static void
nvme_ns_bio_test(void *arg)
{
struct nvme_io_test_internal *io_test = arg;
struct cdevsw *csw;
struct mtx *mtx;
struct bio *bio;
struct cdev *dev;
void *buf;
struct timeval t;
uint64_t offset;
uint32_t idx, io_completed = 0;
#if __FreeBSD_version >= 900017
int ref;
#endif
buf = malloc(io_test->size, M_NVME, M_WAITOK);
idx = atomic_fetchadd_int(&io_test->td_idx, 1);
dev = io_test->ns->cdev;
offset = idx * 2048 * nvme_ns_get_sector_size(io_test->ns);
while (1) {
bio = g_alloc_bio();
memset(bio, 0, sizeof(*bio));
bio->bio_cmd = (io_test->opc == NVME_OPC_READ) ?
BIO_READ : BIO_WRITE;
bio->bio_done = nvme_ns_bio_test_cb;
bio->bio_dev = dev;
bio->bio_offset = offset;
bio->bio_data = buf;
bio->bio_bcount = io_test->size;
if (io_test->flags & NVME_TEST_FLAG_REFTHREAD) {
#if __FreeBSD_version >= 900017
csw = dev_refthread(dev, &ref);
#else
csw = dev_refthread(dev);
#endif
} else
csw = dev->si_devsw;
mtx = mtx_pool_find(mtxpool_sleep, bio);
mtx_lock(mtx);
(*csw->d_strategy)(bio);
msleep(bio, mtx, PRIBIO, "biotestwait", 0);
mtx_unlock(mtx);
if (io_test->flags & NVME_TEST_FLAG_REFTHREAD) {
#if __FreeBSD_version >= 900017
dev_relthread(dev, ref);
#else
dev_relthread(dev);
#endif
}
if ((bio->bio_flags & BIO_ERROR) || (bio->bio_resid > 0))
break;
g_destroy_bio(bio);
io_completed++;
getmicrouptime(&t);
timevalsub(&t, &io_test->start);
if (t.tv_sec >= io_test->time)
break;
offset += io_test->size;
if ((offset + io_test->size) > nvme_ns_get_size(io_test->ns))
offset = 0;
}
io_test->io_completed[idx] = io_completed;
wakeup_one(io_test);
free(buf, M_NVME);
atomic_subtract_int(&io_test->td_active, 1);
mb();
#if __FreeBSD_version >= 800000
kthread_exit();
#else
kthread_exit(0);
#endif
}
int
physio(struct cdev *dev, struct uio *uio, int ioflag)
{
struct cdevsw *csw;
struct buf *pbuf;
struct bio *bp;
struct vm_page **pages;
caddr_t sa;
u_int iolen, poff;
int error, i, npages, maxpages;
vm_prot_t prot;
csw = dev->si_devsw;
npages = 0;
sa = NULL;
/* check if character device is being destroyed */
if (csw == NULL)
return (ENXIO);
/* XXX: sanity check */
if(dev->si_iosize_max < PAGE_SIZE) {
printf("WARNING: %s si_iosize_max=%d, using DFLTPHYS.\n",
devtoname(dev), dev->si_iosize_max);
dev->si_iosize_max = DFLTPHYS;
}
/*
* If the driver does not want I/O to be split, that means that we
* need to reject any requests that will not fit into one buffer.
*/
if (dev->si_flags & SI_NOSPLIT &&
(uio->uio_resid > dev->si_iosize_max || uio->uio_resid > MAXPHYS ||
uio->uio_iovcnt > 1)) {
/*
* Tell the user why his I/O was rejected.
*/
if (uio->uio_resid > dev->si_iosize_max)
uprintf("%s: request size=%zd > si_iosize_max=%d; "
"cannot split request\n", devtoname(dev),
uio->uio_resid, dev->si_iosize_max);
if (uio->uio_resid > MAXPHYS)
uprintf("%s: request size=%zd > MAXPHYS=%d; "
"cannot split request\n", devtoname(dev),
uio->uio_resid, MAXPHYS);
if (uio->uio_iovcnt > 1)
uprintf("%s: request vectors=%d > 1; "
"cannot split request\n", devtoname(dev),
uio->uio_iovcnt);
return (EFBIG);
}
/*
* Keep the process UPAGES from being swapped. Processes swapped
* out while holding pbufs, used by swapper, may lead to deadlock.
*/
PHOLD(curproc);
bp = g_alloc_bio();
if (uio->uio_segflg != UIO_USERSPACE) {
pbuf = NULL;
pages = NULL;
} else if ((dev->si_flags & SI_UNMAPPED) && unmapped_buf_allowed) {
pbuf = NULL;
maxpages = btoc(MIN(uio->uio_resid, MAXPHYS)) + 1;
pages = malloc(sizeof(*pages) * maxpages, M_DEVBUF, M_WAITOK);
} else {
pbuf = uma_zalloc(pbuf_zone, M_WAITOK);
sa = pbuf->b_data;
maxpages = btoc(MAXPHYS);
pages = pbuf->b_pages;
}
prot = VM_PROT_READ;
if (uio->uio_rw == UIO_READ)
prot |= VM_PROT_WRITE; /* Less backwards than it looks */
error = 0;
for (i = 0; i < uio->uio_iovcnt; i++) {
#ifdef RACCT
if (racct_enable) {
PROC_LOCK(curproc);
if (uio->uio_rw == UIO_READ) {
racct_add_force(curproc, RACCT_READBPS,
uio->uio_iov[i].iov_len);
racct_add_force(curproc, RACCT_READIOPS, 1);
} else {
racct_add_force(curproc, RACCT_WRITEBPS,
uio->uio_iov[i].iov_len);
racct_add_force(curproc, RACCT_WRITEIOPS, 1);
}
PROC_UNLOCK(curproc);
}
#endif /* RACCT */
while (uio->uio_iov[i].iov_len) {
g_reset_bio(bp);
if (uio->uio_rw == UIO_READ) {
bp->bio_cmd = BIO_READ;
curthread->td_ru.ru_inblock++;
} else {
bp->bio_cmd = BIO_WRITE;
curthread->td_ru.ru_oublock++;
}
bp->bio_offset = uio->uio_offset;
bp->bio_data = uio->uio_iov[i].iov_base;
bp->bio_length = uio->uio_iov[i].iov_len;
if (bp->bio_length > dev->si_iosize_max)
bp->bio_length = dev->si_iosize_max;
if (bp->bio_length > MAXPHYS)
bp->bio_length = MAXPHYS;
/*
* Make sure the pbuf can map the request.
* The pbuf has kvasize = MAXPHYS, so a request
* larger than MAXPHYS - PAGE_SIZE must be
* page aligned or it will be fragmented.
*/
poff = (vm_offset_t)bp->bio_data & PAGE_MASK;
if (pbuf && bp->bio_length + poff > pbuf->b_kvasize) {
if (dev->si_flags & SI_NOSPLIT) {
uprintf("%s: request ptr %p is not "
"on a page boundary; cannot split "
"request\n", devtoname(dev),
bp->bio_data);
error = EFBIG;
goto doerror;
}
bp->bio_length = pbuf->b_kvasize;
if (poff != 0)
bp->bio_length -= PAGE_SIZE;
}
bp->bio_bcount = bp->bio_length;
bp->bio_dev = dev;
if (pages) {
if ((npages = vm_fault_quick_hold_pages(
&curproc->p_vmspace->vm_map,
(vm_offset_t)bp->bio_data, bp->bio_length,
prot, pages, maxpages)) < 0) {
error = EFAULT;
goto doerror;
}
if (pbuf && sa) {
pmap_qenter((vm_offset_t)sa,
pages, npages);
bp->bio_data = sa + poff;
} else {
bp->bio_ma = pages;
bp->bio_ma_n = npages;
bp->bio_ma_offset = poff;
bp->bio_data = unmapped_buf;
bp->bio_flags |= BIO_UNMAPPED;
}
}
csw->d_strategy(bp);
if (uio->uio_rw == UIO_READ)
biowait(bp, "physrd");
else
biowait(bp, "physwr");
if (pages) {
if (pbuf)
pmap_qremove((vm_offset_t)sa, npages);
vm_page_unhold_pages(pages, npages);
}
iolen = bp->bio_length - bp->bio_resid;
if (iolen == 0 && !(bp->bio_flags & BIO_ERROR))
goto doerror; /* EOF */
uio->uio_iov[i].iov_len -= iolen;
uio->uio_iov[i].iov_base =
(char *)uio->uio_iov[i].iov_base + iolen;
uio->uio_resid -= iolen;
uio->uio_offset += iolen;
if (bp->bio_flags & BIO_ERROR) {
error = bp->bio_error;
goto doerror;
}
}
}
doerror:
if (pbuf)
uma_zfree(pbuf_zone, pbuf);
else if (pages)
free(pages, M_DEVBUF);
g_destroy_bio(bp);
PRELE(curproc);
return (error);
}
