From: Auger Eric <eric.auger@redhat.com>
To: Peter Xu <peterx@redhat.com>
Cc: yang.zhong@intel.com, peter.maydell@linaro.org,
kevin.tian@intel.com, tnowicki@marvell.com, mst@redhat.com,
jean-philippe.brucker@arm.com, quintela@redhat.com,
qemu-devel@nongnu.org, armbru@redhat.com,
bharatb.linux@gmail.com, qemu-arm@nongnu.org,
dgilbert@redhat.com, eric.auger.pro@gmail.com
Subject: Re: [PATCH for-5.0 v11 08/20] virtio-iommu: Implement translate
Date: Thu, 19 Dec 2019 16:09:47 +0100 [thread overview]
Message-ID: <9ec9d0d5-062b-f96b-c72c-4d15865ff9a1@redhat.com> (raw)
In-Reply-To: <20191219144936.GB50561@xz-x1>
Hi Peter, jean,
On 12/19/19 3:49 PM, Peter Xu wrote:
> On Thu, Dec 19, 2019 at 03:38:34PM +0100, Auger Eric wrote:
>> Hi Peter,
>>
>> On 12/19/19 2:33 PM, Peter Xu wrote:
>>> On Thu, Dec 19, 2019 at 11:30:40AM +0100, Auger Eric wrote:
>>>> Hi Peter,
>>>> On 12/10/19 8:33 PM, Peter Xu wrote:
>>>>> On Fri, Nov 22, 2019 at 07:29:31PM +0100, Eric Auger wrote:
>>>>>> This patch implements the translate callback
>>>>>>
>>>>>> Signed-off-by: Eric Auger <eric.auger@redhat.com>
>>>>>>
>>>>>> ---
>>>>>>
>>>>>> v10 -> v11:
>>>>>> - take into account the new value struct and use
>>>>>> g_tree_lookup_extended
>>>>>> - switched to error_report_once
>>>>>>
>>>>>> v6 -> v7:
>>>>>> - implemented bypass-mode
>>>>>>
>>>>>> v5 -> v6:
>>>>>> - replace error_report by qemu_log_mask
>>>>>>
>>>>>> v4 -> v5:
>>>>>> - check the device domain is not NULL
>>>>>> - s/printf/error_report
>>>>>> - set flags to IOMMU_NONE in case of all translation faults
>>>>>> ---
>>>>>> hw/virtio/trace-events | 1 +
>>>>>> hw/virtio/virtio-iommu.c | 63 +++++++++++++++++++++++++++++++++++++++-
>>>>>> 2 files changed, 63 insertions(+), 1 deletion(-)
>>>>>>
>>>>>> diff --git a/hw/virtio/trace-events b/hw/virtio/trace-events
>>>>>> index f25359cee2..de7cbb3c8f 100644
>>>>>> --- a/hw/virtio/trace-events
>>>>>> +++ b/hw/virtio/trace-events
>>>>>> @@ -72,3 +72,4 @@ virtio_iommu_get_endpoint(uint32_t ep_id) "Alloc endpoint=%d"
>>>>>> virtio_iommu_put_endpoint(uint32_t ep_id) "Free endpoint=%d"
>>>>>> virtio_iommu_get_domain(uint32_t domain_id) "Alloc domain=%d"
>>>>>> virtio_iommu_put_domain(uint32_t domain_id) "Free domain=%d"
>>>>>> +virtio_iommu_translate_out(uint64_t virt_addr, uint64_t phys_addr, uint32_t sid) "0x%"PRIx64" -> 0x%"PRIx64 " for sid=%d"
>>>>>> diff --git a/hw/virtio/virtio-iommu.c b/hw/virtio/virtio-iommu.c
>>>>>> index f0a56833a2..a83666557b 100644
>>>>>> --- a/hw/virtio/virtio-iommu.c
>>>>>> +++ b/hw/virtio/virtio-iommu.c
>>>>>> @@ -412,19 +412,80 @@ static IOMMUTLBEntry virtio_iommu_translate(IOMMUMemoryRegion *mr, hwaddr addr,
>>>>>> int iommu_idx)
>>>>>> {
>>>>>> IOMMUDevice *sdev = container_of(mr, IOMMUDevice, iommu_mr);
>>>>>> + viommu_interval interval, *mapping_key;
>>>>>> + viommu_mapping *mapping_value;
>>>>>> + VirtIOIOMMU *s = sdev->viommu;
>>>>>> + viommu_endpoint *ep;
>>>>>> + bool bypass_allowed;
>>>>>> uint32_t sid;
>>>>>> + bool found;
>>>>>> +
>>>>>> + interval.low = addr;
>>>>>> + interval.high = addr + 1;
>>>>>>
>>>>>> IOMMUTLBEntry entry = {
>>>>>> .target_as = &address_space_memory,
>>>>>> .iova = addr,
