Skip to contentSkip to conversation

vhost: disable for OABI

PATCH In review Michael S. Tsirkin 3 participants 4 messages Apr 20, 2020 – Apr 27, 2020

Subsystems (6)
lore
Share & tools
lore

Michael S. Tsirkin

author maintainer
#0 lore raw

PATCH v4 1/1 vhost: disable for OABI

vhost is currently broken on the some ARM configs.

The reason is that the ring element addresses are passed between components with different alignments assumptions. Thus, if guest selects a pointer and host then gets and dereferences it, then alignment assumed by the host's compiler might be greater than the actual alignment of the pointer. compiler on the host from assuming pointer is aligned.

This actually triggers on ARM with -mabi=apcs-gnu - which is a deprecated configuration. With this OABI, compiler assumes that all structures are 4 byte aligned - which is stronger than virtio guarantees for available and used rings, which are merely 2 bytes. Thus a guest without -mabi=apcs-gnu running on top of host with -mabi=apcs-gnu will be broken.

The correct fix is to force alignment of structures - however that is an intrusive fix that's best deferred until the next release.

We didn't previously support such ancient systems at all - this surfaced after vdpa support prompted removing dependency of vhost on VIRTULIZATION. So for now, let's just add something along the lines of

	depends on !ARM || AEABI

to the virtio Kconfig declaration, and add a comment that it has to do with struct member alignment.

Note: we can't make VHOST and VHOST_RING themselves have a dependency since these are selected. Add a new symbol for that.

We should be able to drop this dependency down the road.

Fixes 20c384f1ea1a0bc7 ("vhost: refine vhost and vringh kconfig")

---

changes from v3:
update commit log clarifying the motivation and that it's a temporary fix.

suggested by Christoph Hellwig

 drivers/misc/mic/Kconfig |  2 +-
 drivers/net/caif/Kconfig |  2 +-
 drivers/vdpa/Kconfig     |  2 +-
 drivers/vhost/Kconfig    | 17 +++++++++++++----
 4 files changed, 16 insertions(+), 7 deletions(-)
drivers/misc/mic/Kconfig +1 -1
index 8f201d019f5a..3bfe72c59864 100644
@@ -116,7 +116,7 @@ config MIC_COSM
 
 config VOP
 	tristate "VOP Driver"
-	depends on VOP_BUS
+	depends on VOP_BUS && VHOST_DPN
 	select VHOST_RING
 	select VIRTIO
 	help
drivers/net/caif/Kconfig +1 -1
index 9db0570c5beb..661c25eb1c46 100644
@@ -50,7 +50,7 @@ config CAIF_HSI
 
 config CAIF_VIRTIO
 	tristate "CAIF virtio transport driver"
-	depends on CAIF && HAS_DMA
+	depends on CAIF && HAS_DMA && VHOST_DPN
 	select VHOST_RING
 	select VIRTIO
 	select GENERIC_ALLOCATOR
drivers/vdpa/Kconfig +1 -1
index 3e1ceb8e9f2b..e8140065c8a5 100644
@@ -10,7 +10,7 @@ if VDPA
 
 config VDPA_SIM
 	tristate "vDPA device simulator"
-	depends on RUNTIME_TESTING_MENU && HAS_DMA
+	depends on RUNTIME_TESTING_MENU && HAS_DMA && VHOST_DPN
 	select VHOST_RING
 	default n
 	help
drivers/vhost/Kconfig +13 -4
index 2c75d164b827..c4f273793595 100644
@@ -13,6 +13,15 @@ config VHOST_RING
 	  This option is selected by any driver which needs to access
 	  the host side of a virtio ring.
 
+config VHOST_DPN
+	bool
+	depends on !ARM || AEABI
+	default y
+	help
+	  Anything selecting VHOST or VHOST_RING must depend on VHOST_DPN.
+	  This excludes the deprecated ARM ABI since that forces a 4 byte
+	  alignment on all structs - incompatible with virtio spec requirements.
+
 config VHOST
 	tristate
 	select VHOST_IOTLB
@@ -28,7 +37,7 @@ if VHOST_MENU
 
 config VHOST_NET
 	tristate "Host kernel accelerator for virtio net"
-	depends on NET && EVENTFD && (TUN || !TUN) && (TAP || !TAP)
+	depends on NET && EVENTFD && (TUN || !TUN) && (TAP || !TAP) && VHOST_DPN
 	select VHOST
 	---help---
 	  This kernel module can be loaded in host kernel to accelerate
@@ -40,7 +49,7 @@ config VHOST_NET
 
 config VHOST_SCSI
 	tristate "VHOST_SCSI TCM fabric driver"
-	depends on TARGET_CORE && EVENTFD
+	depends on TARGET_CORE && EVENTFD && VHOST_DPN
 	select VHOST
 	default n
 	---help---
@@ -49,7 +58,7 @@ config VHOST_SCSI
 
 config VHOST_VSOCK
 	tristate "vhost virtio-vsock driver"
-	depends on VSOCKETS && EVENTFD
+	depends on VSOCKETS && EVENTFD && VHOST_DPN
 	select VHOST
 	select VIRTIO_VSOCKETS_COMMON
 	default n
@@ -63,7 +72,7 @@ config VHOST_VSOCK
 
 config VHOST_VDPA
 	tristate "Vhost driver for vDPA-based backend"
-	depends on EVENTFD
+	depends on EVENTFD && VHOST_DPN
 	select VHOST
 	depends on VDPA
 	help
Signature · 1 line
MST
Acked-by on v4 1/1
#1 lore raw
Quoted 138 lines · Michael S. Tsirkin

vhost is currently broken on the some ARM configs.

