Skip to contentSkip to conversation

memory: Delete assertion in memory_region_unregister_iommu_notifier

RFC Superseded Eugenio Pérez 6 participants 38 messages Jun 25, 2020 – Oct 15, 2020

v1 of 8 Superseded Latest is v8 →

lore
Share & tools
lore
#0 lore raw

cover memory: Delete assertion in memory_region_unregister_iommu_notifier

I am able to hit this assertion when a Red Hat 7 guest virtio_net device raises an "Invalidation" of all the TLB entries. This happens in the guest's startup if 'intel_iommu=on' argument is passed to the guest kernel and right IOMMU/ATS devices are declared in qemu's command line.

If we examinate *entry in frame 4 of backtrace:
*entry = {target_as = 0x555556f6c050, iova = 0x0, translated_addr = 0x0, addr_mask = 0xffffffffffffffff, perm = 0x0}

Which (I think) tries to invalidate all the TLB registers of the device.

Just deleting that assert is enough for the VM to start and communicate using IOMMU, but maybe a better alternative is possible. We could move it to the caller functions in other cases than IOMMU invalidation, or make it conditional only if not invalidating.

Any comment would be appreciated. Thanks!

--

Guest kernel version: kernel-3.10.0-1151.el7.x86_64

Command line:
/home/qemu/x86_64-softmmu/qemu-system-x86_64 -name \
guest=rhel7-test,debug-threads=on -machine \
pc-q35-5.1,accel=kvm,usb=off,dump-guest-core=off,kernel_irqchip=split \
-cpu \
Broadwell,vme=on,ss=on,vmx=on,f16c=on,rdrand=on,hypervisor=on,arat=on,tsc-adjust=on,umip=on,arch-capabilities=on,xsaveopt=on,pdpe1gb=on,abm=on,skip-l1dfl-vmentry=on,rtm=on,hle=on \
-m 8096 -realtime mlock=off -smp 2,sockets=2,cores=1,threads=1 -uuid \
d022ecbf-679e-4755-87ce-eb87fc5bbc5d -display none -no-user-config \
-nodefaults -rtc base=utc,driftfix=slew -global \
kvm-pit.lost_tick_policy=delay -no-hpet -no-shutdown -global \
ICH9-LPC.disable_s3=1 -global ICH9-LPC.disable_s4=1 -boot strict=on \
-device intel-iommu,intremap=on,device-iotlb=on -device \
pcie-root-port,port=0x8,chassis=1,id=pci.1,bus=pcie.0,multifunction=on,addr=0x1 \
-device \
pcie-root-port,port=0x9,chassis=2,id=pci.2,bus=pcie.0,addr=0x1.0x1 \
-device \
pcie-root-port,port=0xa,chassis=3,id=pci.3,bus=pcie.0,addr=0x1.0x2 \
-device \
pcie-root-port,port=0xb,chassis=4,id=pci.4,bus=pcie.0,addr=0x1.0x3 \
-device \
pcie-root-port,port=0xc,chassis=5,id=pci.5,bus=pcie.0,addr=0x1.0x4 \
-device \
pcie-root-port,port=0xd,chassis=6,id=pci.6,bus=pcie.0,addr=0x1.0x5 \
-device \
pcie-root-port,port=0xe,chassis=7,id=pci.7,bus=pcie.0,addr=0x1.0x6 \
-device qemu-xhci,p2=15,p3=15,id=usb,bus=pci.2,addr=0x0 -device \
virtio-serial-pci,id=virtio-serial0,bus=pci.3,addr=0x0 -drive \
file=/home/virtio-test2.qcow2,format=qcow2,if=none,id=drive-virtio-disk0 \
-device \
virtio-blk-pci,scsi=off,bus=pci.4,addr=0x0,drive=drive-virtio-disk0,id=virtio-disk0,bootindex=1 \
-netdev tap,id=hostnet0,vhost=on,vhostforce=on -device \
virtio-net-pci,netdev=hostnet0,id=net0,mac=52:54:00:0d:1d:f2,bus=pci.1,addr=0x0,iommu_platform=on,ats=on \
-device virtio-balloon-pci,id=balloon0,bus=pci.5,addr=0x0 -object \
rng-random,id=objrng0,filename=/dev/urandom -device \
virtio-rng-pci,rng=objrng0,id=rng0,bus=pci.6,addr=0x0 -s -msg \
timestamp=on

Full backtrace:

#0 0x00007ffff521370f in raise () at /lib64/libc.so.6
#1 0x00007ffff51fdb25 in abort () at /lib64/libc.so.6
#2 0x00007ffff51fd9f9 in _nl_load_domain.cold.0 () at /lib64/libc.so.6
#3 0x00007ffff520bcc6 in .annobin_assert.c_end () at /lib64/libc.so.6
#4 0x0000555555888171 in memory_region_notify_one (notifier=0x7ffde05dfde8, entry=0x7ffde5dfe200) at /home/qemu/memory.c:1918 #5 0x0000555555888247 in memory_region_notify_iommu (iommu_mr=0x555556f6c0b0, iommu_idx=0, entry=...) at /home/qemu/memory.c:1941 #6 0x0000555555951c8d in vtd_process_device_iotlb_desc (s=0x555557609000, inv_desc=0x7ffde5dfe2d0)

    at /home/qemu/hw/i386/intel_iommu.c:2468

#7 0x0000555555951e6a in vtd_process_inv_desc (s=0x555557609000) at /home/qemu/hw/i386/intel_iommu.c:2531
#8 0x0000555555951fa5 in vtd_fetch_inv_desc (s=0x555557609000) at /home/qemu/hw/i386/intel_iommu.c:2563
#9 0x00005555559520e5 in vtd_handle_iqt_write (s=0x555557609000) at /home/qemu/hw/i386/intel_iommu.c:2590
#10 0x0000555555952b45 in vtd_mem_write (opaque=0x555557609000, addr=136, val=2688, size=4) at /home/qemu/hw/i386/intel_iommu.c:2837 #11 0x0000555555883e17 in memory_region_write_accessor

    (mr=0x555557609330, addr=136, value=0x7ffde5dfe478, size=4, shift=0, mask=4294967295, attrs=...) at /home/qemu/memory.c:483
#12 0x000055555588401d in access_with_adjusted_size
    (addr=136, value=0x7ffde5dfe478, size=4, access_size_min=4, access_size_max=8, access_fn=
    0x555555883d38 <memory_region_write_accessor>, mr=0x555557609330, attrs=...) at /home/qemu/memory.c:544
#13 0x0000555555886f37 in memory_region_dispatch_write (mr=0x555557609330, addr=136, data=2688, op=MO_32, attrs=...)
    at /home/qemu/memory.c:1476
#14 0x0000555555827a03 in flatview_write_continue
    (fv=0x7ffde00935d0, addr=4275634312, attrs=..., ptr=0x7ffff7ff0028, len=4, addr1=136, l=4, mr=0x555557609330) at /home/qemu/exec.c:3146
#15 0x0000555555827b48 in flatview_write (fv=0x7ffde00935d0, addr=4275634312, attrs=..., buf=0x7ffff7ff0028, len=4)
    at /home/qemu/exec.c:3186
#16 0x0000555555827e9d in address_space_write
    (as=0x5555567ca640 <address_space_memory>, addr=4275634312, attrs=..., buf=0x7ffff7ff0028, len=4) at /home/qemu/exec.c:3277
#17 0x0000555555827f0a in address_space_rw
    (as=0x5555567ca640 <address_space_memory>, addr=4275634312, attrs=..., buf=0x7ffff7ff0028, len=4, is_write=true)
    at /home/qemu/exec.c:3287

#18 0x000055555589b633 in kvm_cpu_exec (cpu=0x555556b65640) at /home/qemu/accel/kvm/kvm-all.c:2511 #19 0x0000555555876ba8 in qemu_kvm_cpu_thread_fn (arg=0x555556b65640) at /home/qemu/cpus.c:1284 #20 0x0000555555dafff1 in qemu_thread_start (args=0x555556b8c3b0) at util/qemu-thread-posix.c:521 #21 0x00007ffff55a62de in start_thread () at /lib64/libpthread.so.0
#22 0x00007ffff52d7e83 in clone () at /lib64/libc.so.6

--

Eugenio Pérez (1):
memory: Delete assertion in memory_region_unregister_iommu_notifier

 memory.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
Signature · 1 line
2.18.1
#1 lore raw

RFC 1/1 memory: Delete assertion in memory_region_unregister_iommu_notifier

Bug reference: https://bugs.launchpad.net/qemu/+bug/1885175

It is possible to hit this assertion on rhel7 guests if iommu is properly enabled.

---
 memory.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
memory.c +1 -1
index 2f15a4b250..4db609b843 100644
@@ -1915,7 +1915,7 @@ void memory_region_notify_one(IOMMUNotifier *notifier,
         return;
     }
 
-    assert(entry->iova >= notifier->start && entry_end <= notifier->end);
+    // assert(entry->iova >= notifier->start && entry_end <= notifier->end);
 
     if (entry->perm & IOMMU_RW) {
         request_flags = IOMMU_NOTIFIER_MAP;
Signature · 1 line
2.18.1
#2 lore raw
In reply to Eugenio Pérez
Patchew URL: https://patchew.org/QEMU/[email protected]/

Patchew URL: https://patchew.org/QEMU/[email protected]/

Hi,

This series seems to have some coding style problems. See output below for more information:

Subject: [RFC 0/1] memory: Delete assertion in memory_region_unregister_iommu_notifier
Type: series
Message-id: [email protected]

=== TEST SCRIPT BEGIN ===
#!/bin/bash
git rev-parse base > /dev/null || exit 0
git config --local diff.renamelimit 0
git config --local diff.renames True
git config --local diff.algorithm histogram
./scripts/checkpatch.pl --mailback base..
=== TEST SCRIPT END ===

From https://github.com/patchew-project/qemu
* [new tag] patchew/[email protected] -> patchew/[email protected]
Switched to a new branch 'test'
3dcb120 memory: Delete assertion in memory_region_unregister_iommu_notifier

=== OUTPUT BEGIN ===
ERROR: do not use C99 // comments
#24: FILE: memory.c:1918:
+ // assert(entry->iova >= notifier->start && entry_end <= notifier->end);

total: 1 errors, 0 warnings, 8 lines checked

Commit 3dcb1204223a (memory: Delete assertion in memory_region_unregister_iommu_notifier) has style problems, please review. If any of these errors are false positives report them to the maintainer, see
CHECKPATCH in MAINTAINERS.
=== OUTPUT END ===

Test command exited with code: 1

The full log is available at
http://patchew.org/logs/[email protected]/testing.checkpatch/?type=message.
---
Email generated automatically by Patchew [https://patchew.org/].
Please send your feedback to [email protected]

#4 lore raw

RFC v6 1/13 memory: Rename memory_region_notify_one to memory_region_notify_iommu_one

In reply to Eugenio Pérez

Previous name didn't reflect the iommu operation.

---
 hw/arm/smmu-common.c  | 2 +-
 hw/arm/smmuv3.c       | 2 +-
 hw/i386/intel_iommu.c | 4 ++--
 include/exec/memory.h | 6 +++---
 softmmu/memory.c      | 6 +++---
 5 files changed, 10 insertions(+), 10 deletions(-)
hw/arm/smmu-common.c +1 -1
index e13a5f4a7c..b02ffb8822 100644
@@ -396,7 +396,7 @@ static void smmu_unmap_notifier_range(IOMMUNotifier *n)
     entry.perm = IOMMU_NONE;
     entry.addr_mask = n->end - n->start;
 
-    memory_region_notify_one(n, &entry);
+    memory_region_notify_iommu_one(n, &entry);
 }
 
 /* Unmap all notifiers attached to @mr */
hw/arm/smmuv3.c +1 -1
index 57a79df55b..3bb85ab7e1 100644
@@ -838,7 +838,7 @@ static void smmuv3_notify_iova(IOMMUMemoryRegion *mr,
     entry.addr_mask = (1 << tt->granule_sz) - 1;
     entry.perm = IOMMU_NONE;
 
-    memory_region_notify_one(n, &entry);
+    memory_region_notify_iommu_one(n, &entry);
 }
 
 /* invalidate an asid/iova tuple in all mr's */
hw/i386/intel_iommu.c +2 -2
index 5284bb68b6..2ad6b9d796 100644
@@ -3498,7 +3498,7 @@ static void vtd_address_space_unmap(VTDAddressSpace *as, IOMMUNotifier *n)
         /* This field is meaningless for unmap */
         entry.translated_addr = 0;
 
-        memory_region_notify_one(n, &entry);
+        memory_region_notify_iommu_one(n, &entry);
 
         start += mask;
         remain -= mask;
@@ -3536,7 +3536,7 @@ static void vtd_address_space_refresh_all(IntelIOMMUState *s)
 
 static int vtd_replay_hook(IOMMUTLBEntry *entry, void *private)
 {
-    memory_region_notify_one((IOMMUNotifier *)private, entry);
+    memory_region_notify_iommu_one((IOMMUNotifier *)private, entry);
     return 0;
 }
 
include/exec/memory.h +3 -3
index 0cfe987ab4..22c5f564d1 100644
@@ -226,7 +226,7 @@ enum IOMMUMemoryRegionAttr {
  * The IOMMU implementation must use the IOMMU notifier infrastructure
  * to report whenever mappings are changed, by calling
  * memory_region_notify_iommu() (or, if necessary, by calling
- * memory_region_notify_one() for each registered notifier).
+ * memory_region_notify_iommu_one() for each registered notifier).
  *
  * Conceptually an IOMMU provides a mapping from input address
  * to an output TLB entry. If the IOMMU is aware of memory transaction
@@ -1274,7 +1274,7 @@ void memory_region_notify_iommu(IOMMUMemoryRegion *iommu_mr,
                                 IOMMUTLBEntry entry);
 
 /**
- * memory_region_notify_one: notify a change in an IOMMU translation
+ * memory_region_notify_iommu_one: notify a change in an IOMMU translation
  *                           entry to a single notifier
  *
  * This works just like memory_region_notify_iommu(), but it only
@@ -1285,7 +1285,7 @@ void memory_region_notify_iommu(IOMMUMemoryRegion *iommu_mr,
  *         replaces all old entries for the same virtual I/O address range.
  *         Deleted entries have .@perm == 0.
  */
-void memory_region_notify_one(IOMMUNotifier *notifier,
+void memory_region_notify_iommu_one(IOMMUNotifier *notifier,
                               IOMMUTLBEntry *entry);
 
 /**
softmmu/memory.c +3 -3
index 70b93104e8..961c25b42f 100644
@@ -1890,8 +1890,8 @@ void memory_region_unregister_iommu_notifier(MemoryRegion *mr,
     memory_region_update_iommu_notify_flags(iommu_mr, NULL);
 }
 
-void memory_region_notify_one(IOMMUNotifier *notifier,
-                              IOMMUTLBEntry *entry)
+void memory_region_notify_iommu_one(IOMMUNotifier *notifier,
+                                    IOMMUTLBEntry *entry)
 {
     IOMMUNotifierFlag request_flags;
     hwaddr entry_end = entry->iova + entry->addr_mask;
@@ -1927,7 +1927,7 @@ void memory_region_notify_iommu(IOMMUMemoryRegion *iommu_mr,
 
     IOMMU_NOTIFIER_FOREACH(iommu_notifier, iommu_mr) {
         if (iommu_notifier->iommu_idx == iommu_idx) {
-            memory_region_notify_one(iommu_notifier, &entry);
+            memory_region_notify_iommu_one(iommu_notifier, &entry);
         }
     }
 }
Signature · 1 line
2.18.1
#5 lore raw

RFC v6 2/13 memory: Add IOMMUTLBNotificationType to IOMMUTLBEntry

In reply to Eugenio Pérez

This way we can tell between MAPs and UNMAP, and potentially avoid to send them to a notifier that does not require them.