>>>>>> .translated_addr = addr,
>>>>>> - .addr_mask = ~(hwaddr)0,
>>>>>> + .addr_mask = (1 << ctz32(s->config.page_size_mask)) - 1,
>>>>>> .perm = IOMMU_NONE,
>>>>>> };
>>>>>>
>>>>>> + bypass_allowed = virtio_has_feature(s->acked_features,
>>>>>> + VIRTIO_IOMMU_F_BYPASS);
>>>>>> +
>>>>>
>>>>> Would it be easier to check bypass_allowed here once and then drop the
>>>>> latter [1] and [2] check?
>>>> bypass_allowed does not mean you systematically bypass. You bypass if
>>>> the SID is unknown or if the device is not attached to any domain.
>>>> Otherwise you translate. But maybe I miss your point.
>>>
>>> Ah ok, then could I ask how will this VIRTIO_IOMMU_F_BYPASS be used?
>>> For example, I think VT-d defines passthrough in a totally different
>>> way in that the PT mark will be stored in the per-device context
>>> entries, then we can allow a specific device to be pass-through when
>>> doing DMA. That information is explicit (e.g., unknown SID will
>>> always fail the DMA), and per-device.
>>>
>>> Here do you mean that you just don't put a device into any domain to
>>> show it wants to use PT? Then I'm not sure how do you identify
>>> whether this is a legal PT or a malicious device (e.g., an unknown
>>> device that even does not have any driver bound to it, which will also
>>> satisfy "unknown SID" and "not attached to any domain", iiuc).
>>
>> The virtio-iommu spec currently says:
>>
>> "If the VIRTIO_IOMMU_F_BYPASS feature is negotiated, all accesses from
>> unattached endpoints are
>> allowed and translated by the IOMMU using the identity function. If the
>> feature is not negotiated, any
>> memory access from an unattached endpoint fails. Upon attaching an
>> endpoint in bypass mode to a new
>> domain, any memory access from the endpoint fails, since the domain does
>> not contain any mapping.
>> "
>>
>> I guess this can serve the purpose of devices doing early accesses,
>> before the guest OS gets the hand and maps them?
>
> OK, so there's no global enablement knob for virtio-iommu? Hmm... Then:
well this is a global knob. If this is bot negotiated any unmapped
device can PT.
My assumption above must be wrong as this is a negotiated feature so
anyway the virtio-iommu driver should be involved.
I don't really remember the rationale of the feature bit tbh.
In "[virtio-dev] RE: [RFC] virtio-iommu version 0.4 " Jean discussed
that with Kevein. Sorry I cannot find the link.
" If the endpoint is not attached to any address space,
then the device MAY abort the transaction."
Kevin> From definition of BYPASS, it's orthogonal to whether there is an
address space attached, then should we still allow "May abort" behavior?
Jean> The behavior is left as an implementation choice, and I'm not sure
it's worth enforcing in the architecture. If the endpoint isn't attached
to any domain then (unless VIRTIO_IOMMU_F_BYPASS is negotiated), it
isn't necessarily able to do DMA at all. The virtio-iommu device may
setup DMA mastering lazily, in which case any DMA transaction would
abort, or have setup DMA already, in which case the endpoint can access
MEM_T_BYPASS regions.
Hopefully Jean will remember and comment on this.
