hw/tpm: fix tpm-tis-device backward migration on aarch64
v2 cover hw/tpm: fix tpm-tis-device backward migration on aarch64
46cd2c1050f0 ("hw/tpm: add PPI support to tpm-tis-device for ARM64 virt") unconditionally registers a "tpm-ppi" RAMBlock when the TPM TIS sysbus device is realized. This breaks backward migration: a QEMU with PPI support cannot migrate to an older QEMU without it:
Unknown ramblock "tpm-ppi", cannot accept migration
The failure is 100% reproducible on any aarch64 guest that uses tpm-tis-device when migrating to a host whose QEMU predates the PPI commit.
Fix by gating the PPI memory region behind a new "ppi" device property, and disabling it via hw_compat_11_0[] for machine types <= virt-11.0. Also move the sysbus_init_mmio() calls from instance_init to realizefn so the memory regions are properly initialized before being registered (Peter).
While testing I also found that not exposing the 2nd MMIO slot causes platform_bus_get_mmio_addr() to crash: sysbus_mmio_get_region() returns NULL for the unregistered slot, and the function passes it to memory_region_is_mapped() without a NULL check. Patch 3 adds a one-line guard so callers get -1 instead of a SIGSEGV.
Changes in v2:
- Dropped zero-size memory region hack (Peter)
- Moved sysbus_init_mmio() from initfn to realizefn (Peter)
- Only expose PPI MMIO region when ppi=on (Peter)
- Added NULL guard in platform_bus_get_mmio_addr()
- v1: https://lore.kernel.org/qemu-devel/[email protected]
Mohammadfaiz Bawa (3):
hw/tpm: gate PPI support on tpm-tis-device behind a device property hw/core/machine: disable tpm-tis-device PPI for machine type <= 11.0 hw/core/platform-bus: guard platform_bus_get_mmio_addr() against NULL
hw/core/machine.c | 1 + hw/core/platform-bus.c | 3 +-- hw/tpm/tpm_tis.h | 1 + hw/tpm/tpm_tis_sysbus.c | 18 +++++++++++------- 4 files changed, 14 insertions(+), 9 deletions(-)
Signature · 1 line
2.54.0
PATCH v2 1/3 hw/tpm: gate PPI support on tpm-tis-device behind a device property
Add a "ppi" boolean property (default: true) to tpm-tis-device. When ppi=off the RAMBlock is never registered and the migration stream omits "tpm-ppi", restoring backward compatibility.
Move sysbus_init_mmio() calls from instance_init to realizefn so the memory regions are initialized before being registered. When ppi=off, the PPI MMIO region is simply not exposed.
--- hw/tpm/tpm_tis.h | 1 + hw/tpm/tpm_tis_sysbus.c | 18 +++++++++++------- 2 files changed, 12 insertions(+), 7 deletions(-)
hw/tpm/tpm_tis.h +1 -0
@@ -76,6 +76,7 @@ typedef struct TPMState {size_t be_buffer_size;TPMPPI ppi; +bool ppi_enabled; } TPMState;extern const VMStateDescription vmstate_locty;
hw/tpm/tpm_tis_sysbus.c +11 -7
@@ -94,6 +94,7 @@ static void tpm_tis_sysbus_reset(DeviceState *dev)static const Property tpm_tis_sysbus_properties[] = {DEFINE_PROP_UINT32 ("irq" , TPMStateSysBus, state.irq_num,TPM_TIS_IRQ ),DEFINE_PROP_TPMBE ("tpmdev" , TPMStateSysBus, state.be_driver), +DEFINE_PROP_BOOL ("ppi" , TPMStateSysBus, state.ppi_enabled, true), };static void tpm_tis_sysbus_initfn (Object *obj)
@@ -101,9 +102,7 @@ static void tpm_tis_sysbus_initfn(Object *obj)TPMStateSysBus *sbdev =TPM_TIS_SYSBUS (obj);TPMState *s = &sbdev->state; -sysbus_init_mmio (SYS_BUS_DEVICE (obj), &s->mmio);sysbus_init_irq (SYS_BUS_DEVICE (obj), &s->irq); -sysbus_init_mmio (SYS_BUS_DEVICE (obj), &s->ppi.ram); }static void tpm_tis_sysbus_realizefn (DeviceState *dev,Error **errp)
@@ -122,14 +121,19 @@ static void tpm_tis_sysbus_realizefn(DeviceState *dev, Error **errp)return ; } - s->ppi.buf =qemu_memalign (host_page_size, -ROUND_UP (TPM_PPI_ADDR_SIZE , host_page_size));memory_region_init_io (&s->mmio,OBJECT (dev), &tpm_tis_memory_ops, s,"tpm-tis-mmio" ,TPM_TIS_NUM_LOCALITIES <<TPM_TIS_LOCALITY_SHIFT ); -memory_region_init_ram_device_ptr (&s->ppi.ram,OBJECT (dev),"tpm-ppi" , -TPM_PPI_ADDR_SIZE , s->ppi.buf); -vmstate_register_ram (&s->ppi.ram, dev); +sysbus_init_mmio (SYS_BUS_DEVICE (dev), &s->mmio); + +if (s->ppi_enabled) { + s->ppi.buf =qemu_memalign (host_page_size, +ROUND_UP (TPM_PPI_ADDR_SIZE , host_page_size)); +memory_region_init_ram_device_ptr (&s->ppi.ram,OBJECT (dev),"tpm-ppi" , +TPM_PPI_ADDR_SIZE , s->ppi.buf); +sysbus_init_mmio (SYS_BUS_DEVICE (dev), &s->ppi.ram); +vmstate_register_ram (&s->ppi.ram, dev); + } }static void tpm_tis_sysbus_class_init (ObjectClass *klass,const void *data)
Signature · 1 line
2.54.0
PATCH v2 2/3 hw/core/machine: disable tpm-tis-device PPI for machine type <= 11.0
In reply to Mohammadfaiz BawaSet ppi=off in hw_compat_11_0[] so that older machine types do not register the "tpm-ppi" RAMBlock, restoring backward migration.