---
 include/exec/memory.h | 17 ++++++++++++-----
 1 file changed, 12 insertions(+), 5 deletions(-)
include/exec/memory.h +12 -5
index 22c5f564d1..f6d91c54aa 100644
@@ -59,6 +59,12 @@ struct ReservedRegion {
 
 typedef struct IOMMUTLBEntry IOMMUTLBEntry;
 
+typedef enum {
+    IOMMU_IOTLB_NONE  = 0,
+    IOMMU_IOTLB_UNMAP = 1,
+    IOMMU_IOTLB_MAP   = 2,
+} IOMMUTLBNotificationType;
+
 /* See address_space_translate: bit 0 is read, bit 1 is write.  */
 typedef enum {
     IOMMU_NONE = 0,
@@ -70,11 +76,12 @@ typedef enum {
 #define IOMMU_ACCESS_FLAG(r, w) (((r) ? IOMMU_RO : 0) | ((w) ? IOMMU_WO : 0))
 
 struct IOMMUTLBEntry {
-    AddressSpace    *target_as;
-    hwaddr           iova;
-    hwaddr           translated_addr;
-    hwaddr           addr_mask;  /* 0xfff = 4k translation */
-    IOMMUAccessFlags perm;
+    AddressSpace            *target_as;
+    hwaddr                   iova;
+    hwaddr                   translated_addr;
+    hwaddr                   addr_mask;  /* 0xfff = 4k translation */
+    IOMMUAccessFlags         perm;
+    IOMMUTLBNotificationType type; /* Only valid if it is a notification */
 };
 
 /*
Signature · 1 line
2.18.1
#3 lore raw

v6 cover memory: Delete assertion in memory_region_unregister_iommu_notifier

In reply to Eugenio Pérez

I am able to hit this assertion when a Red Hat 7 guest virtio_net device raises an "Invalidation" of all the TLB entries. This happens in the guest's startup if 'intel_iommu=on' argument is passed to the guest kernel and right IOMMU/ATS devices are declared in qemu's command line.

Command line:
/home/qemu/x86_64-softmmu/qemu-system-x86_64 -name \
guest=rhel7-test,debug-threads=on -machine \
pc-q35-5.1,accel=kvm,usb=off,dump-guest-core=off,kernel_irqchip=split \
-cpu \
Broadwell,vme=on,ss=on,vmx=on,f16c=on,rdrand=on,hypervisor=on,arat=on,tsc-adjust=on,umip=on,arch-capabilities=on,xsaveopt=on,pdpe1gb=on,abm=on,skip-l1dfl-vmentry=on,rtm=on,hle=on \
-m 8096 -realtime mlock=off -smp 2,sockets=2,cores=1,threads=1 -uuid \
d022ecbf-679e-4755-87ce-eb87fc5bbc5d -display none -no-user-config \
-nodefaults -rtc base=utc,driftfix=slew -global \
kvm-pit.lost_tick_policy=delay -no-hpet -no-shutdown -global \
ICH9-LPC.disable_s3=1 -global ICH9-LPC.disable_s4=1 -boot strict=on \
-device intel-iommu,intremap=on,device-iotlb=on -device \
pcie-root-port,port=0x8,chassis=1,id=pci.1,bus=pcie.0,multifunction=on,addr=0x1 \
-device \
pcie-root-port,port=0x9,chassis=2,id=pci.2,bus=pcie.0,addr=0x1.0x1 \
-device \
pcie-root-port,port=0xa,chassis=3,id=pci.3,bus=pcie.0,addr=0x1.0x2 \
-device \
pcie-root-port,port=0xb,chassis=4,id=pci.4,bus=pcie.0,addr=0x1.0x3 \
-device \
pcie-root-port,port=0xc,chassis=5,id=pci.5,bus=pcie.0,addr=0x1.0x4 \
-device \
pcie-root-port,port=0xd,chassis=6,id=pci.6,bus=pcie.0,addr=0x1.0x5 \
-device \
pcie-root-port,port=0xe,chassis=7,id=pci.7,bus=pcie.0,addr=0x1.0x6 \
-device qemu-xhci,p2=15,p3=15,id=usb,bus=pci.2,addr=0x0 -device \
virtio-serial-pci,id=virtio-serial0,bus=pci.3,addr=0x0 -drive \
file=/home/virtio-test2.qcow2,format=qcow2,if=none,id=drive-virtio-disk0 \
-device \
virtio-blk-pci,scsi=off,bus=pci.4,addr=0x0,drive=drive-virtio-disk0,id=virtio-disk0,bootindex=1 \
-netdev tap,id=hostnet0,vhost=on,vhostforce=on -device \
virtio-net-pci,netdev=hostnet0,id=net0,mac=52:54:00:0d:1d:f2,bus=pci.1,addr=0x0,iommu_platform=on,ats=on \
-device virtio-balloon-pci,id=balloon0,bus=pci.5,addr=0x0 -object \
rng-random,id=objrng0,filename=/dev/urandom -device \
virtio-rng-pci,rng=objrng0,id=rng0,bus=pci.6,addr=0x0 -s -msg \
timestamp=on

Full backtrace:

#0  0x00007ffff521370f in raise () at /lib64/libc.so.6
#1  0x00007ffff51fdb25 in abort () at /lib64/libc.so.6
#2  0x00007ffff51fd9f9 in _nl_load_domain.cold.0 () at /lib64/libc.so.6
#3  0x00007ffff520bcc6 in .annobin_assert.c_end () at /lib64/libc.so.6
#4  0x0000555555888171 in memory_region_notify_one (notifier=0x7ffde0487fa8,
                                                    entry=0x7ffde5dfe200)
                          at /home/qemu/memory.c:1918
#5  0x0000555555888247 in memory_region_notify_iommu (iommu_mr=0x555556f6c0b0,
                                                      iommu_idx=0, entry=...)
                          at /home/qemu/memory.c:1941
#6  0x0000555555951c8d in vtd_process_device_iotlb_desc (s=0x555557609000,
                                                       inv_desc=0x7ffde5dfe2d0)
                          at /home/qemu/hw/i386/intel_iommu.c:2468
#7  0x0000555555951e6a in vtd_process_inv_desc (s=0x555557609000)
                          at /home/qemu/hw/i386/intel_iommu.c:2531
#8  0x0000555555951fa5 in vtd_fetch_inv_desc (s=0x555557609000)
                          at /home/qemu/hw/i386/intel_iommu.c:2563
#9  0x00005555559520e5 in vtd_handle_iqt_write (s=0x555557609000)
                          at /home/qemu/hw/i386/intel_iommu.c:2590
#10 0x0000555555952b45 in vtd_mem_write (opaque=0x555557609000, addr=136,
                                         val=2688, size=4)
                          at /home/qemu/hw/i386/intel_iommu.c:2837
#11 0x0000555555883e17 in memory_region_write_accessor (mr=0x555557609330,
                                                        addr=136,
                                                        value=0x7ffde5dfe478,
                                                        size=4,
                                                        shift=0,
                                                        mask=4294967295,
                                                        attrs=...)
                         at /home/qemu/memory.c:483
#12 0x000055555588401d in access_with_adjusted_size (addr=136,
                       value=0x7ffde5dfe478,
                       size=4,
                       access_size_min=4,
                       access_size_max=8,
                       access_fn=0x555555883d38 <memory_region_write_accessor>,
                       mr=0x555557609330,
                       attrs=...)
                       at /home/qemu/memory.c:544
#13 0x0000555555886f37 in memory_region_dispatch_write (mr=0x555557609330,
                                                       addr=136,
                                                       data=2688,
                                                       op=MO_32,
                                                       attrs=...)
                         at /home/qemu/memory.c:1476
#14 0x0000555555827a03 in flatview_write_continue (fv=0x7ffdd8503150,
                                                   addr=4275634312,
                                                   attrs=...,
                                                   ptr=0x7ffff7ff0028,
                                                   len=4,
                                                   addr1=136,
                                                   l=4,
                                                   mr=0x555557609330)
                          at /home/qemu/exec.c:3146
#15 0x0000555555827b48 in flatview_write (fv=0x7ffdd8503150,
                                          addr=4275634312,
                                          attrs=...,
                                          buf=0x7ffff7ff0028,
                                          len=4)
                          at /home/qemu/exec.c:3186
#16 0x0000555555827e9d in address_space_write (
                                      as=0x5555567ca640 <address_space_memory>,
                                      addr=4275634312,
                                      attrs=...,
                                      buf=0x7ffff7ff0028,
                                      len=4)
                          at /home/qemu/exec.c:3277
#17 0x0000555555827f0a in address_space_rw (
                                      as=0x5555567ca640 <address_space_memory>,
                                      addr=4275634312,
                                      attrs=...,
                                      buf=0x7ffff7ff0028,
                                      len=4,
                                      is_write=true)
                          at /home/qemu/exec.c:3287
#18 0x000055555589b633 in kvm_cpu_exec (cpu=0x555556b65640)
                               at /home/qemu/accel/kvm/kvm-all.c:2511
#19 0x0000555555876ba8 in qemu_kvm_cpu_thread_fn (arg=0x555556b65640)
                               at /home/qemu/cpus.c:1284
#20 0x0000555555dafff1 in qemu_thread_start (args=0x555556b8c3b0)
                               at util/qemu-thread-posix.c:521

#21 0x00007ffff55a62de in start_thread () at /lib64/libpthread.so.0 #22 0x00007ffff52d7e83 in clone () at /lib64/libc.so.6

(gdb) frame 4
#4 0x0000555555888171 in memory_region_notify_one

                      (notifier=0x7ffde0487fa8, entry=0x7ffde5dfe200)
                      at /home/qemu/memory.c:1918

1918 assert(entry->iova >= notifier->start && entry_end <=
notifier->end);
(gdb) p *entry
$1 = {target_as = 0x555556f6c050, iova = 0, translated_addr = 0, addr_mask = 18446744073709551615, perm = IOMMU_NONE} --

Tested with vhost-net, host<->guest communication. Still more testing needed, since it touches a few architectures and configurations.

v6: Introduce "type" field for IOMMUTLBEntry. Fill in all uses.
Update tests reports with more fine-tuning (CPU, RPS/XPS tunning).

v5: Skip regular IOTLB notifications in dev_iotlb notifiers

v4: Rename IOMMU_NOTIFIER_IOTLB -> IOMMU_NOTIFIER_DEVIOTLB.
Make vhost-net notifier just IOMMU_NOTIFIER_DEVIOTLB, not IOMMU_NOTIFIER_UNMAP

v3: Skip the assertion in case notifier is a IOTLB one, since they can manage
arbitrary ranges. Using a flag in the notifier for now, as Peter suggested.

v2: Actually delete assertion instead of just commenting out using C99

Eugenio Pérez (13):
memory: Rename memory_region_notify_one to
memory_region_notify_iommu_one
memory: Add IOMMUTLBNotificationType to IOMMUTLBEntry hw/alpha/typhoon: Mark all IOMMUTLBEntry as IOMMU_IOTLB_NONE type
amd_iommu: Mark all IOMMUTLBEntry as IOMMU_IOTLB_NONE type hw/arm/smmu: Fill IOMMUTLBEntry notifier type dma/rc4030: Mark all IOMMUTLBEntry as IOMMU_IOTLB_NONE type
intel_iommu: Mark IOMMUTLBEntry of page notification as
IOMMU_IOTLB_UNMAP type
virtio-iommu: Mark virtio_iommu_translate IOTLB as IOMMU_IOTLB_NONE
type
intel_iommu: Set IOMMUTLBEntry type in vtd_page_walk_level
memory: Notify IOMMU IOTLB based on entry type, not permissions
memory: Add IOMMU_DEVIOTLB_UNMAP IOMMUTLBNotificationType
intel_iommu: Do not notify regular iotlb to device-iotlb notifiers
memory: Skip bad range assertion if notifier is DEVIOTLB type

 hw/alpha/typhoon.c       |  1 +
 hw/arm/smmu-common.c     |  4 +++-
 hw/arm/smmuv3.c          |  4 +++-
 hw/dma/rc4030.c          |  1 +
 hw/i386/amd_iommu.c      |  7 ++++++-
 hw/i386/intel_iommu.c    | 14 +++++++++++--
 hw/virtio/vhost.c        |  2 +-
 hw/virtio/virtio-iommu.c |  1 +
 include/exec/memory.h    | 26 ++++++++++++++++-------
 softmmu/memory.c         | 45 ++++++++++++++++++++++++++++++----------
 10 files changed, 80 insertions(+), 25 deletions(-)
Signature · 1 line
2.18.1
#10 lore raw

RFC v6 4/13 amd_iommu: Mark all IOMMUTLBEntry as IOMMU_IOTLB_NONE type

---
 hw/i386/amd_iommu.c | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)
hw/i386/amd_iommu.c +6 -1
index 087f601666..c2607e9e91 100644
@@ -946,6 +946,7 @@ static void amdvi_page_walk(AMDVIAddressSpace *as, uint64_t *dte,
         ret->translated_addr = (pte & AMDVI_DEV_PT_ROOT_MASK) & page_mask;
         ret->addr_mask = ~page_mask;
         ret->perm = amdvi_get_perms(pte);
+        ret->type = IOMMU_IOTLB_NONE;
         return;
     }
 no_remap:
@@ -953,6 +954,7 @@ no_remap:
     ret->translated_addr = addr & AMDVI_PAGE_MASK_4K;
     ret->addr_mask = ~AMDVI_PAGE_MASK_4K;
     ret->perm = amdvi_get_perms(pte);
+    ret->type = IOMMU_IOTLB_NONE;
 }
 
 static void amdvi_do_translate(AMDVIAddressSpace *as, hwaddr addr,
@@ -970,6 +972,7 @@ static void amdvi_do_translate(AMDVIAddressSpace *as, hwaddr addr,
         ret->translated_addr = iotlb_entry->translated_addr;
         ret->addr_mask = iotlb_entry->page_mask;
         ret->perm = iotlb_entry->perms;
+        ret->type = IOMMU_IOTLB_NONE;
         return;
     }
 
@@ -994,6 +997,7 @@ out:
     ret->translated_addr = addr & AMDVI_PAGE_MASK_4K;
     ret->addr_mask = ~AMDVI_PAGE_MASK_4K;
     ret->perm = IOMMU_RW;
+    ret->type = IOMMU_IOTLB_NONE;
 }
 
 static inline bool amdvi_is_interrupt_addr(hwaddr addr)
@@ -1011,7 +1015,8 @@ static IOMMUTLBEntry amdvi_translate(IOMMUMemoryRegion *iommu, hwaddr addr,
         .iova = addr,
         .translated_addr = 0,
         .addr_mask = ~(hwaddr)0,
-        .perm = IOMMU_NONE
+        .perm = IOMMU_NONE,
+        .type = IOMMU_IOTLB_NONE
     };
 
     if (!s->enabled) {
Signature · 1 line
2.18.1
#11 lore raw

RFC v6 5/13 hw/arm/smmu: Fill IOMMUTLBEntry notifier type

In reply to Eugenio Pérez
---
 hw/arm/smmu-common.c | 2 ++
 hw/arm/smmuv3.c      | 2 ++
 2 files changed, 4 insertions(+)
hw/arm/smmu-common.c +2 -0
index b02ffb8822..88cf1b86ea 100644
@@ -181,6 +181,7 @@ static int smmu_ptw_64(SMMUTransCfg *cfg,
 
     tlbe->iova = iova;
     tlbe->addr_mask = (1 << granule_sz) - 1;
+    tlbe->type = IOMMU_IOTLB_NONE;
 
     while (level <= 3) {
         uint64_t subpage_size = 1ULL << level_shift(level, granule_sz);
@@ -395,6 +396,7 @@ static void smmu_unmap_notifier_range(IOMMUNotifier *n)
     entry.iova = n->start;
     entry.perm = IOMMU_NONE;
     entry.addr_mask = n->end - n->start;
+    entry.type = IOMMU_IOTLB_UNMAP,
 
     memory_region_notify_iommu_one(n, &entry);
 }
hw/arm/smmuv3.c +2 -0
index 3bb85ab7e1..dee987b2b1 100644
@@ -635,6 +635,7 @@ static IOMMUTLBEntry smmuv3_translate(IOMMUMemoryRegion *mr, hwaddr addr,
         .translated_addr = addr,
         .addr_mask = ~(hwaddr)0,
         .perm = IOMMU_NONE,
+        .type = IOMMU_IOTLB_NONE,
     };
     SMMUIOTLBKey key, *new_key;
 
