| Message ID | 1457715275-11925-3-git-send-email-okaya@codeaurora.org (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On 3/11/2016 11:54 AM, Sinan Kaya wrote: > The code was allowing platform devices to be used without a supporting VFIO > reset driver. The hardware can be left in some inconsistent state after a > guest machine abort. > > The reset driver will put the hardware back to safe state and disable > interrupts before returning the control back to the host machine. > > Signed-off-by: Sinan Kaya <okaya@codeaurora.org> > --- > drivers/vfio/platform/vfio_platform_common.c | 16 +++++++++------- > 1 file changed, 9 insertions(+), 7 deletions(-) > I was looking at the code. It looks like this is going to break VFIO AMBA. The common implementation is shared with AMBA and platform devices. I couldn't see a reset function for AMBA devices. Is there anyway to write reset function for it? I have no experience with AMBA devices. Would you include a reset needed flag and just not require it for AMBA? (I honestly don't like this idea)
Hi Sinan, On 03/13/2016 06:25 PM, Sinan Kaya wrote: > On 3/11/2016 11:54 AM, Sinan Kaya wrote: >> The code was allowing platform devices to be used without a supporting VFIO >> reset driver. The hardware can be left in some inconsistent state after a >> guest machine abort. >> >> The reset driver will put the hardware back to safe state and disable >> interrupts before returning the control back to the host machine. >> >> Signed-off-by: Sinan Kaya <okaya@codeaurora.org> >> --- >> drivers/vfio/platform/vfio_platform_common.c | 16 +++++++++------- >> 1 file changed, 9 insertions(+), 7 deletions(-) >> > I was looking at the code. It looks like this is going to break VFIO AMBA. The > common implementation is shared with AMBA and platform devices. > > I couldn't see a reset function for AMBA devices. > > Is there anyway to write reset function for it? I have no experience with > AMBA devices. To my knowledge only the PL330 DMA controller (drivers/dma/pl330.c) was passthrough'ed at some point, rather for development purpose than production purpose. This was done by Virtual Open Systems (ask Baptiste for more details). But I don't think this is really used. > > Would you include a reset needed flag and just not require it for AMBA? > (I honestly don't like this idea) > I think the requirement also makes sense for AMBA. Maybe an option would be to add a module parameter that would allow to use the vfio platform/amba driver without reset module (with explicit opt-in from the user). Maybe this can be done later on. FYI I will not have access to my mailbox until the end of the week. Best Regards Eric
Hi Baptiste, >> I couldn't see a reset function for AMBA devices. >> >> Is there anyway to write reset function for it? I have no experience with >> AMBA devices. > To my knowledge only the PL330 DMA controller (drivers/dma/pl330.c) was > passthrough'ed at some point, rather for development purpose than > production purpose. This was done by Virtual Open Systems (ask Baptiste > for more details). But I don't think this is really used. Are you OK with requiring reset driver for AMBA devices too? The PL330 driver won't work until a reset driver is submitted for it. >> >> Would you include a reset needed flag and just not require it for AMBA? >> (I honestly don't like this idea) >> > I think the requirement also makes sense for AMBA. > > Maybe an option would be to add a module parameter that would allow to > use the vfio platform/amba driver without reset module (with explicit > opt-in from the user). Maybe this can be done later on. I can see this flag useful for testing purposes but it should not be used in production. How about I add a module parameter which is not set by default? When set, I don't perform the reset requirement check. > > FYI I will not have access to my mailbox until the end of the week. OK, no rush. > > Best Regards > > Eric >
Hi Sinan, On 03/16/2016 03:27 PM, Sinan Kaya wrote: > Hi Baptiste, > >>> I couldn't see a reset function for AMBA devices. >>> >>> Is there anyway to write reset function for it? I have no experience with >>> AMBA devices. >> To my knowledge only the PL330 DMA controller (drivers/dma/pl330.c) was >> passthrough'ed at some point, rather for development purpose than >> production purpose. This was done by Virtual Open Systems (ask Baptiste >> for more details). But I don't think this is really used. > > Are you OK with requiring reset driver for AMBA devices too? The PL330 > driver won't work until a reset driver is submitted for it. > >>> >>> Would you include a reset needed flag and just not require it for AMBA? >>> (I honestly don't like this idea) >>> >> I think the requirement also makes sense for AMBA. >> >> Maybe an option would be to add a module parameter that would allow to >> use the vfio platform/amba driver without reset module (with explicit >> opt-in from the user). Maybe this can be done later on. > > I can see this flag useful for testing purposes but it should not be used in > production. > > How about I add a module parameter which is not set by default? When set, > I don't perform the reset requirement check. Yes that's what I meant Best Regards Eric > >> >> FYI I will not have access to my mailbox until the end of the week. > > OK, no rush. > >> >> Best Regards >> >> Eric >> > >
On 3/27/2016 11:21 AM, Eric Auger wrote: >> How about I add a module parameter which is not set by default? When set, >> > I don't perform the reset requirement check. > Yes that's what I meant > > Best Regards > > Eric Got it. I'll post something soon.
diff --git a/drivers/vfio/platform/vfio_platform_common.c b/drivers/vfio/platform/vfio_platform_common.c index b3f6ba2..5535fe1 100644 --- a/drivers/vfio/platform/vfio_platform_common.c +++ b/drivers/vfio/platform/vfio_platform_common.c @@ -165,12 +165,8 @@ static void vfio_platform_release(void *device_data) mutex_lock(&driver_lock); if (!(--vdev->refcnt)) { - if (vdev->reset) { - dev_info(vdev->device, "reset\n"); - vdev->reset(vdev); - } else { - dev_warn(vdev->device, "no reset function found!\n"); - } + dev_info(vdev->device, "reset\n"); + vdev->reset(vdev); vfio_platform_regions_cleanup(vdev); vfio_platform_irq_cleanup(vdev); } @@ -634,7 +630,13 @@ int vfio_platform_probe_common(struct vfio_platform_device *vdev, return ret; } - vfio_platform_get_reset(vdev); + ret = vfio_platform_get_reset(vdev); + if (ret) { + pr_err("vfio: no reset function found for device %s\n", + vdev->name); + iommu_group_put(group); + return ret; + } mutex_init(&vdev->igate); return 0;
The code was allowing platform devices to be used without a supporting VFIO reset driver. The hardware can be left in some inconsistent state after a guest machine abort. The reset driver will put the hardware back to safe state and disable interrupts before returning the control back to the host machine. Signed-off-by: Sinan Kaya <okaya@codeaurora.org> --- drivers/vfio/platform/vfio_platform_common.c | 16 +++++++++------- 1 file changed, 9 insertions(+), 7 deletions(-)