From: Matthew Rosato <mjrosato@linux.ibm.com>
To: Cornelia Huck <cohuck@redhat.com>
Cc: alex.williamson@redhat.com, pmorel@linux.ibm.com,
schnelle@linux.ibm.com, rth@twiddle.net, david@redhat.com,
thuth@redhat.com, pasic@linux.ibm.com, borntraeger@de.ibm.com,
mst@redhat.com, pbonzini@redhat.com, qemu-s390x@nongnu.org,
qemu-devel@nongnu.org, kvm@vger.kernel.org
Subject: Re: [PATCH v3 5/5] s390x/pci: Honor DMA limits set by vfio
Date: Wed, 16 Sep 2020 08:58:39 -0400 [thread overview]
Message-ID: <df001856-dcb3-6d34-7934-0fb0f7c02ba7@linux.ibm.com> (raw)
In-Reply-To: <20200916130524.48e11b26.cohuck@redhat.com>
On 9/16/20 7:05 AM, Cornelia Huck wrote:
> On Tue, 15 Sep 2020 15:14:43 -0400
> Matthew Rosato <mjrosato@linux.ibm.com> wrote:
>
>> When an s390 guest is using lazy unmapping, it can result in a very
>> large number of oustanding DMA requests, far beyond the default
>> limit configured for vfio. Let's track DMA usage similar to vfio
>> in the host, and trigger the guest to flush their DMA mappings
>> before vfio runs out.
>>
>> Signed-off-by: Matthew Rosato <mjrosato@linux.ibm.com>
>> ---
>> hw/s390x/s390-pci-bus.c | 56 +++++++++++++++++++++++++++++++++++++++++++-----
>> hw/s390x/s390-pci-bus.h | 9 ++++++++
>> hw/s390x/s390-pci-inst.c | 34 +++++++++++++++++++++++------
>> hw/s390x/s390-pci-inst.h | 3 +++
>> 4 files changed, 91 insertions(+), 11 deletions(-)
>
> (...)
>
>> @@ -737,6 +740,41 @@ static void s390_pci_iommu_free(S390pciState *s, PCIBus *bus, int32_t devfn)
>> object_unref(OBJECT(iommu));
>> }
>>
>> +static S390PCIDMACount *s390_start_dma_count(S390pciState *s, VFIODevice *vdev)
>
> Should these go into the new vfio-related file?
>
>> +{
>> + int id = vdev->group->container->fd;
>> + S390PCIDMACount *cnt;
>> + uint32_t avail;
>> +
>> + if (!s390_pci_update_dma_avail(id, &avail)) {
>> + return NULL;
>> + }
>> +
>> + QTAILQ_FOREACH(cnt, &s->zpci_dma_limit, link) {
>> + if (cnt->id == id) {
>> + cnt->users++;
>> + return cnt;
>> + }
>> + }
>> +
>> + cnt = g_new0(S390PCIDMACount, 1);
>> + cnt->id = id;
>> + cnt->users = 1;
>> + cnt->avail = avail;
>> + QTAILQ_INSERT_TAIL(&s->zpci_dma_limit, cnt, link);
>> + return cnt;
>> +}
>> +
>> +static void s390_end_dma_count(S390pciState *s, S390PCIDMACount *cnt)
>> +{
>> + assert(cnt);
>> +
>> + cnt->users--;
>> + if (cnt->users == 0) {
>> + QTAILQ_REMOVE(&s->zpci_dma_limit, cnt, link);
>> + }
>> +}
>> +
>> static void s390_pcihost_realize(DeviceState *dev, Error **errp)
>> {
>> PCIBus *b;
>> @@ -764,6 +802,7 @@ static void s390_pcihost_realize(DeviceState *dev, Error **errp)
>> s->bus_no = 0;
>> QTAILQ_INIT(&s->pending_sei);
>> QTAILQ_INIT(&s->zpci_devs);
>> + QTAILQ_INIT(&s->zpci_dma_limit);
>>
>> css_register_io_adapters(CSS_IO_ADAPTER_PCI, true, false,
>> S390_ADAPTER_SUPPRESSIBLE, errp);
>> @@ -902,6 +941,7 @@ static void s390_pcihost_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
>> {
>> S390pciState *s = S390_PCI_HOST_BRIDGE(hotplug_dev);
>> PCIDevice *pdev = NULL;
>> + VFIOPCIDevice *vpdev = NULL;
>> S390PCIBusDevice *pbdev = NULL;
>>
>> if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_BRIDGE)) {
>> @@ -941,17 +981,20 @@ static void s390_pcihost_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
>> }
>> }
>>
>> + pbdev->pdev = pdev;
>> + pbdev->iommu = s390_pci_get_iommu(s, pci_get_bus(pdev), pdev->devfn);
>> + pbdev->iommu->pbdev = pbdev;
>> + pbdev->state = ZPCI_FS_DISABLED;
>> +
>> if (object_dynamic_cast(OBJECT(dev), "vfio-pci")) {
>> pbdev->fh |= FH_SHM_VFIO;
>> + vpdev = container_of(pbdev->pdev, VFIOPCIDevice, pdev);
>> + pbdev->iommu->dma_limit = s390_start_dma_count(s,
>> + &vpdev->vbasedev);
>
> I think you can just pass s and pbdev to that function... that would
> move dealing with vfio specifics from this file.
I had considered this as well, should have went with my gut -- I'll move
them.
>
>> } else {
>> pbdev->fh |= FH_SHM_EMUL;
>> }
>>
>> - pbdev->pdev = pdev;
>> - pbdev->iommu = s390_pci_get_iommu(s, pci_get_bus(pdev), pdev->devfn);
>> - pbdev->iommu->pbdev = pbdev;
>> - pbdev->state = ZPCI_FS_DISABLED;
>> -
>> if (s390_pci_msix_init(pbdev)) {
>> error_setg(errp, "MSI-X support is mandatory "
>> "in the S390 architecture");
>
> (...)
>
>> diff --git a/hw/s390x/s390-pci-inst.c b/hw/s390x/s390-pci-inst.c
>> index 2f7a7d7..cc34b17 100644
>> --- a/hw/s390x/s390-pci-inst.c
>> +++ b/hw/s390x/s390-pci-inst.c
>> @@ -32,6 +32,9 @@
>> } \
>> } while (0)
>>
>> +#define inc_dma_avail(iommu) if (iommu->dma_limit) iommu->dma_limit->avail++;
>
> I was thinking more of something like
>
> static inline void inc_dma_avail(S390PCIIOMMU *iommu)
> {
> if (iommu->dma_limit) {
> iommu->dma_limit->avail++;
> }
> }
>
Ah, I read the 'lowercase' and missed the 'inline function' part of your
previous comment, sorry. Will change.
>> +#define dec_dma_avail(iommu) if (iommu->dma_limit) iommu->dma_limit->avail--;
>> +
>> static void s390_set_status_code(CPUS390XState *env,
>> uint8_t r, uint64_t status_code)
>> {
>
> (...)
>
prev parent reply other threads:[~2020-09-16 20:39 UTC|newest]
Thread overview: 15+ messages / expand[flat|nested] mbox.gz Atom feed top
2020-09-15 19:14 [PATCH v3 0/5] s390x/pci: Accomodate vfio DMA limiting Matthew Rosato
2020-09-15 19:14 ` [PATCH v3 1/5] linux-headers: update against 5.9-rc5 Matthew Rosato
2020-09-15 19:14 ` [PATCH v3 2/5] vfio: Create shared routine for scanning info capabilities Matthew Rosato
2020-09-16 7:15 ` Philippe Mathieu-Daudé
2020-09-16 9:58 ` Cornelia Huck
2020-09-15 19:14 ` [PATCH v3 3/5] vfio: Find DMA available capability Matthew Rosato
2020-09-16 10:01 ` Cornelia Huck
2020-09-15 19:14 ` [PATCH v3 4/5] s390x/pci: Add routine to get the vfio dma available count Matthew Rosato
2020-09-16 7:21 ` Philippe Mathieu-Daudé
2020-09-16 10:27 ` Cornelia Huck
2020-09-16 12:55 ` Matthew Rosato
2020-09-17 9:59 ` Cornelia Huck
2020-09-15 19:14 ` [PATCH v3 5/5] s390x/pci: Honor DMA limits set by vfio Matthew Rosato
2020-09-16 11:05 ` Cornelia Huck
2020-09-16 12:58 ` Matthew Rosato [this message]
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=df001856-dcb3-6d34-7934-0fb0f7c02ba7@linux.ibm.com \
--to=mjrosato@linux.ibm.com \
--cc=alex.williamson@redhat.com \
--cc=borntraeger@de.ibm.com \
--cc=cohuck@redhat.com \
--cc=david@redhat.com \
--cc=kvm@vger.kernel.org \
--cc=mst@redhat.com \
--cc=pasic@linux.ibm.com \
--cc=pbonzini@redhat.com \
--cc=pmorel@linux.ibm.com \
--cc=qemu-devel@nongnu.org \
--cc=qemu-s390x@nongnu.org \
--cc=rth@twiddle.net \
--cc=schnelle@linux.ibm.com \
--cc=thuth@redhat.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).