@@ -837,6 +838,7 @@ static void smmuv3_notify_iova(IOMMUMemoryRegion *mr,
     entry.iova = iova;
     entry.addr_mask = (1 << tt->granule_sz) - 1;
     entry.perm = IOMMU_NONE;
+    entry.type = IOMMU_NOTIFIER_UNMAP;
 
     memory_region_notify_iommu_one(n, &entry);
 }
Signature · 1 line
2.18.1
#8 lore raw

RFC v6 3/13 hw/alpha/typhoon: Mark all IOMMUTLBEntry as IOMMU_IOTLB_NONE type

In reply to Eugenio Pérez
---
 hw/alpha/typhoon.c | 1 +
 1 file changed, 1 insertion(+)
hw/alpha/typhoon.c +1 -0
index 29d44dfb06..b1e6c4e929 100644
@@ -615,6 +615,7 @@ static bool make_iommu_tlbe(hwaddr taddr, hwaddr mask, IOMMUTLBEntry *ret)
         .translated_addr = taddr,
         .addr_mask = mask,
         .perm = IOMMU_RW,
+        .type = IOMMU_IOTLB_NONE,
     };
     return true;
 }
Signature · 1 line
2.18.1
#15 lore raw

RFC v6 9/13 intel_iommu: Set IOMMUTLBEntry type in vtd_page_walk_level

In reply to Eugenio Pérez
---
 hw/i386/intel_iommu.c | 1 +
 1 file changed, 1 insertion(+)
hw/i386/intel_iommu.c +1 -0
index ed83e496b8..0b3399874f 100644
@@ -1251,6 +1251,7 @@ static int vtd_page_walk_level(dma_addr_t addr, uint64_t start,
             entry.addr_mask = ~subpage_mask;
             /* NOTE: this is only meaningful if entry_valid == true */
             entry.translated_addr = vtd_get_slpte_addr(slpte, info->aw);
+            entry.type = entry.perm ? IOMMU_IOTLB_MAP : IOMMU_IOTLB_UNMAP;
             ret = vtd_page_walk_one(&entry, info);
         }
 
Signature · 1 line
2.18.1
#12 lore raw

RFC v6 6/13 dma/rc4030: Mark all IOMMUTLBEntry as IOMMU_IOTLB_NONE type

In reply to Eugenio Pérez
---
 hw/dma/rc4030.c | 1 +
 1 file changed, 1 insertion(+)
hw/dma/rc4030.c +1 -0
index 7eddc9a776..8eee12b1cb 100644
@@ -506,6 +506,7 @@ static IOMMUTLBEntry rc4030_dma_translate(IOMMUMemoryRegion *iommu, hwaddr addr,
         .translated_addr = 0,
         .addr_mask = DMA_PAGESIZE - 1,
         .perm = IOMMU_NONE,
+        .type = DEV_IOTLB_NONE,
     };
     uint64_t i, entry_address;
     dma_pagetable_entry entry;
Signature · 1 line
2.18.1
#16 lore raw

RFC v6 10/13 memory: Notify IOMMU IOTLB based on entry type, not permissions

In reply to Eugenio Pérez

This way the intention is much clearer.

---
 softmmu/memory.c | 27 +++++++++++++++++++--------
 1 file changed, 19 insertions(+), 8 deletions(-)
softmmu/memory.c +19 -8
index 961c25b42f..3e68442ca6 100644
@@ -1890,10 +1890,27 @@ void memory_region_unregister_iommu_notifier(MemoryRegion *mr,
     memory_region_update_iommu_notify_flags(iommu_mr, NULL);
 }
 
+static IOMMUNotifierFlag notifier_type_iommu(const IOMMUNotifier *notifier)
+{
+    return notifier->notifier_flags & IOMMU_NOTIFIER_ALL;
+}
+
+static bool memory_region_notify(const IOMMUNotifier *notifier,
+                                 const IOMMUTLBEntry *entry)
+{
+    switch(entry->type) {
+    case IOMMU_IOTLB_MAP:
+        return notifier_type_iommu(notifier) == IOMMU_NOTIFIER_MAP;
+    case IOMMU_IOTLB_UNMAP:
+        return notifier_type_iommu(notifier) == IOMMU_NOTIFIER_UNMAP;
+    default:
+        return false;
+    };
+}
+
 void memory_region_notify_iommu_one(IOMMUNotifier *notifier,
                                     IOMMUTLBEntry *entry)
 {
-    IOMMUNotifierFlag request_flags;
     hwaddr entry_end = entry->iova + entry->addr_mask;
 
     /*
@@ -1906,13 +1923,7 @@ void memory_region_notify_iommu_one(IOMMUNotifier *notifier,
 
     assert(entry->iova >= notifier->start && entry_end <= notifier->end);
 
-    if (entry->perm & IOMMU_RW) {
-        request_flags = IOMMU_NOTIFIER_MAP;
-    } else {
-        request_flags = IOMMU_NOTIFIER_UNMAP;
-    }
-
-    if (notifier->notifier_flags & request_flags) {
+    if (memory_region_notify(notifier, entry)) {
         notifier->notify(notifier, entry);
     }
 }
Signature · 1 line
2.18.1
#14 lore raw

RFC v6 8/13 virtio-iommu: Mark virtio_iommu_translate IOTLB as IOMMU_IOTLB_NONE type

In reply to Eugenio Pérez
---
 hw/virtio/virtio-iommu.c | 1 +
 1 file changed, 1 insertion(+)
hw/virtio/virtio-iommu.c +1 -0
index 5d56865e56..16267e8096 100644
@@ -619,6 +619,7 @@ static IOMMUTLBEntry virtio_iommu_translate(IOMMUMemoryRegion *mr, hwaddr addr,
         .translated_addr = addr,
         .addr_mask = (1 << ctz32(s->config.page_size_mask)) - 1,
         .perm = IOMMU_NONE,
+        .type = IOMMU_IOTLB_NONE,
     };
 
     bypass_allowed = virtio_vdev_has_feature(&s->parent_obj,
Signature · 1 line
2.18.1
#17 lore raw

RFC v6 11/13 memory: Add IOMMU_DEVIOTLB_UNMAP IOMMUTLBNotificationType

In reply to Eugenio Pérez

Adapt intel and vhost to use this new notification type

---
 hw/i386/intel_iommu.c | 2 +-
 hw/virtio/vhost.c     | 2 +-
 include/exec/memory.h | 9 ++++++---
 softmmu/memory.c      | 5 ++++-
 4 files changed, 12 insertions(+), 6 deletions(-)
hw/i386/intel_iommu.c +1 -1
index 0b3399874f..ddb828da1f 100644
@@ -2467,7 +2467,7 @@ static bool vtd_process_device_iotlb_desc(IntelIOMMUState *s,
     entry.iova = addr;
     entry.perm = IOMMU_NONE;
     entry.translated_addr = 0;
-    entry.type = IOMMU_IOTLB_UNMAP;
+    entry.type = IOMMU_DEVIOTLB_UNMAP;
     memory_region_notify_iommu(&vtd_dev_as->iommu, 0, entry);
 
 done:
hw/virtio/vhost.c +1 -1
index 1a1384e7a6..6ca168b47e 100644
@@ -729,7 +729,7 @@ static void vhost_iommu_region_add(MemoryListener *listener,
     iommu_idx = memory_region_iommu_attrs_to_index(iommu_mr,
                                                    MEMTXATTRS_UNSPECIFIED);
     iommu_notifier_init(&iommu->n, vhost_iommu_unmap_notify,
-                        IOMMU_NOTIFIER_UNMAP,
+                        IOMMU_NOTIFIER_DEVIOTLB,
                         section->offset_within_region,
                         int128_get64(end),
                         iommu_idx);
include/exec/memory.h +6 -3
index f6d91c54aa..477c3af24c 100644
@@ -60,9 +60,10 @@ struct ReservedRegion {
 typedef struct IOMMUTLBEntry IOMMUTLBEntry;
 
 typedef enum {
-    IOMMU_IOTLB_NONE  = 0,
-    IOMMU_IOTLB_UNMAP = 1,
-    IOMMU_IOTLB_MAP   = 2,
+    IOMMU_IOTLB_NONE     = 0,
+    IOMMU_IOTLB_UNMAP    = 1,
+    IOMMU_IOTLB_MAP      = 2,
+    IOMMU_DEVIOTLB_UNMAP = 3,
 } IOMMUTLBNotificationType;
 
 /* See address_space_translate: bit 0 is read, bit 1 is write.  */
@@ -94,6 +95,8 @@ typedef enum {
     IOMMU_NOTIFIER_UNMAP = 0x1,
     /* Notify entry changes (newly created entries) */
     IOMMU_NOTIFIER_MAP = 0x2,
+    /* Notify changes on device IOTLB entries */
+    IOMMU_NOTIFIER_DEVIOTLB = 0x04,
 } IOMMUNotifierFlag;
 
 #define IOMMU_NOTIFIER_ALL (IOMMU_NOTIFIER_MAP | IOMMU_NOTIFIER_UNMAP)
softmmu/memory.c +4 -1
index 3e68442ca6..4ed63f4d0d 100644
@@ -1892,7 +1892,8 @@ void memory_region_unregister_iommu_notifier(MemoryRegion *mr,
 
 static IOMMUNotifierFlag notifier_type_iommu(const IOMMUNotifier *notifier)
 {
-    return notifier->notifier_flags & IOMMU_NOTIFIER_ALL;
+    return notifier->notifier_flags &
+           (IOMMU_NOTIFIER_ALL | IOMMU_NOTIFIER_DEVIOTLB);
 }
 
 static bool memory_region_notify(const IOMMUNotifier *notifier,
@@ -1903,6 +1904,8 @@ static bool memory_region_notify(const IOMMUNotifier *notifier,
         return notifier_type_iommu(notifier) == IOMMU_NOTIFIER_MAP;
     case IOMMU_IOTLB_UNMAP:
         return notifier_type_iommu(notifier) == IOMMU_NOTIFIER_UNMAP;
+    case IOMMU_DEVIOTLB_UNMAP:
+        return notifier_type_iommu(notifier) == IOMMU_NOTIFIER_DEVIOTLB;
     default:
         return false;
     };
Signature · 1 line
2.18.1
#13 lore raw

RFC v6 7/13 intel_iommu: Mark IOMMUTLBEntry of page notification as IOMMU_IOTLB_UNMAP type

In reply to Eugenio Pérez
---
 hw/i386/intel_iommu.c | 3 +++
 1 file changed, 3 insertions(+)
hw/i386/intel_iommu.c +3 -0
index 2ad6b9d796..ed83e496b8 100644
@@ -1999,6 +1999,7 @@ static void vtd_iotlb_page_invalidate_notify(IntelIOMMUState *s,
                     .translated_addr = 0,
                     .addr_mask = size - 1,
                     .perm = IOMMU_NONE,
+                    .type = IOMMU_IOTLB_UNMAP,
                 };
                 memory_region_notify_iommu(&vtd_as->iommu, 0, entry);
             }
@@ -2465,6 +2466,7 @@ static bool vtd_process_device_iotlb_desc(IntelIOMMUState *s,
     entry.iova = addr;
     entry.perm = IOMMU_NONE;
     entry.translated_addr = 0;
+    entry.type = IOMMU_IOTLB_UNMAP;
     memory_region_notify_iommu(&vtd_dev_as->iommu, 0, entry);
 
 done:
@@ -3497,6 +3499,7 @@ static void vtd_address_space_unmap(VTDAddressSpace *as, IOMMUNotifier *n)
         entry.perm = IOMMU_NONE;
         /* This field is meaningless for unmap */
         entry.translated_addr = 0;
+        entry.type = IOMMU_NOTIFIER_UNMAP;
 
         memory_region_notify_iommu_one(n, &entry);
 
Signature · 1 line
2.18.1
#18 lore raw

RFC v6 12/13 intel_iommu: Do not notify regular iotlb to device-iotlb notifiers

In reply to Eugenio Pérez

This improves performance in case of netperf with vhost-net:
* TCP_STREAM: From 9049.59Mbit/s to 9049.59Mbit/s (13%)
* TCP_RR: From 8464.73 trans/s to 8932.703333 trans/s (5.5%)
* UDP_RR: From 8562.08 trans/s to 9005.62/s (5.1%)
* UDP_STREAM: No change observed (insignificant 0.1% improvement)

---
 hw/i386/intel_iommu.c | 6 ++++++
 1 file changed, 6 insertions(+)
hw/i386/intel_iommu.c +6 -0
index ddb828da1f..7620a1abbf 100644
@@ -1960,6 +1960,12 @@ static void vtd_iotlb_domain_invalidate(IntelIOMMUState *s, uint16_t domain_id)
     vtd_iommu_unlock(s);
 
     QLIST_FOREACH(vtd_as, &s->vtd_as_with_notifiers, next) {
+        if (vtd_as->iommu.iommu_notify_flags & IOMMU_NOTIFIER_DEVIOTLB) {
+            /* If IOMMU memory region is DEVICE IOTLB type, it does not make
+             * sense to send regular IOMMU notifications. */
+            continue;
+        }
+
         if (!vtd_dev_to_context_entry(s, pci_bus_num(vtd_as->bus),
                                       vtd_as->devfn, &ce) &&
             domain_id == vtd_get_domain_id(s, &ce)) {
Signature · 1 line
2.18.1
#21 lore raw

RFC v6 13/13 memory: Skip bad range assertion if notifier is DEVIOTLB type

In reply to Eugenio Pérez
---
 softmmu/memory.c | 13 +++++++++++--
 1 file changed, 11 insertions(+), 2 deletions(-)
softmmu/memory.c +11 -2
index 4ed63f4d0d..d2797e996a 100644
@@ -1915,6 +1915,7 @@ void memory_region_notify_iommu_one(IOMMUNotifier *notifier,
                                     IOMMUTLBEntry *entry)
 {
     hwaddr entry_end = entry->iova + entry->addr_mask;
+    IOMMUTLBEntry tmp = *entry;
 
     /*
      * Skip the notification if the notification does not overlap
@@ -1924,10 +1925,18 @@ void memory_region_notify_iommu_one(IOMMUNotifier *notifier,
         return;
     }
 
-    assert(entry->iova >= notifier->start && entry_end <= notifier->end);
+    if (notifier->notifier_flags & IOMMU_NOTIFIER_DEVIOTLB) {
+        /* Crop (iova, addr_mask) to range */
+        tmp.iova = MAX(tmp.iova, notifier->start);
+        tmp.addr_mask = MIN(entry_end, notifier->end) - tmp.iova;
+        /* Confirm no underflow */
+        assert(MIN(entry_end, notifier->end) >= tmp.iova);
+    } else {
+        assert(entry->iova >= notifier->start && entry_end <= notifier->end);
+    }
 
     if (memory_region_notify(notifier, entry)) {
-        notifier->notify(notifier, entry);
+        notifier->notify(notifier, &tmp);
     }
 }
 
Signature · 1 line
2.18.1
#22 lore raw
In reply to Eugenio Pérez

Hi!

Sending v6 to see if that is on the same page as what you meant. Making each setting of "type" explicitly IOMMU_IOTLB_NONE if not used in notifications. This is done in different commits in case this helps review of different architectures.

