked-by on v3 1/1
#20 lore raw
In reply to Olaf Hering
Quoted 6 lines · Olaf Hering

# HG changeset patch
# User Olaf Hering <[email protected]>
# Date 1367593457 -7200
# Node ID b8af60cf8282bfddb13cc10e4ffaf0c396a15104 # Parent 9df019eef776d129c2abb130d1458914fe1ecac4
xen: handle paged gfn in wrmsr_hypervisor_regs

Quoted 142 lines

If xenpaging is started very early for a guest the gfn for the hypercall page may be paged-out already. This leads to a guest crash:

...
(XEN) HVM10: Allocated Xen hypercall page at 169ff000
(XEN) traps.c:654:d10 Bad GMFN 169ff (MFN 3e900000000) to MSR 40000000 (XEN) HVM10: Detected Xen v4.3
(XEN) io.c:201:d10 MMIO emulation failed @ 0008:c2c2c2c2: 18 7c 55 6d 03 83 ff ff 10 7c
(XEN) hvm.c:1253:d10 Triple fault on VCPU0 - invoking HVM shutdown action 1. (XEN) HVM11: HVM Loader
...

Update return codes of wrmsr_hypervisor_regs, update callers to deal with the new return codes:
0: not handled
1: handled
-EINVAL: error during handling
-EAGAIN: retry

Also update the gdprintk to handle a page value of NULL to avoid printing a bogus MFN value. Update also computing of MSR value in gdprintk, the idx was always zero.

diff -r 9df019eef776 -r b8af60cf8282 xen/arch/x86/hvm/svm/svm.c

xen/arch/x86/hvm/svm/svm.c +13 -3
@@ -1569,7 +1569,7 @@ static int svm_msr_read_intercept(unsign
 
 static int svm_msr_write_intercept(unsigned int msr, uint64_t msr_content)
 {
-    int ret;
+    int ret, result = X86EMUL_OKAY;
     struct vcpu *v = current;
     struct vmcb_struct *vmcb = v->arch.hvm_svm.vmcb;
     int sync = 0;
@@ -1682,14 +1682,24 @@ static int svm_msr_write_intercept(unsig
         if ( wrmsr_viridian_regs(msr, msr_content) )
             break;
 
-        wrmsr_hypervisor_regs(msr, msr_content);
+        switch ( wrmsr_hypervisor_regs(msr, msr_content) )
+        {
+            case -EAGAIN:
+                result = X86EMUL_RETRY;
+                break;
+            case 0:
+            case 1:
+                break;
+            default:
+         goto gpf;
+        }
         break;
     }
 
     if ( sync )
         svm_vmload(vmcb);
 
-    return X86EMUL_OKAY;
+    return result;
 
  gpf:
     hvm_inject_hw_exception(TRAP_gp_fault, 0);

diff -r 9df019eef776 -r b8af60cf8282 xen/arch/x86/hvm/vmx/vmx.c

xen/arch/x86/hvm/vmx/vmx.c +10 -1
@@ -2088,7 +2088,16 @@ static int vmx_msr_write_intercept(unsig
             case HNDL_unhandled:
                 if ( (vmx_write_guest_msr(msr, msr_content) != 0) &&
                      !is_last_branch_msr(msr) )
-                    wrmsr_hypervisor_regs(msr, msr_content);
+                    switch ( wrmsr_hypervisor_regs(msr, msr_content) )
+                    {
+                        case -EAGAIN:
+                            return X86EMUL_RETRY;
+                        case 0:
+                        case 1:
+                            break;
+                        default:
+                            goto gp_fault;
+                    }
                 break;
             case HNDL_exception_raised:
                 return X86EMUL_EXCEPTION;

diff -r 9df019eef776 -r b8af60cf8282 xen/arch/x86/traps.c

xen/arch/x86/traps.c +13 -5
@@ -634,25 +634,33 @@ int wrmsr_hypervisor_regs(uint32_t idx,
         unsigned long gmfn = val >> 12;
         unsigned int idx  = val & 0xfff;
         struct page_info *page;
+        p2m_type_t t;
 
         if ( idx > 0 )
         {
             gdprintk(XENLOG_WARNING,
                      "Out of range index %u to MSR %08x\n",
                      idx, 0x40000000);
-            return 0;
+            return -EINVAL;
         }
 
