xen/arm: Inject an exception to the guest rather than crashing it
cover xen/arm: Inject an exception to the guest rather than crashing it
Hi all,
This small series replaces all call to domain_crash_synchronous by injecting an exception to the guest.
This will result to a nicer trace from the guest (no need to manually walk the stack) and give a chance to the guest to give a bit more information on what it was doing.
Cheers,
Julien Grall (3):
xen/arm: io: Distinguish unhandled IO from aborted one xen/arm: Don't crash domain on bad MMIO emulation xen/arm: Don't crash the domain on invalid HVC immediate
xen/arch/arm/io.c | 24 +++++++++++++---------- xen/arch/arm/traps.c | 47 ++++++++++++++++++++++++++++++---------------- xen/arch/arm/vgic-v2.c | 2 -- xen/arch/arm/vgic-v3-its.c | 3 --- xen/arch/arm/vgic-v3.c | 8 -------- xen/arch/arm/vpl011.c | 2 -- xen/include/asm-arm/mmio.h | 9 ++++++++- 7 files changed, 53 insertions(+), 42 deletions(-)
Signature · 7 lines
2.11.0 _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
PATCH 1/3 xen/arm: io: Distinguish unhandled IO from aborted one
Currently, Xen is considering that an IO could either be handled or unhandled. When unhandled, the stage-2 abort function will try another way to resolve the abort.
However, the MMIO emulation may return unhandled when the address belongs to an emulated range but was not correct. In that case, Xen should avodi to try another way and directly inject a guest data abort.
Introduce a tri-state return to distinguish the following state:
* IO_ABORT: The IO was handled but resulted to an abort
* IO_HANDLED: The IO was handled
* IO_UNHANDLED: The IO was unhandled
For now, it is considered that an IO belonging to an emulated range could either be handled or inject an abort. This could be revisit in the future if overlapped region exist (or we want to try another way to resolve the abort).
--- xen/arch/arm/io.c | 24 ++++++++++++++---------- xen/arch/arm/traps.c | 34 +++++++++++++++++++++++----------- xen/include/asm-arm/mmio.h | 9 ++++++++- 3 files changed, 45 insertions(+), 22 deletions(-)
xen/arch/arm/io.c +14 -10
@@ -23,8 +23,9 @@#include <asm/current.h> #include <asm/mmio.h> -static int handle_read (const struct mmio_handler *handler,struct vcpu *v, -mmio_info_t *info) +static enum io_state handle_read (const struct mmio_handler *handler, +struct vcpu *v, +mmio_info_t *info) {const struct hsr_dabt dabt = info->dabt;struct cpu_user_regs *regs =guest_cpu_user_regs ();
@@ -37,7 +38,7 @@ static int handle_read(const struct mmio_handler *handler, struct vcpu *v,uint8_t size = (1 << dabt.size) *8 ;if ( !handler->ops->read (v, info, &r, handler->priv) ) -return 0 ; +return IO_ABORT ;/* * Sign extend if required.
@@ -57,17 +58,20 @@ static int handle_read(const struct mmio_handler *handler, struct vcpu *v,set_user_reg (regs, dabt.reg, r); -return 1 ; +return IO_HANDLED ; } -static int handle_write (const struct mmio_handler *handler,struct vcpu *v, -mmio_info_t *info) +static enum io_state handle_write (const struct mmio_handler *handler, +struct vcpu *v, +mmio_info_t *info) {const struct hsr_dabt dabt = info->dabt;struct cpu_user_regs *regs =guest_cpu_user_regs (); +int ret; -return handler->ops->write (v, info,get_user_reg (regs, dabt.reg), - handler->priv); + ret = handler->ops->write (v, info,get_user_reg (regs, dabt.reg), + handler->priv); +return ( ret ) ?IO_HANDLED :IO_ABORT ; }/* This function assumes that mmio regions are not overlapped */
@@ -100,14 +104,14 @@ static const struct mmio_handler *find_mmio_handler(struct domain *d,return handler; } -int handle_mmio (mmio_info_t *info) +enum io_state handle_mmio (mmio_info_t *info) {struct vcpu *v = current;const struct mmio_handler *handler =NULL ; handler =find_mmio_handler (v->domain, info->gpa);if ( !handler ) -return 0 ; +return IO_UNHANDLED ;if ( info->dabt.write )return handle_write (handler, v, info);
xen/arch/arm/traps.c +23 -11
@@ -1864,10 +1864,10 @@ static inline bool hpfar_is_valid(bool s1ptw, uint8_t fsc)return s1ptw || (fsc ==FSC_FLT_TRANS && !check_workaround_834220 ()); } -static bool try_handle_mmio (struct cpu_user_regs *regs, -const union hsr hsr, -paddr_t gpa) -{ +static enum io_state try_handle_mmio (struct cpu_user_regs *regs, +const union hsr hsr, +paddr_t gpa) + {const struct hsr_dabt dabt = hsr.dabt;mmio_info_t info = { .gpa = gpa,
@@ -1879,11 +1879,11 @@ static bool try_handle_mmio(struct cpu_user_regs *regs,/* stage-1 page table should never live in an emulated MMIO region */ if ( dabt.s1ptw ) -return false; +return IO_UNHANDLED ;/* All the instructions used on emulated MMIO region should be valid */ if ( !dabt.valid ) -return false; +return IO_UNHANDLED ;/* * Erratum 766422: Thumb store translation fault to Hypervisor may
@@ -1896,11 +1896,11 @@ static bool try_handle_mmio(struct cpu_user_regs *regs,if ( rc ) {gprintk (XENLOG_DEBUG ,"Unable to decode instruction\n" ); -return false; +return IO_ABORT ; } } -return !!handle_mmio (&info); +return handle_mmio (&info); }/*
@@ -2005,10 +2005,21 @@ static void do_trap_stage2_abort_guest(struct cpu_user_regs *regs,* * Note that emulated region cannot be executed */ -if ( is_data &&try_handle_mmio (regs, hsr, gpa) ) +if ( is_data ) { -advance_pc (regs, hsr); -return ; +enum io_state state =try_handle_mmio (regs, hsr, gpa); + +switch ( state ) + { +case IO_ABORT : + goto inject_abt; +case IO_HANDLED : +advance_pc (regs, hsr); +return ; +case IO_UNHANDLED : +/* Nothing to do */ +break ; + } }/*
@@ -2029,6 +2040,7 @@ static void do_trap_stage2_abort_guest(struct cpu_user_regs *regs, hsr.bits, xabt.fsc); } +inject_abt:gdprintk (XENLOG_DEBUG ,"HSR=0x%x pc=%#" PRIregister" gva=%#" PRIvaddr" gpa=%#" PRIpaddr"\n" , hsr.bits, regs->pc, gva, gpa);if ( is_data )
xen/include/asm-arm/mmio.h +8 -1
@@ -32,6 +32,13 @@ typedef structpaddr_t gpa; } mmio_info_t; +enum io_state +{ +IO_ABORT ,/* The IO was handled by the helper and lead to an abort. */ +IO_HANDLED ,/* The IO was successfully handled by the helper. */ +IO_UNHANDLED ,/* The IO was not handled by the helper. */ +}; +typedef int (*mmio_read_t )(struct vcpu *v,mmio_info_t *info,register_t *r,void *priv); typedef int (*mmio_write_t)(struct vcpu *v, mmio_info_t *info,
@@ -56,7 +63,7 @@ struct vmmio {struct mmio_handler *handlers; }; -extern int handle_mmio (mmio_info_t *info); +extern enum io_state handle_mmio (mmio_info_t *info);void register_mmio_handler (struct domain *d,const struct mmio_handler_ops *ops,paddr_t addr,paddr_t size,void *priv);
Signature · 7 lines
2.11.0 _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH 1/3] xen/arm: io: Distinguish unhandled IO from aborted one
Quoted 7 lines · Julien Grall
Currently, Xen is considering that an IO could either be handled or unhandled. When unhandled, the stage-2 abort function will try another way to resolve the abort.
However, the MMIO emulation may return unhandled when the address belongs to an emulated range but was not correct. In that case, Xen should avodi to try another way and directly inject a guest data abort.
^ avoid
Quoted 2 lines
Introduce a tri-state return to distinguish the following state:
*IO_ABORT: The IO was handled but resulted to an abort
^ in an abort
Quoted 117 lines
*
IO_HANDLED: The IO was handled
*IO_UNHANDLED: The IO was unhandledFor now, it is considered that an IO belonging to an emulated range could either be handled or inject an abort. This could be revisit in the future if overlapped region exist (or we want to try another way to resolve the abort).
