git: af20f017ec0b - main - firewire: add shared helpers for ISO receive drivers

From: Adrian Chadd <adrian_at_FreeBSD.org>
Date: Thu, 25 Jun 2026 02:21:27 UTC
The branch main has been updated by adrian:

URL: https://cgit.FreeBSD.org/src/commit/?id=af20f017ec0b75456d9031059bc79be01f141c3b

commit af20f017ec0b75456d9031059bc79be01f141c3b
Author:     Abdelkader Boudih <freebsd@seuros.com>
AuthorDate: 2026-06-25 02:18:52 +0000
Commit:     Adrian Chadd <adrian@FreeBSD.org>
CommitDate: 2026-06-25 02:18:55 +0000

    firewire: add shared helpers for ISO receive drivers
    
    Add fw_helpers.h with common static inline helpers for FireWire ISO
    receive drivers: async xfer wait with timeout and tlabel cleanup,
    quadlet read/write, and ISO mbuf management.
    
    Reviewed by:    adrian
    Differential Revision:  https://reviews.freebsd.org/D57684
---
 sys/dev/firewire/fw_helpers.h | 290 ++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 290 insertions(+)

diff --git a/sys/dev/firewire/fw_helpers.h b/sys/dev/firewire/fw_helpers.h
new file mode 100644
index 000000000000..863b06441217
--- /dev/null
+++ b/sys/dev/firewire/fw_helpers.h
@@ -0,0 +1,290 @@
+/*
+ * Copyright (c) 2026 Abdelkader Boudih <freebsd@seuros.com>
+ *
+ * SPDX-License-Identifier: BSD-2-Clause
+ */
+
+#ifndef _DEV_FIREWIRE_FW_HELPERS_H_
+#define _DEV_FIREWIRE_FW_HELPERS_H_
+
+static __inline int
+fw_xfer_timeout_cancel(struct fw_xfer *xfer)
+{
+	struct firewire_comm *fc = xfer->fc;
+	struct fw_xfer *txfer;
+	int can_cancel_sent, cancelled, remove_tlabel;
+
+	can_cancel_sent = 0;
+	cancelled = 0;
+	remove_tlabel = 0;
+
+	/*
+	 * FWXF_INQ xfers are still on the software AT queue and can be
+	 * removed safely.  FWXF_SENT xfers have left the controller and are
+	 * waiting only for a split response, so removing their tlabel is enough
+	 * to make a late response miss this xfer.  FWXF_START xfers are still
+	 * owned by the controller descriptor ring and must not be cancelled.
+	 */
+	FW_GLOCK(fc);
+	if ((xfer->flag & FWXF_INQ) != 0) {
+		STAILQ_REMOVE(&xfer->q->q, xfer, fw_xfer, link);
+		xfer->flag &= ~FWXF_INQ;
+		cancelled = 1;
+		remove_tlabel = 1;
+	} else if ((xfer->flag & FWXF_SENT) != 0) {
+		can_cancel_sent = 1;
+	}
+	FW_GUNLOCK(fc);
+
+	if (remove_tlabel || can_cancel_sent) {
+		mtx_lock(&fc->tlabel_lock);
+		if (xfer->tl >= 0) {
+			STAILQ_FOREACH(txfer, &fc->tlabels[xfer->tl], tlabel) {
+				if (txfer == xfer)
+					break;
+			}
+			if (txfer == xfer) {
+				STAILQ_REMOVE(&fc->tlabels[xfer->tl], xfer,
+				    fw_xfer, tlabel);
+				cancelled = 1;
+			}
+			xfer->tl = -1;
+		}
+		mtx_unlock(&fc->tlabel_lock);
+	}
+
+	if (cancelled) {
+		mtx_lock(&fc->wait_lock);
+		xfer->resp = ETIMEDOUT;
+		xfer->flag |= FWXF_WAKE;
+		mtx_unlock(&fc->wait_lock);
+		wakeup(xfer);
+	}
+
+	return (cancelled);
+}
+
+/*
+ * Wait for an async transfer to complete, with timeout.
+ *
+ * If the timeout fires while the controller still owns the transfer, keep the
+ * xfer alive and continue waiting for the core completion path.  Returning
