Fix for gfx1151 (Strix Halo) ROCm "out of memory" abort on hipStreamCreate

libfix_strix_cwsr.so

 /*
     * Fix for gfx1151 (Strix Halo) ROCm "out of memory" abort on hipStreamCreate:
     *
     * Issue: amdkfd enforces ctx_save_restore_area_size >= topo cwsr_size (0x124c000 / 19.2 MB).
     *        hsakmt hardcodes 0xd4c000 (13.9 MB), causing AMDKFD_IOC_CREATE_QUEUE to return -EINVAL,
     *        which ROCr surfaces to HIP as hipErrorOutOfMemory.
     *
     * Solution: Intercept mmap/ioctl to expand CWSR buffer sizes to match kernel requirements.
     *
     * Build: gcc -shared -fPIC -O3 fix_strix_cwsr.c -o libfix_strix_cwsr.so -ldl
     * Usage: LD_PRELOAD=./libfix_strix_cwsr.so HIP_LAUNCH_BLOCKING=1 llama-server ...
     */
  
    #define _GNU_SOURCE
    #include <stdio.h>
    #include <stdarg.h>
    #include <dlfcn.h>
    #include <sys/ioctl.h>
    #include <sys/mman.h>
    #include <stdint.h>
    #include <linux/kfd_ioctl.h>
  
    #define CWSR_KERNEL_SIZE 0x124c000 // 19,185,664 bytes (cwsr_size on gfx1151)
    #define CWSR_ALLOC_SIZE  0x1800000 // 24 MB (2MB-aligned buffer)
    #define CWSR_GUARD_SIZE  0x2000000 // 32 MB guard area
  
    static int (*real_ioctl)(int, unsigned long, ...) = NULL;
    static void* (*real_mmap)(void *, size_t, int, int, int, off_t) = NULL;
  
    void* mmap(void *addr, size_t length, int prot, int flags, int fd, off_t offset) {
        if (!real_mmap) real_mmap = dlsym(RTLD_NEXT, "mmap");
        // Expand hsakmt CWSR buffer reservations
        if (length == 0xf55000) length = CWSR_GUARD_SIZE;
        else if (length == 0xd56000) length = CWSR_ALLOC_SIZE;
        return real_mmap(addr, length, prot, flags, fd, offset);
    }
  
    int ioctl(int fd, unsigned long request, ...) {
        if (!real_ioctl) real_ioctl = dlsym(RTLD_NEXT, "ioctl");
        va_list args;
        va_start(args, request);
        void *argp = va_arg(args, void *);
        va_end(args);
  
        unsigned int nr = request & 0xff;
        if (nr == 0x20) { // AMDKFD_IOC_SVM: expand buffer allocation
            struct kfd_ioctl_svm_args *s = (struct kfd_ioctl_svm_args *)argp;
            if (s->size == 0xd56000) s->size = CWSR_ALLOC_SIZE;
        } else if (nr == 0x02) { // AMDKFD_IOC_CREATE_QUEUE: override undersized CWSR area
            struct kfd_ioctl_create_queue_args *q = (struct kfd_ioctl_create_queue_args *)argp;
            if (q->ctx_save_restore_size == 0xd4c000) q->ctx_save_restore_size = CWSR_KERNEL_SIZE;
        }
  
        return real_ioctl(fd, request, argp);
    }