9d4cdb44fc
fastfetch reported 3 seconds of uptime after 15 real ones, `time` claimed
3.12s for a 20-second command, and ping's one-second interval was closer to
six. One cause, not three: `ticks` was advancing at ~14.5 Hz instead of 100.
It was not the timer. virtio-gpu-pci shares IRQ 11 with the e1000 and
registered no interrupt handler at all. PCI INTx is level-triggered, so the
device held the line asserted, the kernel signalled EOI, and it fired again
immediately -- 72,000 interrupts a second, 83% of CPU0, starving the tick.
The e1000's handler read ICR as zero half a million times, correctly
answering "not mine", which is what made it look like a network-free problem.
Reading virtio's ISR byte clears it and drops the line. That is the fix.
irq_dispatch now measures each line's rate against the TSC -- not ticks,
which is the first casualty of a storm -- and names the attached handlers
once above 20000/s. It reproduces the original fault as a single line.
Found along the way, and fixed because they were real:
- lapicinit loaded TICR with xv6's uncalibrated 10000000, whose own comment
admitted it was never measured. calibrate_timers() now measures the LAPIC
and the TSC against the 8254 PIT. (It came back at 999.9 MHz, so this was
not the bug -- but the next machine is not obliged to be 1 GHz.)
- microdelay() was an empty function body. Every microdelay(1000) in the
xHCI reset path and the USB hub power-on sequence returned instantly.
- HZ is a constant in param.h now, not the literal 100 written out at a
dozen ticks-to-seconds conversions. USER_HZ is separate and deliberately
does not follow it: musl answers sysconf(_SC_CLK_TCK) with 100 always, so
times() and /proc/stat scale at the syscall boundary the way Linux does.
- TIOCGWINSZ answered a hardcoded 24x80. gfx_cols()/gfx_rows() have known
the real geometry all along; at 1024x768 the console is 128x48, so every
program that lays out a screen was formatting for the wrong one.
- TCFLSH fell through to the catch-all `return 0`, claiming to discard
input without discarding any.
uptime now advances 5.01 seconds per 5 real seconds. schfs-check passes.
Full investigation, including the two wrong hypotheses and the measurement
flaw that protected them, in docs/history/kernel.md section 13.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VQ7Qx8E1q1M1bbCq8QLHS2
67 lines
3.5 KiB
C
67 lines
3.5 KiB
C
#pragma once
|
|
|
|
// The timer interrupt rate the kernel is built around. Everything that turns
|
|
// ticks into a duration divides by this. It used to be the literal 100 written
|
|
// out at each site, which made it look like a fact about the hardware; it is
|
|
// not. It is a target, and kernel/arch/x86_64/lapic.c calibrates the LAPIC
|
|
// timer against the PIT to actually hit it.
|
|
#define HZ 100 // timer interrupts per second
|
|
|
|
// The tick rate the *user-space ABI* is defined in, which is a different thing
|
|
// from HZ and must not follow it. glibc and musl both answer sysconf(_SC_CLK_TCK)
|
|
// with a hardcoded 100, so times() and /proc/stat are read as hundredths of a
|
|
// second no matter what the kernel's internal rate is. Linux calls this USER_HZ
|
|
// and scales at the syscall boundary; so does SchOS, in the two places below.
|
|
#define USER_HZ 100
|
|
|
|
// Converting a POSIX timespec/timeval into ticks. These used to appear as the
|
|
// bare literals 10000000 and 10000 at each call site, where they looked like
|
|
// magic numbers rather than what they are: the length of one tick, expressed
|
|
// in the units the caller happens to be using.
|
|
#define NSEC_PER_TICK (1000000000 / HZ)
|
|
#define USEC_PER_TICK (1000000 / HZ)
|
|
#define MSEC_PER_TICK (1000 / HZ)
|
|
|
|
#define NPROC 128 // maximum number of processes
|
|
#define KSTACKSIZE 4096 // size of per-process kernel stack (increased for x64)
|
|
#define KSTACK_CANARY 0xDEADC0DEDEADC0DEULL // sentinel at kstack[0], see kstack_check_canary()
|
|
#define NCPU 16 // maximum number of CPUs
|
|
#define NOFILE 128 // open files per process
|
|
#define NFILE 512 // open files per system
|
|
#define NINODE 200 // maximum number of active i-nodes
|
|
#define NDEV 10 // maximum major device number
|
|
#define RAMDISK_DEV 2 // memory-backed block device (kernel/dev/ramdisk.c)
|
|
#define USBDISK_DEV 3 // USB mass storage (kernel/drivers/usb_storage.c)
|
|
|
|
// Device the root filesystem is mounted from. Building with RAMROOT=1 defines
|
|
// RAMDISK_ROOT and boots from the SCHFS image linked into the kernel instead
|
|
// of the IDE disk, leaving the physical disk free to be reformatted from a
|
|
// running system. See kernel/dev/ramdisk.c.
|
|
#ifdef RAMDISK_ROOT
|
|
#define ROOTDEV RAMDISK_DEV
|
|
#else
|
|
#define ROOTDEV 1 // device number of file system root disk
|
|
#endif
|
|
#define MAXARG 32 // max exec arguments
|
|
#define MAXPATH 128 // max path
|
|
#define MAXOPBLOCKS 30 // max # of blocks any FS op writes
|
|
#define LOGSIZE (MAXOPBLOCKS*3) // max data blocks in on-disk log
|
|
// Disk block cache, in blocks. 256 * BSIZE = 1 MB.
|
|
//
|
|
// Was MAXOPBLOCKS*3 = 90 blocks, and at the old 1 KB BSIZE that is a 90 KB
|
|
// cache -- smaller than the toybox binary that every single command execs. So
|
|
// nothing was ever still resident on the next command and each one paid the
|
|
// full read again. 1 MB holds toybox with room to spare, which is what makes
|
|
// the *second* command fast rather than just the first.
|
|
//
|
|
// Sized rather than maximised: these are statically allocated in bio.c, so
|
|
// this is memory the system always spends. 1 MB is affordable on a machine
|
|
// with tens of megabytes; 8 MB would not be.
|
|
#define NBUF 256
|
|
#define FSSIZE 150000 // size of file system in blocks
|
|
|
|
// Network interfaces the stack will attach. Two is already an ordinary machine
|
|
// -- a laptop with Ethernet plus a virtual adapter -- and each one costs a
|
|
// struct netif, so the limit is about keeping a fixed array honest rather than
|
|
// about memory.
|
|
#define NETIF_MAX 4
|