rumpk/hal/virtio_net.zig

483 lines
15 KiB
Zig

// SPDX-License-Identifier: LCL-1.0
// Copyright (c) 2026 Markus Maiwald
// Stewardship: Self Sovereign Society Foundation
//
// This file is part of the Nexus Commonwealth.
// See legal/LICENSE_COMMONWEALTH.md for license terms.
//! Rumpk Layer 0: VirtIO-Net Driver (Sovereign Edition)
//!
//! Implements a zero-copy network interface using ION slabs for RX/TX.
//! Supports both Legacy and Modern VirtIO via the universal transport layer.
//!
//! SAFETY: Uses volatile pointers for hardware rings and memory barriers (fences)
//! to ensure correct synchronization with the virtual device.
const std = @import("std");
const uart = @import("uart.zig");
const pci = @import("virtio_pci.zig");
// External Nim functions
extern fn net_ingest_packet(data: [*]const u8, len: usize) bool;
// External C/Zig stubs
extern fn malloc(size: usize) ?*anyopaque;
extern fn ion_alloc_raw(out_id: *u16) u64;
extern fn ion_free_raw(id: u16) void;
extern fn ion_ingress(id: u16, len: u16) void;
extern fn ion_get_virt(id: u16) [*]u8;
extern fn ion_get_phys(id: u16) u64;
extern fn ion_tx_pop(out_id: *u16, out_len: *u16) bool;
var global_driver: ?VirtioNetDriver = null;
var poll_count: u32 = 0;
export fn virtio_net_poll() void {
poll_count += 1;
// Periodic debug: show queue state (SILENCED FOR PRODUCTION)
// if (poll_count == 1 or (poll_count % 1000000 == 0)) {
// if (global_driver) |*d| {
// if (d.rx_queue) |_| {
// asm volatile ("fence" ::: .{ .memory = true });
// uart.print("[VirtIO] Poll #");
// uart.print_hex(poll_count);
// uart.print("\n");
// }
// }
// }
if (global_driver) |*d| {
if (d.rx_queue) |q| {
d.rx_poll(q);
}
if (d.tx_queue) |q| {
d.tx_poll(q);
// Fast Path Egress
var budget: usize = 16;
while (budget > 0) : (budget -= 1) {
var slab_id: u16 = 0;
var len: u16 = 0;
if (ion_tx_pop(&slab_id, &len)) {
d.send_slab(slab_id, len);
} else {
break;
}
}
}
}
}
export fn virtio_net_send(data: [*]const u8, len: usize) void {
uart.print("[VirtIO] virtio_net_send called, len=");
uart.print_hex(len);
uart.print("\n");
if (global_driver) |*d| {
d.send_packet(data, len);
}
}
pub fn init() void {
if (VirtioNetDriver.probe()) |_| {
uart.print("[Rumpk L0] Networking initialized (Sovereign).\n");
}
}
pub const VirtioNetDriver = struct {
transport: pci.VirtioTransport,
irq: u32,
rx_queue: ?*Virtqueue = null,
tx_queue: ?*Virtqueue = null,
pub fn init(base: usize, irq_num: u32) VirtioNetDriver {
return .{
.transport = pci.VirtioTransport.init(base),
.irq = irq_num,
.rx_queue = null,
.tx_queue = null,
};
}
pub fn probe() ?VirtioNetDriver {
uart.print("[VirtIO] Probing PCI for networking device...\n");
const PCI_ECAM_BASE: usize = 0x30000000;
const bus: u8 = 0;
const func: u8 = 0;
var i: u8 = 1;
while (i <= 8) : (i += 1) {
const addr = PCI_ECAM_BASE | (@as(usize, bus) << 20) | (@as(usize, i) << 15) | (@as(usize, func) << 12);