--- hw/core/machine.c | 1 + 1 file changed, 1 insertion(+)
hw/core/machine.c +1 -0
@@ -44,6 +44,7 @@ GlobalProperty hw_compat_11_0[] = { { "chardev-vc", "encoding", "cp437" }, { "tpm-crb", "cap-chunk", "off" }, { "tpm-crb", "x-allow-chunk-migration", "off" }, + { "tpm-tis-device", "ppi", "off" }, { TYPE_ARM_SMMUV3, "ats", "off" }, { TYPE_ARM_SMMUV3, "ril", "on" }, { TYPE_ARM_SMMUV3, "ssidsize", "0" },
Signature · 1 line
2.54.0
PATCH v2 3/3 hw/core/platform-bus: guard platform_bus_get_mmio_addr() against NULL
In reply to Mohammadfaiz Bawasysbus_mmio_get_region() returns NULL when a device has fewer MMIO regions than the requested slot index. platform_bus_get_mmio_addr() passes the result directly to memory_region_is_mapped() without a NULL check, causing a SIGSEGV.
Return -1 early when the region pointer is NULL, consistent with the existing "not mapped" path.
--- hw/core/platform-bus.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-)
hw/core/platform-bus.c +1 -2
@@ -59,8 +59,7 @@ hwaddr platform_bus_get_mmio_addr(PlatformBusDevice *pbus, SysBusDevice *sbdev,Object *pbus_mr_obj =OBJECT (pbus_mr);Object *parent_mr; -if (!memory_region_is_mapped (sbdev_mr)) { -/* Region is not mapped? */ +if (!sbdev_mr || !memory_region_is_mapped (sbdev_mr)) {return -1 ; }
Signature · 1 line
2.54.0
Re: [PATCH v2 3/3] hw/core/platform-bus: guard platform_bus_get_mmio_addr() against NULL
Quoted 19 lines · Mohammadfaiz Bawa
sysbus_mmio_get_region()returns NULL when a device has fewer MMIO regions than the requested slot index.platform_bus_get_mmio_addr()passes the result directly tomemory_region_is_mapped()without a NULL check, causing a SIGSEGV.Return -1 early when the region pointer is NULL, consistent with the existing "not mapped" path.
Signed-off-by Mohammadfaiz Bawa <[email protected]>--- hw/core/platform-bus.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-)hw/core/platform-bus.c +0
-0
hw/core/platform-bus.c hwaddr platform_bus_get_mmio_addr(PlatformBusDevice *pbus, SysBusDevice *sbdev, · line 64
Quoted 8 lines
hw/core/platform-bus.c +1
-2@@ -59,8 +59,7 @@ hwaddr platform_bus_get_mmio_addr(PlatformBusDevice *pbus, SysBusDevice *sbdev,Object *pbus_mr_obj =OBJECT (pbus_mr);Object *parent_mr; -if (!memory_region_is_mapped (sbdev_mr)) { -/* Region is not mapped? */ +if (!sbdev_mr || !memory_region_is_mapped (sbdev_mr)) {return -1 ; }
but doesn't this need to come earlier in the series, so that the code can handle the NULL pointer before we change the device/machine to not expose the 2nd MR?
thanks
-- PMM
Re: [PATCH v2 3/3] hw/core/platform-bus: guard platform_bus_get_mmio_addr() against NULL
Quoted 32 lines · Peter Maydell
Quoted 19 lines · Mohammadfaiz Bawa
sysbus_mmio_get_region()returns NULL when a device has fewer MMIO regions than the requested slot index.platform_bus_get_mmio_addr()passes the result directly tomemory_region_is_mapped()without a NULL check, causing a SIGSEGV.Return -1 early when the region pointer is NULL, consistent with the existing "not mapped" path.
Signed-off-by Mohammadfaiz Bawa <[email protected]>--- hw/core/platform-bus.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-)hw/core/platform-bus.c +0
-0hw/core/platform-bus.c hwaddr platform_bus_get_mmio_addr(PlatformBusDevice *pbus, SysBusDevice *sbdev, · line 64
Quoted 8 lines
hw/core/platform-bus.c +1
-2@@ -59,8 +59,7 @@ hwaddr platform_bus_get_mmio_addr(PlatformBusDevice *pbus, SysBusDevice *sbdev,Object *pbus_mr_obj =OBJECT (pbus_mr);Object *parent_mr; -if (!memory_region_is_mapped (sbdev_mr)) { -/* Region is not mapped? */ +if (!sbdev_mr || !memory_region_is_mapped (sbdev_mr)) {return -1 ; }Reviewed-by Peter Maydell <[email protected]>
Quoted 8 lines
but doesn't this need to come earlier in the series, so that the code can handle the NULL pointer before we change the device/machine to not expose the 2nd MR?
thanks
-- PMM
Right, I didn't think about bisectability,
I put the patches in the order I made the changes. v3 then
Thanks,
Faiz