Message ID | 1398885355-13200-1-git-send-email-julien.grall@linaro.org |
---|---|
State | Accepted, archived |
Headers | show |
On Wed, 2014-04-30 at 20:15 +0100, Julien Grall wrote: > While I was adding new failing code at the end of the function, I've noticed > that the vtimers are not freed which mess all the timers and will crash Xen > quickly when the page will be reused. > > Currently neither vcpu_vgic_init nor vcpu_vtimer_init fail, so we > are safe for now. With the new GICv3 code, the former function will be able > to fail. This will result to a memory leak. > > Call vcpu_destroy if the initialization has failed. We also need to add a > boolean to know if the vtimers are correctly setup as the timer common code > doesn't have safe guard against removing non-initialized timer. > > Signed-off-by: Julien Grall <julien.grall@linaro.org> I was about to acked + apply but it failed to build on arm64 with: domain.c: In function 'alloc_vcpu_struct': /local/scratch/ianc/devel/committer.git/xen/include/xen/lib.h:19:31: error: static assertion failed: "!(sizeof(*v) > PAGE_SIZE)" #define BUILD_BUG_ON(cond) ({ _Static_assert(!(cond), "!(" #cond ")"); }) ^ domain.c:415:5: note: in expansion of macro 'BUILD_BUG_ON' BUILD_BUG_ON(sizeof(*v) > PAGE_SIZE); ^ struct arch_vcpu is apparently now too large. I had also reworded your commit message somewhat: xen/arm: vcpu: Correctly release resources when a VCPU fails to initialize While I was adding new failing code at the end of the function, I noticed that the vtimers are not freed which messes up all the timers and will crash Xen quickly when the page s reused. Currently neither vcpu_vgic_init nor vcpu_vtimer_init fails, so we are safe for now. With the new GICv3 code, the former function will be able to fail. This will result in a memory leak. Call vcpu_destroy if the initialization has failed. We also need to add a boolean to know if the vtimers are correctly setup as the timer common code doesn't have any safeguard against removing a non-initialized timer. Ian. > > --- > Changes in v2: > - Update commit message > --- > xen/arch/arm/domain.c | 8 ++++++-- > xen/arch/arm/vtimer.c | 5 +++++ > xen/include/asm-arm/domain.h | 1 + > 3 files changed, 12 insertions(+), 2 deletions(-) > > diff --git a/xen/arch/arm/domain.c b/xen/arch/arm/domain.c > index ccccb77..c47db4a 100644 > --- a/xen/arch/arm/domain.c > +++ b/xen/arch/arm/domain.c > @@ -468,12 +468,16 @@ int vcpu_initialise(struct vcpu *v) > processor_vcpu_initialise(v); > > if ( (rc = vcpu_vgic_init(v)) != 0 ) > - return rc; > + goto fail; > > if ( (rc = vcpu_vtimer_init(v)) != 0 ) > - return rc; > + goto fail; > > return rc; > + > +fail: > + vcpu_destroy(v); > + return rc; > } > > void vcpu_destroy(struct vcpu *v) > diff --git a/xen/arch/arm/vtimer.c b/xen/arch/arm/vtimer.c > index cb690bb..c515e7e 100644 > --- a/xen/arch/arm/vtimer.c > +++ b/xen/arch/arm/vtimer.c > @@ -77,11 +77,16 @@ int vcpu_vtimer_init(struct vcpu *v) > : GUEST_TIMER_VIRT_PPI; > t->v = v; > > + v->arch.vtimer_initialized = 1; > + > return 0; > } > > void vcpu_timer_destroy(struct vcpu *v) > { > + if ( !v->arch.vtimer_initialized ) > + return; > + > kill_timer(&v->arch.virt_timer.timer); > kill_timer(&v->arch.phys_timer.timer); > } > diff --git a/xen/include/asm-arm/domain.h b/xen/include/asm-arm/domain.h > index ec66a4e..1be3da2 100644 > --- a/xen/include/asm-arm/domain.h > +++ b/xen/include/asm-arm/domain.h > @@ -285,6 +285,7 @@ struct arch_vcpu > > struct vtimer phys_timer; > struct vtimer virt_timer; > + bool_t vtimer_initialized; > } __cacheline_aligned; > > void vcpu_show_execution_state(struct vcpu *);