Thanks
Eric
>
> - This flag is a must for all virtio-iommu emulation, right?
> (otherwise I can't see how system bootstraps..)
>
> - Should this flag be gone right after OS starts (otherwise I think
> we still have the issue that any malicious device can be seen as
> in PT mode as default)? How is that done?
>
> Thanks,
>
next prev parent reply other threads:[~2019-12-19 15:11 UTC|newest]
Thread overview: 89+ messages / expand[flat|nested] mbox.gz Atom feed top
2019-11-22 18:29 [PATCH for-5.0 v11 00/20] VIRTIO-IOMMU device Eric Auger
2019-11-22 18:29 ` [PATCH for-5.0 v11 01/20] migration: Support QLIST migration Eric Auger
2019-11-27 11:46 ` Dr. David Alan Gilbert
2020-01-08 13:19 ` Juan Quintela
2020-01-08 13:40 ` Auger Eric
2020-01-08 13:51 ` Juan Quintela
2020-01-08 14:02 ` Auger Eric
2019-11-22 18:29 ` [PATCH for-5.0 v11 02/20] virtio-iommu: Add skeleton Eric Auger
2019-12-10 16:31 ` Jean-Philippe Brucker
2019-12-19 10:31 ` Auger Eric
2019-11-22 18:29 ` [PATCH for-5.0 v11 03/20] virtio-iommu: Decode the command payload Eric Auger
2019-12-10 16:32 ` Jean-Philippe Brucker
2019-12-10 19:14 ` Peter Xu
2019-11-22 18:29 ` [PATCH for-5.0 v11 04/20] virtio-iommu: Add the iommu regions Eric Auger
2019-12-10 16:34 ` Jean-Philippe Brucker
2019-12-19 18:11 ` Auger Eric
2019-12-10 19:18 ` Peter Xu
2019-11-22 18:29 ` [PATCH for-5.0 v11 05/20] virtio-iommu: Endpoint and domains structs and helpers Eric Auger
2019-12-10 16:37 ` Jean-Philippe Brucker
2019-12-19 18:31 ` Auger Eric
2019-12-20 17:00 ` Jean-Philippe Brucker
2019-12-23 9:11 ` Auger Eric
2019-11-22 18:29 ` [PATCH for-5.0 v11 06/20] virtio-iommu: Implement attach/detach command Eric Auger
2019-12-10 16:41 ` Jean-Philippe Brucker
2019-12-23 9:14 ` Auger Eric
2019-11-22 18:29 ` [PATCH for-5.0 v11 07/20] virtio-iommu: Implement map/unmap Eric Auger
2019-12-10 16:43 ` Jean-Philippe Brucker
2019-12-23 9:42 ` Auger Eric
2019-11-22 18:29 ` [PATCH for-5.0 v11 08/20] virtio-iommu: Implement translate Eric Auger
2019-12-10 16:43 ` Jean-Philippe Brucker
2019-12-10 19:33 ` Peter Xu
2019-12-19 10:30 ` Auger Eric
2019-12-19 13:33 ` Peter Xu
2019-12-19 14:38 ` Auger Eric
2019-12-19 14:49 ` Peter Xu
2019-12-19 15:09 ` Auger Eric [this message]
2019-12-20 16:26 ` Jean-Philippe Brucker
2019-12-20 16:51 ` Peter Xu
2020-01-06 17:06 ` Jean-Philippe Brucker
2020-01-06 17:58 ` Peter Xu
2020-01-07 10:10 ` Jean-Philippe Brucker
2020-01-08 16:55 ` Auger Eric
2020-01-09 8:47 ` Jean-Philippe Brucker
2020-01-09 8:58 ` Auger Eric
2020-01-09 10:40 ` Jean-Philippe Brucker
2020-01-09 11:01 ` Auger Eric
2020-01-09 11:15 ` Jean-Philippe Brucker
2020-01-09 11:32 ` Auger Eric
2019-11-22 18:29 ` [PATCH for-5.0 v11 09/20] virtio-iommu: Implement fault reporting Eric Auger
2019-12-10 16:44 ` Jean-Philippe Brucker