The reason is that the ring element addresses are passed between components with different alignments assumptions. Thus, if guest selects a pointer and host then gets and dereferences it, then alignment assumed by the host's compiler might be greater than the actual alignment of the pointer. compiler on the host from assuming pointer is aligned.

This actually triggers on ARM with -mabi=apcs-gnu - which is a deprecated configuration. With this OABI, compiler assumes that all structures are 4 byte aligned - which is stronger than virtio guarantees for available and used rings, which are merely 2 bytes. Thus a guest without -mabi=apcs-gnu running on top of host with -mabi=apcs-gnu will be broken.

The correct fix is to force alignment of structures - however that is an intrusive fix that's best deferred until the next release.

We didn't previously support such ancient systems at all - this surfaced after vdpa support prompted removing dependency of vhost on VIRTULIZATION. So for now, let's just add something along the lines of

	depends on !ARM || AEABI

to the virtio Kconfig declaration, and add a comment that it has to do with struct member alignment.

Note: we can't make VHOST and VHOST_RING themselves have a dependency since these are selected. Add a new symbol for that.

We should be able to drop this dependency down the road.

Fixes 20c384f1ea1a0bc7 ("vhost: refine vhost and vringh kconfig")

---

changes from v3:
update commit log clarifying the motivation and that it's a temporary fix.

suggested by Christoph Hellwig

  drivers/misc/mic/Kconfig |  2 +-
  drivers/net/caif/Kconfig |  2 +-
  drivers/vdpa/Kconfig     |  2 +-
  drivers/vhost/Kconfig    | 17 +++++++++++++----
  4 files changed, 16 insertions(+), 7 deletions(-)
drivers/misc/mic/Kconfig +1 -1
index 8f201d019f5a..3bfe72c59864 100644
@@ -116,7 +116,7 @@ config MIC_COSM
 
  config VOP
  	tristate "VOP Driver"
-	depends on VOP_BUS
+	depends on VOP_BUS && VHOST_DPN
  	select VHOST_RING
  	select VIRTIO
  	help
drivers/net/caif/Kconfig +1 -1
index 9db0570c5beb..661c25eb1c46 100644
@@ -50,7 +50,7 @@ config CAIF_HSI
 
  config CAIF_VIRTIO
  	tristate "CAIF virtio transport driver"
-	depends on CAIF && HAS_DMA
+	depends on CAIF && HAS_DMA && VHOST_DPN
  	select VHOST_RING
  	select VIRTIO
  	select GENERIC_ALLOCATOR
drivers/vdpa/Kconfig +1 -1
index 3e1ceb8e9f2b..e8140065c8a5 100644
@@ -10,7 +10,7 @@ if VDPA
 
  config VDPA_SIM
  	tristate "vDPA device simulator"
-	depends on RUNTIME_TESTING_MENU && HAS_DMA
+	depends on RUNTIME_TESTING_MENU && HAS_DMA && VHOST_DPN
  	select VHOST_RING
  	default n
  	help
drivers/vhost/Kconfig +12 -3
index 2c75d164b827..c4f273793595 100644
@@ -13,6 +13,15 @@ config VHOST_RING
  	  This option is selected by any driver which needs to access
  	  the host side of a virtio ring.
 
+config VHOST_DPN
+	bool
+	depends on !ARM || AEABI
+	default y
+	help
+	  Anything selecting VHOST or VHOST_RING must depend on VHOST_DPN.
+	  This excludes the deprecated ARM ABI since that forces a 4 byte
+	  alignment on all structs - incompatible with virtio spec requirements.
+
  config VHOST
  	tristate
  	select VHOST_IOTLB
@@ -28,7 +37,7 @@ if VHOST_MENU
 
  config VHOST_NET
  	tristate "Host kernel accelerator for virtio net"
-	depends on NET && EVENTFD && (TUN || !TUN) && (TAP || !TAP)
+	depends on NET && EVENTFD && (TUN || !TUN) && (TAP || !TAP) && VHOST_DPN
  	select VHOST
  	---help---
  	  This kernel module can be loaded in host kernel to accelerate
@@ -40,7 +49,7 @@ config VHOST_NET
 