On 05/02/2014 01:25 PM, Ian Campbell wrote: > On Wed, 2014-04-30 at 20:15 +0100, Julien Grall wrote: >> While I was adding new failing code at the end of the function, I've noticed >> that the vtimers are not freed which mess all the timers and will crash Xen >> quickly when the page will be reused. >> >> Currently neither vcpu_vgic_init nor vcpu_vtimer_init fail, so we >> are safe for now. With the new GICv3 code, the former function will be able >> to fail. This will result to a memory leak. >> >> Call vcpu_destroy if the initialization has failed. We also need to add a >> boolean to know if the vtimers are correctly setup as the timer common code >> doesn't have safe guard against removing non-initialized timer. >> >> Signed-off-by: Julien Grall <julien.grall@linaro.org> > > I was about to acked + apply but it failed to build on arm64 with: > > domain.c: In function 'alloc_vcpu_struct': > /local/scratch/ianc/devel/committer.git/xen/include/xen/lib.h:19:31: error: static assertion failed: "!(sizeof(*v) > PAGE_SIZE)" > #define BUILD_BUG_ON(cond) ({ _Static_assert(!(cond), "!(" #cond ")"); }) > ^ > domain.c:415:5: note: in expansion of macro 'BUILD_BUG_ON' > BUILD_BUG_ON(sizeof(*v) > PAGE_SIZE); > ^ > struct arch_vcpu is apparently now too large. Hmmm... I'm not sure what is the best solution. Can: 1) Allocate 2 pages for the VCPU structure 2) Allocate vgic structure outside. Any opinions? Regards,
On 02/05/14 15:09, Julien Grall wrote: > On 05/02/2014 01:25 PM, Ian Campbell wrote: >> On Wed, 2014-04-30 at 20:15 +0100, Julien Grall wrote: >>> While I was adding new failing code at the end of the function, I've noticed >>> that the vtimers are not freed which mess all the timers and will crash Xen >>> quickly when the page will be reused. >>> >>> Currently neither vcpu_vgic_init nor vcpu_vtimer_init fail, so we >>> are safe for now. With the new GICv3 code, the former function will be able >>> to fail. This will result to a memory leak. >>> >>> Call vcpu_destroy if the initialization has failed. We also need to add a >>> boolean to know if the vtimers are correctly setup as the timer common code >>> doesn't have safe guard against removing non-initialized timer. >>> >>> Signed-off-by: Julien Grall <julien.grall@linaro.org> >> I was about to acked + apply but it failed to build on arm64 with: >> >> domain.c: In function 'alloc_vcpu_struct': >> /local/scratch/ianc/devel/committer.git/xen/include/xen/lib.h:19:31: error: static assertion failed: "!(sizeof(*v) > PAGE_SIZE)" >> #define BUILD_BUG_ON(cond) ({ _Static_assert(!(cond), "!(" #cond ")"); }) >> ^ >> domain.c:415:5: note: in expansion of macro 'BUILD_BUG_ON' >> BUILD_BUG_ON(sizeof(*v) > PAGE_SIZE); >> ^ >> struct arch_vcpu is apparently now too large. > Hmmm... I'm not sure what is the best solution. Can: > 1) Allocate 2 pages for the VCPU structure > 2) Allocate vgic structure outside. > > Any opinions? > > Regards, > 2) The reason structs vcpu/domain were reduced to this size to was avoid needing multi-page allocations, which risk allocation failures on systems with sufficiently fragmented memory. ~Andrew
