qemu-devel.nongnu.org archive mirror
 help / color / mirror / Atom feed
From: "Cédric Le Goater" <clg@redhat.com>
To: Amit Machhiwal <amachhiw@linux.ibm.com>,
	qemu-ppc@nongnu.org, Nicholas Piggin <npiggin@gmail.com>,
	Harsh Prateek Bora <harshpb@linux.ibm.com>
Cc: qemu-devel@nongnu.org, Vaibhav Jain <vaibhav@linux.ibm.com>,
	Shivaprasad G Bhat <sbhat@linux.ibm.com>,
	Daniel Henrique Barboza <danielhb413@gmail.com>,
	Alex Williamson <alex.williamson@redhat.com>
Subject: Re: [PATCH] vfio/spapr: Fix L2 crash with PCI device passthrough with L2 guest memory > 128G
Date: Fri, 4 Apr 2025 13:29:04 +0200	[thread overview]
Message-ID: <1beef03c-ac75-4f25-8b39-0abf01384549@redhat.com> (raw)
In-Reply-To: <20250404091721.2653539-1-amachhiw@linux.ibm.com>

On 4/4/25 11:17, Amit Machhiwal wrote:
> An L2 KVM guest fails to boot inside a pSeries LPAR when booted with a
> memory more than 128 GB and PCI device passthrough. The L2 guest also
> crashes when it is booted with a memory greater than 128 GB and a PCI
> device is hotplugged later.
> 
> The issue arises from a conditional check for `levels > 1` in
> `spapr_tce_create_table()` within L1 KVM. This check is meant to prevent
> multi-level TCEs, which are not supported by the PowerVM hypervisor. As
> a result, when QEMU makes a `VFIO_IOMMU_SPAPR_TCE_CREATE` ioctl call
> with `levels > 1`, it triggers the conditional check and returns
> `EINVAL`, causing the guest to crash with the following errors:
> 
>   2025-03-04T06:36:36.133117Z qemu-system-ppc64: Failed to create a window, ret = -1 (Invalid argument)
>   2025-03-04T06:36:36.133176Z qemu-system-ppc64: Failed to create SPAPR window: Invalid argument
>   qemu: hardware error: vfio: DMA mapping failed, unable to continue
> 
> Fix this by checking the supported DDW "levels" returned by the
> VFIO_IOMMU_SPAPR_TCE_GET_INFO ioctl before attempting the TCE create
> ioctl in KVM.
> 
> The patch has been tested on KVM guests with memory configurations of up
> to 390GB, and 450GB on PowerVM and bare-metal environments respectively.
> > Signed-off-by: Amit Machhiwal <amachhiw@linux.ibm.com>
> ---
>   hw/vfio/spapr.c | 35 ++++++++++++++++++++++++++---------
>   1 file changed, 26 insertions(+), 9 deletions(-)
> 
> diff --git a/hw/vfio/spapr.c b/hw/vfio/spapr.c
> index 1a5d1611f2cd..07498218fea9 100644
> --- a/hw/vfio/spapr.c
> +++ b/hw/vfio/spapr.c
> @@ -26,6 +26,7 @@ typedef struct VFIOSpaprContainer {
>       VFIOContainer container;
>       MemoryListener prereg_listener;
>       QLIST_HEAD(, VFIOHostDMAWindow) hostwin_list;
> +    unsigned int levels;
>   } VFIOSpaprContainer;
>   
>   OBJECT_DECLARE_SIMPLE_TYPE(VFIOSpaprContainer, VFIO_IOMMU_SPAPR);
> @@ -236,9 +237,11 @@ static int vfio_spapr_create_window(VFIOContainer *container,
>   {
>       int ret = 0;
>       VFIOContainerBase *bcontainer = &container->bcontainer;
> +    VFIOSpaprContainer *scontainer = container_of(container, VFIOSpaprContainer,
> +                                                  container);
>       IOMMUMemoryRegion *iommu_mr = IOMMU_MEMORY_REGION(section->mr);
>       uint64_t pagesize = memory_region_iommu_get_min_page_size(iommu_mr), pgmask;
> -    unsigned entries, bits_total, bits_per_level, max_levels;
> +    unsigned entries, bits_total, bits_per_level, max_levels, ddw_levels;
>       struct vfio_iommu_spapr_tce_create create = { .argsz = sizeof(create) };
>       long rampagesize = qemu_minrampagesize();
>   
> @@ -291,16 +294,28 @@ static int vfio_spapr_create_window(VFIOContainer *container,
>        */
>       bits_per_level = ctz64(qemu_real_host_page_size()) + 8;
>       create.levels = bits_total / bits_per_level;
> -    if (bits_total % bits_per_level) {
> -        ++create.levels;
> -    }
> -    max_levels = (64 - create.page_shift) / ctz64(qemu_real_host_page_size());
> -    for ( ; create.levels <= max_levels; ++create.levels) {
> -        ret = ioctl(container->fd, VFIO_IOMMU_SPAPR_TCE_CREATE, &create);
> -        if (!ret) {
> -            break;
> +
> +    ddw_levels = scontainer->levels;
> +    if (ddw_levels > 1) {
> +        if (bits_total % bits_per_level) {
> +            ++create.levels;
>           }
> +        max_levels = (64 - create.page_shift) / ctz64(qemu_real_host_page_size());
> +        for ( ; create.levels <= max_levels; ++create.levels) {
> +            ret = ioctl(container->fd, VFIO_IOMMU_SPAPR_TCE_CREATE, &create);
> +            if (!ret) {
> +                break;
> +            }
> +        }
> +    } else { /* ddw_levels == 1 */
> +        if (create.levels > ddw_levels) {
> +            error_report("Host doesn't support multi-level TCE tables. "
> +                         "Use larger IO page size. Supported mask is 0x%lx",
> +                         bcontainer->pgsizes);

While at it, please modify vfio_spapr_create_window(), add an 'Error **'
parameter to report errors to the caller with error_setg(errp ...)

Thanks,

C.




> +        }
> +        ret = ioctl(container->fd, VFIO_IOMMU_SPAPR_TCE_CREATE, &create);
>       }
> +
>       if (ret) {
>           error_report("Failed to create a window, ret = %d (%m)", ret);
>           return -errno;
> @@ -502,6 +517,8 @@ static bool vfio_spapr_container_setup(VFIOContainerBase *bcontainer,
>           goto listener_unregister_exit;
>       }
>   
> +    scontainer->levels = info.ddw.levels;
> +
>       if (v2) {
>           bcontainer->pgsizes = info.ddw.pgsizes;
>           /*
> 
> base-commit: 0adf626718bc0ca9c46550249a76047f8e45da15



  reply	other threads:[~2025-04-04 11:30 UTC|newest]

Thread overview: 3+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2025-04-04  9:17 [PATCH] vfio/spapr: Fix L2 crash with PCI device passthrough with L2 guest memory > 128G Amit Machhiwal
2025-04-04 11:29 ` Cédric Le Goater [this message]
2025-04-07  8:45   ` Amit Machhiwal

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=1beef03c-ac75-4f25-8b39-0abf01384549@redhat.com \
    --to=clg@redhat.com \
    --cc=alex.williamson@redhat.com \
    --cc=amachhiw@linux.ibm.com \
    --cc=danielhb413@gmail.com \
    --cc=harshpb@linux.ibm.com \
    --cc=npiggin@gmail.com \
    --cc=qemu-devel@nongnu.org \
    --cc=qemu-ppc@nongnu.org \
    --cc=sbhat@linux.ibm.com \
    --cc=vaibhav@linux.ibm.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).