I think that this way we have too much freedom between entry flags (currently only access type, RW) and notification type. Since not all of them are valid nor used in the same context, I think this adds complexity. I'm wondering if:

Option a) We could make it private to memory.c, and make it a flag of memory_region_notify_iommu (like "bool deviotlb_type)". IOW, instead of making it a member of IOMMUTLBEntry, wrap the "entry" parameter of memory_region_notify_iommu in a new private structure defined in memory.c that adds that flag.

Option b) We could keep the IOMMUTLBNotificationType enum (open to suggestions for a better name :)), but not embed it in the struct, like:

include/exec/memory.h +11 -3
index 477c3af24c..d9150e7b7e 100644
@@ -72,7 +72,8 @@ typedef enum {
     IOMMU_RO   = 1,
     IOMMU_WO   = 2,
     IOMMU_RW   = 3,
-} IOMMUAccessFlags;
+    IOMMU_DEVIOTLB = 4,
+} IOMMUEntryFlags;
 
 #define IOMMU_ACCESS_FLAG(r, w) (((r) ? IOMMU_RO : 0) | ((w) ? IOMMU_WO : 0))
 
@@ -81,10 +82,18 @@ struct IOMMUTLBEntry {
     hwaddr                   iova;
     hwaddr                   translated_addr;
     hwaddr                   addr_mask;  /* 0xfff = 4k translation */
-    IOMMUAccessFlags         perm;
+    IOMMUEntryFlags          flags;
     IOMMUTLBNotificationType type; /* Only valid if it is a notification */
 };
 
+IOMMUTLBNotificationType iommu_tlb_entry_type(struct IOMMUTLBEntry *s) {
+    if (s->flags & IOMMU_DEVIOTLB)
+        return IOMMU_DEVIOTLB_UNMAP;
+    else if (s->flags & IOMMU_RW)
+        return IOMMU_IOTLB_MAP;
+    return IOMMU_IOTLB_UNMAP;
+}
+
--
 

Thanks!

Quoted 191 lines · Eugenio Pérez

I am able to hit this assertion when a Red Hat 7 guest virtio_net device raises an "Invalidation" of all the TLB entries. This happens in the guest's startup if 'intel_iommu=on' argument is passed to the guest kernel and right IOMMU/ATS devices are declared in qemu's command line.

Command line:
/home/qemu/x86_64-softmmu/qemu-system-x86_64 -name \
guest=rhel7-test,debug-threads=on -machine \
pc-q35-5.1,accel=kvm,usb=off,dump-guest-core=off,kernel_irqchip=split \
-cpu \
Broadwell,vme=on,ss=on,vmx=on,f16c=on,rdrand=on,hypervisor=on,arat=on,tsc-adjust=on,umip=on,arch-capabilities=on,xsaveopt=on,pdpe1gb=on,abm=on,skip-l1dfl-vmentry=on,rtm=on,hle=on \
-m 8096 -realtime mlock=off -smp 2,sockets=2,cores=1,threads=1 -uuid \
d022ecbf-679e-4755-87ce-eb87fc5bbc5d -display none -no-user-config \
-nodefaults -rtc base=utc,driftfix=slew -global \
kvm-pit.lost_tick_policy=delay -no-hpet -no-shutdown -global \
ICH9-LPC.disable_s3=1 -global ICH9-LPC.disable_s4=1 -boot strict=on \
-device intel-iommu,intremap=on,device-iotlb=on -device \
pcie-root-port,port=0x8,chassis=1,id=pci.1,bus=pcie.0,multifunction=on,addr=0x1 \
-device \
pcie-root-port,port=0x9,chassis=2,id=pci.2,bus=pcie.0,addr=0x1.0x1 \
-device \
pcie-root-port,port=0xa,chassis=3,id=pci.3,bus=pcie.0,addr=0x1.0x2 \
-device \
pcie-root-port,port=0xb,chassis=4,id=pci.4,bus=pcie.0,addr=0x1.0x3 \
-device \
pcie-root-port,port=0xc,chassis=5,id=pci.5,bus=pcie.0,addr=0x1.0x4 \
-device \
pcie-root-port,port=0xd,chassis=6,id=pci.6,bus=pcie.0,addr=0x1.0x5 \
-device \
pcie-root-port,port=0xe,chassis=7,id=pci.7,bus=pcie.0,addr=0x1.0x6 \
-device qemu-xhci,p2=15,p3=15,id=usb,bus=pci.2,addr=0x0 -device \
virtio-serial-pci,id=virtio-serial0,bus=pci.3,addr=0x0 -drive \
file=/home/virtio-test2.qcow2,format=qcow2,if=none,id=drive-virtio-disk0 \
-device \
virtio-blk-pci,scsi=off,bus=pci.4,addr=0x0,drive=drive-virtio-disk0,id=virtio-disk0,bootindex=1 \
-netdev tap,id=hostnet0,vhost=on,vhostforce=on -device \
virtio-net-pci,netdev=hostnet0,id=net0,mac=52:54:00:0d:1d:f2,bus=pci.1,addr=0x0,iommu_platform=on,ats=on \
-device virtio-balloon-pci,id=balloon0,bus=pci.5,addr=0x0 -object \
rng-random,id=objrng0,filename=/dev/urandom -device \
virtio-rng-pci,rng=objrng0,id=rng0,bus=pci.6,addr=0x0 -s -msg \
timestamp=on

Full backtrace:

#0  0x00007ffff521370f in raise () at /lib64/libc.so.6
#1  0x00007ffff51fdb25 in abort () at /lib64/libc.so.6
#2  0x00007ffff51fd9f9 in _nl_load_domain.cold.0 () at /lib64/libc.so.6
#3  0x00007ffff520bcc6 in .annobin_assert.c_end () at /lib64/libc.so.6
#4  0x0000555555888171 in memory_region_notify_one (notifier=0x7ffde0487fa8,
                                                    entry=0x7ffde5dfe200)
                          at /home/qemu/memory.c:1918
#5  0x0000555555888247 in memory_region_notify_iommu (iommu_mr=0x555556f6c0b0,
                                                      iommu_idx=0, entry=...)
                          at /home/qemu/memory.c:1941
#6  0x0000555555951c8d in vtd_process_device_iotlb_desc (s=0x555557609000,
                                                       inv_desc=0x7ffde5dfe2d0)
                          at /home/qemu/hw/i386/intel_iommu.c:2468
#7  0x0000555555951e6a in vtd_process_inv_desc (s=0x555557609000)
                          at /home/qemu/hw/i386/intel_iommu.c:2531
#8  0x0000555555951fa5 in vtd_fetch_inv_desc (s=0x555557609000)
                          at /home/qemu/hw/i386/intel_iommu.c:2563
#9  0x00005555559520e5 in vtd_handle_iqt_write (s=0x555557609000)
                          at /home/qemu/hw/i386/intel_iommu.c:2590
#10 0x0000555555952b45 in vtd_mem_write (opaque=0x555557609000, addr=136,
                                         val=2688, size=4)
                          at /home/qemu/hw/i386/intel_iommu.c:2837
#11 0x0000555555883e17 in memory_region_write_accessor (mr=0x555557609330,
                                                        addr=136,
                                                        value=0x7ffde5dfe478,
                                                        size=4,
                                                        shift=0,
                                                        mask=4294967295,
                                                        attrs=...)
                         at /home/qemu/memory.c:483
#12 0x000055555588401d in access_with_adjusted_size (addr=136,
                       value=0x7ffde5dfe478,
                       size=4,
                       access_size_min=4,
                       access_size_max=8,
                       access_fn=0x555555883d38 <memory_region_write_accessor>,
                       mr=0x555557609330,
                       attrs=...)
                       at /home/qemu/memory.c:544
#13 0x0000555555886f37 in memory_region_dispatch_write (mr=0x555557609330,
                                                       addr=136,
                                                       data=2688,
                                                       op=MO_32,
                                                       attrs=...)
                         at /home/qemu/memory.c:1476
#14 0x0000555555827a03 in flatview_write_continue (fv=0x7ffdd8503150,
                                                   addr=4275634312,
                                                   attrs=...,
                                                   ptr=0x7ffff7ff0028,
                                                   len=4,
                                                   addr1=136,
                                                   l=4,
                                                   mr=0x555557609330)
                          at /home/qemu/exec.c:3146
#15 0x0000555555827b48 in flatview_write (fv=0x7ffdd8503150,
                                          addr=4275634312,
                                          attrs=...,
                                          buf=0x7ffff7ff0028,
                                          len=4)
                          at /home/qemu/exec.c:3186
#16 0x0000555555827e9d in address_space_write (
                                      as=0x5555567ca640 <address_space_memory>,
                                      addr=4275634312,
                                      attrs=...,
                                      buf=0x7ffff7ff0028,
                                      len=4)
                          at /home/qemu/exec.c:3277
#17 0x0000555555827f0a in address_space_rw (
                                      as=0x5555567ca640 <address_space_memory>,
                                      addr=4275634312,
                                      attrs=...,
                                      buf=0x7ffff7ff0028,
                                      len=4,
                                      is_write=true)
                          at /home/qemu/exec.c:3287
#18 0x000055555589b633 in kvm_cpu_exec (cpu=0x555556b65640)
                               at /home/qemu/accel/kvm/kvm-all.c:2511
#19 0x0000555555876ba8 in qemu_kvm_cpu_thread_fn (arg=0x555556b65640)
                               at /home/qemu/cpus.c:1284
#20 0x0000555555dafff1 in qemu_thread_start (args=0x555556b8c3b0)
                               at util/qemu-thread-posix.c:521

#21 0x00007ffff55a62de in start_thread () at /lib64/libpthread.so.0 #22 0x00007ffff52d7e83 in clone () at /lib64/libc.so.6

(gdb) frame 4
#4 0x0000555555888171 in memory_region_notify_one

                      (notifier=0x7ffde0487fa8, entry=0x7ffde5dfe200)
                      at /home/qemu/memory.c:1918

1918 assert(entry->iova >= notifier->start && entry_end <=
notifier->end);
(gdb) p *entry
$1 = {target_as = 0x555556f6c050, iova = 0, translated_addr = 0, addr_mask = 18446744073709551615, perm = IOMMU_NONE} --

Tested with vhost-net, host<->guest communication. Still more testing needed, since it touches a few architectures and configurations.

v6: Introduce "type" field for IOMMUTLBEntry. Fill in all uses.
Update tests reports with more fine-tuning (CPU, RPS/XPS tunning).

v5: Skip regular IOTLB notifications in dev_iotlb notifiers

v4: Rename IOMMU_NOTIFIER_IOTLB -> IOMMU_NOTIFIER_DEVIOTLB.
Make vhost-net notifier just IOMMU_NOTIFIER_DEVIOTLB, not IOMMU_NOTIFIER_UNMAP

v3: Skip the assertion in case notifier is a IOTLB one, since they can manage
arbitrary ranges. Using a flag in the notifier for now, as Peter suggested.

v2: Actually delete assertion instead of just commenting out using C99

Eugenio Pérez (13):
memory: Rename memory_region_notify_one to
memory_region_notify_iommu_one
memory: Add IOMMUTLBNotificationType to IOMMUTLBEntry hw/alpha/typhoon: Mark all IOMMUTLBEntry as IOMMU_IOTLB_NONE type
amd_iommu: Mark all IOMMUTLBEntry as IOMMU_IOTLB_NONE type hw/arm/smmu: Fill IOMMUTLBEntry notifier type dma/rc4030: Mark all IOMMUTLBEntry as IOMMU_IOTLB_NONE type
intel_iommu: Mark IOMMUTLBEntry of page notification as
IOMMU_IOTLB_UNMAP type
virtio-iommu: Mark virtio_iommu_translate IOTLB as IOMMU_IOTLB_NONE
type
intel_iommu: Set IOMMUTLBEntry type in vtd_page_walk_level
memory: Notify IOMMU IOTLB based on entry type, not permissions
memory: Add IOMMU_DEVIOTLB_UNMAP IOMMUTLBNotificationType
intel_iommu: Do not notify regular iotlb to device-iotlb notifiers
memory: Skip bad range assertion if notifier is DEVIOTLB type

 hw/alpha/typhoon.c       |  1 +
 hw/arm/smmu-common.c     |  4 +++-
 hw/arm/smmuv3.c          |  4 +++-
 hw/dma/rc4030.c          |  1 +
 hw/i386/amd_iommu.c      |  7 ++++++-
 hw/i386/intel_iommu.c    | 14 +++++++++++--
 hw/virtio/vhost.c        |  2 +-
 hw/virtio/virtio-iommu.c |  1 +
 include/exec/memory.h    | 26 ++++++++++++++++-------
 softmmu/memory.c         | 45 ++++++++++++++++++++++++++++++----------
 10 files changed, 80 insertions(+), 25 deletions(-)

--
2.18.1

#6 lore raw

Re: [RFC v6 02/13] memory: Add IOMMUTLBNotificationType to IOMMUTLBEntry

In reply to Eugenio Pérez
Quoted 13 lines · Eugenio Pérez

This way we can tell between MAPs and UNMAP, and potentially avoid to send them to a notifier that does not require them.

---
 include/exec/memory.h | 17 ++++++++++++-----
 1 file changed, 12 insertions(+), 5 deletions(-)
include/exec/memory.h +0 -0
index 22c5f564d1..f6d91c54aa 100644

include/exec/memory.h struct ReservedRegion { · line 66

Quoted 8 lines
include/exec/memory.h +5 -0
@@ -59,6 +59,12 @@ struct ReservedRegion {
 
 typedef struct IOMMUTLBEntry IOMMUTLBEntry;
 
+typedef enum {
+    IOMMU_IOTLB_NONE  = 0,
+    IOMMU_IOTLB_UNMAP = 1,
+    IOMMU_IOTLB_MAP   = 2,
+} IOMMUTLBNotificationType;

Can we directly use IOMMUNotifierFlag as the type rather than introducing a similar enumeration? Thanks,

Signature · 1 line
Peter Xu
#9 lore raw

Re: [RFC v6 03/13] hw/alpha/typhoon: Mark all IOMMUTLBEntry as IOMMU_IOTLB_NONE type

In reply to Eugenio Pérez
Quoted 10 lines · Eugenio Pérez
---
 hw/alpha/typhoon.c | 1 +
 1 file changed, 1 insertion(+)
hw/alpha/typhoon.c +0 -0
index 29d44dfb06..b1e6c4e929 100644

hw/alpha/typhoon.c static bool make_iommu_tlbe(hwaddr taddr, hwaddr mask, IOMMUTLBEntry *ret) · line 618

Quoted 4 lines
hw/alpha/typhoon.c +1 -0
@@ -615,6 +615,7 @@ static bool make_iommu_tlbe(hwaddr taddr, hwaddr mask, IOMMUTLBEntry *ret)
         .translated_addr = taddr,
         .addr_mask = mask,
         .perm = IOMMU_RW,
+        .type = IOMMU_IOTLB_NONE,

IMHO we don't need to touch all the IOMMUTLBEntry users, but only the callers of memory_region_notify_iommu*(). We should also comment at the type field that it's meaningless except when used for IOMMU notifications, because these are really two different things: IOMMUTLBEntry was originally a translated entry out of IOMMU hardware, so in those case it does not need a "type" field.

To make it clearer (depending on your preference...), we can introduce IOMMUTLBEvent to be:

  struct IOMMUTLBEvent {
    IOMMUTLBEntry entry;
    IOMMUTLBType type;
  };

Then it'll be clear on which is which. Though you'll need to touch more things (all the callers and all the registerered notifiers).

Thanks,

Signature · 1 line
Peter Xu
#23 lore raw
In reply to Eugenio Perez Martin
Quoted 6 lines · Eugenio Perez Martin

Hi!

Sending v6 to see if that is on the same page as what you meant. Making each setting of "type" explicitly IOMMU_IOTLB_NONE if not used in notifications. This is done in different commits in case this helps review of different architectures.

I've also proposed IOMMUTLBEvent in the other reply, that might help too.