const ptr: *volatile u32 = @ptrFromInt(addr);
const id = ptr.*;
uart.print("[VirtIO-Net] Probing dev ");
uart.print_hex(i);
uart.print(", ID: ");
uart.print_hex(id);
uart.print("\n");
if (id == 0x10001af4 or id == 0x10411af4) {
uart.print("[VirtIO] Found VirtIO-Net device at PCI 00:0");
uart.print_hex(i);
uart.print(".0\n");
var self = VirtioNetDriver.init(addr, 33);
if (self.init_device()) {
return self;
}
}
}
return null;
}
pub fn init_device(self: *VirtioNetDriver) bool {
uart.print("[VirtIO] Initializing device (Sovereign Mode)...\n");
// 1. Probe Capabilities
if (!self.transport.probe()) {
uart.print("[VirtIO] Probe Failed. Aborting.\n");
return false;
}
// 2. Reset Device
uart.print("[VirtIO] Resetting device...\n");
self.transport.reset();
// 3. Acknowledge & Sense Driver
self.transport.add_status(1); // ACKNOWLEDGE
self.transport.add_status(2); // DRIVER
// 4. Feature Negotiation
// 5. Setup RX Queue (0)
self.transport.select_queue(0);
const rx_count = self.transport.get_queue_size();
uart.print("[VirtIO] RX Queue Size: ");
uart.print_hex(rx_count);
uart.print("\n");
if (rx_count == 0 or rx_count == 0xFFFF) {
uart.print("[VirtIO] Invalid RX Queue Size. Aborting.\n");
return false;
}
self.rx_queue = self.setup_queue(0, rx_count) catch {
uart.print("[VirtIO] Failed to setup RX queue\n");
return false;
};
// KICK THE RX QUEUE
uart.print("[VirtIO] Kicking RX Queue to activate...\n");
self.transport.notify(0);
// 6. Setup TX Queue (1)
self.transport.select_queue(1);
const tx_count = self.transport.get_queue_size();
uart.print("[VirtIO] TX Queue Size: ");
uart.print_hex(tx_count);
uart.print("\n");
if (tx_count == 0 or tx_count == 0xFFFF) {
uart.print("[VirtIO] Invalid TX Queue Size. Aborting.\n");
return false;
}
self.tx_queue = self.setup_queue(1, tx_count) catch {
uart.print("[VirtIO] Failed to setup TX queue\n");
return false;
};
// 7. Driver OK
self.transport.add_status(4); // DRIVER_OK
global_driver = self.*;
const end_status = self.transport.get_status();
uart.print("[VirtIO] Initialization Complete. Status: ");
uart.print_hex(end_status);
uart.print("\n");
return true;
}
fn setup_queue(self: *VirtioNetDriver, index: u16, count: u16) !*Virtqueue {
// Layout: [Descriptors (16*N)] [Avail (6 + 2*N)] [Padding] [Used (6 + 8*N)]
const desc_size = 16 * @as(usize, count);
const avail_size = 6 + 2 * @as(usize, count);
const used_offset = (desc_size + avail_size + 4095) & ~@as(usize, 4095);
const used_size = 6 + 8 * @as(usize, count);
const total_size = used_offset + used_size;
const raw_ptr = malloc(total_size + 4096) orelse return error.OutOfMemory;
const aligned_addr = (@intFromPtr(raw_ptr) + 4095) & ~@as(usize, 4095);
const q_ptr_raw = malloc(@sizeOf(Virtqueue)) orelse return error.OutOfMemory;
const q_ptr: *Virtqueue = @ptrCast(@alignCast(q_ptr_raw));