Signed-off-by Julien Grall <[email protected]>--- xen/arch/arm/io.c | 24 ++++++++++++++---------- xen/arch/arm/traps.c | 34 +++++++++++++++++++++++----------- xen/include/asm-arm/mmio.h | 9 ++++++++- 3 files changed, 45 insertions(+), 22 deletions(-)xen/arch/arm/io.c +14
-10@@ -23,8 +23,9 @@#include <asm/current.h> #include <asm/mmio.h> -static int handle_read (const struct mmio_handler *handler,struct vcpu *v, -mmio_info_t *info) +static enum io_state handle_read (const struct mmio_handler *handler, +struct vcpu *v, +mmio_info_t *info) {const struct hsr_dabt dabt = info->dabt;struct cpu_user_regs *regs =guest_cpu_user_regs ();@@ -37,7 +38,7 @@ static int handle_read(const struct mmio_handler *handler, struct vcpu *v,uint8_t size = (1 << dabt.size) *8 ;if ( !handler->ops->read (v, info, &r, handler->priv) ) -return 0 ; +return IO_ABORT ;/* * Sign extend if required. @@ -57,17 +58,20 @@ static int handle_read(const struct mmio_handler *handler, struct vcpu *v,set_user_reg (regs, dabt.reg, r); -return 1 ; +return IO_HANDLED ; } -static int handle_write (const struct mmio_handler *handler,struct vcpu *v, -mmio_info_t *info) +static enum io_state handle_write (const struct mmio_handler *handler, +struct vcpu *v, +mmio_info_t *info) {const struct hsr_dabt dabt = info->dabt;struct cpu_user_regs *regs =guest_cpu_user_regs (); +int ret; -return handler->ops->write (v, info,get_user_reg (regs, dabt.reg), - handler->priv); + ret = handler->ops->write (v, info,get_user_reg (regs, dabt.reg), + handler->priv); +return ( ret ) ?IO_HANDLED :IO_ABORT ; }/* This function assumes that mmio regions are not overlapped */ @@ -100,14 +104,14 @@ static const struct mmio_handler *find_mmio_handler(struct domain *d,return handler; } -int handle_mmio (mmio_info_t *info) +enum io_state handle_mmio (mmio_info_t *info) {struct vcpu *v = current;const struct mmio_handler *handler =NULL ; handler =find_mmio_handler (v->domain, info->gpa);if ( !handler ) -return 0 ; +return IO_UNHANDLED ;if ( info->dabt.write )return handle_write (handler, v, info);xen/arch/arm/traps.c +6
-6@@ -1864,10 +1864,10 @@ static inline bool hpfar_is_valid(bool s1ptw, uint8_t fsc)return s1ptw || (fsc ==FSC_FLT_TRANS && !check_workaround_834220 ()); } -static bool try_handle_mmio (struct cpu_user_regs *regs, -const union hsr hsr, -paddr_t gpa) -{ +static enum io_state try_handle_mmio (struct cpu_user_regs *regs, +const union hsr hsr, +paddr_t gpa) + {const struct hsr_dabt dabt = hsr.dabt;mmio_info_t info = { .gpa = gpa,@@ -1879,11 +1879,11 @@ static bool try_handle_mmio(struct cpu_user_regs *regs,/* stage-1 page table should never live in an emulated MMIO region */ if ( dabt.s1ptw ) -return false; +return IO_UNHANDLED ;/* All the instructions used on emulated MMIO region should be valid */ if ( !dabt.valid ) -return false; +return IO_UNHANDLED ;/* * Erratum 766422: Thumb store translation fault to Hypervisor may
xen/arch/arm/traps.c static bool try_handle_mmio(struct cpu_user_regs *regs, · line 1901
Quoted 7 lines
xen/arch/arm/traps.c +1
-1@@ -1896,11 +1896,11 @@ static bool try_handle_mmio(struct cpu_user_regs *regs,if ( rc ) {gprintk (XENLOG_DEBUG ,"Unable to decode instruction\n" ); -return false; +return IO_ABORT ; } }
Why do the first two error checks result in IO_UNHANDLED, while the third result in IO_ABORT? Specifically in relation to pagetable walk failures due to someone else changing stage-2 entry simultaneously (see comment toward the end of do_trap_stage2_abort_guest)?
Quoted 43 lines
xen/arch/arm/traps.c +16
-4-return !!handle_mmio (&info); +return handle_mmio (&info); }/* @@ -2005,10 +2005,21 @@ static void do_trap_stage2_abort_guest(struct cpu_user_regs *regs,* * Note that emulated region cannot be executed */ -if ( is_data &&try_handle_mmio (regs, hsr, gpa) ) +if ( is_data ) { -advance_pc (regs, hsr); -return ; +enum io_state state =try_handle_mmio (regs, hsr, gpa); + +switch ( state ) + { +case IO_ABORT : + goto inject_abt; +case IO_HANDLED : +advance_pc (regs, hsr); +return ; +case IO_UNHANDLED : +/* Nothing to do */ +break ; + } }/* @@ -2029,6 +2040,7 @@ static void do_trap_stage2_abort_guest(struct cpu_user_regs *regs, hsr.bits, xabt.fsc); } +inject_abt:gdprintk (XENLOG_DEBUG ,"HSR=0x%x pc=%#" PRIregister" gva=%#" PRIvaddr" gpa=%#" PRIpaddr"\n" , hsr.bits, regs->pc, gva, gpa);if ( is_data )xen/include/asm-arm/mmio.h +0
-0
xen/include/asm-arm/mmio.h typedef struct · line 37
Quoted 6 lines
xen/include/asm-arm/mmio.h +3
-0@@ -32,6 +32,13 @@ typedef structpaddr_t gpa; } mmio_info_t; +enum io_state +{ +IO_ABORT ,/* The IO was handled by the helper and lead to an abort. */
^ led
Quoted 19 lines
xen/include/asm-arm/mmio.h +5
-1+IO_HANDLED ,/* The IO was successfully handled by the helper. */ + IO_UNHANDLED, /* The IO was not handled by the helper. */ +}; +typedef int (*mmio_read_t )(struct vcpu *v,mmio_info_t *info,register_t *r,void *priv); typedef int (*mmio_write_t)(struct vcpu *v, mmio_info_t *info,@@ -56,7 +63,7 @@ struct vmmio {struct mmio_handler *handlers; }; -extern int handle_mmio (mmio_info_t *info); +extern enum io_state handle_mmio (mmio_info_t *info);void register_mmio_handler (struct domain *d,const struct mmio_handler_ops *ops,paddr_t addr,paddr_t size,void *priv);Signature · 1 line
2.11.0
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH 1/3] xen/arm: io: Distinguish unhandled IO from aborted one
Hi Stefano,
Quoted 130 lines · Stefano Stabellini
Quoted 7 lines · Julien Grall
Currently, Xen is considering that an IO could either be handled or unhandled. When unhandled, the stage-2 abort function will try another way to resolve the abort.
However, the MMIO emulation may return unhandled when the address belongs to an emulated range but was not correct. In that case, Xen should avodi to try another way and directly inject a guest data abort.
^ avoid
Quoted 2 lines
Introduce a tri-state return to distinguish the following state:
*IO_ABORT: The IO was handled but resulted to an abort^ in an abort
Quoted 114 lines
*
IO_HANDLED: The IO was handled
*IO_UNHANDLED: The IO was unhandledFor now, it is considered that an IO belonging to an emulated range could either be handled or inject an abort. This could be revisit in the future if overlapped region exist (or we want to try another way to resolve the abort).
Signed-off-by Julien Grall <[email protected]>--- xen/arch/arm/io.c | 24 ++++++++++++++---------- xen/arch/arm/traps.c | 34 +++++++++++++++++++++++----------- xen/include/asm-arm/mmio.h | 9 ++++++++- 3 files changed, 45 insertions(+), 22 deletions(-)xen/arch/arm/io.c +14
-10@@ -23,8 +23,9 @@#include <asm/current.h> #include <asm/mmio.h> -static int handle_read (const struct mmio_handler *handler,struct vcpu *v, -mmio_info_t *info) +static enum io_state handle_read (const struct mmio_handler *handler, +struct vcpu *v, +mmio_info_t *info) {const struct hsr_dabt dabt = info->dabt;struct cpu_user_regs *regs =guest_cpu_user_regs ();@@ -37,7 +38,7 @@ static int handle_read(const struct mmio_handler *handler, struct vcpu *v,uint8_t size = (1 << dabt.size) *8 ;if ( !handler->ops->read (v, info, &r, handler->priv) ) -return 0 ; +return IO_ABORT ;/* * Sign extend if required. @@ -57,17 +58,20 @@ static int handle_read(const struct mmio_handler *handler, struct vcpu *v,set_user_reg (regs, dabt.reg, r); -return 1 ; +return IO_HANDLED ; } -static int handle_write (const struct mmio_handler *handler,struct vcpu *v, -mmio_info_t *info) +static enum io_state handle_write (const struct mmio_handler *handler, +struct vcpu *v, +mmio_info_t *info) {const struct hsr_dabt dabt = info->dabt;struct cpu_user_regs *regs =guest_cpu_user_regs (); +int ret; -return handler->ops->write (v, info,get_user_reg (regs, dabt.reg), - handler->priv); + ret = handler->ops->write (v, info,get_user_reg (regs, dabt.reg), + handler->priv); +return ( ret ) ?IO_HANDLED :IO_ABORT ; }/* This function assumes that mmio regions are not overlapped */ @@ -100,14 +104,14 @@ static const struct mmio_handler *find_mmio_handler(struct domain *d,return handler; } -int handle_mmio (mmio_info_t *info) +enum io_state handle_mmio (mmio_info_t *info) {struct vcpu *v = current;const struct mmio_handler *handler =NULL ; handler =find_mmio_handler (v->domain, info->gpa);if ( !handler ) -return 0 ; +return IO_UNHANDLED ;if ( info->dabt.write )return handle_write (handler, v, info);xen/arch/arm/traps.c +6
-6@@ -1864,10 +1864,10 @@ static inline bool hpfar_is_valid(bool s1ptw, uint8_t fsc)return s1ptw || (fsc ==FSC_FLT_TRANS && !check_workaround_834220 ()); } -static bool try_handle_mmio (struct cpu_user_regs *regs, -const union hsr hsr, -paddr_t gpa) -{ +static enum io_state try_handle_mmio (struct cpu_user_regs *regs, +const union hsr hsr, +paddr_t gpa) + {const struct hsr_dabt dabt = hsr.dabt;mmio_info_t info = { .gpa = gpa,@@ -1879,11 +1879,11 @@ static bool try_handle_mmio(struct cpu_user_regs *regs,/* stage-1 page table should never live in an emulated MMIO region */ if ( dabt.s1ptw ) -return false; +return IO_UNHANDLED ;/* All the instructions used on emulated MMIO region should be valid */ if ( !dabt.valid ) -return false; +return IO_UNHANDLED ;
Quoted 15 lines
xen/arch/arm/traps.c static bool try_handle_mmio(struct cpu_user_regs *regs, · line 1901
Quoted 10 lines
xen/arch/arm/traps.c +1
-1/* * Erratum 766422: Thumb store translation fault to Hypervisor may @@ -1896,11 +1896,11 @@ static bool try_handle_mmio(struct cpu_user_regs *regs,if ( rc ) {gprintk (XENLOG_DEBUG ,"Unable to decode instruction\n" ); -return false; +return IO_ABORT ; } }Why do the first two error checks result in
IO_UNHANDLED, while the third result inIO_ABORT? Specifically in relation to pagetable walk failures due to someone else changing stage-2 entry simultaneously (see comment toward the end ofdo_trap_stage2_abort_guest)?
Good question. Somehow I considered the first two as part of looking up the proper handler and not the device itself.
But I guess that at this stage we know that IO was targeting an emulated region. So we can return IO_ABORT.
On a side note, it looks like the check dabt.s1ptw is unnecessary because a stage 2 abort on stage 1 translation table lookup will not return a valid instruction syndrome (see B3-1433 in DDI 0406C.c and D10-2460 in DDI 0487C.a).
Cheers,
Signature · 6 lines
Julien Grall _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH 1/3] xen/arm: io: Distinguish unhandled IO from aborted one
Quoted 145 lines · Julien Grall
Hi Stefano,
Quoted 133 lines · Stefano Stabellini
Quoted 7 lines · Julien Grall
Currently, Xen is considering that an IO could either be handled or unhandled. When unhandled, the stage-2 abort function will try another way to resolve the abort.
However, the MMIO emulation may return unhandled when the address belongs to an emulated range but was not correct. In that case, Xen should avodi to try another way and directly inject a guest data abort.
^ avoid
Quoted 2 lines
Introduce a tri-state return to distinguish the following state:
*IO_ABORT: The IO was handled but resulted to an abort^ in an abort
Quoted 117 lines
*
IO_HANDLED: The IO was handled
*IO_UNHANDLED: The IO was unhandledFor now, it is considered that an IO belonging to an emulated range could either be handled or inject an abort. This could be revisit in the future if overlapped region exist (or we want to try another way to resolve the abort).
Signed-off-by Julien Grall <[email protected]>--- xen/arch/arm/io.c | 24 ++++++++++++++---------- xen/arch/arm/traps.c | 34 +++++++++++++++++++++++----------- xen/include/asm-arm/mmio.h | 9 ++++++++- 3 files changed, 45 insertions(+), 22 deletions(-)xen/arch/arm/io.c +14
-5@@ -23,8 +23,9 @@#include <asm/current.h> #include <asm/mmio.h> -static int handle_read(const struct mmio_handler *handler, struct vcpu *v, -mmio_info_t *info) +static enum io_state handle_read (const struct mmio_handler *handler, +struct vcpu *v, + mmio_info_t *info) {const struct hsr_dabt dabt = info->dabt;struct cpu_user_regs *regs =guest_cpu_user_regs ();@@ -37,7 +38,7 @@ static int handle_read(const struct mmio_handler *handler, struct vcpu *v, uint8_t size = (1 << dabt.size) * 8;if ( !handler->ops->read (v, info, &r, handler->priv) ) -return 0 ; +return IO_ABORT ;/* * Sign extend if required. @@ -57,17 +58,20 @@ static int handle_read(const struct mmio_handler *handler, struct vcpu *v,set_user_reg (regs, dabt.reg, r); - return 1; +return IO_HANDLED ; } -static int handle_write(const struct mmio_handler *handler, struct vcpu *v, -mmio_info_t *info) +static enum io_state handle_write (const struct mmio_handler *handler, +struct vcpu *v, + mmio_info_t *info) {const struct hsr_dabt dabt = info->dabt;struct cpu_user_regs *regs =guest_cpu_user_regs (); +int ret; - return handler->ops->write(v, info, get_user_reg(regs, dabt.reg), - handler->priv); + ret = handler->ops->write (v, info,get_user_reg (regs, dabt.reg), + handler->priv); +return ( ret ) ?IO_HANDLED :IO_ABORT ; }/* This function assumes that mmio regions are not overlapped */ @@ -100,14 +104,14 @@ static const struct mmio_handler *find_mmio_handler(struct domain *d, return handler; } -int handle_mmio(mmio_info_t *info) +enum io_state handle_mmio(mmio_info_t *info) { struct vcpu *v = current;const struct mmio_handler *handler =NULL ; handler =find_mmio_handler (v->domain, info->gpa);if ( !handler ) - return 0; +return IO_UNHANDLED ;if ( info->dabt.write )return handle_write (handler, v, info);xen/arch/arm/traps.c +5
-4@@ -1864,10 +1864,10 @@ static inline bool hpfar_is_valid(bool s1ptw, uint8_t fsc)return s1ptw || (fsc ==FSC_FLT_TRANS && !check_workaround_834220 ()); } -static bool try_handle_mmio(struct cpu_user_regs *regs, -const union hsr hsr, -paddr_t gpa) -{ +static enum io_state try_handle_mmio (struct cpu_user_regs *regs, +const union hsr hsr, + paddr_t gpa) + {const struct hsr_dabt dabt = hsr.dabt;mmio_info_t info = { .gpa = gpa,@@ -1879,11 +1879,11 @@ static bool try_handle_mmio(struct cpu_user_regs *regs,/* stage-1 page table should never live in an emulated MMIO region */ if ( dabt.s1ptw ) -return false; +return IO_UNHANDLED ;/* All the instructions used on emulated MMIO region should be valid */
if ( !dabt.valid )xen/arch/arm/traps.c +1
-2-return false; +return IO_UNHANDLED ;/* * Erratum 766422: Thumb store translation fault to Hypervisor may @@ -1896,11 +1896,11 @@ static bool try_handle_mmio(struct cpu_user_regs *regs, if ( rc ) {gprintk (XENLOG_DEBUG ,"Unable to decode instruction\n" ); -return false;Quoted 8 lines
xen/arch/arm/traps.c static bool try_handle_mmio(struct cpu_user_regs *regs, · line 1901
Quoted 3 lines
xen/arch/arm/traps.c +1
-0+return IO_ABORT ; } }Why do the first two error checks result in
IO_UNHANDLED, while the third result inIO_ABORT? Specifically in relation to pagetable walk failures due to someone else changing stage-2 entry simultaneously (see comment toward the end ofdo_trap_stage2_abort_guest)?
Quoted 5 lines
Good question. Somehow I considered the first two as part of looking up the proper handler and not the device itself.
But I guess that at this stage we know that IO was targeting an emulated region. So we can return
IO_ABORT.
That is what I thought as well
Quoted 3 lines
On a side note, it looks like the check dabt.s1ptw is unnecessary because a stage 2 abort on stage 1 translation table lookup will not return a valid instruction syndrome (see B3-1433 in DDI
0406C.cand D10-2460 in DDI 0487C.a).
in that case, go ahead and remove it as part of this patch, mention it in the commit message
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH 1/3] xen/arm: io: Distinguish unhandled IO from aborted one
Hi Stefano,
Quoted 50 lines · Stefano Stabellini
Quoted 49 lines · Julien Grall
Quoted 48 lines
Quoted 35 lines
xen/arch/arm/traps.c +5
-4@@ -1864,10 +1864,10 @@ static inline bool hpfar_is_valid(bool s1ptw, uint8_t fsc)return s1ptw || (fsc ==FSC_FLT_TRANS && !check_workaround_834220 ()); } -static bool try_handle_mmio(struct cpu_user_regs *regs, -const union hsr hsr, -paddr_t gpa) -{ +static enum io_state try_handle_mmio (struct cpu_user_regs *regs, +const union hsr hsr, + paddr_t gpa) + {const struct hsr_dabt dabt = hsr.dabt;mmio_info_t info = { .gpa = gpa,@@ -1879,11 +1879,11 @@ static bool try_handle_mmio(struct cpu_user_regs *regs,/* stage-1 page table should never live in an emulated MMIO region */ if ( dabt.s1ptw ) -return false; +return IO_UNHANDLED ;/* All the instructions used on emulated MMIO region should be valid */
if ( !dabt.valid )xen/arch/arm/traps.c +1
-1-return false; +return IO_UNHANDLED ;/* * Erratum 766422: Thumb store translation fault to Hypervisor may xen/arch/arm/traps.c static bool try_handle_mmio(struct cpu_user_regs · line 1902
Quoted 8 lines
xen/arch/arm/traps.c +1
-1@@ -1896,11 +1896,11 @@ static bool try_handle_mmio(struct cpu_user_regs *regs, if ( rc ) {gprintk (XENLOG_DEBUG ,"Unable to decode instruction\n" ); -return false; +return IO_ABORT ; } }Why do the first two error checks result in
IO_UNHANDLED, while the third result inIO_ABORT? Specifically in relation to pagetable walk failures due to someone else changing stage-2 entry simultaneously (see comment toward the end ofdo_trap_stage2_abort_guest)?
Quoted 7 lines
Quoted 5 lines
Good question. Somehow I considered the first two as part of looking up the proper handler and not the device itself.
But I guess that at this stage we know that IO was targeting an emulated region. So we can return
IO_ABORT.That is what I thought as well
Actually, I have said something completely wrong yesterday. Must have been too tired :/.
At the time you call try_handle_mmio, you still don't know whether the fault was because of accessing an emulated MMIO region. You will only be sure when find_mmio_handler() has returned a non-NULL pointer (see handle_mmio()).
So returning IO_UNHANDLED here is correct as you want to try another method to handle the fault.
However, it also means that even bad access to emulated region will result to fallback on another method. While this should not be issue, I don't think this is future proof (I am mostly worry on the ACPI case where MMIO are mapped on-demand).
So I will send a patch to fold try_handle_mmio() into handle_mmio().
Quoted 8 lines
Quoted 5 lines
On a side note, it looks like the check dabt.s1ptw is unnecessary because a stage 2 abort on stage 1 translation table lookup will not return a valid instruction syndrome (see B3-1433 in DDI
0406C.cand D10-2460 in DDI 0487C.a).in that case, go ahead and remove it as part of this patch, mention it in the commit message
I will do that in a patch that fold try_handle_mmio() in handle_mmio().
Cheers,
Signature · 6 lines
Julien Grall _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH 1/3] xen/arm: io: Distinguish unhandled IO from aborted one
Quoted 78 lines · Julien Grall
Hi Stefano,
Quoted 56 lines · Stefano Stabellini
Quoted 55 lines · Julien Grall
Quoted 54 lines
Quoted 45 lines
xen/arch/arm/traps.c +4
-3@@ -1864,10 +1864,10 @@ static inline bool hpfar_is_valid(bool s1ptw, uint8_t fsc)return s1ptw || (fsc ==FSC_FLT_TRANS && !check_workaround_834220 ()); } -static bool try_handle_mmio(struct cpu_user_regs *regs, -const union hsr hsr, -paddr_t gpa) -{ +static enum io_state try_handle_mmio (struct cpu_user_regs *regs, +const union hsr hsr, + paddr_t gpa) + {const struct hsr_dabt dabt = hsr.dabt;mmio_info_t info = { .gpa = gpa,@@ -1879,11 +1879,11 @@ static bool try_handle_mmio(struct
cpu_user_regs
*regs,/* stage-1 page table should never live in an emulated MMIOregion
*/
if ( dabt.s1ptw )xen/arch/arm/traps.c +1
-1-return false; +return IO_UNHANDLED ;/* All the instructions used on emulated MMIO region should be
valid */
if ( !dabt.valid )xen/arch/arm/traps.c +1
-1-return false; +return IO_UNHANDLED ;/* * Erratum 766422: Thumb store translation fault to Hypervisor may
xen/arch/arm/traps.c +0
-0@@ -1896,11 +1896,11 @@ static bool try_handle_mmio(struct
cpu_user_regs
*regs,
if ( rc )
{
gprintk(XENLOG_DEBUG, "Unable to decode
instruction\n");xen/arch/arm/traps.c static bool try_handle_mmio(struct cpu_user_regs *regs, · line 1901
Quoted 4 lines
xen/arch/arm/traps.c +1
-1-return false; +return IO_ABORT ; } }Why do the first two error checks result in
IO_UNHANDLED, while the third result inIO_ABORT? Specifically in relation to pagetable walk failures due to someone else changing stage-2 entry simultaneously (see comment toward the end ofdo_trap_stage2_abort_guest)?Quoted 8 lines
Quoted 6 lines
Good question. Somehow I considered the first two as part of looking up the
proper handler and not the device itself.But I guess that at this stage we know that IO was targeting an emulated region. So we can return
IO_ABORT.That is what I thought as well
Actually, I have said something completely wrong yesterday. Must have been too tired :/.
At the time you call
try_handle_mmio, you still don't know whether the fault was because of accessing an emulated MMIO region. You will only be sure whenfind_mmio_handler()has returned a non-NULL pointer (seehandle_mmio()).So returning
IO_UNHANDLEDhere is correct as you want to try another method to handle the fault.
Quoted 6 lines
However, it also means that even bad access to emulated region will result to fallback on another method. While this should not be issue, I don't think this is future proof (I am mostly worry on the ACPI case where MMIO are mapped on-demand).
So I will send a patch to fold
try_handle_mmio()intohandle_mmio().
All right
Quoted 8 lines
Quoted 8 lines
Quoted 7 lines
On a side note, it looks like the check dabt.s1ptw is unnecessary because a
stage 2 abort on stage 1 translation table lookup will not return a valid instruction syndrome (see B3-1433 in DDI0406C.cand D10-2460 in DDI 0487C.a).
Quoted 10 lines
Quoted 2 lines
in that case, go ahead and remove it as part of this patch, mention it in the commit message
I will do that in a patch that fold
try_handle_mmio()inhandle_mmio().Cheers,
Signature · 1 line
Julien Grall
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
PATCH 2/3 xen/arm: Don't crash domain on bad MMIO emulation
In reply to Julien GrallNow the MMIO emulation is able to distinguish unhandled IO from aborted one, there are no need to crash the domain when the region is access with a bad width.
Instead let Xen inject a data abort to the guest and decide what to do.
--- xen/arch/arm/vgic-v2.c | 2 -- xen/arch/arm/vgic-v3-its.c | 3 --- xen/arch/arm/vgic-v3.c | 8 -------- xen/arch/arm/vpl011.c | 2 -- 4 files changed, 15 deletions(-)
xen/arch/arm/vgic-v2.c +0 -2
@@ -348,7 +348,6 @@ static int vgic_v2_distr_mmio_read(struct vcpu *v, mmio_info_t *info, bad_width:printk (XENLOG_G_ERR "%pv: vGICD: bad read width %d r%d offset %#08x\n" , v, dabt.size, dabt.reg, gicd_reg); -domain_crash_synchronous ();return 0 ; read_as_zero_32:
@@ -613,7 +612,6 @@ bad_width:printk (XENLOG_G_ERR "%pv: vGICD: bad write width %d r%d=%" PRIregister" offset %#08x\n" , v, dabt.size, dabt.reg, r, gicd_reg); -domain_crash_synchronous ();return 0 ; write_ignore_32:
xen/arch/arm/vgic-v3-its.c +0 -3
@@ -1136,7 +1136,6 @@ read_reserved: bad_width:printk (XENLOG_G_ERR "vGITS: bad read width %d r%d offset %#04lx\n" , info->dabt.size, info->dabt.reg, (unsigned long )info->gpa &0xffff ); -domain_crash_synchronous ();return 0 ; }
@@ -1446,8 +1445,6 @@ bad_width:printk (XENLOG_G_ERR "vGITS: bad write width %d r%d offset %#08lx\n" , info->dabt.size, info->dabt.reg, (unsigned long )info->gpa &0xffff ); -domain_crash_synchronous (); -return 0 ; }
xen/arch/arm/vgic-v3.c +0 -8
@@ -328,7 +328,6 @@ static int __vgic_v3_rdistr_rd_mmio_read(struct vcpu *v, mmio_info_t *info, bad_width:printk (XENLOG_G_ERR "%pv vGICR: bad read width %d r%d offset %#08x\n" , v, dabt.size, dabt.reg, gicr_reg); -domain_crash_synchronous ();return 0 ; read_as_zero_64:
@@ -648,7 +647,6 @@ bad_width:printk (XENLOG_G_ERR "%pv: vGICR: bad write width %d r%d=%" PRIregister" offset %#08x\n" , v, dabt.size, dabt.reg, r, gicr_reg); -domain_crash_synchronous ();return 0 ; write_ignore_64:
@@ -760,7 +758,6 @@ static int __vgic_v3_distr_common_mmio_read(const char *name, struct vcpu *v, bad_width:printk (XENLOG_G_ERR "%pv: %s: bad read width %d r%d offset %#08x\n" , v, name, dabt.size, dabt.reg, reg); -domain_crash_synchronous ();return 0 ; read_as_zero:
@@ -876,7 +873,6 @@ bad_width:printk (XENLOG_G_ERR "%pv: %s: bad write width %d r%d=%" PRIregister" offset %#08x\n" , v, name, dabt.size, dabt.reg, r, reg); -domain_crash_synchronous ();return 0 ; write_ignore_32:
@@ -937,7 +933,6 @@ static int vgic_v3_rdistr_sgi_mmio_read(struct vcpu *v, mmio_info_t *info, bad_width:printk (XENLOG_G_ERR "%pv: vGICR: SGI: bad read width %d r%d offset %#08x\n" , v, dabt.size, dabt.reg, gicr_reg); -domain_crash_synchronous ();return 0 ; read_as_zero_32:
@@ -1017,7 +1012,6 @@ bad_width:printk (XENLOG_G_ERR "%pv: vGICR: SGI: bad write width %d r%d=%" PRIregister" offset %#08x\n" , v, dabt.size, dabt.reg, r, gicr_reg); -domain_crash_synchronous ();return 0 ; write_ignore_32:
@@ -1268,7 +1262,6 @@ static int vgic_v3_distr_mmio_read(struct vcpu *v, mmio_info_t *info, bad_width:printk (XENLOG_G_ERR "%pv: vGICD: bad read width %d r%d offset %#08x\n" , v, dabt.size, dabt.reg, gicd_reg); -domain_crash_synchronous ();return 0 ; read_as_zero_32:
@@ -1456,7 +1449,6 @@ bad_width:printk (XENLOG_G_ERR "%pv: vGICD: bad write width %d r%d=%" PRIregister" offset %#08x\n" , v, dabt.size, dabt.reg, r, gicd_reg); -domain_crash_synchronous ();return 0 ; write_ignore_32:
xen/arch/arm/vpl011.c +0 -2
@@ -296,7 +296,6 @@ static int vpl011_mmio_read(struct vcpu *v, bad_width:gprintk (XENLOG_ERR ,"vpl011: bad read width %d r%d offset %#08x\n" , dabt.size, dabt.reg, vpl011_reg); -domain_crash_synchronous ();return 0 ; }
@@ -366,7 +365,6 @@ write_ignore: bad_width:gprintk (XENLOG_ERR ,"vpl011: bad write width %d r%d offset %#08x\n" , dabt.size, dabt.reg, vpl011_reg); -domain_crash_synchronous ();return 0 ; }
Signature · 7 lines
2.11.0 _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH 2/3] xen/arm: Don't crash domain on bad MMIO emulation
Quoted 7 lines · Julien Grall
Now the MMIO emulation is able to distinguish unhandled IO from aborted one, there are no need to crash the domain when the region is access with a bad width.
Instead let Xen inject a data abort to the guest and decide what to do.
Signed-off-by Julien Grall <[email protected]>
Nice!
Quoted 139 lines
--- xen/arch/arm/vgic-v2.c | 2 -- xen/arch/arm/vgic-v3-its.c | 3 --- xen/arch/arm/vgic-v3.c | 8 -------- xen/arch/arm/vpl011.c | 2 -- 4 files changed, 15 deletions(-)xen/arch/arm/vgic-v2.c +0
-2@@ -348,7 +348,6 @@ static int vgic_v2_distr_mmio_read(struct vcpu *v, mmio_info_t *info, bad_width:printk (XENLOG_G_ERR "%pv: vGICD: bad read width %d r%d offset %#08x\n" , v, dabt.size, dabt.reg, gicd_reg); -domain_crash_synchronous ();return 0 ; read_as_zero_32:@@ -613,7 +612,6 @@ bad_width:printk (XENLOG_G_ERR "%pv: vGICD: bad write width %d r%d=%" PRIregister" offset %#08x\n" , v, dabt.size, dabt.reg, r, gicd_reg); -domain_crash_synchronous ();return 0 ; write_ignore_32:xen/arch/arm/vgic-v3-its.c +0
-3@@ -1136,7 +1136,6 @@ read_reserved: bad_width:printk (XENLOG_G_ERR "vGITS: bad read width %d r%d offset %#04lx\n" , info->dabt.size, info->dabt.reg, (unsigned long )info->gpa &0xffff ); -domain_crash_synchronous ();return 0 ; }@@ -1446,8 +1445,6 @@ bad_width:printk (XENLOG_G_ERR "vGITS: bad write width %d r%d offset %#08lx\n" , info->dabt.size, info->dabt.reg, (unsigned long )info->gpa &0xffff ); -domain_crash_synchronous (); -return 0 ; }xen/arch/arm/vgic-v3.c +0
-8@@ -328,7 +328,6 @@ static int __vgic_v3_rdistr_rd_mmio_read(struct vcpu *v, mmio_info_t *info, bad_width:printk (XENLOG_G_ERR "%pv vGICR: bad read width %d r%d offset %#08x\n" , v, dabt.size, dabt.reg, gicr_reg); -domain_crash_synchronous ();return 0 ; read_as_zero_64:@@ -648,7 +647,6 @@ bad_width:printk (XENLOG_G_ERR "%pv: vGICR: bad write width %d r%d=%" PRIregister" offset %#08x\n" , v, dabt.size, dabt.reg, r, gicr_reg); -domain_crash_synchronous ();return 0 ; write_ignore_64:@@ -760,7 +758,6 @@ static int __vgic_v3_distr_common_mmio_read(const char *name, struct vcpu *v, bad_width:printk (XENLOG_G_ERR "%pv: %s: bad read width %d r%d offset %#08x\n" , v, name, dabt.size, dabt.reg, reg); -domain_crash_synchronous ();return 0 ; read_as_zero:@@ -876,7 +873,6 @@ bad_width:printk (XENLOG_G_ERR "%pv: %s: bad write width %d r%d=%" PRIregister" offset %#08x\n" , v, name, dabt.size, dabt.reg, r, reg); -domain_crash_synchronous ();return 0 ; write_ignore_32:@@ -937,7 +933,6 @@ static int vgic_v3_rdistr_sgi_mmio_read(struct vcpu *v, mmio_info_t *info, bad_width:printk (XENLOG_G_ERR "%pv: vGICR: SGI: bad read width %d r%d offset %#08x\n" , v, dabt.size, dabt.reg, gicr_reg); -domain_crash_synchronous ();return 0 ; read_as_zero_32:@@ -1017,7 +1012,6 @@ bad_width:printk (XENLOG_G_ERR "%pv: vGICR: SGI: bad write width %d r%d=%" PRIregister" offset %#08x\n" , v, dabt.size, dabt.reg, r, gicr_reg); -domain_crash_synchronous ();return 0 ; write_ignore_32:@@ -1268,7 +1262,6 @@ static int vgic_v3_distr_mmio_read(struct vcpu *v, mmio_info_t *info, bad_width:printk (XENLOG_G_ERR "%pv: vGICD: bad read width %d r%d offset %#08x\n" , v, dabt.size, dabt.reg, gicd_reg); -domain_crash_synchronous ();return 0 ; read_as_zero_32:@@ -1456,7 +1449,6 @@ bad_width:printk (XENLOG_G_ERR "%pv: vGICD: bad write width %d r%d=%" PRIregister" offset %#08x\n" , v, dabt.size, dabt.reg, r, gicd_reg); -domain_crash_synchronous ();return 0 ; write_ignore_32:xen/arch/arm/vpl011.c +0
-2@@ -296,7 +296,6 @@ static int vpl011_mmio_read(struct vcpu *v, bad_width:gprintk (XENLOG_ERR ,"vpl011: bad read width %d r%d offset %#08x\n" , dabt.size, dabt.reg, vpl011_reg); -domain_crash_synchronous ();return 0 ; }@@ -366,7 +365,6 @@ write_ignore: bad_width:gprintk (XENLOG_ERR ,"vpl011: bad write width %d r%d offset %#08x\n" , dabt.size, dabt.reg, vpl011_reg); -domain_crash_synchronous ();return 0 ; }Signature · 1 line
2.11.0
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
PATCH 3/3 xen/arm: Don't crash the domain on invalid HVC immediate
In reply to Julien Gralldomain_crash_synchronous() should only be used when something went wrong in Xen. It is better to inject to the guest as it will be in better position to provide helpful information (stack trace...).
---
We potentially want to return -1 instead. This would make Xen more future-proof if we decide to implement the other HVC immediate.
--- xen/arch/arm/traps.c | 13 ++++++++----- 1 file changed, 8 insertions(+), 5 deletions(-)
xen/arch/arm/traps.c +8 -5
@@ -1473,14 +1473,17 @@ static void do_debug_trap(struct cpu_user_regs *regs, unsigned int code)#endif static void do_trap_hypercall (struct cpu_user_regs *regs,register_t *nr, -unsigned long iss) +const union hsr hsr) {arm_hypercall_fn_t call =NULL ;BUILD_BUG_ON (NR_hypercalls <ARRAY_SIZE (arm_hypercall_table) ); -if ( iss !=XEN_HYPERCALL_TAG ) -domain_crash_synchronous (); +if ( hsr.iss !=XEN_HYPERCALL_TAG ) + { +gprintk (XENLOG_WARNING ,"Invalid HVC imm 0x%x\n" , hsr.iss); +return inject_undef_exception (regs, hsr); + }if ( *nr >=ARRAY_SIZE (arm_hypercall_table) ) {
@@ -2150,7 +2153,7 @@ void do_trap_guest_sync(struct cpu_user_regs *regs)if ( hsr.iss ==0 )return do_trap_hvc_smccc (regs); nr = regs->r12; -do_trap_hypercall (regs, &nr, hsr.iss); +do_trap_hypercall (regs, &nr, hsr); regs->r12 = (uint32_t )nr;break ; }
@@ -2164,7 +2167,7 @@ void do_trap_guest_sync(struct cpu_user_regs *regs)#endif if ( hsr.iss ==0 )return do_trap_hvc_smccc (regs); -do_trap_hypercall (regs, ®s->x16, hsr.iss); +do_trap_hypercall (regs, ®s->x16, hsr);break ;case HSR_EC_SMC64 :/*
Signature · 7 lines
2.11.0 _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH 3/3] xen/arm: Don't crash the domain on invalid HVC immediate
Quoted 2 lines · Julien Grall
domain_crash_synchronous()should only be used when something went wrong in Xen. It is better to inject to the guest as it will be in better
^ a better
Quoted 23 lines
position to provide helpful information (stack trace...).
Signed-off-by Julien Grall <[email protected]>---
We potentially want to return -1 instead. This would make Xen more future-proof if we decide to implement the other HVC immediate.
--- xen/arch/arm/traps.c | 13 ++++++++----- 1 file changed, 8 insertions(+), 5 deletions(-)xen/arch/arm/traps.c +1
-1@@ -1473,14 +1473,17 @@ static void do_debug_trap(struct cpu_user_regs *regs, unsigned int code)#endif static void do_trap_hypercall (struct cpu_user_regs *regs,register_t *nr, -unsigned long iss) +const union hsr hsr) {
xen/arch/arm/traps.c static void do_debug_trap(struct cpu_user_regs *regs, unsigned int code) · line 1485
Quoted 10 lines
xen/arch/arm/traps.c +4
-2arm_hypercall_fn_t call =NULL ;BUILD_BUG_ON (NR_hypercalls <ARRAY_SIZE (arm_hypercall_table) ); -if ( iss !=XEN_HYPERCALL_TAG ) -domain_crash_synchronous (); +if ( hsr.iss !=XEN_HYPERCALL_TAG ) + { +gprintk (XENLOG_WARNING ,"Invalid HVC imm 0x%x\n" , hsr.iss); +return inject_undef_exception (regs, hsr);
It looks a bit weird, given that inject_undef_exception doesn't return anything. This is just code style so anyway:
Quoted 25 lines
xen/arch/arm/traps.c +3
-2+ }if ( *nr >=ARRAY_SIZE (arm_hypercall_table) ) {@@ -2150,7 +2153,7 @@ void do_trap_guest_sync(struct cpu_user_regs *regs)if ( hsr.iss ==0 )return do_trap_hvc_smccc (regs); nr = regs->r12; -do_trap_hypercall (regs, &nr, hsr.iss); +do_trap_hypercall (regs, &nr, hsr); regs->r12 = (uint32_t )nr;break ; }@@ -2164,7 +2167,7 @@ void do_trap_guest_sync(struct cpu_user_regs *regs)#endif if ( hsr.iss ==0 )return do_trap_hvc_smccc (regs); -do_trap_hypercall (regs, ®s->x16, hsr.iss); +do_trap_hypercall (regs, ®s->x16, hsr);break ;case HSR_EC_SMC64 :/* Signature · 1 line
2.11.0
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH 3/3] xen/arm: Don't crash the domain on invalid HVC immediate
Quoted 33 lines · Stefano Stabellini
Quoted 2 lines · Julien Grall
domain_crash_synchronous()should only be used when something went wrong in Xen. It is better to inject to the guest as it will be in better^ a better
Quoted 27 lines
position to provide helpful information (stack trace...).
Signed-off-by Julien Grall <[email protected]>---
We potentially want to return -1 instead. This would make Xen more future-proof if we decide to implement the other HVC immediate.
--- xen/arch/arm/traps.c | 13 ++++++++----- 1 file changed, 8 insertions(+), 5 deletions(-)xen/arch/arm/traps.c +1
-1@@ -1473,14 +1473,17 @@ static void do_debug_trap(struct cpu_user_regs *regs, unsigned int code)#endif static void do_trap_hypercall (struct cpu_user_regs *regs,register_t *nr, -unsigned long iss) +const union hsr hsr) {arm_hypercall_fn_t call =NULL ;BUILD_BUG_ON (NR_hypercalls <ARRAY_SIZE (arm_hypercall_table) );
Quoted 9 lines
xen/arch/arm/traps.c static void do_debug_trap(struct cpu_user_regs *regs, unsigned int code) · line 1485
Quoted 6 lines
xen/arch/arm/traps.c +4
-2-if ( iss !=XEN_HYPERCALL_TAG ) -domain_crash_synchronous (); +if ( hsr.iss !=XEN_HYPERCALL_TAG ) + { +gprintk (XENLOG_WARNING ,"Invalid HVC imm 0x%x\n" , hsr.iss); +return inject_undef_exception (regs, hsr);It looks a bit weird, given that
inject_undef_exceptiondoesn't return anything. This is just code style so anyway:
I followed coding style in other part of the emulation (see do_sysreg for instance). I am happy to change that though.
Cheers,
Quoted 29 lines
Reviewed-by Stefano Stabellini <[email protected]>Quoted 25 lines
xen/arch/arm/traps.c +3
-2+ }if ( *nr >=ARRAY_SIZE (arm_hypercall_table) ) {@@ -2150,7 +2153,7 @@ void do_trap_guest_sync(struct cpu_user_regs *regs)if ( hsr.iss ==0 )return do_trap_hvc_smccc (regs); nr = regs->r12; -do_trap_hypercall (regs, &nr, hsr.iss); +do_trap_hypercall (regs, &nr, hsr); regs->r12 = (uint32_t )nr;break ; }@@ -2164,7 +2167,7 @@ void do_trap_guest_sync(struct cpu_user_regs *regs)#endif if ( hsr.iss ==0 )return do_trap_hvc_smccc (regs); -do_trap_hypercall (regs, ®s->x16, hsr.iss); +do_trap_hypercall (regs, ®s->x16, hsr);break ;case HSR_EC_SMC64 :/* Signature · 1 line
2.11.0
Signature · 6 lines
Julien Grall _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
Quoted 9 lines · Julien Grall
Hi all,
This small series replaces all call to
domain_crash_synchronousby injecting an exception to the guest.This will result to a nicer trace from the guest (no need to manually walk the stack) and give a chance to the guest to give a bit more information on what it was doing.
Quoted 6 lines
Cheers,
Julien Grall (3):
xen/arm: io: Distinguish unhandled IO from aborted one xen/arm: Don't crash domain on bad MMIO emulation xen/arm: Don't crash the domain on invalid HVC immediate
Thanks.
I don't feel qualified to review these, but some notes.
Patch 1. s/avodi/avoid/ in the commit message
Patches 2 and 3. You probably want to convert the printks to gdprintk()s, otherwise guests can choke up the ratelimited log. Doing so will also mean that the vcpu will be identified consistently, which it isn't currently.
~Andrew
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Quoted 19 lines · Andrew Cooper
Quoted 9 lines · Julien Grall
Hi all,
This small series replaces all call to
domain_crash_synchronousby injecting an exception to the guest.This will result to a nicer trace from the guest (no need to manually walk the stack) and give a chance to the guest to give a bit more information on what it was doing.
Quoted 6 lines
Cheers,
Julien Grall (3):
xen/arm: io: Distinguish unhandled IO from aborted one xen/arm: Don't crash domain on bad MMIO emulation xen/arm: Don't crash the domain on invalid HVC immediateThanks.
Quoted 8 lines
I don't feel qualified to review these, but some notes.
Patch 1. s/avodi/avoid/ in the commit message
Patches 2 and 3. You probably want to convert the printks to gdprintk()s, otherwise guests can choke up the ratelimited log. Doing so will also mean that the vcpu will be identified consistently, which it isn't currently.
We didn't use g*printk because it would be more confusing to print the current vCPU in some cases (e.g when accessing the re-distributor of another vCPU) or does not matter (e.g for ITS).
The problem with the debug version is those information are actually quite useful in non-debug build. We found quite a few issues thanks to them.
I think it would make more sense for Xen to provide per-guest ratelimited than hiding those messages in non-debug build.
Quoted 3 lines
~Andrew
Signature · 6 lines
Julien Grall _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
Quoted 29 lines · Julien Grall
Quoted 26 lines · Andrew Cooper
Quoted 12 lines · Julien Grall
Hi all,
This small series replaces all call to
domain_crash_synchronousby injecting
an exception to the guest.This will result to a nicer trace from the guest (no need to manually walk
the stack) and give a chance to the guest to give a bit more information on
what it was doing.Quoted 6 lines
Cheers,
Julien Grall (3):
xen/arm: io: Distinguish unhandled IO from aborted one xen/arm: Don't crash domain on bad MMIO emulation xen/arm: Don't crash the domain on invalid HVC immediateThanks.
I don't feel qualified to review these, but some notes.
Patch 1. s/avodi/avoid/ in the commit message
Quoted 7 lines
Quoted 4 lines
Patches 2 and 3. You probably want to convert the printks to gdprintk()s, otherwise guests can choke up the ratelimited log. Doing so will also mean that the vcpu will be identified consistently, which it isn't currently.
We didn't use g*printk because it would be more confusing to print the current vCPU in some cases (e.g when accessing the re-distributor of another vCPU) or does not matter (e.g for ITS).
In the former case, you'd want to print both current, and the target vcpu. The latter still matters what current is if something goes wrong.
We have plenty of similar cases in x86, but at the point you are printing an diagnostic message, ignoring current is almost always the wrong think to do.
Quoted 7 lines
The problem with the debug version is those information are actually quite useful in non-debug build. We found quite a few issues thanks to them.
I think it would make more sense for Xen to provide per-guest ratelimited than hiding those messages in non-debug build.
Per guest is quite a lot more complicated than global, and would still require a global limit to prevent a concerted attack from multiple guests to avoid DoSing the system.
Debug vs unilateral is your prerogative as a maintainer, but as you've said yourself, the are used for debugging purposes, which proves my point.
~Andrew
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Hi,
Quoted 38 lines · Andrew Cooper
Quoted 29 lines · Julien Grall
Quoted 26 lines · Andrew Cooper
Quoted 12 lines · Julien Grall
Hi all,
This small series replaces all call to
domain_crash_synchronousby injecting
an exception to the guest.This will result to a nicer trace from the guest (no need to manually walk
the stack) and give a chance to the guest to give a bit more information on
what it was doing.Quoted 6 lines
Cheers,
Julien Grall (3):
xen/arm: io: Distinguish unhandled IO from aborted one xen/arm: Don't crash domain on bad MMIO emulation xen/arm: Don't crash the domain on invalid HVC immediateThanks.
I don't feel qualified to review these, but some notes.
Patch 1. s/avodi/avoid/ in the commit message
Quoted 7 lines
Quoted 4 lines
Patches 2 and 3. You probably want to convert the printks to gdprintk()s, otherwise guests can choke up the ratelimited log. Doing so will also mean that the vcpu will be identified consistently, which it isn't currently.
We didn't use g*printk because it would be more confusing to print the current vCPU in some cases (e.g when accessing the re-distributor of another vCPU) or does not matter (e.g for ITS).
Quoted 6 lines
In the former case, you'd want to print both current, and the target vcpu. The latter still matters what current is if something goes wrong.
We have plenty of similar cases in x86, but at the point you are printing an diagnostic message, ignoring current is almost always the wrong think to do.
I will look at it on another series.
Quoted 9 lines
Quoted 8 lines
The problem with the debug version is those information are actually quite useful in non-debug build. We found quite a few issues thanks to them.
I think it would make more sense for Xen to provide per-guest ratelimited than hiding those messages in non-debug build.
Quoted 6 lines
Per guest is quite a lot more complicated than global, and would still require a global limit to prevent a concerted attack from multiple guests to avoid DoSing the system.
Debug vs unilateral is your prerogative as a maintainer, but as you've said yourself, the are used for debugging purposes, which proves my point.
So on x86, you always request the user to reproduce it with debug build enable?
Stefano, what's your opinion here?
Cheers,
Signature · 6 lines
Julien Grall _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
Quoted 41 lines · Julien Grall
Hi,
Quoted 38 lines · Andrew Cooper
Quoted 36 lines · Julien Grall
Quoted 22 lines · Andrew Cooper
Quoted 12 lines · Julien Grall
Hi all,
This small series replaces all call to
domain_crash_synchronousby injecting
an exception to the guest.This will result to a nicer trace from the guest (no need to manually walk
the stack) and give a chance to the guest to give a bit more information on
what it was doing.Quoted 6 lines
Cheers,
Julien Grall (3):
xen/arm: io: Distinguish unhandled IO from aborted one xen/arm: Don't crash domain on bad MMIO emulation xen/arm: Don't crash the domain on invalid HVC immediateThanks.
Quoted 8 lines
I don't feel qualified to review these, but some notes.
Patch 1. s/avodi/avoid/ in the commit message
Patches 2 and 3. You probably want to convert the printks to gdprintk()s, otherwise guests can choke up the ratelimited log. Doing so will also mean that the vcpu will be identified consistently, which it isn't currently.
We didn't use g*printk because it would be more confusing to print the current vCPU in some cases (e.g when accessing the re-distributor of another vCPU) or does not matter (e.g for ITS).
Quoted 8 lines
Quoted 6 lines
In the former case, you'd want to print both current, and the target vcpu. The latter still matters what current is if something goes wrong.
We have plenty of similar cases in x86, but at the point you are printing an diagnostic message, ignoring current is almost always the wrong think to do.
I will look at it on another series.
Fair enough.
Quoted 14 lines
Quoted 14 lines
Quoted 9 lines
The problem with the debug version is those information are actually quite useful in non-debug build. We found quite a few issues thanks to them.
I think it would make more sense for Xen to provide per-guest ratelimited than hiding those messages in non-debug build.
Per guest is quite a lot more complicated than global, and would still require a global limit to prevent a concerted attack from multiple guests to avoid DoSing the system.
Quoted 6 lines
Quoted 3 lines
Debug vs unilateral is your prerogative as a maintainer, but as you've said yourself, the are used for debugging purposes, which proves my point.
So on x86, you always request the user to reproduce it with debug build enable?
That is very specific to what goes wrong, but no, we generally don't have release-build messages for "the guest screwed up and we hit it expected way for doing so". We do (well - are trying to, which is how I found this domain_crash_sync issue in the first place) get consistent at printing useful release-build messages for abnormal termination of the guest.