Since at it, there's also another trick to use - we don't need to touch those "type" as long as the default type is "zero", so as long as we make sure the default type (IOMMU_NOTIFIER_NONE) is zero, then we don't need to set it everywhere either.

Quoted 11 lines

I think that this way we have too much freedom between entry flags (currently only access type, RW) and notification type. Since not all of them are valid nor used in the same context, I think this adds complexity. I'm wondering if:

Option a) We could make it private to memory.c, and make it a flag of memory_region_notify_iommu (like "bool deviotlb_type)". IOW, instead of making it a member of IOMMUTLBEntry, wrap the "entry" parameter of memory_region_notify_iommu in a new private structure defined in memory.c that adds that flag.

No strong preference from me. But since you posted the series before you provide the options... Maybe continue with what we have can be easier. :)

Quoted 10 lines

Option b) We could keep the IOMMUTLBNotificationType enum (open to suggestions for a better name :)), but not embed it in the struct, like:

include/exec/memory.h +0 -0
index 477c3af24c..d9150e7b7e 100644

include/exec/memory.h typedef enum { · line 76

Quoted 6 lines
include/exec/memory.h +2 -1
@@ -72,7 +72,8 @@ typedef enum {
     IOMMU_RO   = 1,
     IOMMU_WO   = 2,
     IOMMU_RW   = 3,
-} IOMMUAccessFlags;
+    IOMMU_DEVIOTLB = 4,
+} IOMMUEntryFlags;

Just in case you didn't notice - IOMMUAccessFlags is actaully a bitmap. :)

IMHO we can keep the IOMMUAccessFlags scemantics, since it's still correct for a general translated IOMMUTLBEntry object.

Thanks,

Signature · 1 line
Peter Xu
#19 lore raw

Re: [RFC v6 12/13] intel_iommu: Do not notify regular iotlb to device-iotlb notifiers

In reply to Eugenio Pérez
Quoted 3 lines · Eugenio Pérez

This improves performance in case of netperf with vhost-net:
* TCP_STREAM: From 9049.59Mbit/s to 9049.59Mbit/s (13%)

What's improvement ? they are the same

Quoted 33 lines

* TCP_RR: From 8464.73 trans/s to 8932.703333 trans/s (5.5%)
* UDP_RR: From 8562.08 trans/s to 9005.62/s (5.1%)
* UDP_STREAM: No change observed (insignificant 0.1% improvement)

---
 hw/i386/intel_iommu.c | 6 ++++++
 1 file changed, 6 insertions(+)
hw/i386/intel_iommu.c +2 -0
index ddb828da1f..7620a1abbf 100644
@@ -1960,6 +1960,12 @@ static void
 vtd_iotlb_domain_invalidate(IntelIOMMUState *s, uint16_t domain_id)
     vtd_iommu_unlock(s);
 
     QLIST_FOREACH(vtd_as, &s->vtd_as_with_notifiers, next) {
+        if (vtd_as->iommu.iommu_notify_flags & IOMMU_NOTIFIER_DEVIOTLB) {
+            /* If IOMMU memory region is DEVICE IOTLB type, it does not

make

hw/i386/intel_iommu.c +4 -1
+             * sense to send regular IOMMU notifications. */
+            continue;
+        }
+
         if (!vtd_dev_to_context_entry(s, pci_bus_num(vtd_as->bus),
                                       vtd_as->devfn, &ce) &&
             domain_id == vtd_get_domain_id(s, &ce)) {
--

2.18.1

Signature · 6 lines
         此致
礼
罗勇刚
Yours
    sincerely,
Yonggang Luo
#7 lore raw

Re: [RFC v6 02/13] memory: Add IOMMUTLBNotificationType to IOMMUTLBEntry

In reply to Peter Xu
Quoted 18 lines · Peter Xu
Quoted 16 lines · Eugenio Pérez

This way we can tell between MAPs and UNMAP, and potentially avoid to send them to a notifier that does not require them.

---
 include/exec/memory.h | 17 ++++++++++++-----
 1 file changed, 12 insertions(+), 5 deletions(-)
include/exec/memory.h +0 -0
index 22c5f564d1..f6d91c54aa 100644
@@ -59,6 +59,12 @@ struct ReservedRegion {
 
 typedef struct IOMMUTLBEntry IOMMUTLBEntry;
 
Quoted 9 lines

include/exec/memory.h struct ReservedRegion { · line 66

Quoted 5 lines
include/exec/memory.h +5 -0
+typedef enum {
+    IOMMU_IOTLB_NONE  = 0,
+    IOMMU_IOTLB_UNMAP = 1,
+    IOMMU_IOTLB_MAP   = 2,
+} IOMMUTLBNotificationType;

Can we directly use IOMMUNotifierFlag as the type rather than introducing a similar enumeration? Thanks,

Right, I think it's simpler your way.

Thanks!

Quoted 3 lines

--
Peter Xu

#24 lore raw
In reply to Peter Xu
Quoted 11 lines · Peter Xu
Quoted 6 lines · Eugenio Perez Martin

Hi!

Sending v6 to see if that is on the same page as what you meant. Making each setting of "type" explicitly IOMMU_IOTLB_NONE if not used in notifications. This is done in different commits in case this helps review of different architectures.

I've also proposed IOMMUTLBEvent in the other reply, that might help too.

I think IOMMUTLBEvent would be way clearer, yes :).

Quoted 5 lines

Since at it, there's also another trick to use - we don't need to touch those "type" as long as the default type is "zero", so as long as we make sure the default type (IOMMU_NOTIFIER_NONE) is zero, then we don't need to set it everywhere either.

Right, I was just making it explicit.

Quoted 6 lines
Quoted 6 lines

I think that this way we have too much freedom between entry flags (currently only access type, RW) and notification type. Since not all of them are valid nor used in the same context, I think this adds complexity. I'm wondering if:

Quoted 9 lines
Quoted 5 lines

Option a) We could make it private to memory.c, and make it a flag of memory_region_notify_iommu (like "bool deviotlb_type)". IOW, instead of making it a member of IOMMUTLBEntry, wrap the "entry" parameter of memory_region_notify_iommu in a new private structure defined in memory.c that adds that flag.

No strong preference from me. But since you posted the series before you provide the options... Maybe continue with what we have can be easier. :)

I just wanted to be sure I understood your proposal before comparing :)

Quoted 5 lines
Quoted 5 lines

Option b) We could keep the IOMMUTLBNotificationType enum (open to suggestions for a better name :)), but not embed it in the struct, like:

Quoted 14 lines

include/exec/memory.h typedef enum { · line 76

Quoted 11 lines
include/exec/memory.h +2 -1
index 477c3af24c..d9150e7b7e 100644
@@ -72,7 +72,8 @@ typedef enum {
     IOMMU_RO   = 1,
     IOMMU_WO   = 2,
     IOMMU_RW   = 3,
-} IOMMUAccessFlags;
+    IOMMU_DEVIOTLB = 4,
+} IOMMUEntryFlags;

Just in case you didn't notice - IOMMUAccessFlags is actaully a bitmap. :)

I know I know, that is why I compared with IOMMU_RW in proposed iommu_tlb_entry_type.

Thanks!

Quoted 8 lines

IMHO we can keep the IOMMUAccessFlags scemantics, since it's still correct for a general translated IOMMUTLBEntry object.

Thanks,

--
Peter Xu

#20 lore raw

Re: [RFC v6 12/13] intel_iommu: Do not notify regular iotlb to device-iotlb notifiers

In reply to 罗勇刚(Yonggang Luo)
Quoted 10 lines · 罗勇刚(Yonggang Luo)
Quoted 3 lines · Eugenio Pérez

This improves performance in case of netperf with vhost-net:
* TCP_STREAM: From 9049.59Mbit/s to 9049.59Mbit/s (13%)

What's improvement ? they are the same

Ouch, it was from 1923.6 Mbit/s to 2175.13 Mbit/s. Thanks for notify it, will fix in next revision!

Quoted 40 lines
Quoted 28 lines

* TCP_RR: From 8464.73 trans/s to 8932.703333 trans/s (5.5%)
* UDP_RR: From 8562.08 trans/s to 9005.62/s (5.1%)
* UDP_STREAM: No change observed (insignificant 0.1% improvement)

---
 hw/i386/intel_iommu.c | 6 ++++++
 1 file changed, 6 insertions(+)
hw/i386/intel_iommu.c +6 -1
index ddb828da1f..7620a1abbf 100644
@@ -1960,6 +1960,12 @@ static void vtd_iotlb_domain_invalidate(IntelIOMMUState *s, uint16_t domain_id)
     vtd_iommu_unlock(s);
 
     QLIST_FOREACH(vtd_as, &s->vtd_as_with_notifiers, next) {
+        if (vtd_as->iommu.iommu_notify_flags & IOMMU_NOTIFIER_DEVIOTLB) {
+            /* If IOMMU memory region is DEVICE IOTLB type, it does not make
+             * sense to send regular IOMMU notifications. */
+            continue;
+        }
+
         if (!vtd_dev_to_context_entry(s, pci_bus_num(vtd_as->bus),
                                       vtd_as->devfn, &ce) &&
             domain_id == vtd_get_domain_id(s, &ce)) {
--
Quoted 3 lines

2.18.1

--
此致
礼
罗勇刚
Yours
sincerely,
Yonggang Luo

#25 lore raw

cover memory: Skip assertion in memory_region_unregister_iommu_notifier

In reply to Eugenio Pérez

I am able to hit this assertion when a Red Hat 7 guest virtio_net device raises an "Invalidation" of all the TLB entries. This happens in the guest's startup if 'intel_iommu=on' argument is passed to the guest kernel and right IOMMU/ATS devices are declared in qemu's command line.

Command line:
/home/qemu/x86_64-softmmu/qemu-system-x86_64 -name \
guest=rhel7-test,debug-threads=on -machine \
pc-q35-5.1,accel=kvm,usb=off,dump-guest-core=off,kernel_irqchip=split \
-cpu \
Broadwell,vme=on,ss=on,vmx=on,f16c=on,rdrand=on,hypervisor=on,arat=on,tsc-adjust=on,umip=on,arch-capabilities=on,xsaveopt=on,pdpe1gb=on,abm=on,skip-l1dfl-vmentry=on,rtm=on,hle=on \
-m 8096 -realtime mlock=off -smp 2,sockets=2,cores=1,threads=1 -uuid \
d022ecbf-679e-4755-87ce-eb87fc5bbc5d -display none -no-user-config \
-nodefaults -rtc base=utc,driftfix=slew -global \
kvm-pit.lost_tick_policy=delay -no-hpet -no-shutdown -global \
ICH9-LPC.disable_s3=1 -global ICH9-LPC.disable_s4=1 -boot strict=on \
-device intel-iommu,intremap=on,device-iotlb=on -device \
pcie-root-port,port=0x8,chassis=1,id=pci.1,bus=pcie.0,multifunction=on,addr=0x1 \
-device \
pcie-root-port,port=0x9,chassis=2,id=pci.2,bus=pcie.0,addr=0x1.0x1 \
-device \
pcie-root-port,port=0xa,chassis=3,id=pci.3,bus=pcie.0,addr=0x1.0x2 \
-device \
pcie-root-port,port=0xb,chassis=4,id=pci.4,bus=pcie.0,addr=0x1.0x3 \
-device \
pcie-root-port,port=0xc,chassis=5,id=pci.5,bus=pcie.0,addr=0x1.0x4 \
-device \
pcie-root-port,port=0xd,chassis=6,id=pci.6,bus=pcie.0,addr=0x1.0x5 \
-device \
pcie-root-port,port=0xe,chassis=7,id=pci.7,bus=pcie.0,addr=0x1.0x6 \
-device qemu-xhci,p2=15,p3=15,id=usb,bus=pci.2,addr=0x0 -device \
virtio-serial-pci,id=virtio-serial0,bus=pci.3,addr=0x0 -drive \
file=/home/virtio-test2.qcow2,format=qcow2,if=none,id=drive-virtio-disk0 \
-device \
virtio-blk-pci,scsi=off,bus=pci.4,addr=0x0,drive=drive-virtio-disk0,id=virtio-disk0,bootindex=1 \
-netdev tap,id=hostnet0,vhost=on,vhostforce=on -device \
virtio-net-pci,netdev=hostnet0,id=net0,mac=52:54:00:0d:1d:f2,bus=pci.1,addr=0x0,iommu_platform=on,ats=on \
-device virtio-balloon-pci,id=balloon0,bus=pci.5,addr=0x0 -object \
rng-random,id=objrng0,filename=/dev/urandom -device \
virtio-rng-pci,rng=objrng0,id=rng0,bus=pci.6,addr=0x0 -s -msg \
timestamp=on

Full backtrace:

 #0  0x00007ffff521370f in raise () at /lib64/libc.so.6
 #1  0x00007ffff51fdb25 in abort () at /lib64/libc.so.6
 #2  0x00007ffff51fd9f9 in _nl_load_domain.cold.0 () at /lib64/libc.so.6
 #3  0x00007ffff520bcc6 in .annobin_assert.c_end () at /lib64/libc.so.6
 #4  0x0000555555888171 in memory_region_notify_one (notifier=0x7ffde0487fa8,
                                                    entry=0x7ffde5dfe200)
                          at /home/qemu/memory.c:1918
 #5  0x0000555555888247 in memory_region_notify_iommu (iommu_mr=0x555556f6c0b0,
                                                      iommu_idx=0, entry=...)
                          at /home/qemu/memory.c:1941
 #6  0x0000555555951c8d in vtd_process_device_iotlb_desc (s=0x555557609000,
                                                       inv_desc=0x7ffde5dfe2d0)
                          at /home/qemu/hw/i386/intel_iommu.c:2468
 #7  0x0000555555951e6a in vtd_process_inv_desc (s=0x555557609000)
                          at /home/qemu/hw/i386/intel_iommu.c:2531
 #8  0x0000555555951fa5 in vtd_fetch_inv_desc (s=0x555557609000)
                          at /home/qemu/hw/i386/intel_iommu.c:2563
 #9  0x00005555559520e5 in vtd_handle_iqt_write (s=0x555557609000)
                          at /home/qemu/hw/i386/intel_iommu.c:2590
 #10 0x0000555555952b45 in vtd_mem_write (opaque=0x555557609000, addr=136,
                                         val=2688, size=4)
                          at /home/qemu/hw/i386/intel_iommu.c:2837
 #11 0x0000555555883e17 in memory_region_write_accessor (mr=0x555557609330,
                                                        addr=136,
                                                        value=0x7ffde5dfe478,
                                                        size=4,
                                                        shift=0,
                                                        mask=4294967295,
                                                        attrs=...)
                         at /home/qemu/memory.c:483
 #12 0x000055555588401d in access_with_adjusted_size (addr=136,
                       value=0x7ffde5dfe478,
                       size=4,
                       access_size_min=4,
                       access_size_max=8,
                       access_fn=0x555555883d38 <memory_region_write_accessor>,
                       mr=0x555557609330,
                       attrs=...)
                       at /home/qemu/memory.c:544
 #13 0x0000555555886f37 in memory_region_dispatch_write (mr=0x555557609330,
                                                       addr=136,
                                                       data=2688,
                                                       op=MO_32,
                                                       attrs=...)
                         at /home/qemu/memory.c:1476
 #14 0x0000555555827a03 in flatview_write_continue (fv=0x7ffdd8503150,
                                                   addr=4275634312,
                                                   attrs=...,
                                                   ptr=0x7ffff7ff0028,
                                                   len=4,
                                                   addr1=136,
                                                   l=4,
                                                   mr=0x555557609330)
                          at /home/qemu/exec.c:3146
 #15 0x0000555555827b48 in flatview_write (fv=0x7ffdd8503150,
                                          addr=4275634312,
                                          attrs=...,
                                          buf=0x7ffff7ff0028,
                                          len=4)
                          at /home/qemu/exec.c:3186
 #16 0x0000555555827e9d in address_space_write (
                                      as=0x5555567ca640 <address_space_memory>,
                                      addr=4275634312,
                                      attrs=...,
                                      buf=0x7ffff7ff0028,
                                      len=4)
                          at /home/qemu/exec.c:3277
 #17 0x0000555555827f0a in address_space_rw (
                                      as=0x5555567ca640 <address_space_memory>,
                                      addr=4275634312,
                                      attrs=...,
                                      buf=0x7ffff7ff0028,
                                      len=4,
                                      is_write=true)
                          at /home/qemu/exec.c:3287
 #18 0x000055555589b633 in kvm_cpu_exec (cpu=0x555556b65640)
                               at /home/qemu/accel/kvm/kvm-all.c:2511
 #19 0x0000555555876ba8 in qemu_kvm_cpu_thread_fn (arg=0x555556b65640)
                               at /home/qemu/cpus.c:1284
 #20 0x0000555555dafff1 in qemu_thread_start (args=0x555556b8c3b0)
                               at util/qemu-thread-posix.c:521
 #21 0x00007ffff55a62de in start_thread () at /lib64/libpthread.so.0
 #22 0x00007ffff52d7e83 in clone () at /lib64/libc.so.6