q_ptr.num = count;
q_ptr.index = 0;
q_ptr.desc = @ptrFromInt(aligned_addr);
q_ptr.avail = @ptrFromInt(aligned_addr + desc_size);
q_ptr.used = @ptrFromInt(aligned_addr + used_offset);
// Allocate ID tracking array
const ids_size = @as(usize, count) * @sizeOf(u16);
const ids_ptr = malloc(ids_size) orelse return error.OutOfMemory;
q_ptr.ids = @ptrCast(@alignCast(ids_ptr));
// Pre-allocate buffers for descriptors
const is_rx = (index == 0);
const avail_ring = get_avail_ring(q_ptr.avail);
for (0..count) |i| {
var slab_id: u16 = 0;
var phys_addr: u64 = 0;
if (is_rx) {
// RX: Allocate Initial Slabs
phys_addr = ion_alloc_raw(&slab_id);
if (phys_addr == 0) {
uart.print("[VirtIO] RX ION Alloc Failed. OOM.\n");
return error.OutOfMemory;
}
} else {
// TX: Start empty
phys_addr = 0;
slab_id = 0;
}
q_ptr.desc[i].addr = phys_addr;
q_ptr.desc[i].len = 2048; // Slab Size
q_ptr.desc[i].flags = if (is_rx) @as(u16, 2) else @as(u16, 0); // 2 = VRING_DESC_F_WRITE
q_ptr.desc[i].next = 0;
q_ptr.ids[i] = slab_id;
if (is_rx) {
avail_ring[i] = @intCast(i);
}
}
// Initialize flags to 0 to avoid event suppression
q_ptr.avail.flags = 0;
q_ptr.used.flags = 0;
asm volatile ("fence w, w" ::: .{ .memory = true });
if (is_rx) {
q_ptr.avail.idx = count;
asm volatile ("fence w, w" ::: .{ .memory = true });
}
const phys_addr = aligned_addr;
self.transport.select_queue(index);
if (self.transport.is_modern) {
self.transport.setup_modern_queue(phys_addr, phys_addr + desc_size, phys_addr + used_offset);
} else {
const pfn = @as(u32, @intCast(phys_addr >> 12));
self.transport.setup_legacy_queue(pfn);
}
uart.print("[VirtIO] Queue setup complete\n");
return q_ptr;
}
pub fn rx_poll(self: *VirtioNetDriver, q: *Virtqueue) void {
asm volatile ("fence" ::: .{ .memory = true });
const used = q.used;
const hw_idx = used.idx;
const drv_idx = q.index;
if (hw_idx == drv_idx) {
return;
}
const avail_ring = get_avail_ring(q.avail);
const used_ring = get_used_ring(used);
var replenished: bool = false;
while (q.index != hw_idx) {
// uart.print("[VirtIO RX] Processing Packet...\n");
const elem = used_ring[q.index % q.num];
const desc_idx = elem.id;
const slab_id = q.ids[desc_idx];
const len = elem.len;
// uart.print(" Desc: ");
// uart.print_hex(@intCast(desc_idx));
// uart.print(" Len: ");
// uart.print_hex(len);
// uart.print(" Slab: ");
// uart.print_hex(slab_id);
// uart.print("\n");
const header_len: u32 = 10;
if (len > header_len) {
// Call ION - Pass only the Ethernet Frame (Skip VirtIO Header)
// ion_ingress receives slab_id which contains full buffer.
// We need to tell it the offset.
// Hack: Pass `len - header_len` as the actual Ethernet frame length.
// The NPL must then offset into the buffer by 10 to get to Ethernet.
// OR: We adjust here. Let's adjust here by storing offset.
// Simplest: Pass len directly, NPL will skip first 10 bytes.
ion_ingress(slab_id, @intCast(len - header_len));
} else {
uart.print(" [Warn] Packet too short/empty\n");
ion_free_raw(slab_id);
}
// Replenish
var new_id: u16 = 0;
const new_phys = ion_alloc_raw(&new_id);
if (new_phys != 0) {
q.desc[desc_idx].addr = new_phys;
q.ids[desc_idx] = new_id;
asm volatile ("fence" ::: .{ .memory = true });
avail_ring[q.avail.idx % q.num] = @intCast(desc_idx);
q.avail.idx +%= 1;
replenished = true;
} else {
uart.print(" [Crit] OOM during replenish!\n");
}
q.index +%= 1;
}
if (replenished) {
asm volatile ("fence" ::: .{ .memory = true });
self.transport.notify(0);
}
}
pub fn tx_poll(self: *VirtioNetDriver, q: *Virtqueue) void {
_ = self;
asm volatile ("fence" ::: .{ .memory = true });
const used = q.used;
const used_idx = used.idx;
const used_ring = get_used_ring(used);
while (q.index != used_idx) {
const elem = used_ring[q.index % q.num];
const desc_idx = elem.id;
const slab_id = q.ids[desc_idx];
// Free the TX slab
ion_free_raw(slab_id);
q.index +%= 1;
}
}
pub fn send_slab(self: *VirtioNetDriver, slab_id: u16, len: u16) void {
// Zero-Copy Transmit
const q = self.tx_queue orelse return;
const avail_phase = q.avail.idx;
const avail_ring = get_avail_ring(q.avail);
const idx = avail_phase % q.num;
const phys_addr = ion_get_phys(slab_id);
const desc = &q.desc[idx];
desc.addr = phys_addr;
desc.len = @intCast(len);
desc.flags = 0; // No NEXT, No WRITE
q.ids[idx] = slab_id;
asm volatile ("fence" ::: .{ .memory = true });
avail_ring[idx] = @intCast(idx);
asm volatile ("fence" ::: .{ .memory = true });
q.avail.idx +%= 1;
asm volatile ("fence" ::: .{ .memory = true });
self.transport.notify(1);
uart.print("[VirtIO TX-Slab] Sent ");
uart.print_hex(len);
uart.print(" bytes\n");
}
pub fn send_packet(self: *VirtioNetDriver, data: [*]const u8, len: usize) void {
const q = self.tx_queue orelse return;
const avail_ring = get_avail_ring(q.avail);
var slab_id: u16 = 0;
const phys = ion_alloc_raw(&slab_id);
if (phys == 0) {
uart.print("[VirtIO] TX OOM\n");
return;
}
const buf_ptr = ion_get_virt(slab_id);
const idx = q.avail.idx % q.num;
const desc_idx = idx;
const desc = &q.desc[desc_idx];
q.ids[desc_idx] = slab_id;
const header_len: usize = 10;
@memset(buf_ptr[0..header_len], 0);
const copy_len = if (len > 2000) 2000 else len;
@memcpy(buf_ptr[header_len .. header_len + copy_len], data[0..copy_len]);
desc.addr = phys;
desc.len = @intCast(header_len + copy_len);
desc.flags = 0;
asm volatile ("fence" ::: .{ .memory = true });
avail_ring[idx] = @intCast(desc_idx);
asm volatile ("fence" ::: .{ .memory = true });
q.avail.idx +%= 1;
asm volatile ("fence" ::: .{ .memory = true });
self.transport.notify(1);
uart.print("[VirtIO TX] Queued & Notified Len=");
uart.print_hex(header_len + copy_len);
uart.print("\n");
}
const Virtqueue = struct {
desc: [*]volatile VirtioDesc,
avail: *volatile VirtioAvail,
used: *volatile VirtioUsed,
ids: [*]u16, // Shadow array to track Slab ID per descriptor
num: u16,
index: u16,
};
const VirtioDesc = struct {
addr: u64,
len: u32,
flags: u16,
next: u16,
};
const VirtioAvail = extern struct {
flags: u16,
idx: u16,
};
inline fn get_avail_ring(avail: *volatile VirtioAvail) [*]volatile u16 {
return @ptrFromInt(@intFromPtr(avail) + 4);
}
const VirtioUsed = extern struct {
flags: u16,
idx: u16,
};
inline fn get_used_ring(used: *volatile VirtioUsed) [*]volatile VirtioUsedElem {
return @ptrFromInt(@intFromPtr(used) + 4);
}
const VirtioUsedElem = extern struct {
id: u32,
len: u32,
};
};