| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <CAMvTesCyXOgHiz_UrGApWYUZdS31dEdcZznRRtjo8Czx6iz_WA@mail.gmail.com>
Date: Thu, 27 Nov 2025 17:16:38 +0800
From: Tianyu Lan <ltykernel@...il.com>
To: vdso@...bites.dev
Cc: decui@...rosoft.com, haiyangz@...rosoft.com, kys@...rosoft.com,
linux-hyperv@...r.kernel.org, linux-kernel@...r.kernel.org,
longli@...rosoft.com, tiala@...rosoft.com, wei.liu@...nel.org
Subject: Re: [RFC PATCH] Drivers: hv: Confidential VMBus exernal memory support
On Thu, Nov 27, 2025 at 1:16 PM <vdso@...bites.dev> wrote:
>
> From: Roman Kisel <vdso@...bites.dev>
>
> Tianyu Lan wrote:
>
> > In CVM(Confidential VM), system memory is encrypted
> > by default. Device drivers typically use the swiotlb
> > bounce buffer for DMA memory, which is decrypted
> > and shared between the guest and host. Confidential
> > Vmbus, however, supports a confidential channel
> > that employs encrypted memory for the Vmbus ring
> > buffer and external DMA memory. The support for
> > the confidential ring buffer has already been
> > integrated.
> >
> > In CVM, device drivers usually employ the standard
> > DMA API to map DMA memory with the bounce buffer,
> > which remains transparent to the device driver.
> > For external DMA memory support, Hyper-V specific
> > DMA operations are introduced, bypassing the bounce
> > buffer when the confidential external memory flag
> > is set. These DMA operations might also be reused
> > for TDISP devices in the future, which also support
> > DMA operations with encrypted memory.
> >
> > The DMA operations used are global architecture
> > DMA operations (for details, see get_arch_dma_ops()
> > and get_dma_ops()), and there is no need to set up
> > for each device individually.
> >
> > Signed-off-by: Tianyu Lan <tiala@...rosoft.com>
>
> Tinayu,
>
> Looks great to me!
>
> Reviewed-by: Roman Kisel <vdso@...bites.dev>
>
> P.S. For the inclined reader, here is how the old, only for
> storage and not satisfactory in other ways my attempt to solve this:
>
> https://lore.kernel.org/linux-hyperv/20250409000835.285105-6-romank@linux.microsoft.com/
> https://lore.kernel.org/linux-hyperv/20250409000835.285105-7-romank@linux.microsoft.com/
>
> Maybe it'd be a good idea to CC folks who provided feedback back then.
>
Hi Roman:
Thanks for your review. I will follow your suggestion.
Thanks
> > ---
> > drivers/hv/vmbus_drv.c | 90 +++++++++++++++++++++++++++++++++++++++++-
> > 1 file changed, 89 insertions(+), 1 deletion(-)
> >
> > diff --git a/drivers/hv/vmbus_drv.c b/drivers/hv/vmbus_drv.c
> > index 0dc4692b411a..ca31231b2c32 100644
> > --- a/drivers/hv/vmbus_drv.c
> > +++ b/drivers/hv/vmbus_drv.c
> > @@ -39,6 +39,9 @@
> > #include <clocksource/hyperv_timer.h>
> > #include <asm/mshyperv.h>
> > #include "hyperv_vmbus.h"
> > +#include "../../kernel/dma/direct.h"
> > +
> > +extern const struct dma_map_ops *dma_ops;
> >
> > struct vmbus_dynid {
> > struct list_head node;
> > @@ -1429,6 +1432,88 @@ static int vmbus_alloc_synic_and_connect(void)
> > return -ENOMEM;
> > }
> >
> > +
> > +static bool hyperv_private_memory_dma(struct device *dev)
> > +{
> > + struct hv_device *hv_dev = device_to_hv_device(dev);
> > +
> > + if (hv_dev && hv_dev->channel && hv_dev->channel->co_external_memory)
> > + return true;
> > + else
> > + return false;
> > +}
> > +
> > +static dma_addr_t hyperv_dma_map_page(struct device *dev, struct page *page,
> > + unsigned long offset, size_t size,
> > + enum dma_data_direction dir,
> > + unsigned long attrs)
> > +{
> > + phys_addr_t phys = page_to_phys(page) + offset;
> > +
> > + if (hyperv_private_memory_dma(dev))
> > + return __phys_to_dma(dev, phys);
> > + else
> > + return dma_direct_map_phys(dev, phys, size, dir, attrs);
> > +}
> > +
> > +static void hyperv_dma_unmap_page(struct device *dev, dma_addr_t dma_handle,
> > + size_t size, enum dma_data_direction dir, unsigned long attrs)
> > +{
> > + if (!hyperv_private_memory_dma(dev))
> > + dma_direct_unmap_phys(dev, dma_handle, size, dir, attrs);
> > +}
> > +
> > +static int hyperv_dma_map_sg(struct device *dev, struct scatterlist *sgl,
> > + int nelems, enum dma_data_direction dir,
> > + unsigned long attrs)
> > +{
> > + struct scatterlist *sg;
> > + dma_addr_t dma_addr;
> > + int i;
> > +
> > + if (hyperv_private_memory_dma(dev)) {
> > + for_each_sg(sgl, sg, nelems, i) {
> > + dma_addr = __phys_to_dma(dev, sg_phys(sg));
> > + sg_dma_address(sg) = dma_addr;
> > + sg_dma_len(sg) = sg->length;
> > + }
> > +
> > + return nelems;
> > + } else {
> > + return dma_direct_map_sg(dev, sgl, nelems, dir, attrs);
> > + }
> > +}
> > +
> > +static void hyperv_dma_unmap_sg(struct device *dev, struct scatterlist *sgl,
> > + int nelems, enum dma_data_direction dir, unsigned long attrs)
> > +{
> > + if (!hyperv_private_memory_dma(dev))
> > + dma_direct_unmap_sg(dev, sgl, nelems, dir, attrs);
> > +}
> > +
> > +static int hyperv_dma_supported(struct device *dev, u64 mask)
> > +{
> > + dev->coherent_dma_mask = mask;
> > + return 1;
> > +}
> > +
> > +static size_t hyperv_dma_max_mapping_size(struct device *dev)
> > +{
> > + if (hyperv_private_memory_dma(dev))
> > + return SIZE_MAX;
> > + else
> > + return swiotlb_max_mapping_size(dev);
> > +}
> > +
> > +const struct dma_map_ops hyperv_dma_ops = {
> > + .map_page = hyperv_dma_map_page,
> > + .unmap_page = hyperv_dma_unmap_page,
> > + .map_sg = hyperv_dma_map_sg,
> > + .unmap_sg = hyperv_dma_unmap_sg,
> > + .dma_supported = hyperv_dma_supported,
> > + .max_mapping_size = hyperv_dma_max_mapping_size,
> > +};
> > +
> > /*
> > * vmbus_bus_init -Main vmbus driver initialization routine.
> > *
> > @@ -1479,8 +1564,11 @@ static int vmbus_bus_init(void)
> > * doing that on each VP while initializing SynIC's wastes time.
> > */
> > is_confidential = ms_hyperv.confidential_vmbus_available;
> > - if (is_confidential)
> > + if (is_confidential) {
> > + dma_ops = &hyperv_dma_ops;
> > pr_info("Establishing connection to the confidential VMBus\n");
> > + }
> > +
> > hv_para_set_sint_proxy(!is_confidential);
> > ret = vmbus_alloc_synic_and_connect();
> > if (ret)
> > --
> > 2.50.1
--
Thanks
Tianyu Lan
Powered by blists - more mailing lists