-        page = get_page_from_gfn(d, gmfn, NULL, P2M_ALLOC);
+        page = get_page_from_gfn(d, gmfn, &t, P2M_ALLOC);
 
         if ( !page || !get_page_type(page, PGT_writable_page) )
         {
             if ( page )
                 put_page(page);
+
+            if ( p2m_is_paging(t) )
+            {
+                p2m_mem_paging_populate(d, gmfn);
+                return -EAGAIN;
+            }
+
             gdprintk(XENLOG_WARNING,
                      "Bad GMFN %lx (MFN %lx) to MSR %08x\n",
-                     gmfn, page_to_mfn(page), base + idx);
-            return 0;
+                     gmfn, page ? page_to_mfn(page) : -1UL, base);
+            return -EINVAL;
         }
 
         hypercall_page = __map_domain_page(page);
@@ -2490,7 +2498,7 @@ static int emulate_privileged_op(struct
                 goto fail;
             break;
         default:
-            if ( wrmsr_hypervisor_regs(regs->ecx, msr_content) )
+            if ( wrmsr_hypervisor_regs(regs->ecx, msr_content) == 1 )
                 break;
 
             rc = vmce_wrmsr(regs->ecx, msr_content);

_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel

#21 lore raw

PATCH v4 1/1 xen: handle paged gfn in wrmsr_hypervisor_regs

In reply to Olaf Hering

# HG changeset patch
# User Olaf Hering <[email protected]>
# Date 1367595698 -7200
# Node ID 4b7f3f754c015e0c2012d4b8ef4faf89ba18a3bf # Parent 9df019eef776d129c2abb130d1458914fe1ecac4
xen: handle paged gfn in wrmsr_hypervisor_regs

If xenpaging is started very early for a guest the gfn for the hypercall page may be paged-out already. This leads to a guest crash:

...
(XEN) HVM10: Allocated Xen hypercall page at 169ff000
(XEN) traps.c:654:d10 Bad GMFN 169ff (MFN 3e900000000) to MSR 40000000 (XEN) HVM10: Detected Xen v4.3
(XEN) io.c:201:d10 MMIO emulation failed @ 0008:c2c2c2c2: 18 7c 55 6d 03 83 ff ff 10 7c (XEN) hvm.c:1253:d10 Triple fault on VCPU0 - invoking HVM shutdown action 1. (XEN) HVM11: HVM Loader
...

Update return codes of wrmsr_hypervisor_regs, update callers to deal with the new return codes:
0: not handled
1: handled
-EINVAL: error during handling
-EAGAIN: retry

Also update the gdprintk to handle a page value of NULL to avoid printing a bogus MFN value. Update also computing of MSR value in gdprintk, the idx was always zero.

xen/arch/x86/hvm/svm/svm.c +13 -3
diff -r 9df019eef776 -r 4b7f3f754c01 xen/arch/x86/hvm/svm/svm.c
@@ -1569,7 +1569,7 @@ static int svm_msr_read_intercept(unsign
 
 static int svm_msr_write_intercept(unsigned int msr, uint64_t msr_content)
 {
-    int ret;
+    int ret, result = X86EMUL_OKAY;
     struct vcpu *v = current;
     struct vmcb_struct *vmcb = v->arch.hvm_svm.vmcb;
     int sync = 0;
@@ -1682,14 +1682,24 @@ static int svm_msr_write_intercept(unsig
         if ( wrmsr_viridian_regs(msr, msr_content) )
             break;
 
-        wrmsr_hypervisor_regs(msr, msr_content);
+        switch ( wrmsr_hypervisor_regs(msr, msr_content) )
+        {
+            case -EAGAIN:
+                result = X86EMUL_RETRY;
+                break;
+            case 0:
+            case 1:
+                break;
+            default:
+                goto gpf;
+        }
         break;
     }
 
     if ( sync )
         svm_vmload(vmcb);
 
-    return X86EMUL_OKAY;
+    return result;
 
  gpf:
     hvm_inject_hw_exception(TRAP_gp_fault, 0);
xen/arch/x86/hvm/vmx/vmx.c +10 -1
diff -r 9df019eef776 -r 4b7f3f754c01 xen/arch/x86/hvm/vmx/vmx.c
@@ -2088,7 +2088,16 @@ static int vmx_msr_write_intercept(unsig
             case HNDL_unhandled:
                 if ( (vmx_write_guest_msr(msr, msr_content) != 0) &&
                      !is_last_branch_msr(msr) )
-                    wrmsr_hypervisor_regs(msr, msr_content);
+                    switch ( wrmsr_hypervisor_regs(msr, msr_content) )
+                    {
+                        case -EAGAIN:
+                            return X86EMUL_RETRY;
+                        case 0:
+                        case 1:
+                            break;
+                        default:
+                            goto gp_fault;
+                    }
                 break;
             case HNDL_exception_raised:
                 return X86EMUL_EXCEPTION;
xen/arch/x86/traps.c +11 -3
diff -r 9df019eef776 -r 4b7f3f754c01 xen/arch/x86/traps.c
@@ -634,6 +634,7 @@ int wrmsr_hypervisor_regs(uint32_t idx, 
         unsigned long gmfn = val >> 12;
         unsigned int idx  = val & 0xfff;
         struct page_info *page;
+        p2m_type_t t;
 
         if ( idx > 0 )
         {
@@ -643,15 +644,22 @@ int wrmsr_hypervisor_regs(uint32_t idx, 
             return 0;
         }
 
-        page = get_page_from_gfn(d, gmfn, NULL, P2M_ALLOC);
+        page = get_page_from_gfn(d, gmfn, &t, P2M_ALLOC);
 
         if ( !page || !get_page_type(page, PGT_writable_page) )
         {
             if ( page )
                 put_page(page);
+
+            if ( p2m_is_paging(t) )
+            {
+                p2m_mem_paging_populate(d, gmfn);
+                return -EAGAIN;
+            }
+
             gdprintk(XENLOG_WARNING,
                      "Bad GMFN %lx (MFN %lx) to MSR %08x\n",
-                     gmfn, page_to_mfn(page), base + idx);
+                     gmfn, page ? page_to_mfn(page) : -1UL, base);
             return 0;
         }
 
@@ -2490,7 +2498,7 @@ static int emulate_privileged_op(struct 
                 goto fail;
             break;
         default:
-            if ( wrmsr_hypervisor_regs(regs->ecx, msr_content) )
+            if ( wrmsr_hypervisor_regs(regs->ecx, msr_content) == 1 )
                 break;
 
             rc = vmce_wrmsr(regs->ecx, msr_content);
#22 lore raw

PATCH v5 1/1 xen: handle paged gfn in wrmsr_hypervisor_regs

In reply to Olaf Hering

# HG changeset patch
# User Olaf Hering <[email protected]>
# Date 1367598534 -7200
# Node ID e04344e05ce4ab608df4570d7c081770cc40f3cf # Parent 9df019eef776d129c2abb130d1458914fe1ecac4
xen: handle paged gfn in wrmsr_hypervisor_regs

If xenpaging is started very early for a guest the gfn for the hypercall page may be paged-out already. This leads to a guest crash:

...
(XEN) HVM10: Allocated Xen hypercall page at 169ff000
(XEN) traps.c:654:d10 Bad GMFN 169ff (MFN 3e900000000) to MSR 40000000 (XEN) HVM10: Detected Xen v4.3
(XEN) io.c:201:d10 MMIO emulation failed @ 0008:c2c2c2c2: 18 7c 55 6d 03 83 ff ff 10 7c (XEN) hvm.c:1253:d10 Triple fault on VCPU0 - invoking HVM shutdown action 1. (XEN) HVM11: HVM Loader
...

Update return codes of wrmsr_hypervisor_regs, update callers to deal with the new return codes:
0: not handled
1: handled
-EAGAIN: retry

Currently wrmsr_hypervisor_regs will not return the following error, it will be added in a separate patch:
-EINVAL: error during handling

Also update the gdprintk to handle a page value of NULL to avoid printing a bogus MFN value. Update also computing of MSR value in gdprintk, the idx was always zero.

xen/arch/x86/hvm/svm/svm.c +13 -3
diff -r 9df019eef776 -r e04344e05ce4 xen/arch/x86/hvm/svm/svm.c
@@ -1569,7 +1569,7 @@ static int svm_msr_read_intercept(unsign
 
 static int svm_msr_write_intercept(unsigned int msr, uint64_t msr_content)
 {
-    int ret;
+    int ret, result = X86EMUL_OKAY;
     struct vcpu *v = current;
     struct vmcb_struct *vmcb = v->arch.hvm_svm.vmcb;
     int sync = 0;
@@ -1682,14 +1682,24 @@ static int svm_msr_write_intercept(unsig
         if ( wrmsr_viridian_regs(msr, msr_content) )
             break;
 
-        wrmsr_hypervisor_regs(msr, msr_content);
+        switch ( wrmsr_hypervisor_regs(msr, msr_content) )
+        {
+        case -EAGAIN:
+            result = X86EMUL_RETRY;
+            break;
+        case 0:
+        case 1:
+            break;
+        default:
+            goto gpf;
+        }
         break;
     }
 
     if ( sync )
         svm_vmload(vmcb);
 