On Fri, 2014-05-02 at 15:17 +0100, Andrew Cooper wrote: > On 02/05/14 15:09, Julien Grall wrote: > > On 05/02/2014 01:25 PM, Ian Campbell wrote: > >> On Wed, 2014-04-30 at 20:15 +0100, Julien Grall wrote: > >>> While I was adding new failing code at the end of the function, I've noticed > >>> that the vtimers are not freed which mess all the timers and will crash Xen > >>> quickly when the page will be reused. > >>> > >>> Currently neither vcpu_vgic_init nor vcpu_vtimer_init fail, so we > >>> are safe for now. With the new GICv3 code, the former function will be able > >>> to fail. This will result to a memory leak. > >>> > >>> Call vcpu_destroy if the initialization has failed. We also need to add a > >>> boolean to know if the vtimers are correctly setup as the timer common code > >>> doesn't have safe guard against removing non-initialized timer. > >>> > >>> Signed-off-by: Julien Grall <julien.grall@linaro.org> > >> I was about to acked + apply but it failed to build on arm64 with: > >> > >> domain.c: In function 'alloc_vcpu_struct': > >> /local/scratch/ianc/devel/committer.git/xen/include/xen/lib.h:19:31: error: static assertion failed: "!(sizeof(*v) > PAGE_SIZE)" > >> #define BUILD_BUG_ON(cond) ({ _Static_assert(!(cond), "!(" #cond ")"); }) > >> ^ > >> domain.c:415:5: note: in expansion of macro 'BUILD_BUG_ON' > >> BUILD_BUG_ON(sizeof(*v) > PAGE_SIZE); > >> ^ > >> struct arch_vcpu is apparently now too large. > > Hmmm... I'm not sure what is the best solution. Can: > > 1) Allocate 2 pages for the VCPU structure > > 2) Allocate vgic structure outside. > > > > Any opinions? > > > > Regards, > > > > 2) > > The reason structs vcpu/domain were reduced to this size to was avoid > needing multi-page allocations, which risk allocation failures on > systems with sufficiently fragmented memory. Ack. #2 (with s/vgic/anything suitably self contained/) is the answer. Ian.
On Fri, 2014-05-02 at 16:15 +0100, Ian Campbell wrote: > On Fri, 2014-05-02 at 15:17 +0100, Andrew Cooper wrote: > > On 02/05/14 15:09, Julien Grall wrote: > > > On 05/02/2014 01:25 PM, Ian Campbell wrote: > > >> On Wed, 2014-04-30 at 20:15 +0100, Julien Grall wrote: > > >>> While I was adding new failing code at the end of the function, I've noticed > > >>> that the vtimers are not freed which mess all the timers and will crash Xen > > >>> quickly when the page will be reused. > > >>> > > >>> Currently neither vcpu_vgic_init nor vcpu_vtimer_init fail, so we > > >>> are safe for now. With the new GICv3 code, the former function will be able > > >>> to fail. This will result to a memory leak. > > >>> > > >>> Call vcpu_destroy if the initialization has failed. We also need to add a > > >>> boolean to know if the vtimers are correctly setup as the timer common code > > >>> doesn't have safe guard against removing non-initialized timer. > > >>> > > >>> Signed-off-by: Julien Grall <julien.grall@linaro.org> > > >> I was about to acked + apply but it failed to build on arm64 with: > > >> > > >> domain.c: In function 'alloc_vcpu_struct': > > >> /local/scratch/ianc/devel/committer.git/xen/include/xen/lib.h:19:31: error: static assertion failed: "!(sizeof(*v) > PAGE_SIZE)" > > >> #define BUILD_BUG_ON(cond) ({ _Static_assert(!(cond), "!(" #cond ")"); }) > > >> ^ > > >> domain.c:415:5: note: in expansion of macro 'BUILD_BUG_ON' > > >> BUILD_BUG_ON(sizeof(*v) > PAGE_SIZE); > > >> ^ > > >> struct arch_vcpu is apparently now too large. > > > Hmmm... I'm not sure what is the best solution. Can: > > > 1) Allocate 2 pages for the VCPU structure > > > 2) Allocate vgic structure outside. > > > > > > Any opinions? > > > > > > Regards, > > > > > > > 2) > > > > The reason structs vcpu/domain were reduced to this size to was avoid > > needing multi-page allocations, which risk allocation failures on > > systems with sufficiently fragmented memory. > > Ack. #2 (with s/vgic/anything suitably self contained/) is the answer. Was Vijay not moving the vgtic stuff out in one of the gicv3 patches?
On 05/02/2014 04:27 PM, Ian Campbell wrote: > On Fri, 2014-05-02 at 16:15 +0100, Ian Campbell wrote: >> On Fri, 2014-05-02 at 15:17 +0100, Andrew Cooper wrote: >>> On 02/05/14 15:09, Julien Grall wrote: >>>> On 05/02/2014 01:25 PM, Ian Campbell wrote: >>>>> On Wed, 2014-04-30 at 20:15 +0100, Julien Grall wrote: >>>>>> While I was adding new failing code at the end of the function, I've noticed >>>>>> that the vtimers are not freed which mess all the timers and will crash Xen >>>>>> quickly when the page will be reused. >>>>>> >>>>>> Currently neither vcpu_vgic_init nor vcpu_vtimer_init fail, so we >>>>>> are safe for now. With the new GICv3 code, the former function will be able >>>>>> to fail. This will result to a memory leak. >>>>>> >>>>>> Call vcpu_destroy if the initialization has failed. We also need to add a >>>>>> boolean to know if the vtimers are correctly setup as the timer common code >>>>>> doesn't have safe guard against removing non-initialized timer. >>>>>> >>>>>> Signed-off-by: Julien Grall <julien.grall@linaro.org> >>>>> I was about to acked + apply but it failed to build on arm64 with: >>>>> >>>>> domain.c: In function 'alloc_vcpu_struct': >>>>> /local/scratch/ianc/devel/committer.git/xen/include/xen/lib.h:19:31: error: static assertion failed: "!(sizeof(*v) > PAGE_SIZE)" >>>>> #define BUILD_BUG_ON(cond) ({ _Static_assert(!(cond), "!(" #cond ")"); }) >>>>> ^ >>>>> domain.c:415:5: note: in expansion of macro 'BUILD_BUG_ON' >>>>> BUILD_BUG_ON(sizeof(*v) > PAGE_SIZE); >>>>> ^ >>>>> struct arch_vcpu is apparently now too large. >>>> Hmmm... I'm not sure what is the best solution. Can: >>>> 1) Allocate 2 pages for the VCPU structure >>>> 2) Allocate vgic structure outside. >>>> >>>> Any opinions? >>>> >>>> Regards, >>>> >>> >>> 2) >>> >>> The reason structs vcpu/domain were reduced to this size to was avoid >>> needing multi-page allocations, which risk allocation failures on >>> systems with sufficiently fragmented memory. >> >> Ack. #2 (with s/vgic/anything suitably self contained/) is the answer. > > Was Vijay not moving the vgtic stuff out in one of the gicv3 patches? (CC him) IIRC, he only moves the private_irqs field. I think we should move the whole structure, to give more space for the future. This patch will be necessary for the GICv3 serie has vcpu_vgic_init will be able to fail (see patch #10). Ideally, for bisection purpose, it should be applied before the patch #10. Regards,
On 02/05/14 16:36, Julien Grall wrote: >> Was Vijay not moving the vgtic stuff out in one of the gicv3 patches? > > (CC him) > > IIRC, he only moves the private_irqs field. I think we should move the > whole structure, to give more space for the future. > > This patch will be necessary for the GICv3 serie has vcpu_vgic_init will > be able to fail (see patch #10). > > Ideally, for bisection purpose, it should be applied before the patch #10. After thinking, both patch are interdependent. If I want to move out some structure Xen will have to allocate memory. Which mean the function can fail and leak some resources (before this patch is applied). As we are fine for now, I will wait GICv3 patches is pushed (or another patch related to arch vcpu initialization) before sending again this. Regards,
Hi Ian, On 05/02/2014 01:25 PM, Ian Campbell wrote: > On Wed, 2014-04-30 at 20:15 +0100, Julien Grall wrote: >> While I was adding new failing code at the end of the function, I've noticed >> that the vtimers are not freed which mess all the timers and will crash Xen >> quickly when the page will be reused. >> >> Currently neither vcpu_vgic_init nor vcpu_vtimer_init fail, so we >> are safe for now. With the new GICv3 code, the former function will be able >> to fail. This will result to a memory leak. >> >> Call vcpu_destroy if the initialization has failed. We also need to add a >> boolean to know if the vtimers are correctly setup as the timer common code >> doesn't have safe guard against removing non-initialized timer. >> >> Signed-off-by: Julien Grall <julien.grall@linaro.org> > > I was about to acked + apply but it failed to build on arm64 with: > > domain.c: In function 'alloc_vcpu_struct': > /local/scratch/ianc/devel/committer.git/xen/include/xen/lib.h:19:31: error: static assertion failed: "!