2019-11-22 18:29 ` [PATCH for-5.0 v11 10/20] virtio-iommu-pci: Add virtio iommu pci support Eric Auger
2019-12-10 16:44 ` Jean-Philippe Brucker
2019-11-22 18:29 ` [PATCH for-5.0 v11 11/20] hw/arm/virt: Add the virtio-iommu device tree mappings Eric Auger
2019-12-10 16:45 ` Jean-Philippe Brucker
2019-11-22 18:29 ` [PATCH for-5.0 v11 12/20] qapi: Introduce DEFINE_PROP_INTERVAL Eric Auger
2019-11-22 19:03 ` Dr. David Alan Gilbert
2019-11-25 13:12 ` Auger Eric
2019-12-12 12:17 ` Markus Armbruster
2019-12-12 15:13 ` Auger Eric
2019-12-13 10:03 ` Markus Armbruster
2019-11-22 18:29 ` [PATCH for-5.0 v11 13/20] virtio-iommu: Implement probe request Eric Auger
2019-12-10 16:46 ` Jean-Philippe Brucker
2019-12-10 19:36 ` Peter Xu
2019-11-22 18:29 ` [PATCH for-5.0 v11 14/20] virtio-iommu: Handle reserved regions in the translation process Eric Auger
2019-12-10 16:46 ` Jean-Philippe Brucker
2019-12-10 19:39 ` Peter Xu
2019-11-22 18:29 ` [PATCH for-5.0 v11 15/20] virtio-iommu-pci: Add array of Interval properties Eric Auger
2019-12-10 16:47 ` Jean-Philippe Brucker
2019-11-22 18:29 ` [PATCH for-5.0 v11 16/20] hw/arm/virt-acpi-build: Introduce fill_iort_idmap helper Eric Auger
2019-12-10 16:47 ` Jean-Philippe Brucker
2019-11-22 18:29 ` [PATCH for-5.0 v11 17/20] hw/arm/virt-acpi-build: Add virtio-iommu node in IORT table Eric Auger
2019-12-10 16:48 ` Jean-Philippe Brucker
2019-11-22 18:29 ` [PATCH for-5.0 v11 18/20] virtio-iommu: Support migration Eric Auger
2019-11-27 12:06 ` Dr. David Alan Gilbert
2019-12-10 16:50 ` Jean-Philippe Brucker
2019-12-19 11:03 ` Auger Eric
2019-12-10 20:01 ` Peter Xu
2019-12-24 7:39 ` Auger Eric
2019-11-22 18:29 ` [PATCH for-5.0 v11 19/20] pc: Add support for virtio-iommu-pci Eric Auger
2019-12-10 16:50 ` Jean-Philippe Brucker
2019-12-24 7:39 ` Auger Eric
2020-01-09 12:02 ` Michael S. Tsirkin
2020-01-09 13:34 ` Auger Eric
2019-11-22 18:29 ` [PATCH for-5.0 v11 20/20] tests: Add virtio-iommu test Eric Auger
2019-11-22 21:56 ` [PATCH for-5.0 v11 00/20] VIRTIO-IOMMU device no-reply
2019-12-11 16:40 ` Michael S. Tsirkin
2019-12-11 16:48 ` Auger Eric
2019-12-11 20:40 ` Michael S. Tsirkin
2019-12-12 15:05 ` Auger Eric
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=9ec9d0d5-062b-f96b-c72c-4d15865ff9a1@redhat.com \
--to=eric.auger@redhat.com \
--cc=armbru@redhat.com \
--cc=bharatb.linux@gmail.com \
--cc=dgilbert@redhat.com \
--cc=eric.auger.pro@gmail.com \
--cc=jean-philippe.brucker@arm.com \
--cc=kevin.tian@intel.com \
--cc=mst@redhat.com \
--cc=peter.maydell@linaro.org \
--cc=peterx@redhat.com \
--cc=qemu-arm@nongnu.org \
--cc=qemu-devel@nongnu.org \
--cc=quintela@redhat.com \
--cc=tnowicki@marvell.com \
--cc=yang.zhong@intel.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).