-    return X86EMUL_OKAY;
+    return result;
 
  gpf:
     hvm_inject_hw_exception(TRAP_gp_fault, 0);
xen/arch/x86/hvm/vmx/vmx.c +10 -1
diff -r 9df019eef776 -r e04344e05ce4 xen/arch/x86/hvm/vmx/vmx.c
@@ -2088,7 +2088,16 @@ static int vmx_msr_write_intercept(unsig
             case HNDL_unhandled:
                 if ( (vmx_write_guest_msr(msr, msr_content) != 0) &&
                      !is_last_branch_msr(msr) )
-                    wrmsr_hypervisor_regs(msr, msr_content);
+                    switch ( wrmsr_hypervisor_regs(msr, msr_content) )
+                    {
+                    case -EAGAIN:
+                        return X86EMUL_RETRY;
+                    case 0:
+                    case 1:
+                        break;
+                    default:
+                        goto gp_fault;
+                    }
                 break;
             case HNDL_exception_raised:
                 return X86EMUL_EXCEPTION;
xen/arch/x86/traps.c +11 -3
diff -r 9df019eef776 -r e04344e05ce4 xen/arch/x86/traps.c
@@ -634,6 +634,7 @@ int wrmsr_hypervisor_regs(uint32_t idx, 
         unsigned long gmfn = val >> 12;
         unsigned int idx  = val & 0xfff;
         struct page_info *page;
+        p2m_type_t t;
 
         if ( idx > 0 )
         {
@@ -643,15 +644,22 @@ int wrmsr_hypervisor_regs(uint32_t idx, 
             return 0;
         }
 
-        page = get_page_from_gfn(d, gmfn, NULL, P2M_ALLOC);
+        page = get_page_from_gfn(d, gmfn, &t, P2M_ALLOC);
 
         if ( !page || !get_page_type(page, PGT_writable_page) )
         {
             if ( page )
                 put_page(page);
+
+            if ( p2m_is_paging(t) )
+            {
+                p2m_mem_paging_populate(d, gmfn);
+                return -EAGAIN;
+            }
+
             gdprintk(XENLOG_WARNING,
                      "Bad GMFN %lx (MFN %lx) to MSR %08x\n",
-                     gmfn, page_to_mfn(page), base + idx);
+                     gmfn, page ? page_to_mfn(page) : -1UL, base);
             return 0;
         }
 
@@ -2490,7 +2498,7 @@ static int emulate_privileged_op(struct 
                 goto fail;
             break;
         default:
-            if ( wrmsr_hypervisor_regs(regs->ecx, msr_content) )
+            if ( wrmsr_hypervisor_regs(regs->ecx, msr_content) == 1 )
                 break;
 
             rc = vmce_wrmsr(regs->ecx, msr_content);
#23 lore raw
Quoted 2 lines
Quoted 1 line
Quoted 1 line · Olaf Hering

xen: handle paged gfn in wrmsr_hypervisor_regs

Does this need any backporting?

Jan

#24 lore raw
Quoted 4 lines · Jan Beulich
Quoted 2 lines
Quoted 1 line
Quoted 1 line · Olaf Hering

xen: handle paged gfn in wrmsr_hypervisor_regs

Does this need any backporting?

4.2 has the same issue, if I read the code correctly.

Olaf

#25 lore raw
Quoted 9 lines
Quoted 1 line
Quoted 1 line · Olaf Hering
Quoted 4 lines · Jan Beulich
Quoted 2 lines
Quoted 1 line
Quoted 1 line · Olaf Hering

xen: handle paged gfn in wrmsr_hypervisor_regs

Does this need any backporting?

4.2 has the same issue, if I read the code correctly.

I understand that - the question is whether paging is usable enough in 4.2 to warrant this being backported.

Jan

#26 lore raw
Quoted 4 lines · Jan Beulich
Quoted 4 lines
Quoted 1 line
Quoted 1 line · Olaf Hering
Quoted 1 line · Jan Beulich
Quoted 1 line
Quoted 1 line
Quoted 1 line · Olaf Hering
Quoted 8 lines
Quoted 5 lines
Quoted 3 lines
Quoted 1 line

xen: handle paged gfn in wrmsr_hypervisor_regs

Does this need any backporting?

4.2 has the same issue, if I read the code correctly.

I understand that - the question is whether paging is usable enough in 4.2 to warrant this being backported.

It is, but for some reason I did not run into this issue.

Olaf