  config VHOST_SCSI
  	tristate "VHOST_SCSI TCM fabric driver"
-	depends on TARGET_CORE && EVENTFD
+	depends on TARGET_CORE && EVENTFD && VHOST_DPN
  	select VHOST
  	default n
  	---help---
@@ -49,7 +58,7 @@ config VHOST_SCSI
 
  config VHOST_VSOCK
  	tristate "vhost virtio-vsock driver"
-	depends on VSOCKETS && EVENTFD
+	depends on VSOCKETS && EVENTFD && VHOST_DPN
  	select VHOST
  	select VIRTIO_VSOCKETS_COMMON
  	default n

drivers/vhost/Kconfig config VHOST_VSOCK · line 78

Quoted 8 lines
drivers/vhost/Kconfig +1 -1
@@ -63,7 +72,7 @@ config VHOST_VSOCK
 
  config VHOST_VDPA
  	tristate "Vhost driver for vDPA-based backend"
-	depends on EVENTFD
+	depends on EVENTFD && VHOST_DPN
  	select VHOST
  	depends on VDPA
  	help

Thanks

#2 lore raw
In reply to Michael S. Tsirkin

Hi Michael,

Thanks for your patch!

Quoted 23 lines · Michael S. Tsirkin

vhost is currently broken on the some ARM configs.

The reason is that the ring element addresses are passed between components with different alignments assumptions. Thus, if guest selects a pointer and host then gets and dereferences it, then alignment assumed by the host's compiler might be greater than the actual alignment of the pointer. compiler on the host from assuming pointer is aligned.

This actually triggers on ARM with -mabi=apcs-gnu - which is a deprecated configuration. With this OABI, compiler assumes that all structures are 4 byte aligned - which is stronger than virtio guarantees for available and used rings, which are merely 2 bytes. Thus a guest without -mabi=apcs-gnu running on top of host with -mabi=apcs-gnu will be broken.

The correct fix is to force alignment of structures - however that is an intrusive fix that's best deferred until the next release.

We didn't previously support such ancient systems at all - this surfaced after vdpa support prompted removing dependency of vhost on VIRTULIZATION. So for now, let's just add something along the lines of

Quoted 7 lines
        depends on !ARM || AEABI

to the virtio Kconfig declaration, and add a comment that it has to do with struct member alignment.

Note: we can't make VHOST and VHOST_RING themselves have a dependency since these are selected. Add a new symbol for that.

Adding the dependencies to VHOST and VHOST_RING themselves is indeed not sufficient. But IMHO you should still add VHOST_DPN dependencies t
these two symbols, so any driver selecting them without fulfilling the
VHOST_DPN dependency will trigger a Kconfig warning. Else the issue will be ignored silently.

Quoted 6 lines

We should be able to drop this dependency down the road.

Fixes 20c384f1ea1a0bc7 ("vhost: refine vhost and vringh kconfig")

Gr{oetje,eeting}s,

Geert

Signature · 5 lines
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- [email protected]

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds
author maintainer
#3 lore raw
Quoted 33 lines · Geert Uytterhoeven

Hi Michael,

Thanks for your patch!

Quoted 28 lines · Michael S. Tsirkin

vhost is currently broken on the some ARM configs.

The reason is that the ring element addresses are passed between components with different alignments assumptions. Thus, if guest selects a pointer and host then gets and dereferences it, then alignment assumed by the host's compiler might be greater than the actual alignment of the pointer. compiler on the host from assuming pointer is aligned.

This actually triggers on ARM with -mabi=apcs-gnu - which is a deprecated configuration. With this OABI, compiler assumes that all structures are 4 byte aligned - which is stronger than virtio guarantees for available and used rings, which are merely 2 bytes. Thus a guest without -mabi=apcs-gnu running on top of host with -mabi=apcs-gnu will be broken.

The correct fix is to force alignment of structures - however that is an intrusive fix that's best deferred until the next release.

We didn't previously support such ancient systems at all - this surfaced after vdpa support prompted removing dependency of vhost on VIRTULIZATION. So for now, let's just add something along the lines of

        depends on !ARM || AEABI

to the virtio Kconfig declaration, and add a comment that it has to do with struct member alignment.

Quoted 8 lines
Quoted 2 lines

Note: we can't make VHOST and VHOST_RING themselves have a dependency since these are selected. Add a new symbol for that.

Adding the dependencies to VHOST and VHOST_RING themselves is indeed not sufficient. But IMHO you should still add VHOST_DPN dependencies t
these two symbols, so any driver selecting them without fulfilling the
VHOST_DPN dependency will trigger a Kconfig warning. Else the issue will be ignored silently.

Good point.
For now I'm trying to just get rid of this work around. If I can't I will add the suggested change. Thanks!

Quoted 17 lines
Quoted 6 lines

We should be able to drop this dependency down the road.

Fixes 20c384f1ea1a0bc7 ("vhost: refine vhost and vringh kconfig")

Gr{oetje,eeting}s,

Geert

Signature · 5 lines
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- [email protected]

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds