Re: [PATCH vhost v6 08/11] virtio_ring: introduce virtqueue_dma_dev()

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Mon, 10 Apr 2023 13:14:13 +0800, Jason Wang <jasowang@xxxxxxxxxx> wrote:
> On Mon, Apr 10, 2023 at 11:48 AM Xuan Zhuo <xuanzhuo@xxxxxxxxxxxxxxxxx> wrote:
> >
> > On Fri, 7 Apr 2023 07:02:34 -0400, "Michael S. Tsirkin" <mst@xxxxxxxxxx> wrote:
> > > On Thu, Apr 06, 2023 at 10:20:09AM -0700, Guenter Roeck wrote:
> > > > Hi,
> > > >
> > > > On Mon, Mar 27, 2023 at 12:05:33PM +0800, Xuan Zhuo wrote:
> > > > > Added virtqueue_dma_dev() to get DMA device for virtio. Then the
> > > > > caller can do dma operation in advance. The purpose is to keep memory
> > > > > mapped across multiple add/get buf operations.
> > > > >
> > > > > Signed-off-by: Xuan Zhuo <xuanzhuo@xxxxxxxxxxxxxxxxx>
> > > > > Acked-by: Jason Wang <jasowang@xxxxxxxxxx>
> > > >
> > > > On sparc64, this patch results in
> > > >
> > > > [    4.364643] Unable to handle kernel NULL pointer dereference
> > > > [    4.365157] tsk->{mm,active_mm}->context = 0000000000000000
> > > > [    4.365394] tsk->{mm,active_mm}->pgd = fffff80000402000
> > > > [    4.365611]               \|/ ____ \|/
> > > > [    4.365611]               "@'/ .. \`@"
> > > > [    4.365611]               /_| \__/ |_\
> > > > [    4.365611]                  \__U_/
> > > > [    4.366165] swapper/0(1): Oops [#1]
> > > > [    4.366630] CPU: 0 PID: 1 Comm: swapper/0 Tainted: G                 N 6.3.0-rc5-next-20230405 #1
> > > > [    4.367121] TSTATE: 0000004411001606 TPC: 00000000004375c0 TNPC: 00000000004375c4 Y: 00000002    Tainted: G                 N
> > > > [    4.367548] TPC: <dma_4u_supported+0x20/0x40>
> > > > [    4.368111] g0: 0000000000000000 g1: 0000000001a93a50 g2: 0000000000000000 g3: 0000000001a96170
> > > > [    4.368439] g4: fffff8000416e9a0 g5: fffff8001dc98000 g6: fffff80004170000 g7: 0000000000000005
> > > > [    4.368769] o0: 0000000000000000 o1: ffffffffffffffff o2: 0000000000000001 o3: fffff800040b78d8
> > > > [    4.369095] o4: 0000000000000000 o5: 0000000000000000 sp: fffff80004172d51 ret_pc: 00000000004375ac
> > > > [    4.369934] RPC: <dma_4u_supported+0xc/0x40>
> > > > [    4.370160] l0: 0000000000000002 l1: 0000000000000002 l2: 0000000000000000 l3: 00000000ffffffff
> > > > [    4.370493] l4: 0000000000000001 l5: 000000000119d2b0 l6: fffff8000416e9a0 l7: 00000000018df000
> > > > [    4.370820] i0: 0000000000000001 i1: ffffffffffffffff i2: fffff80004657758 i3: 000000000125ac00
> > > > [    4.371145] i4: 0000000002362400 i5: 0000000000000000 i6: fffff80004172e01 i7: 000000000050e288
> > > > [    4.371473] I7: <dma_set_mask+0x28/0x80>
> > > > [    4.371683] Call Trace:
> > > > [    4.371864] [<000000000050e288>] dma_set_mask+0x28/0x80
> > > > [    4.372123] [<0000000000a83234>] virtio_dev_probe+0x14/0x400
> > > > [    4.372348] [<0000000000ac7a18>] really_probe+0xb8/0x340
> > > > [    4.372555] [<0000000000ac7d64>] driver_probe_device+0x24/0x120
> > > > [    4.372794] [<0000000000ac8010>] __driver_attach+0x90/0x1a0
> > > > [    4.373012] [<0000000000ac5b4c>] bus_for_each_dev+0x4c/0xa0
> > > > [    4.373226] [<0000000000ac6d80>] bus_add_driver+0x140/0x1e0
> > > > [    4.373440] [<0000000000ac8d94>] driver_register+0x74/0x120
> > > > [    4.373653] [<0000000001b2a690>] virtio_net_driver_init+0x74/0xa8
> > > > [    4.373886] [<0000000000427ee0>] do_one_initcall+0x60/0x340
> > > > [    4.374099] [<0000000001b02f50>] kernel_init_freeable+0x1bc/0x228
> > > > [    4.374330] [<0000000000f37264>] kernel_init+0x18/0x134
> > > > [    4.374534] [<00000000004060e8>] ret_from_fork+0x1c/0x2c
> > > > [    4.374738] [<0000000000000000>] 0x0
> > > > [    4.374981] Disabling lock debugging due to kernel taint
> > > > [    4.375237] Caller[000000000050e288]: dma_set_mask+0x28/0x80
> > > > [    4.375477] Caller[0000000000a83234]: virtio_dev_probe+0x14/0x400
> > > > [    4.375704] Caller[0000000000ac7a18]: really_probe+0xb8/0x340
> > > > [    4.375916] Caller[0000000000ac7d64]: driver_probe_device+0x24/0x120
> > > > [    4.376146] Caller[0000000000ac8010]: __driver_attach+0x90/0x1a0
> > > > [    4.376365] Caller[0000000000ac5b4c]: bus_for_each_dev+0x4c/0xa0
> > > > [    4.376583] Caller[0000000000ac6d80]: bus_add_driver+0x140/0x1e0
> > > > [    4.376805] Caller[0000000000ac8d94]: driver_register+0x74/0x120
> > > > [    4.377025] Caller[0000000001b2a690]: virtio_net_driver_init+0x74/0xa8
> > > > [    4.377262] Caller[0000000000427ee0]: do_one_initcall+0x60/0x340
> > > > [    4.377480] Caller[0000000001b02f50]: kernel_init_freeable+0x1bc/0x228
> > > > [    4.377721] Caller[0000000000f37264]: kernel_init+0x18/0x134
> > > > [    4.377930] Caller[00000000004060e8]: ret_from_fork+0x1c/0x2c
> > > > [    4.378140] Caller[0000000000000000]: 0x0
> > > > [    4.378309] Instruction DUMP:
> > > > [    4.378339]  80a22000
> > > > [    4.378556]  12400006
> > > > [    4.378654]  b0102001
> > > > [    4.378746] <c2076658>
> > > > [    4.378838]  b0102000
> > > > [    4.378930]  80a04019
> > > > [    4.379022]  b1653001
> > > > [    4.379115]  81cfe008
> > > > [    4.379507]  913a2000
> > > > [    4.379617]
> > > > [    4.380504] Kernel panic - not syncing: Attempted to kill init! exitcode=0x00000009
> > > >
> > > > Reverting it fixes the problem. Bisect log attached.
> > > >
> > > > The reason is that dma_supported() calls dma_4u_supported(), which
> > > > expects that dev->archdata.iommu has been initialized. However,
> > > > this is not the case for the virtio device. Result is a null pointer
> > > > dereference in dma_4u_supported().
> > > >
> > > > static int dma_4u_supported(struct device *dev, u64 device_mask)
> > > > {
> > > >         struct iommu *iommu = dev->archdata.iommu;
> > > >
> > > >         if (ali_sound_dma_hack(dev, device_mask))
> > > >                 return 1;
> > > >
> > > >         if (device_mask < iommu->dma_addr_mask)
> > > >                       ^^^^^^^^^^^^^^^^^^^^ Crash location
> > > >                 return 0;
> > > >         return 1;
> > > > }
> > > >
> > > > Guenter
> > >
> > > This is from the unconditional call to dma_set_mask_and_coherent, right?
> >
> > Maybe not.
> >
> > The problem is that the dma_4u_supported() will be called in dma_supported().
> > The premise of this function is that IOMMU has been initialized.
> >
> > We hope to turn virtio dev into a direct dma device by dev->ops is null.
> >
> > The first step in DMA frame is obtaining ops. In many cases, get_arch_dma_ops()
> > returns null. When ops is null, dma frame will use direct dma.
> >
> > static inline const struct dma_map_ops *get_dma_ops(struct device *dev)
> > {
> >         if (dev->dma_ops)
> >                 return dev->dma_ops;
> >         return get_arch_dma_ops();
> > }
> >
> > But rethink, this is unreliable. Some platforms have returned their own ops,
> > including X86.
> >
> > Because the priority of dev->dma_ops is higher than get_arch_dma_ops(), we
> > should not let dev->dma_ops be null. We should define a set of dma_ops to
> > actively implement direct dMA.
>
> Then this will duplicate with existing DMA API helpers. Could we tweak
> the sparc or introduce a new flag for the device structure then the
> DMA API knows it needs to use direct mapping?

If we add a new flag will better. This is not the problem of sparc, x86 may
occurs this also.

Thanks.


>
> Adding Christoph for more comments.
>
> >
> >
> > >
> > > Xuan Zhuo, I feel there should an explanation in the commit log
> > > about making this unconditional call. Why are you making it
> > > in probe?
> > >
> > > I note that different virtio transports handle this differently.
> > > And looking at this more I feel this should maybe be reverted for now:
> > >
> > > Your patch does:
> > >
> > > @@ -243,6 +244,11 @@ static int virtio_dev_probe(struct device *_d)
> > >         u64 driver_features;
> > >         u64 driver_features_legacy;
> > >
> > > +       _d->dma_mask = &_d->coherent_dma_mask;
> > > +       err = dma_set_mask_and_coherent(_d, DMA_BIT_MASK(64));
> > > +       if (err)
> > > +               return err;
> > > +
> > >         /* We have a driver! */
> > >
> > >
> > > but for example virtio PCI has a 32 bit fallback.
> > >
> > > For example Virtio legacy also can't access 64 bit addresses at all,
> > > so it does:
> > >
> > >         rc = dma_set_mask(&pci_dev->dev, DMA_BIT_MASK(64));
> >
> > dma_set_mask() will also call dma_supported().
> >
> >
> > >         if (rc) {
> > >                 rc = dma_set_mask_and_coherent(&pci_dev->dev, DMA_BIT_MASK(32));
> > >         } else {
> > >                 /*
> > >                  * The virtio ring base address is expressed as a 32-bit PFN,
> > >                  * with a page size of 1 << VIRTIO_PCI_QUEUE_ADDR_SHIFT.
> > >                  */
> > >                 dma_set_coherent_mask(&pci_dev->dev,
> > >                                 DMA_BIT_MASK(32 + VIRTIO_PCI_QUEUE_ADDR_SHIFT));
> > >         }
> > >
> > >
> > > mmio tries a fallback to 32 bit if 64 bit fails which does not make
> > > sense to me, but maybe I am wrong.
> > >
> > >
> >
> > These physical devices have IOMMU, so it is normal, but Virtio Dev does not have
> > any physical IOMMU, so there are problems. I borrow from VDPA. So I think
> > there are similar problems with VDPA.
>
> Yes.
>
> Thanks
>
> >
> >
> > Thanks.
> >
> >
> > >
> > >
> > >
> > >
> > > > ---
> > > > # bad: [8417c8f5007bf4567ccffda850a3157c7d905f67] Add linux-next specific files for 20230405
> > > > # good: [7e364e56293bb98cae1b55fd835f5991c4e96e7d] Linux 6.3-rc5
> > > > git bisect start 'HEAD' 'v6.3-rc5'
> > > > # good: [699c146eb5a03458f29e94cfde4d7dd7a36deeb4] Merge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/herbert/cryptodev-2.6.git
> > > > git bisect good 699c146eb5a03458f29e94cfde4d7dd7a36deeb4
> > > > # good: [efe74e6476a9f04263288009910f07a26597386f] Merge branch 'for-next' of git://git.kernel.org/pub/scm/linux/kernel/git/krzk/linux-dt.git
> > > > git bisect good efe74e6476a9f04263288009910f07a26597386f
> > > > # good: [24be607eedb226c1627973190a0b65cab39440b9] Merge branch 'char-misc-next' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/char-misc.git
> > > > git bisect good 24be607eedb226c1627973190a0b65cab39440b9
> > > > # good: [5213285e3f212cf266c085c1c11041adda2bc63f] Merge branch 'for-next' of git://git.kernel.org/pub/scm/linux/kernel/git/mkp/scsi.git
> > > > git bisect good 5213285e3f212cf266c085c1c11041adda2bc63f
> > > > # bad: [e8f5d7e1787104da5977773ba5b3e1d502fdb9e3] Merge branch 'for-next' of git://git.kernel.org/pub/scm/linux/kernel/git/thierry.reding/linux-pwm.git
> > > > git bisect bad e8f5d7e1787104da5977773ba5b3e1d502fdb9e3
> > > > # bad: [c08b5ad3300122790cef1bf7c1f51c554c778e4d] Merge branch 'gpio/for-next' of git://git.kernel.org/pub/scm/linux/kernel/git/brgl/linux.git
> > > > git bisect bad c08b5ad3300122790cef1bf7c1f51c554c778e4d
> > > > # good: [4ea0c97776bf8c63805eb0f8182d9c20072219d9] gpiolib: Check array_info for NULL only once in gpiod_get_array()
> > > > git bisect good 4ea0c97776bf8c63805eb0f8182d9c20072219d9
> > > > # good: [b8e4bb0f05bef8334e689618c75cfed122f3a292] vdpa_sim: use kthread worker
> > > > git bisect good b8e4bb0f05bef8334e689618c75cfed122f3a292
> > > > # bad: [ac1fa106f52a3ea2d2718a0de7d532d3d6c03e4e] Merge branch 'linux-next' of git://git.kernel.org/pub/scm/linux/kernel/git/mst/vhost.git
> > > > git bisect bad ac1fa106f52a3ea2d2718a0de7d532d3d6c03e4e
> > > > # good: [56c8cf37fb0800ad66c926eb3d044e50691afa6e] virtio_ring: packed: separate dma codes
> > > > git bisect good 56c8cf37fb0800ad66c926eb3d044e50691afa6e
> > > > # good: [f19986b93ab7aa23f76dedf9fd8657f865324f78] virtio_ring: update document for virtqueue_add_*
> > > > git bisect good f19986b93ab7aa23f76dedf9fd8657f865324f78
> > > > # bad: [60602b367dc928e2b3d6c1f21df5d05f90e37fa3] virtio_ring: correct the expression of the description of virtqueue_resize()
> > > > git bisect bad 60602b367dc928e2b3d6c1f21df5d05f90e37fa3
> > > > # bad: [3a06353479111e1c9e072825bb0d0730e3a0f4e7] virtio_ring: introduce virtqueue_dma_dev()
> > > > git bisect bad 3a06353479111e1c9e072825bb0d0730e3a0f4e7
> > > > # first bad commit: [3a06353479111e1c9e072825bb0d0730e3a0f4e7] virtio_ring: introduce virtqueue_dma_dev()
> > >
> >
>
_______________________________________________
Virtualization mailing list
Virtualization@xxxxxxxxxxxxxxxxxxxxxxxxxx
https://lists.linuxfoundation.org/mailman/listinfo/virtualization




[Index of Archives]     [KVM Development]     [Libvirt Development]     [Libvirt Users]     [CentOS Virtualization]     [Netdev]     [Ethernet Bridging]     [Linux Wireless]     [Kernel Newbies]     [Security]     [Linux for Hams]     [Netfilter]     [Bugtraq]     [Yosemite Forum]     [MIPS Linux]     [ARM Linux]     [Linux RAID]     [Linux Admin]     [Samba]

  Powered by Linux