(sizeof(*v) > PAGE_SIZE)" > #define BUILD_BUG_ON(cond) ({ _Static_assert(!(cond), "!(" #cond ")"); }) > ^ > domain.c:415:5: note: in expansion of macro 'BUILD_BUG_ON' > BUILD_BUG_ON(sizeof(*v) > PAGE_SIZE); > ^ > struct arch_vcpu is apparently now too large. > > I had also reworded your commit message somewhat: > xen/arm: vcpu: Correctly release resources when a VCPU fails to initialize > > While I was adding new failing code at the end of the function, I noticed > that the vtimers are not freed which messes up all the timers and will crash > Xen quickly when the page s reused. > > Currently neither vcpu_vgic_init nor vcpu_vtimer_init fails, so we > are safe for now. With the new GICv3 code, the former function will be able > to fail. This will result in a memory leak. > > Call vcpu_destroy if the initialization has failed. We also need to add a > boolean to know if the vtimers are correctly setup as the timer common code > doesn't have any safeguard against removing a non-initialized timer. The commit 6fedf29 "xen/arm: Drop event_mask in arch_vcpu" which makes this patch compiles on both arm32 and arm64. I think you can safely push this patch now :). Thanks,
On Wed, 2014-05-21 at 13:39 +0100, Julien Grall wrote: > > The commit 6fedf29 "xen/arm: Drop event_mask in arch_vcpu" which makes > this patch compiles on both arm32 and arm64. > > I think you can safely push this patch now :). Acked + applied.
diff --git a/xen/arch/arm/domain.c b/xen/arch/arm/domain.c index ccccb77..c47db4a 100644 --- a/xen/arch/arm/domain.c +++ b/xen/arch/arm/domain.c @@ -468,12 +468,16 @@ int vcpu_initialise(struct vcpu *v) processor_vcpu_initialise(v); if ( (rc = vcpu_vgic_init(v)) != 0 ) - return rc; + goto fail; if ( (rc = vcpu_vtimer_init(v)) != 0 ) - return rc; + goto fail; return rc; + +fail: + vcpu_destroy(v); + return rc; } void vcpu_destroy(struct vcpu *v) diff --git a/xen/arch/arm/vtimer.c b/xen/arch/arm/vtimer.c index cb690bb..c515e7e 100644 --- a/xen/arch/arm/vtimer.c +++ b/xen/arch/arm/vtimer.c @@ -77,11 +77,16 @@ int vcpu_vtimer_init(struct vcpu *v) : GUEST_TIMER_VIRT_PPI; t->v = v; + v->arch.vtimer_initialized = 1; + return 0; } void vcpu_timer_destroy(struct vcpu *v) { + if ( !v->arch.vtimer_initialized ) + return; + kill_timer(&v->arch.virt_timer.timer); kill_timer(&v->arch.phys_timer.timer); } diff --git a/xen/include/asm-arm/domain.h b/xen/include/asm-arm/domain.h index ec66a4e..1be3da2 100644 --- a/xen/include/asm-arm/domain.h +++ b/xen/include/asm-arm/domain.h @@ -285,6 +285,7 @@ struct arch_vcpu struct vtimer phys_timer; struct vtimer virt_timer; + bool_t vtimer_initialized; } __cacheline_aligned; void vcpu_show_execution_state(struct vcpu *);
While I was adding new failing code at the end of the function, I've noticed that the vtimers are not freed which mess all the timers and will crash Xen quickly when the page will be reused. Currently neither vcpu_vgic_init nor vcpu_vtimer_init fail, so we are safe for now. With the new GICv3 code, the former function will be able to fail. This will result to a memory leak. Call vcpu_destroy if the initialization has failed. We also need to add a boolean to know if the vtimers are correctly setup as the timer common code doesn't have safe guard against removing non-initialized timer. Signed-off-by: Julien Grall <julien.grall@linaro.org> --- Changes in v2: - Update commit message --- xen/arch/arm/domain.c | 8 ++++++-- xen/arch/arm/vtimer.c | 5 +++++ xen/include/asm-arm/domain.h | 1 + 3 files changed, 12 insertions(+), 2 deletions(-)