(gdb) frame 4
#4 0x0000555555888171 in memory_region_notify_one

                      (notifier=0x7ffde0487fa8, entry=0x7ffde5dfe200)
                      at /home/qemu/memory.c:1918

1918 assert(entry->iova >= notifier->start && entry_end <=
notifier->end);
(gdb) p *entry
$1 = {target_as = 0x555556f6c050, iova = 0, translated_addr = 0, addr_mask = 18446744073709551615, perm = IOMMU_NONE} --

Tested with vhost-net and qemu driver, host<->guest communication.

v1: * IOMMU_NOTIFIER_ALL now includes IOMMU_NOTIFIER_DEVIOTLB_EVENTS
also. VFIO IOMMU notifier will register for all events (as before of the patching)
* Cosmetic changes, like:
- Expand commit messages
- Better naming and checks
- Fix alignment issues
- Avoid an already present casting from void *

RFC v8: Fix use of "tmp" notification in memory.c:memory_region_notify_iommu_one

v7: Add IOMMUTLBNotification, and move introduced "type" from
IOMMUTLBEntry to the former.

v6: Introduce "type" field for IOMMUTLBEntry. Fill in all uses.
Update tests reports with more fine-tuning (CPU, RPS/XPS tunning).

v5: Skip regular IOTLB notifications in dev_iotlb notifiers

v4: Rename IOMMU_NOTIFIER_IOTLB -> IOMMU_NOTIFIER_DEVIOTLB.
Make vhost-net notifier just IOMMU_NOTIFIER_DEVIOTLB, not IOMMU_NOTIFIER_UNMAP

v3: Skip the assertion in case notifier is a IOTLB one, since they can manage
arbitrary ranges. Using a flag in the notifier for now, as Peter suggested.

v2: Actually delete assertion instead of just commenting out using C99

Eugenio Pérez (5):
memory: Rename memory_region_notify_one to
memory_region_notify_iommu_one
memory: Add IOMMUTLBEvent
memory: Add IOMMU_NOTIFIER_DEVIOTLB_UNMAP IOMMUTLBNotificationType
intel_iommu: Skip page walking on device iotlb invalidations
memory: Skip bad range assertion if notifier is DEVIOTLB_UNMAP type

 include/exec/memory.h | 40 ++++++++++---------
 hw/arm/smmu-common.c  | 13 +++---
 hw/arm/smmuv3.c       | 13 +++---
 hw/i386/intel_iommu.c | 92 +++++++++++++++++++++++++------------------
 hw/misc/tz-mpc.c      | 32 ++++++++-------
 hw/ppc/spapr_iommu.c  | 15 +++----
 hw/virtio/vhost.c     |  2 +-
 softmmu/memory.c      | 31 +++++++++------
 8 files changed, 135 insertions(+), 103 deletions(-)
Signature · 1 line
2.18.1
#26 lore raw

PATCH 1/5 memory: Rename memory_region_notify_one to memory_region_notify_iommu_one

Previous name didn't reflect the iommu operation.

---
 include/exec/memory.h | 6 +++---
 hw/arm/smmu-common.c  | 2 +-
 hw/arm/smmuv3.c       | 2 +-
 hw/i386/intel_iommu.c | 4 ++--
 softmmu/memory.c      | 6 +++---
 5 files changed, 10 insertions(+), 10 deletions(-)
include/exec/memory.h +3 -3
index 0cfe987ab4..22c5f564d1 100644
@@ -226,7 +226,7 @@ enum IOMMUMemoryRegionAttr {
  * The IOMMU implementation must use the IOMMU notifier infrastructure
  * to report whenever mappings are changed, by calling
  * memory_region_notify_iommu() (or, if necessary, by calling
- * memory_region_notify_one() for each registered notifier).
+ * memory_region_notify_iommu_one() for each registered notifier).
  *
  * Conceptually an IOMMU provides a mapping from input address
  * to an output TLB entry. If the IOMMU is aware of memory transaction
@@ -1274,7 +1274,7 @@ void memory_region_notify_iommu(IOMMUMemoryRegion *iommu_mr,
                                 IOMMUTLBEntry entry);
 
 /**
- * memory_region_notify_one: notify a change in an IOMMU translation
+ * memory_region_notify_iommu_one: notify a change in an IOMMU translation
  *                           entry to a single notifier
  *
  * This works just like memory_region_notify_iommu(), but it only
@@ -1285,7 +1285,7 @@ void memory_region_notify_iommu(IOMMUMemoryRegion *iommu_mr,
  *         replaces all old entries for the same virtual I/O address range.
  *         Deleted entries have .@perm == 0.
  */
-void memory_region_notify_one(IOMMUNotifier *notifier,
+void memory_region_notify_iommu_one(IOMMUNotifier *notifier,
                               IOMMUTLBEntry *entry);
 
 /**
hw/arm/smmu-common.c +1 -1
index 3838db1395..88d2c454f0 100644
@@ -472,7 +472,7 @@ static void smmu_unmap_notifier_range(IOMMUNotifier *n)
     entry.perm = IOMMU_NONE;
     entry.addr_mask = n->end - n->start;
 
-    memory_region_notify_one(n, &entry);
+    memory_region_notify_iommu_one(n, &entry);
 }
 
 /* Unmap all notifiers attached to @mr */
hw/arm/smmuv3.c +1 -1
index 0122700e72..0a893ae918 100644
@@ -827,7 +827,7 @@ static void smmuv3_notify_iova(IOMMUMemoryRegion *mr,
     entry.addr_mask = num_pages * (1 << granule) - 1;
     entry.perm = IOMMU_NONE;
 
-    memory_region_notify_one(n, &entry);
+    memory_region_notify_iommu_one(n, &entry);
 }
 
 /* invalidate an asid/iova range tuple in all mr's */
hw/i386/intel_iommu.c +2 -2
index 5284bb68b6..2ad6b9d796 100644
@@ -3498,7 +3498,7 @@ static void vtd_address_space_unmap(VTDAddressSpace *as, IOMMUNotifier *n)
         /* This field is meaningless for unmap */
         entry.translated_addr = 0;
 
-        memory_region_notify_one(n, &entry);
+        memory_region_notify_iommu_one(n, &entry);
 
         start += mask;
         remain -= mask;
@@ -3536,7 +3536,7 @@ static void vtd_address_space_refresh_all(IntelIOMMUState *s)
 
 static int vtd_replay_hook(IOMMUTLBEntry *entry, void *private)
 {
-    memory_region_notify_one((IOMMUNotifier *)private, entry);
+    memory_region_notify_iommu_one((IOMMUNotifier *)private, entry);
     return 0;
 }
 
softmmu/memory.c +3 -3
index 70b93104e8..961c25b42f 100644
@@ -1890,8 +1890,8 @@ void memory_region_unregister_iommu_notifier(MemoryRegion *mr,
     memory_region_update_iommu_notify_flags(iommu_mr, NULL);
 }
 
-void memory_region_notify_one(IOMMUNotifier *notifier,
-                              IOMMUTLBEntry *entry)
+void memory_region_notify_iommu_one(IOMMUNotifier *notifier,
+                                    IOMMUTLBEntry *entry)
 {
     IOMMUNotifierFlag request_flags;
     hwaddr entry_end = entry->iova + entry->addr_mask;
@@ -1927,7 +1927,7 @@ void memory_region_notify_iommu(IOMMUMemoryRegion *iommu_mr,
 
     IOMMU_NOTIFIER_FOREACH(iommu_notifier, iommu_mr) {
         if (iommu_notifier->iommu_idx == iommu_idx) {
-            memory_region_notify_one(iommu_notifier, &entry);
+            memory_region_notify_iommu_one(iommu_notifier, &entry);
         }
     }
 }
Signature · 1 line
2.18.1
#27 lore raw

PATCH 2/5 memory: Add IOMMUTLBEvent

In reply to Eugenio Pérez

This way we can tell between regular IOMMUTLBEntry (entry of IOMMU hardware) and notifications.

In the notifications, we set explicitly if it is a MAPs or an UNMAP, instead of trusting in entry permissions to differentiate them.

---
 include/exec/memory.h | 27 +++++++------
 hw/arm/smmu-common.c  | 13 ++++---
 hw/arm/smmuv3.c       | 13 ++++---
 hw/i386/intel_iommu.c | 88 ++++++++++++++++++++++++-------------------
 hw/misc/tz-mpc.c      | 32 +++++++++-------
 hw/ppc/spapr_iommu.c  | 15 ++++----
 softmmu/memory.c      | 20 +++++-----
 7 files changed, 111 insertions(+), 97 deletions(-)
include/exec/memory.h +13 -14
index 22c5f564d1..ab8d1ac9c9 100644
@@ -106,6 +106,11 @@ struct IOMMUNotifier {
 };
 typedef struct IOMMUNotifier IOMMUNotifier;
 
+typedef struct IOMMUTLBEvent {
+    IOMMUNotifierFlag type;
+    IOMMUTLBEntry entry;
+} IOMMUTLBEvent;
+
 /* RAM is pre-allocated and passed into qemu_ram_alloc_from_ptr */
 #define RAM_PREALLOC   (1 << 0)
 
@@ -1254,24 +1259,18 @@ uint64_t memory_region_iommu_get_min_page_size(IOMMUMemoryRegion *iommu_mr);
 /**
  * memory_region_notify_iommu: notify a change in an IOMMU translation entry.
  *
- * The notification type will be decided by entry.perm bits:
- *
- * - For UNMAP (cache invalidation) notifies: set entry.perm to IOMMU_NONE.
- * - For MAP (newly added entry) notifies: set entry.perm to the
- *   permission of the page (which is definitely !IOMMU_NONE).
- *
  * Note: for any IOMMU implementation, an in-place mapping change
  * should be notified with an UNMAP followed by a MAP.
  *
  * @iommu_mr: the memory region that was changed
  * @iommu_idx: the IOMMU index for the translation table which has changed
- * @entry: the new entry in the IOMMU translation table.  The entry
- *         replaces all old entries for the same virtual I/O address range.
- *         Deleted entries have .@perm == 0.
+ * @event: TLB event with the new entry in the IOMMU translation table.
+ *         The entry replaces all old entries for the same virtual I/O address
+ *         range.
  */
 void memory_region_notify_iommu(IOMMUMemoryRegion *iommu_mr,
                                 int iommu_idx,
-                                IOMMUTLBEntry entry);
+                                IOMMUTLBEvent event);
 
 /**
  * memory_region_notify_iommu_one: notify a change in an IOMMU translation
@@ -1281,12 +1280,12 @@ void memory_region_notify_iommu(IOMMUMemoryRegion *iommu_mr,
  * notifies a specific notifier, not all of them.
  *
  * @notifier: the notifier to be notified
- * @entry: the new entry in the IOMMU translation table.  The entry
- *         replaces all old entries for the same virtual I/O address range.
- *         Deleted entries have .@perm == 0.
+ * @event: TLB event with the new entry in the IOMMU translation table.
+ *         The entry replaces all old entries for the same virtual I/O address
+ *         range.
  */
 void memory_region_notify_iommu_one(IOMMUNotifier *notifier,
-                              IOMMUTLBEntry *entry);
+                                    IOMMUTLBEvent *event);
 
 /**
  * memory_region_register_iommu_notifier: register a notifier for changes to
hw/arm/smmu-common.c +7 -6
index 88d2c454f0..405d5c5325 100644
@@ -465,14 +465,15 @@ IOMMUMemoryRegion *smmu_iommu_mr(SMMUState *s, uint32_t sid)
 /* Unmap the whole notifier's range */
 static void smmu_unmap_notifier_range(IOMMUNotifier *n)
 {
-    IOMMUTLBEntry entry;
+    IOMMUTLBEvent event;
 
-    entry.target_as = &address_space_memory;
-    entry.iova = n->start;
-    entry.perm = IOMMU_NONE;
-    entry.addr_mask = n->end - n->start;
+    event.type = IOMMU_NOTIFIER_UNMAP;
+    event.entry.target_as = &address_space_memory;
+    event.entry.iova = n->start;
+    event.entry.perm = IOMMU_NONE;
+    event.entry.addr_mask = n->end - n->start;
 
-    memory_region_notify_iommu_one(n, &entry);
+    memory_region_notify_iommu_one(n, &event);
 }
 
 /* Unmap all notifiers attached to @mr */
hw/arm/smmuv3.c +7 -6
index 0a893ae918..62b0e289ca 100644
@@ -799,7 +799,7 @@ static void smmuv3_notify_iova(IOMMUMemoryRegion *mr,
                                uint8_t tg, uint64_t num_pages)
 {
     SMMUDevice *sdev = container_of(mr, SMMUDevice, iommu);
-    IOMMUTLBEntry entry;
+    IOMMUTLBEvent event;
     uint8_t granule = tg;
 
     if (!tg) {
@@ -822,12 +822,13 @@ static void smmuv3_notify_iova(IOMMUMemoryRegion *mr,
         granule = tt->granule_sz;
     }
 
-    entry.target_as = &address_space_memory;
-    entry.iova = iova;
-    entry.addr_mask = num_pages * (1 << granule) - 1;
-    entry.perm = IOMMU_NONE;
+    event.type = IOMMU_NOTIFIER_UNMAP;
+    event.entry.target_as = &address_space_memory;
+    event.entry.iova = iova;
+    event.entry.addr_mask = num_pages * (1 << granule) - 1;
+    event.entry.perm = IOMMU_NONE;
 
-    memory_region_notify_iommu_one(n, &entry);
+    memory_region_notify_iommu_one(n, &event);
 }
 
 /* invalidate an asid/iova range tuple in all mr's */
hw/i386/intel_iommu.c +49 -39

198 lines not on this page · raw message

hw/misc/tz-mpc.c +18 -14
index 98f151237f..30481e1c90 100644
@@ -82,8 +82,10 @@ static void tz_mpc_iommu_notify(TZMPC *s, uint32_t lutidx,
     /* Called when the LUT word at lutidx has changed from oldlut to newlut;
      * must call the IOMMU notifiers for the changed blocks.
      */
-    IOMMUTLBEntry entry = {
-        .addr_mask = s->blocksize - 1,
+    IOMMUTLBEvent event = {
+        .entry = {
+            .addr_mask = s->blocksize - 1,
+        }
     };
     hwaddr addr = lutidx * s->blocksize * 32;
     int i;
@@ -100,26 +102,28 @@ static void tz_mpc_iommu_notify(TZMPC *s, uint32_t lutidx,
         block_is_ns = newlut & (1 << i);
 
         trace_tz_mpc_iommu_notify(addr);
-        entry.iova = addr;
-        entry.translated_addr = addr;
+        event.entry.iova = addr;
+        event.entry.translated_addr = addr;
 