In terms of making it easier to debug, XenServer always ships with a release and debug hypervisor built from the same source so customers can trivially switch into the debug version and collect logs if we think that is a useful thing to do, but this is only used in a fraction of cases in the first place.
Anything more complicated is definitely going to require an experienced human to investigate, at which point they can do whatever they want.
~Andrew
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Quoted 63 lines · Julien Grall
Hi,
Quoted 38 lines · Andrew Cooper
Quoted 29 lines · Julien Grall
Quoted 26 lines · Andrew Cooper
Quoted 12 lines · Julien Grall
Hi all,
This small series replaces all call to
domain_crash_synchronousby injecting
an exception to the guest.This will result to a nicer trace from the guest (no need to manually walk
the stack) and give a chance to the guest to give a bit more information on
what it was doing.Quoted 6 lines
Cheers,
Julien Grall (3):
xen/arm: io: Distinguish unhandled IO from aborted one xen/arm: Don't crash domain on bad MMIO emulation xen/arm: Don't crash the domain on invalid HVC immediateThanks.
I don't feel qualified to review these, but some notes.
Patch 1. s/avodi/avoid/ in the commit message
Quoted 7 lines
Quoted 4 lines
Patches 2 and 3. You probably want to convert the printks to gdprintk()s, otherwise guests can choke up the ratelimited log. Doing so will also mean that the vcpu will be identified consistently, which it isn't currently.
We didn't use g*printk because it would be more confusing to print the current vCPU in some cases (e.g when accessing the re-distributor of another vCPU) or does not matter (e.g for ITS).
Quoted 6 lines
In the former case, you'd want to print both current, and the target vcpu. The latter still matters what current is if something goes wrong.
We have plenty of similar cases in x86, but at the point you are printing an diagnostic message, ignoring current is almost always the wrong think to do.
I will look at it on another series.
Quoted 13 lines
Quoted 8 lines
The problem with the debug version is those information are actually quite useful in non-debug build. We found quite a few issues thanks to them.
I think it would make more sense for Xen to provide per-guest ratelimited than hiding those messages in non-debug build.
Per guest is quite a lot more complicated than global, and would still require a global limit to prevent a concerted attack from multiple guests to avoid DoSing the system.
Quoted 7 lines
Quoted 2 lines
Debug vs unilateral is your prerogative as a maintainer, but as you've said yourself, the are used for debugging purposes, which proves my point.
So on x86, you always request the user to reproduce it with debug build enable?
Stefano, what's your opinion here?
I think it would be great to have per-guest rate limiting.
On ARM, it would be impossible to ask users to repro with debug enabled, given that many deployments are embedded (set-top boxes, etc).
I think it is OK to keep the XENLOG_DEBUG, they should be filtered out by loglvl. But we need to be careful about the others. We might want to convert the XENLOG_WARNING in traps.c to XENLOG_DEBUG: they are warning about guests misbehaving, but from the hypervisor point of view, it's not a problem, guests can shoot themselves if they want to; it's OK.
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Quoted 65 lines · Stefano Stabellini
Quoted 56 lines · Julien Grall
Hi,
Quoted 35 lines · Andrew Cooper
Quoted 27 lines · Julien Grall
Quoted 25 lines · Andrew Cooper
Quoted 12 lines · Julien Grall
Hi all,
This small series replaces all call to
domain_crash_synchronousby injecting
an exception to the guest.This will result to a nicer trace from the guest (no need to manually walk
the stack) and give a chance to the guest to give a bit more information on
what it was doing.Quoted 6 lines
Cheers,
Julien Grall (3):
xen/arm: io: Distinguish unhandled IO from aborted one xen/arm: Don't crash domain on bad MMIO emulation xen/arm: Don't crash the domain on invalid HVC immediateThanks.
I don't feel qualified to review these, but some notes.
Patch 1. s/avodi/avoid/ in the commit message
Quoted 7 lines
Quoted 4 lines
Patches 2 and 3. You probably want to convert the printks to gdprintk()s, otherwise guests can choke up the ratelimited log. Doing so will also mean that the vcpu will be identified consistently, which it isn't currently.
We didn't use g*printk because it would be more confusing to print the current vCPU in some cases (e.g when accessing the re-distributor of another vCPU) or does not matter (e.g for ITS).
Quoted 6 lines
In the former case, you'd want to print both current, and the target vcpu. The latter still matters what current is if something goes wrong.
We have plenty of similar cases in x86, but at the point you are printing an diagnostic message, ignoring current is almost always the wrong think to do.
I will look at it on another series.
Quoted 10 lines
Quoted 6 lines
The problem with the debug version is those information are actually quite useful in non-debug build. We found quite a few issues thanks to them.
I think it would make more sense for Xen to provide per-guest ratelimited than hiding those messages in non-debug build.
Per guest is quite a lot more complicated than global, and would still require a global limit to prevent a concerted attack from multiple guests to avoid DoSing the system.
Quoted 6 lines
Quoted 2 lines
Debug vs unilateral is your prerogative as a maintainer, but as you've said yourself, the are used for debugging purposes, which proves my point.
So on x86, you always request the user to reproduce it with debug build enable?
Stefano, what's your opinion here?
I think it would be great to have per-guest rate limiting.
Quoted 8 lines
On ARM, it would be impossible to ask users to repro with debug enabled, given that many deployments are embedded (set-top boxes, etc).
I think it is OK to keep the
XENLOG_DEBUG, they should be filtered out by loglvl. But we need to be careful about the others. We might want to convert theXENLOG_WARNINGintraps.ctoXENLOG_DEBUG: they are warning about guests misbehaving, but from the hypervisor point of view, it's not a problem, guests can shoot themselves if they want to; it's OK.
This is a valid argument, but to do it consistently, you'll want to provide an arch-specific version of gdprintk() so that the common code printk()s don't evaporate into /dev/null.
~Andrew
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Hi Stefano,
Quoted 74 lines · Stefano Stabellini
Quoted 63 lines · Julien Grall
Hi,
Quoted 38 lines · Andrew Cooper
Quoted 29 lines · Julien Grall
Quoted 26 lines · Andrew Cooper
Quoted 12 lines · Julien Grall
Hi all,
This small series replaces all call to
domain_crash_synchronousby injecting
an exception to the guest.This will result to a nicer trace from the guest (no need to manually walk
the stack) and give a chance to the guest to give a bit more information on
what it was doing.Quoted 6 lines
Cheers,
Julien Grall (3):
xen/arm: io: Distinguish unhandled IO from aborted one xen/arm: Don't crash domain on bad MMIO emulation xen/arm: Don't crash the domain on invalid HVC immediateThanks.
I don't feel qualified to review these, but some notes.
Patch 1. s/avodi/avoid/ in the commit message
Quoted 7 lines
Quoted 4 lines
Patches 2 and 3. You probably want to convert the printks to gdprintk()s, otherwise guests can choke up the ratelimited log. Doing so will also mean that the vcpu will be identified consistently, which it isn't currently.
We didn't use g*printk because it would be more confusing to print the current vCPU in some cases (e.g when accessing the re-distributor of another vCPU) or does not matter (e.g for ITS).
Quoted 6 lines
In the former case, you'd want to print both current, and the target vcpu. The latter still matters what current is if something goes wrong.
We have plenty of similar cases in x86, but at the point you are printing an diagnostic message, ignoring current is almost always the wrong think to do.
I will look at it on another series.
Quoted 13 lines
Quoted 8 lines
The problem with the debug version is those information are actually quite useful in non-debug build. We found quite a few issues thanks to them.
I think it would make more sense for Xen to provide per-guest ratelimited than hiding those messages in non-debug build.
Per guest is quite a lot more complicated than global, and would still require a global limit to prevent a concerted attack from multiple guests to avoid DoSing the system.
Quoted 7 lines
Quoted 2 lines
Debug vs unilateral is your prerogative as a maintainer, but as you've said yourself, the are used for debugging purposes, which proves my point.
So on x86, you always request the user to reproduce it with debug build enable?
Stefano, what's your opinion here?
I think it would be great to have per-guest rate limiting.
Quoted 8 lines
On ARM, it would be impossible to ask users to repro with debug enabled, given that many deployments are embedded (set-top boxes, etc).
I think it is OK to keep the
XENLOG_DEBUG, they should be filtered out by loglvl. But we need to be careful about the others. We might want to convert theXENLOG_WARNINGintraps.ctoXENLOG_DEBUG: they are warning about guests misbehaving, but from the hypervisor point of view, it's not a problem, guests can shoot themselves if they want to; it's OK.
Not really. Some of the XENLOG_WARNING may happen because Xen does not handle a trap (see the one on do_trap_guest_sync). That has nothing to do with a guest misbehaving and could happen if Xen boots on newer hardware.
Cheers,
Signature · 6 lines
Julien Grall _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel