Implement SHM region management and re-enable DAX in virtiofs

include/libkrun.h

@@ -121,6 +121,24 @@ int32_t krun_add_virtiofs(uint32_t ctx_id,
                           const char *c_tag,
                           const char *c_path);
 
+/**
+ * Adds an independent virtio-fs device pointing to a host's directory with a tag. This
+ * variant allows specifying the size of the DAX window.
+ *
+ * Arguments:
+ *  "ctx_id"         - the configuration context ID.
+ *  "c_tag"          - tag to identify the filesystem in the guest.
+ *  "c_path"         - full path to the directory in the host to be exposed to the guest.
+ *  "shm_size"       - size of the DAX SHM window in bytes.
+ *
+ * Returns:
+ *  Zero on success or a negative error number on failure.
+ */
+int32_t krun_add_virtiofs2(uint32_t ctx_id,
+                           const char *c_tag,
+                           const char *c_path,
+                           uint64_t shm_size);
+
 /**
  * Configures the networking to use passt.
  * Call to this function disables TSI backend to use passt instead.
@@ -154,7 +172,7 @@ int32_t krun_set_passt_fd(uint32_t ctx_id, int fd);
  * Returns:
  *  Zero on success or a negative error number on failure.
  */
-int32_t krun_set_gvproxy_path(uint32_t ctx_id, char* c_path);
+int32_t krun_set_gvproxy_path(uint32_t ctx_id, char *c_path);
 
 /**
  * Sets the MAC address for the virtio-net device when using the passt backend.
@@ -196,17 +214,17 @@ int32_t krun_set_net_mac(uint32_t ctx_id, uint8_t *const c_mac);
 int32_t krun_set_port_map(uint32_t ctx_id, const char *const port_map[]);
 
 /* Flags for virglrenderer.  Copied from virglrenderer bindings. */
-#define VIRGLRENDERER_USE_EGL            1 << 0
-#define VIRGLRENDERER_THREAD_SYNC        1 << 1
-#define VIRGLRENDERER_USE_GLX            1 << 2
-#define VIRGLRENDERER_USE_SURFACELESS    1 << 3
-#define VIRGLRENDERER_USE_GLES           1 << 4
-#define VIRGLRENDERER_USE_EXTERNAL_BLOB  1 << 5
-#define VIRGLRENDERER_VENUS              1 << 6
-#define VIRGLRENDERER_NO_VIRGL           1 << 7
+#define VIRGLRENDERER_USE_EGL 1 << 0
+#define VIRGLRENDERER_THREAD_SYNC 1 << 1
+#define VIRGLRENDERER_USE_GLX 1 << 2
+#define VIRGLRENDERER_USE_SURFACELESS 1 << 3
+#define VIRGLRENDERER_USE_GLES 1 << 4
+#define VIRGLRENDERER_USE_EXTERNAL_BLOB 1 << 5
+#define VIRGLRENDERER_VENUS 1 << 6
+#define VIRGLRENDERER_NO_VIRGL 1 << 7
 #define VIRGLRENDERER_USE_ASYNC_FENCE_CB 1 << 8
-#define VIRGLRENDERER_RENDER_SERVER      1 << 9
-#define VIRGLRENDERER_DRM                1 << 10
+#define VIRGLRENDERER_RENDER_SERVER 1 << 9
+#define VIRGLRENDERER_DRM 1 << 10
 /**
  * Enables and configures a virtio-gpu device.
  *
@@ -219,6 +237,22 @@ int32_t krun_set_port_map(uint32_t ctx_id, const char *const port_map[]);
  */
 int32_t krun_set_gpu_options(uint32_t ctx_id, uint32_t virgl_flags);
 
+/**
+ * Enables and configures a virtio-gpu device. This variant allows specifying
+ * the size of the host window (acting as vRAM in the guest).
+ *
+ * Arguments:
+ *  "ctx_id"      - the configuration context ID.
+ *  "virgl_flags" - flags to pass to virglrenderer.
+ *  "shm_size"    - size of the SHM host window in bytes.
+ *
+ * Returns:
+ *  Zero on success or a negative error number on failure.
+ */
+int32_t krun_set_gpu_options2(uint32_t ctx_id,
+                              uint32_t virgl_flags,
+                              uint64_t shm_size);
+
 /**
  * Enables or disables a virtio-snd device.
  *
@@ -339,7 +373,6 @@ int32_t krun_add_vsock_port(uint32_t ctx_id,
  */
 int32_t krun_get_shutdown_eventfd(uint32_t ctx_id);
 
-
 /**
  * Configures the console device to ignore stdin and write the output to "c_filepath".
  *

src/arch/src/aarch64/mod.rs

@@ -16,12 +16,11 @@ pub mod macos;
 #[cfg(target_os = "macos")]
 pub use self::macos::*;
 
-use std::cmp::min;
 use std::collections::HashMap;
 use std::fmt::Debug;
 
 use self::gic::GICDevice;
-use crate::ArchMemoryInfo;
+use crate::{round_up, ArchMemoryInfo};
 use vm_memory::{Address, GuestAddress, GuestMemory, GuestMemoryMmap};
 
 #[cfg(feature = "efi")]
@@ -42,35 +41,31 @@ pub enum Error {
 
 /// The start of the memory area reserved for MMIO devices.
 pub const MMIO_MEM_START: u64 = layout::MAPPED_IO_START;
-/// The size of the MMIO shared memory area used by virtio-fs DAX.
-pub const MMIO_SHM_SIZE: u64 = 1 << 33;
 
 pub use self::fdt::DeviceInfoForFDT;
 use crate::DeviceType;
 
 /// Returns a Vec of the valid memory addresses for aarch64.
 /// See [`layout`](layout) module for a drawing of the specific memory model for this platform.
 pub fn arch_memory_regions(size: usize) -> (ArchMemoryInfo, Vec<(GuestAddress, usize)>) {
-    let dram_size = min(size as u64, layout::DRAM_MEM_MAX_SIZE) as usize;
+    let page_size: usize = unsafe { libc::sysconf(libc::_SC_PAGESIZE).try_into().unwrap() };
+    let dram_size = round_up(size, page_size);
     let ram_last_addr = layout::DRAM_MEM_START + (dram_size as u64);
     let shm_start_addr = ((ram_last_addr / 0x4000_0000) + 1) * 0x4000_0000;
+
     let info = ArchMemoryInfo {
         ram_last_addr,
         shm_start_addr,
-        shm_size: MMIO_SHM_SIZE,
+        page_size,
     };
     let regions = if cfg!(feature = "efi") {
         vec![
             // Space for loading EDK2 and its variables
             (GuestAddress(0u64), 0x800_0000),
             (GuestAddress(layout::DRAM_MEM_START), dram_size),
-            (GuestAddress(shm_start_addr), MMIO_SHM_SIZE as usize),
         ]
     } else {
-        vec![
-            (GuestAddress(layout::DRAM_MEM_START), dram_size),
-            (GuestAddress(shm_start_addr), MMIO_SHM_SIZE as usize),
-        ]
+        vec![(GuestAddress(layout::DRAM_MEM_START), dram_size)]
     };
 
     (info, regions)

src/arch/src/lib.rs

@@ -11,7 +11,7 @@ use std::result;
 pub struct ArchMemoryInfo {
     pub ram_last_addr: u64,
     pub shm_start_addr: u64,
-    pub shm_size: u64,
+    pub page_size: usize,
 }
 
 /// Module for aarch64 related functionality.
@@ -22,7 +22,6 @@ pub mod aarch64;
 pub use aarch64::{
     arch_memory_regions, configure_system, get_kernel_start, initrd_load_addr,
     layout::CMDLINE_MAX_SIZE, layout::IRQ_BASE, layout::IRQ_MAX, Error, MMIO_MEM_START,
-    MMIO_SHM_SIZE,
 };
 
 /// Module for x86_64 related functionality.
@@ -33,7 +32,7 @@ pub mod x86_64;
 pub use crate::x86_64::{
     arch_memory_regions, configure_system, get_kernel_start, initrd_load_addr,
     layout::CMDLINE_MAX_SIZE, layout::IRQ_BASE, layout::IRQ_MAX, Error, BIOS_SIZE, BIOS_START,
-    MMIO_MEM_START, MMIO_SHM_SIZE, RESET_VECTOR,
+    MMIO_MEM_START, RESET_VECTOR,
 };
 
 /// Type for returning public functions outcome.
@@ -66,6 +65,15 @@ pub struct InitrdConfig {
 /// Default (smallest) memory page size for the supported architectures.
 pub const PAGE_SIZE: usize = 4096;
 
+pub fn round_up(size: usize, align: usize) -> usize {
+    let page_mask = align - 1;
+    (size + page_mask) & !page_mask
+}
+pub fn round_down(size: usize, align: usize) -> usize {
+    let page_mask = !(align - 1);
+    size & page_mask
+}
+
 impl fmt::Display for DeviceType {
     fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
         write!(f, "{self:?}")

src/arch/src/x86_64/mod.rs

@@ -17,8 +17,7 @@ pub mod msr;
 /// Logic for configuring x86_64 registers.
 pub mod regs;
 
-use crate::ArchMemoryInfo;
-use crate::InitrdConfig;
+use crate::{round_up, ArchMemoryInfo, InitrdConfig};
 use arch_gen::x86::bootparam::{boot_params, E820_RAM};
 use vm_memory::Bytes;
 use vm_memory::{
@@ -60,8 +59,6 @@ const FIRST_ADDR_PAST_32BITS: u64 = 1 << 32;
 const MEM_32BIT_GAP_SIZE: u64 = 768 << 20;
 /// The start of the memory area reserved for MMIO devices.
 pub const MMIO_MEM_START: u64 = FIRST_ADDR_PAST_32BITS - MEM_32BIT_GAP_SIZE;
-/// The size of the MMIO shared memory area used by virtio-fs DAX.
-pub const MMIO_SHM_SIZE: u64 = 1 << 33;
 
 /// Returns a Vec of the valid memory addresses.
 /// These should be used to configure the GuestMemoryMmap structure for the platform.
@@ -73,6 +70,9 @@ pub fn arch_memory_regions(
     kernel_load_addr: u64,
     kernel_size: usize,
 ) -> (ArchMemoryInfo, Vec<(GuestAddress, usize)>) {
+    let page_size: usize = unsafe { libc::sysconf(libc::_SC_PAGESIZE).try_into().unwrap() };
+
+    let size = round_up(size, page_size);
     if size < (kernel_load_addr + kernel_size as u64) as usize {
         panic!("Kernel doesn't fit in RAM");
     }
@@ -90,7 +90,6 @@ pub fn arch_memory_regions(
                 vec![
                     (GuestAddress(0), kernel_load_addr as usize),
                     (GuestAddress(kernel_load_addr + kernel_size as u64), size),
-                    (GuestAddress(FIRST_ADDR_PAST_32BITS), MMIO_SHM_SIZE as usize),
                 ],
             )
         }
@@ -108,15 +107,14 @@ pub fn arch_memory_regions(
                         (MMIO_MEM_START - (kernel_load_addr + kernel_size as u64)) as usize,
                     ),
                     (GuestAddress(FIRST_ADDR_PAST_32BITS), remaining),
-                    (GuestAddress(shm_start_addr), MMIO_SHM_SIZE as usize),
                 ],
             )
         }
     };
     let info = ArchMemoryInfo {
         ram_last_addr,
         shm_start_addr,
-        shm_size: MMIO_SHM_SIZE,
+        page_size,
     };
     (info, regions)
 }
@@ -132,6 +130,9 @@ pub fn arch_memory_regions(
     kernel_load_addr: u64,
     kernel_size: usize,
 ) -> (ArchMemoryInfo, Vec<(GuestAddress, usize)>) {
+    let page_size: usize = unsafe { libc::sysconf(libc::_SC_PAGESIZE).try_into().unwrap() };
+
+    let size = round_up(size, page_size);
     if size < (kernel_load_addr + kernel_size as u64) as usize {
         panic!("Kernel doesn't fit in RAM");
     }
@@ -170,7 +171,7 @@ pub fn arch_memory_regions(
     let info = ArchMemoryInfo {
         ram_last_addr,
         shm_start_addr,
-        shm_size: 0,
+        page_size,
     };
     (info, regions)
 }
@@ -319,7 +320,7 @@ mod tests {
     #[test]
     fn regions_lt_4gb() {
         let (_info, regions) = arch_memory_regions(1usize << 29, KERNEL_LOAD_ADDR, KERNEL_SIZE);
-        assert_eq!(3, regions.len());
+        assert_eq!(2, regions.len());
         assert_eq!(GuestAddress(0), regions[0].0);
         assert_eq!(KERNEL_LOAD_ADDR as usize, regions[0].1);
         assert_eq!(
@@ -333,7 +334,7 @@ mod tests {
     fn regions_gt_4gb() {
         let (_info, regions) =
             arch_memory_regions((1usize << 32) + 0x8000, KERNEL_LOAD_ADDR, KERNEL_SIZE);
-        assert_eq!(4, regions.len());
+        assert_eq!(3, regions.len());
         assert_eq!(GuestAddress(0), regions[0].0);
         assert_eq!(KERNEL_LOAD_ADDR as usize, regions[0].1);
         assert_eq!(

src/devices/src/virtio/fs/device.rs

@@ -1,9 +1,13 @@
+#[cfg(target_os = "macos")]
+use crossbeam_channel::Sender;
 use std::cmp;
 use std::io::Write;
 use std::sync::atomic::AtomicUsize;
 use std::sync::{Arc, Mutex};
 use std::thread::JoinHandle;
 
+#[cfg(target_os = "macos")]
+use hvf::MemoryMapping;
 use utils::eventfd::{EventFd, EFD_NONBLOCK};
 use virtio_bindings::{virtio_config::VIRTIO_F_VERSION_1, virtio_ring::VIRTIO_RING_F_EVENT_IDX};
 use vm_memory::{ByteValued, GuestMemoryMmap};
@@ -49,6 +53,8 @@ pub struct Fs {
     passthrough_cfg: passthrough::Config,
     worker_thread: Option<JoinHandle<()>>,
     worker_stopfd: EventFd,
+    #[cfg(target_os = "macos")]
+    map_sender: Option<Sender<MemoryMapping>>,
 }
 
 impl Fs {
@@ -90,6 +96,8 @@ impl Fs {
             passthrough_cfg: fs_cfg,
             worker_thread: None,
             worker_stopfd: EventFd::new(EFD_NONBLOCK).map_err(FsError::EventFd)?,
+            #[cfg(target_os = "macos")]
+            map_sender: None,
         })
     }
 
@@ -112,6 +120,11 @@ impl Fs {
     pub fn set_shm_region(&mut self, shm_region: VirtioShmRegion) {
         self.shm_region = Some(shm_region);
     }
+
+    #[cfg(target_os = "macos")]
+    pub fn set_map_sender(&mut self, map_sender: Sender<MemoryMapping>) {
+        self.map_sender = Some(map_sender);
+    }
 }
 
 impl VirtioDevice for Fs {
@@ -199,8 +212,11 @@ impl VirtioDevice for Fs {
             self.intc.clone(),
             self.irq_line,
             mem.clone(),
+            self.shm_region.clone(),
             self.passthrough_cfg.clone(),
             self.worker_stopfd.try_clone().unwrap(),
+            #[cfg(target_os = "macos")]
+            self.map_sender.clone(),
         );
         self.worker_thread = Some(worker.run());
 

src/devices/src/virtio/fs/filesystem.rs

@@ -2,6 +2,11 @@
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
+#[cfg(target_os = "macos")]
+use crossbeam_channel::Sender;
+#[cfg(target_os = "macos")]
+use hvf::MemoryMapping;
+
 use std::convert::TryInto;
 use std::ffi::{CStr, CString};
 use std::fs::File;
@@ -1121,6 +1126,7 @@ pub trait FileSystem {
         moffset: u64,
         host_shm_base: u64,
         shm_size: u64,
+        #[cfg(target_os = "macos")] map_sender: &Option<Sender<MemoryMapping>>,
     ) -> io::Result<()> {
         Err(io::Error::from_raw_os_error(libc::ENOSYS))
     }
@@ -1131,6 +1137,7 @@ pub trait FileSystem {
         requests: Vec<RemovemappingOne>,
         host_shm_base: u64,
         shm_size: u64,
+        #[cfg(target_os = "macos")] map_sender: &Option<Sender<MemoryMapping>>,
     ) -> io::Result<()> {
         Err(io::Error::from_raw_os_error(libc::ENOSYS))
     }