-        entry.perm = IOMMU_NONE;
-        memory_region_notify_iommu(&s->upstream, IOMMU_IDX_S, entry);
-        memory_region_notify_iommu(&s->upstream, IOMMU_IDX_NS, entry);
+        event.type = IOMMU_NOTIFIER_UNMAP;
+        event.entry.perm = IOMMU_NONE;
+        memory_region_notify_iommu(&s->upstream, IOMMU_IDX_S, event);
+        memory_region_notify_iommu(&s->upstream, IOMMU_IDX_NS, event);
 
-        entry.perm = IOMMU_RW;
+        event.type = IOMMU_NOTIFIER_MAP;
+        event.entry.perm = IOMMU_RW;
         if (block_is_ns) {
-            entry.target_as = &s->blocked_io_as;
+            event.entry.target_as = &s->blocked_io_as;
         } else {
-            entry.target_as = &s->downstream_as;
+            event.entry.target_as = &s->downstream_as;
         }
-        memory_region_notify_iommu(&s->upstream, IOMMU_IDX_S, entry);
+        memory_region_notify_iommu(&s->upstream, IOMMU_IDX_S, event);
         if (block_is_ns) {
-            entry.target_as = &s->downstream_as;
+            event.entry.target_as = &s->downstream_as;
         } else {
-            entry.target_as = &s->blocked_io_as;
+            event.entry.target_as = &s->blocked_io_as;
         }
-        memory_region_notify_iommu(&s->upstream, IOMMU_IDX_NS, entry);
+        memory_region_notify_iommu(&s->upstream, IOMMU_IDX_NS, event);
     }
 }
 
hw/ppc/spapr_iommu.c +8 -7
index 0fecabc135..8bc3cff05d 100644
@@ -445,7 +445,7 @@ static void spapr_tce_reset(DeviceState *dev)
 static target_ulong put_tce_emu(SpaprTceTable *tcet, target_ulong ioba,
                                 target_ulong tce)
 {
-    IOMMUTLBEntry entry;
+    IOMMUTLBEvent event;
     hwaddr page_mask = IOMMU_PAGE_MASK(tcet->page_shift);
     unsigned long index = (ioba - tcet->bus_offset) >> tcet->page_shift;
 
@@ -457,12 +457,13 @@ static target_ulong put_tce_emu(SpaprTceTable *tcet, target_ulong ioba,
 
     tcet->table[index] = tce;
 
-    entry.target_as = &address_space_memory,
-    entry.iova = (ioba - tcet->bus_offset) & page_mask;
-    entry.translated_addr = tce & page_mask;
-    entry.addr_mask = ~page_mask;
-    entry.perm = spapr_tce_iommu_access_flags(tce);
-    memory_region_notify_iommu(&tcet->iommu, 0, entry);
+    event.entry.target_as = &address_space_memory,
+    event.entry.iova = (ioba - tcet->bus_offset) & page_mask;
+    event.entry.translated_addr = tce & page_mask;
+    event.entry.addr_mask = ~page_mask;
+    event.entry.perm = spapr_tce_iommu_access_flags(tce);
+    event.type = entry.perm ? IOMMU_NOTIFIER_MAP : IOMMU_NOTIFIER_UNMAP;
+    memory_region_notify_iommu(&tcet->iommu, 0, event);
 
     return H_SUCCESS;
 }
softmmu/memory.c +9 -11
index 961c25b42f..8694fc7cf7 100644
@@ -1891,11 +1891,15 @@ void memory_region_unregister_iommu_notifier(MemoryRegion *mr,
 }
 
 void memory_region_notify_iommu_one(IOMMUNotifier *notifier,
-                                    IOMMUTLBEntry *entry)
+                                    IOMMUTLBEvent *event)
 {
-    IOMMUNotifierFlag request_flags;
+    IOMMUTLBEntry *entry = &event->entry;
     hwaddr entry_end = entry->iova + entry->addr_mask;
 
+    if (event->type == IOMMU_NOTIFIER_UNMAP) {
+        assert(entry->perm == IOMMU_NONE);
+    }
+
     /*
      * Skip the notification if the notification does not overlap
      * with registered range.
@@ -1906,20 +1910,14 @@ void memory_region_notify_iommu_one(IOMMUNotifier *notifier,
 
     assert(entry->iova >= notifier->start && entry_end <= notifier->end);
 
-    if (entry->perm & IOMMU_RW) {
-        request_flags = IOMMU_NOTIFIER_MAP;
-    } else {
-        request_flags = IOMMU_NOTIFIER_UNMAP;
-    }
-
-    if (notifier->notifier_flags & request_flags) {
+    if (event->type & notifier->notifier_flags) {
         notifier->notify(notifier, entry);
     }
 }
 
 void memory_region_notify_iommu(IOMMUMemoryRegion *iommu_mr,
                                 int iommu_idx,
-                                IOMMUTLBEntry entry)
+                                IOMMUTLBEvent event)
 {
     IOMMUNotifier *iommu_notifier;
 
@@ -1927,7 +1925,7 @@ void memory_region_notify_iommu(IOMMUMemoryRegion *iommu_mr,
 
     IOMMU_NOTIFIER_FOREACH(iommu_notifier, iommu_mr) {
         if (iommu_notifier->iommu_idx == iommu_idx) {
-            memory_region_notify_iommu_one(iommu_notifier, &entry);
+            memory_region_notify_iommu_one(iommu_notifier, &event);
         }
     }
 }
Signature · 1 line
2.18.1
#31 lore raw

PATCH 5/5 memory: Skip bad range assertion if notifier is DEVIOTLB_UNMAP type

In reply to Eugenio Pérez

Device IOTLB invalidations can unmap arbitrary ranges, eiter outside of the memory region or even [0, ~0ULL] for all the space. The assertion could be hit by a guest, and rhel7 guest effectively hit it.

---
 softmmu/memory.c | 13 +++++++++++--
 1 file changed, 11 insertions(+), 2 deletions(-)
softmmu/memory.c +11 -2
index 8694fc7cf7..e723fcbaa1 100644
@@ -1895,6 +1895,7 @@ void memory_region_notify_iommu_one(IOMMUNotifier *notifier,
 {
     IOMMUTLBEntry *entry = &event->entry;
     hwaddr entry_end = entry->iova + entry->addr_mask;
+    IOMMUTLBEntry tmp = *entry;
 
     if (event->type == IOMMU_NOTIFIER_UNMAP) {
         assert(entry->perm == IOMMU_NONE);
@@ -1908,10 +1909,18 @@ void memory_region_notify_iommu_one(IOMMUNotifier *notifier,
         return;
     }
 
-    assert(entry->iova >= notifier->start && entry_end <= notifier->end);
+    if (notifier->notifier_flags & IOMMU_NOTIFIER_DEVIOTLB_UNMAP) {
+        /* Crop (iova, addr_mask) to range */
+        tmp.iova = MAX(tmp.iova, notifier->start);
+        tmp.addr_mask = MIN(entry_end, notifier->end) - tmp.iova;
+        /* Confirm no underflow */
+        assert(MIN(entry_end, notifier->end) >= tmp.iova);
+    } else {
+        assert(entry->iova >= notifier->start && entry_end <= notifier->end);
+    }
 
     if (event->type & notifier->notifier_flags) {
-        notifier->notify(notifier, entry);
+        notifier->notify(notifier, &tmp);
     }
 }
 
Signature · 1 line
2.18.1
#28 lore raw

PATCH 3/5 memory: Add IOMMU_NOTIFIER_DEVIOTLB_UNMAP IOMMUTLBNotificationType

In reply to Eugenio Pérez

This allows us to differentiate between regular IOMMU map/unmap events and DEVIOTLB unmap. Doing so, notifiers that only need device IOTLB invalidations will not receive regular IOMMU unmappings.

Adapt intel and vhost to use it.

---
 include/exec/memory.h | 7 ++++++-
 hw/i386/intel_iommu.c | 2 +-
 hw/virtio/vhost.c     | 2 +-
 3 files changed, 8 insertions(+), 3 deletions(-)
include/exec/memory.h +6 -1
index ab8d1ac9c9..77afeda60c 100644
@@ -87,9 +87,14 @@ typedef enum {
     IOMMU_NOTIFIER_UNMAP = 0x1,
     /* Notify entry changes (newly created entries) */
     IOMMU_NOTIFIER_MAP = 0x2,
+    /* Notify changes on device IOTLB entries */
+    IOMMU_NOTIFIER_DEVIOTLB_UNMAP = 0x04,
 } IOMMUNotifierFlag;
 
-#define IOMMU_NOTIFIER_ALL (IOMMU_NOTIFIER_MAP | IOMMU_NOTIFIER_UNMAP)
+#define IOMMU_NOTIFIER_IOTLB_EVENTS (IOMMU_NOTIFIER_MAP | IOMMU_NOTIFIER_UNMAP)
+#define IOMMU_NOTIFIER_DEVIOTLB_EVENTS IOMMU_NOTIFIER_DEVIOTLB_UNMAP
+#define IOMMU_NOTIFIER_ALL (IOMMU_NOTIFIER_IOTLB_EVENTS | \
+                            IOMMU_NOTIFIER_DEVIOTLB_EVENTS)
 
 struct IOMMUNotifier;
 typedef void (*IOMMUNotify)(struct IOMMUNotifier *notifier,
hw/i386/intel_iommu.c +1 -1
index 19661bba3e..ab6833d5a0 100644
@@ -2468,7 +2468,7 @@ static bool vtd_process_device_iotlb_desc(IntelIOMMUState *s,
         sz = VTD_PAGE_SIZE;
     }
 
-    event.type = IOMMU_NOTIFIER_UNMAP;
+    event.type = IOMMU_NOTIFIER_DEVIOTLB_UNMAP;
     event.entry.target_as = &vtd_dev_as->as;
     event.entry.addr_mask = sz - 1;
     event.entry.iova = addr;
hw/virtio/vhost.c +1 -1
index 1a1384e7a6..1bf5bbc82e 100644
@@ -729,7 +729,7 @@ static void vhost_iommu_region_add(MemoryListener *listener,
     iommu_idx = memory_region_iommu_attrs_to_index(iommu_mr,
                                                    MEMTXATTRS_UNSPECIFIED);
     iommu_notifier_init(&iommu->n, vhost_iommu_unmap_notify,
-                        IOMMU_NOTIFIER_UNMAP,
+                        IOMMU_NOTIFIER_DEVIOTLB_UNMAP,
                         section->offset_within_region,
                         int128_get64(end),
                         iommu_idx);
Signature · 1 line
2.18.1
#29 lore raw

PATCH 4/5 intel_iommu: Skip page walking on device iotlb invalidations

In reply to Eugenio Pérez

Although they didn't reach the notifier because of the filtering in memory_region_notify_iommu_one, the vt-d was still splitting huge memory invalidations in chunks. Skipping it.

This improves performance in case of netperf with vhost-net:
* TCP_STREAM: From 1923.6Mbit/s to 2175.13Mbit/s (13%)
* TCP_RR: From 8464.73 trans/s to 8932.703333 trans/s (5.5%)
* UDP_RR: From 8562.08 trans/s to 9005.62/s (5.1%)
* UDP_STREAM: No change observed (insignificant 0.1% improvement)

---
 hw/i386/intel_iommu.c | 4 ++++
 1 file changed, 4 insertions(+)
hw/i386/intel_iommu.c +4 -0
index ab6833d5a0..fbbda0c87e 100644
@@ -1478,6 +1478,10 @@ static int vtd_sync_shadow_page_table(VTDAddressSpace *vtd_as)
     VTDContextEntry ce;
     IOMMUNotifier *n;
 
+    if (!(vtd_as->iommu.iommu_notify_flags & IOMMU_NOTIFIER_IOTLB_EVENTS)) {
+        return 0;
+    }
+
     ret = vtd_dev_to_context_entry(vtd_as->iommu_state,
                                    pci_bus_num(vtd_as->bus),
                                    vtd_as->devfn, &ce);
Signature · 1 line
2.18.1
#32 lore raw

Re: [PATCH 5/5] memory: Skip bad range assertion if notifier is DEVIOTLB_UNMAP type

In reply to Eugenio Pérez
Quoted 24 lines · Eugenio Pérez

Device IOTLB invalidations can unmap arbitrary ranges, eiter outside of the memory region or even [0, ~0ULL] for all the space. The assertion could be hit by a guest, and rhel7 guest effectively hit it.

---
  softmmu/memory.c | 13 +++++++++++--
  1 file changed, 11 insertions(+), 2 deletions(-)
softmmu/memory.c +1 -0
index 8694fc7cf7..e723fcbaa1 100644
@@ -1895,6 +1895,7 @@ void memory_region_notify_iommu_one(IOMMUNotifier *notifier,
  {
      IOMMUTLBEntry *entry = &event->entry;
      hwaddr entry_end = entry->iova + entry->addr_mask;
+    IOMMUTLBEntry tmp = *entry;
 
      if (event->type == IOMMU_NOTIFIER_UNMAP) {
          assert(entry->perm == IOMMU_NONE);

softmmu/memory.c void memory_region_notify_iommu_one(IOMMUNotifier *notifier, · line 1917

Quoted 10 lines
softmmu/memory.c +6 -1
@@ -1908,10 +1909,18 @@ void memory_region_notify_iommu_one(IOMMUNotifier *notifier,
          return;
      }
 
-    assert(entry->iova >= notifier->start && entry_end <= notifier->end);
+    if (notifier->notifier_flags & IOMMU_NOTIFIER_DEVIOTLB_UNMAP) {
+        /* Crop (iova, addr_mask) to range */
+        tmp.iova = MAX(tmp.iova, notifier->start);
+        tmp.addr_mask = MIN(entry_end, notifier->end) - tmp.iova;
+        /* Confirm no underflow */
+        assert(MIN(entry_end, notifier->end) >= tmp.iova);

It's still not clear to me why we need such assert. Consider notifier->end is the possible IOVA range but not possible device IOTLB invalidation range (e.g it allows [0, ULLONG_MAX]).

Thanks

Quoted 10 lines
softmmu/memory.c +4 -1
+    } else {
+        assert(entry->iova >= notifier->start && entry_end <= notifier->end);
+    }
 
      if (event->type & notifier->notifier_flags) {
-        notifier->notify(notifier, entry);
+        notifier->notify(notifier, &tmp);
      }
  }
 
Reviewed-by on v1 4/5
#30 lore raw

Re: [PATCH 4/5] intel_iommu: Skip page walking on device iotlb invalidations

In reply to Eugenio Pérez
Quoted 11 lines · Eugenio Pérez

Although they didn't reach the notifier because of the filtering in memory_region_notify_iommu_one, the vt-d was still splitting huge memory invalidations in chunks. Skipping it.

This improves performance in case of netperf with vhost-net:
* TCP_STREAM: From 1923.6Mbit/s to 2175.13Mbit/s (13%)
* TCP_RR: From 8464.73 trans/s to 8932.703333 trans/s (5.5%)
* UDP_RR: From 8562.08 trans/s to 9005.62/s (5.1%)
* UDP_STREAM: No change observed (insignificant 0.1% improvement)

Signature · 1 line
Peter Xu
#33 lore raw

Re: [PATCH 5/5] memory: Skip bad range assertion if notifier is DEVIOTLB_UNMAP type

In reply to Jason Wang
Quoted 43 lines · Jason Wang
Quoted 24 lines · Eugenio Pérez

Device IOTLB invalidations can unmap arbitrary ranges, eiter outside of the memory region or even [0, ~0ULL] for all the space. The assertion could be hit by a guest, and rhel7 guest effectively hit it.

---
  softmmu/memory.c | 13 +++++++++++--
  1 file changed, 11 insertions(+), 2 deletions(-)
softmmu/memory.c +1 -0
index 8694fc7cf7..e723fcbaa1 100644
@@ -1895,6 +1895,7 @@ void memory_region_notify_iommu_one(IOMMUNotifier *notifier,
  {
      IOMMUTLBEntry *entry = &event->entry;
      hwaddr entry_end = entry->iova + entry->addr_mask;
+    IOMMUTLBEntry tmp = *entry;
 
      if (event->type == IOMMU_NOTIFIER_UNMAP) {
          assert(entry->perm == IOMMU_NONE);

softmmu/memory.c void memory_region_notify_iommu_one(IOMMUNotifier *notifier, · line 1917

Quoted 10 lines
softmmu/memory.c +6 -1
@@ -1908,10 +1909,18 @@ void memory_region_notify_iommu_one(IOMMUNotifier *notifier,
          return;
      }
 
-    assert(entry->iova >= notifier->start && entry_end <= notifier->end);
+    if (notifier->notifier_flags & IOMMU_NOTIFIER_DEVIOTLB_UNMAP) {
+        /* Crop (iova, addr_mask) to range */
+        tmp.iova = MAX(tmp.iova, notifier->start);
+        tmp.addr_mask = MIN(entry_end, notifier->end) - tmp.iova;
+        /* Confirm no underflow */
+        assert(MIN(entry_end, notifier->end) >= tmp.iova);

It's still not clear to me why we need such assert. Consider notifier->end is the possible IOVA range but not possible device IOTLB invalidation range (e.g it allows [0, ULLONG_MAX]).

Quoted 2 lines

Thanks

As far as I understood the device should admit that out of bounds notifications in that case,
and the assert just makes sure that there was no underflow in tmp.addr_mask, i.e., that something
very wrong that should never happen in production happened.

Peter, would you mind to confirm/correct it?

Is there anything else needed to pull this patch?

Thanks!

Quoted 12 lines
Quoted 11 lines
softmmu/memory.c +4 -1
+    } else {
+        assert(entry->iova >= notifier->start && entry_end <= notifier->end);
+    }
 
      if (event->type & notifier->notifier_flags) {
-        notifier->notify(notifier, entry);
+        notifier->notify(notifier, &tmp);
      }
  }
 
#34 lore raw

Re: [PATCH 5/5] memory: Skip bad range assertion if notifier is DEVIOTLB_UNMAP type

Quoted 47 lines · Eugenio Perez Martin
Quoted 43 lines · Jason Wang
Quoted 24 lines · Eugenio Pérez

Device IOTLB invalidations can unmap arbitrary ranges, eiter outside of the memory region or even [0, ~0ULL] for all the space. The assertion could be hit by a guest, and rhel7 guest effectively hit it.

---
  softmmu/memory.c | 13 +++++++++++--
  1 file changed, 11 insertions(+), 2 deletions(-)
softmmu/memory.c +1 -0
index 8694fc7cf7..e723fcbaa1 100644
@@ -1895,6 +1895,7 @@ void memory_region_notify_iommu_one(IOMMUNotifier *notifier,
  {
      IOMMUTLBEntry *entry = &event->entry;
      hwaddr entry_end = entry->iova + entry->addr_mask;
+    IOMMUTLBEntry tmp = *entry;
 
      if (event->type == IOMMU_NOTIFIER_UNMAP) {
          assert(entry->perm == IOMMU_NONE);

softmmu/memory.c void memory_region_notify_iommu_one(IOMMUNotifier *notifier, · line 1917

Quoted 10 lines
softmmu/memory.c +6 -1
@@ -1908,10 +1909,18 @@ void memory_region_notify_iommu_one(IOMMUNotifier *notifier,
          return;
      }
 
-    assert(entry->iova >= notifier->start && entry_end <= notifier->end);
+    if (notifier->notifier_flags & IOMMU_NOTIFIER_DEVIOTLB_UNMAP) {
+        /* Crop (iova, addr_mask) to range */
+        tmp.iova = MAX(tmp.iova, notifier->start);
+        tmp.addr_mask = MIN(entry_end, notifier->end) - tmp.iova;
+        /* Confirm no underflow */
+        assert(MIN(entry_end, notifier->end) >= tmp.iova);

It's still not clear to me why we need such assert. Consider notifier->end is the possible IOVA range but not possible device IOTLB invalidation range (e.g it allows [0, ULLONG_MAX]).

Quoted 2 lines

Thanks

Quoted 7 lines

As far as I understood the device should admit that out of bounds notifications in that case,
and the assert just makes sure that there was no underflow in tmp.addr_mask, i.e., that something
very wrong that should never happen in production happened.

Peter, would you mind to confirm/correct it?

I think Jason is right - since we have checked at the entry that the two regions cross over each other:

    /*
     * Skip the notification if the notification does not overlap
     * with registered range.
     */
    if (notifier->start > entry_end || notifier->end < entry->iova) {
        return;
    }

Then I don't see how this assertion can fail any more.

But imho not a big problem either, and it shouldn't hurt to even keep the assertion of above isn't that straightforward.

Quoted 2 lines

Is there anything else needed to pull this patch?

I didn't post a pull for this only because I shouldn't :) - the plan was that all vt-d patches will still go via Michael's tree, iiuc. Though at least to me I think this series is acceptable for merging.

Though it would always be good too if Jason would still like to review it.

Jason, what's your opinion?

Thanks,

Signature · 1 line
Peter Xu
#35 lore raw

Re: [PATCH 5/5] memory: Skip bad range assertion if notifier is DEVIOTLB_UNMAP type

Quoted 73 lines · Peter Xu
Quoted 47 lines · Eugenio Perez Martin
Quoted 43 lines · Jason Wang
Quoted 24 lines · Eugenio Pérez

Device IOTLB invalidations can unmap arbitrary ranges, eiter outside of the memory region or even [0, ~0ULL] for all the space. The assertion could be hit by a guest, and rhel7 guest effectively hit it.

---
  softmmu/memory.c | 13 +++++++++++--
  1 file changed, 11 insertions(+), 2 deletions(-)
softmmu/memory.c +1 -0
index 8694fc7cf7..e723fcbaa1 100644
@@ -1895,6 +1895,7 @@ void memory_region_notify_iommu_one(IOMMUNotifier *notifier,
  {
      IOMMUTLBEntry *entry = &event->entry;
      hwaddr entry_end = entry->iova + entry->addr_mask;
+    IOMMUTLBEntry tmp = *entry;
 
      if (event->type == IOMMU_NOTIFIER_UNMAP) {
          assert(entry->perm == IOMMU_NONE);

softmmu/memory.c void memory_region_notify_iommu_one(IOMMUNotifier *notifier, · line 1917

Quoted 10 lines
softmmu/memory.c +6 -1
@@ -1908,10 +1909,18 @@ void memory_region_notify_iommu_one(IOMMUNotifier *notifier,
          return;
      }
 
-    assert(entry->iova >= notifier->start && entry_end <= notifier->end);
+    if (notifier->notifier_flags & IOMMU_NOTIFIER_DEVIOTLB_UNMAP) {
+        /* Crop (iova, addr_mask) to range */
+        tmp.iova = MAX(tmp.iova, notifier->start);
+        tmp.addr_mask = MIN(entry_end, notifier->end) - tmp.iova;
+        /* Confirm no underflow */
+        assert(MIN(entry_end, notifier->end) >= tmp.iova);

It's still not clear to me why we need such assert. Consider notifier->end is the possible IOVA range but not possible device IOTLB invalidation range (e.g it allows [0, ULLONG_MAX]).

Quoted 2 lines

Thanks

Quoted 7 lines

As far as I understood the device should admit that out of bounds notifications in that case,
and the assert just makes sure that there was no underflow in tmp.addr_mask, i.e., that something
very wrong that should never happen in production happened.

Peter, would you mind to confirm/correct it?

I think Jason is right - since we have checked at the entry that the two regions cross over each other:

    /*
     * Skip the notification if the notification does not overlap
     * with registered range.
     */
    if (notifier->start > entry_end || notifier->end < entry->iova) {
        return;
    }

Then I don't see how this assertion can fail any more.

But imho not a big problem either, and it shouldn't hurt to even keep the assertion of above isn't that straightforward.

Quoted 1 line
Quoted 5 lines
Quoted 1 line

Is there anything else needed to pull this patch?

I didn't post a pull for this only because I shouldn't :) - the plan was that all vt-d patches will still go via Michael's tree, iiuc. Though at least to me I think this series is acceptable for merging.

Sure, that's ok.

Eugenio, you sent patch 0 as a response to another series, which made me miss the series. Pls don't do that in the future.

Looks like Jason reviewed the series - Jason, is that right? - so I'd like his ack if possible.

Quoted 8 lines

Though it would always be good too if Jason would still like to review it.

Jason, what's your opinion?

Thanks,

Signature · 1 line
Peter Xu
#36 lore raw

Re: [PATCH 5/5] memory: Skip bad range assertion if notifier is DEVIOTLB_UNMAP type

Quoted 77 lines · Michael S. Tsirkin
Quoted 75 lines · Peter Xu
Quoted 47 lines · Eugenio Perez Martin
Quoted 43 lines · Jason Wang
Quoted 24 lines · Eugenio Pérez

Device IOTLB invalidations can unmap arbitrary ranges, eiter outside of the memory region or even [0, ~0ULL] for all the space. The assertion could be hit by a guest, and rhel7 guest effectively hit it.

---
  softmmu/memory.c | 13 +++++++++++--
  1 file changed, 11 insertions(+), 2 deletions(-)
softmmu/memory.c +1 -0
index 8694fc7cf7..e723fcbaa1 100644
@@ -1895,6 +1895,7 @@ void memory_region_notify_iommu_one(IOMMUNotifier *notifier,
  {
      IOMMUTLBEntry *entry = &event->entry;
      hwaddr entry_end = entry->iova + entry->addr_mask;
+    IOMMUTLBEntry tmp = *entry;
 
      if (event->type == IOMMU_NOTIFIER_UNMAP) {
          assert(entry->perm == IOMMU_NONE);

softmmu/memory.c void memory_region_notify_iommu_one(IOMMUNotifier *notifier, · line 1917

Quoted 10 lines
softmmu/memory.c +6 -1
@@ -1908,10 +1909,18 @@ void memory_region_notify_iommu_one(IOMMUNotifier *notifier,
          return;
      }
 
-    assert(entry->iova >= notifier->start && entry_end <= notifier->end);
+    if (notifier->notifier_flags & IOMMU_NOTIFIER_DEVIOTLB_UNMAP) {
+        /* Crop (iova, addr_mask) to range */
+        tmp.iova = MAX(tmp.iova, notifier->start);
+        tmp.addr_mask = MIN(entry_end, notifier->end) - tmp.iova;
+        /* Confirm no underflow */
+        assert(MIN(entry_end, notifier->end) >= tmp.iova);

It's still not clear to me why we need such assert. Consider notifier->end is the possible IOVA range but not possible device IOTLB invalidation range (e.g it allows [0, ULLONG_MAX]).

Quoted 2 lines

Thanks

Quoted 7 lines

As far as I understood the device should admit that out of bounds notifications in that case,
and the assert just makes sure that there was no underflow in tmp.addr_mask, i.e., that something
very wrong that should never happen in production happened.

Peter, would you mind to confirm/correct it?

I think Jason is right - since we have checked at the entry that the two regions cross over each other:

    /*
     * Skip the notification if the notification does not overlap
     * with registered range.
     */
    if (notifier->start > entry_end || notifier->end < entry->iova) {
        return;
    }

Then I don't see how this assertion can fail any more.

But imho not a big problem either, and it shouldn't hurt to even keep the assertion of above isn't that straightforward.

Quoted 2 lines

Is there anything else needed to pull this patch?

Quoted 9 lines
Quoted 3 lines

I didn't post a pull for this only because I shouldn't :) - the plan was that all vt-d patches will still go via Michael's tree, iiuc. Though at least to me I think this series is acceptable for merging.

Sure, that's ok.

Eugenio, you sent patch 0 as a response to another series, which made me miss the series. Pls don't do that in the future.

Sorry, noted for the next time.

Thanks!

Quoted 13 lines

Looks like Jason reviewed the series - Jason, is that right? - so I'd like his ack if possible.

Quoted 8 lines

Though it would always be good too if Jason would still like to review it.

Jason, what's your opinion?

Thanks,

--
Peter Xu

#37 lore raw

Re: [PATCH 5/5] memory: Skip bad range assertion if notifier is DEVIOTLB_UNMAP type

In reply to Michael S. Tsirkin
Quoted 73 lines · Michael S. Tsirkin
Quoted 65 lines · Peter Xu
Quoted 44 lines · Eugenio Perez Martin
Quoted 41 lines · Jason Wang
Quoted 24 lines · Eugenio Pérez

Device IOTLB invalidations can unmap arbitrary ranges, eiter outside of the memory region or even [0, ~0ULL] for all the space. The assertion could be hit by a guest, and rhel7 guest effectively hit it.

---
   softmmu/memory.c | 13 +++++++++++--
   1 file changed, 11 insertions(+), 2 deletions(-)
softmmu/memory.c +1 -0
index 8694fc7cf7..e723fcbaa1 100644
@@ -1895,6 +1895,7 @@ void memory_region_notify_iommu_one(IOMMUNotifier *notifier,
   {
       IOMMUTLBEntry *entry = &event->entry;
       hwaddr entry_end = entry->iova + entry->addr_mask;
+    IOMMUTLBEntry tmp = *entry;
 
       if (event->type == IOMMU_NOTIFIER_UNMAP) {
           assert(entry->perm == IOMMU_NONE);

softmmu/memory.c void memory_region_notify_iommu_one(IOMMUNotifier *notifier, · line 1917

Quoted 10 lines
softmmu/memory.c +6 -1
@@ -1908,10 +1909,18 @@ void memory_region_notify_iommu_one(IOMMUNotifier *notifier,
           return;
       }
 
-    assert(entry->iova >= notifier->start && entry_end <= notifier->end);
+    if (notifier->notifier_flags & IOMMU_NOTIFIER_DEVIOTLB_UNMAP) {
+        /* Crop (iova, addr_mask) to range */
+        tmp.iova = MAX(tmp.iova, notifier->start);
+        tmp.addr_mask = MIN(entry_end, notifier->end) - tmp.iova;
+        /* Confirm no underflow */
+        assert(MIN(entry_end, notifier->end) >= tmp.iova);

It's still not clear to me why we need such assert. Consider notifier->end is the possible IOVA range but not possible device IOTLB invalidation range (e.g it allows [0, ULLONG_MAX]).

Quoted 2 lines

Thanks

Quoted 7 lines

As far as I understood the device should admit that out of bounds notifications in that case,
and the assert just makes sure that there was no underflow in tmp.addr_mask, i.e., that something
very wrong that should never happen in production happened.

Peter, would you mind to confirm/correct it?

I think Jason is right - since we have checked at the entry that the two regions cross over each other:

     /*
      * Skip the notification if the notification does not overlap
      * with registered range.
      */
     if (notifier->start > entry_end || notifier->end < entry->iova) {
         return;
     }

Then I don't see how this assertion can fail any more.

Quoted 7 lines

But imho not a big problem either, and it shouldn't hurt to even keep the assertion of above isn't that straightforward.

Quoted 1 line

Is there anything else needed to pull this patch?

I didn't post a pull for this only because I shouldn't :) - the plan was that all vt-d patches will still go via Michael's tree, iiuc. Though at least to me I think this series is acceptable for merging.

Quoted 7 lines

Sure, that's ok.

Eugenio, you sent patch 0 as a response to another series, which made me miss the series. Pls don't do that in the future.

Looks like Jason reviewed the series - Jason, is that right? - so I'd like his ack if possible.

Right.

Euenio. If it's possible I would prefer to remove the assertion but it's ok it you leave it.

And please repost the series.

Thanks

Quoted 11 lines
Quoted 10 lines

Though it would always be good too if Jason would still like to review it.

Jason, what's your opinion?

Thanks,

Signature · 1 line
Peter Xu