+ * early in that state lets callers free an xfer still referenced by the
+ * controller descriptor ring.
+ */
+static __inline int
+fw_xferwait_timo(struct fw_xfer *xfer, int timo)
+{
+	struct firewire_comm *fc = xfer->fc;
+	struct mtx *lock = &fc->wait_lock;
+	int err = 0;
+	int timedout = 0;
+
+	mtx_lock(lock);
+	while ((xfer->flag & FWXF_WAKE) == 0) {
+		err = msleep(xfer, lock, PWAIT, "fwxfer", timo);
+		if (err == EWOULDBLOCK) {
+			mtx_unlock(lock);
+			if (fw_xfer_timeout_cancel(xfer))
+				return (ETIMEDOUT);
+			timedout = 1;
+			mtx_lock(lock);
+			continue;
+		}
+		if (err) {
+			mtx_unlock(lock);
+			return (err);
+		}
+	}
+	mtx_unlock(lock);
+	if (timedout && xfer->resp == ETIMEDOUT)
+		return (ETIMEDOUT);
+	return (0);
+}
+
+/*
+ * Submit an async request and wait for completion with timeout.
+ */
+static __inline int
+fw_xfer_request_wait(struct firewire_comm *fc, struct fw_xfer *xfer, int timo)
+{
+	int err;
+
+	err = fw_asyreq(fc, -1, xfer);
+	if (err != 0)
+		return (err);
+	return (fw_xferwait_timo(xfer, timo));
+}
+
+static __inline int
+fw_read_quadlet(struct firewire_comm *fc, struct malloc_type *mtype,
+    uint16_t dst, uint8_t spd, uint16_t addr_hi, uint32_t addr_lo,
+    uint32_t *val)
+{
+	struct fw_xfer *xfer;
+	struct fw_pkt *fp;
+	int err;
+
+	xfer = fw_xfer_alloc_buf(mtype, 0, 4);
+	if (xfer == NULL)
+		return (ENOMEM);
+
+	xfer->send.spd = spd;
+	xfer->fc = fc;
+	xfer->hand = fw_xferwake;
+
+	fp = &xfer->send.hdr;
+	fp->mode.rreqq.tcode = FWTCODE_RREQQ;
+	fp->mode.rreqq.dst = dst;
+	fp->mode.rreqq.dest_hi = addr_hi;
+	fp->mode.rreqq.dest_lo = addr_lo;
+
+	err = fw_xfer_request_wait(fc, xfer, 2 * hz);
+	if (err != 0)
+		goto out;
+
+	if (xfer->resp == 0 &&
+	    xfer->recv.hdr.mode.rresq.rtcode == FWRCODE_COMPLETE)
+		*val = ntohl(xfer->recv.hdr.mode.rresq.data);
+	else
+		err = EIO;
+out:
+	fw_xfer_free_buf(xfer);
+	return (err);
+}
+
+static __inline int
+fw_write_quadlet(struct firewire_comm *fc, struct malloc_type *mtype,
+    uint16_t dst, uint8_t spd, uint16_t addr_hi, uint32_t addr_lo,
+    uint32_t val)
+{
+	struct fw_xfer *xfer;
+	struct fw_pkt *fp;
+	int err;
+
+	xfer = fw_xfer_alloc_buf(mtype, 0, 0);
+	if (xfer == NULL)
+		return (ENOMEM);
+
+	xfer->send.spd = spd;
+	xfer->fc = fc;
+	xfer->hand = fw_xferwake;
+
+	fp = &xfer->send.hdr;
+	fp->mode.wreqq.tcode = FWTCODE_WREQQ;
+	fp->mode.wreqq.dst = dst;
+	fp->mode.wreqq.dest_hi = addr_hi;
+	fp->mode.wreqq.dest_lo = addr_lo;
+	fp->mode.wreqq.data = htonl(val);
+
+	err = fw_xfer_request_wait(fc, xfer, 2 * hz);
+	if (err != 0)
+		goto out;
+
+	if (xfer->resp != 0 ||
+	    xfer->recv.hdr.mode.wres.rtcode != FWRCODE_COMPLETE)
+		err = EIO;
+out:
+	fw_xfer_free_buf(xfer);
+	return (err);
+}
+
+static __inline void
+fw_iso_init_chunks(struct fw_xferq *xferq)
+{
+	struct mbuf *m;
+	int i;
+
+	STAILQ_INIT(&xferq->stvalid);
+	STAILQ_INIT(&xferq->stfree);
+	STAILQ_INIT(&xferq->stdma);
+	xferq->stproc = NULL;
+
+	for (i = 0; i < xferq->bnchunk; i++) {
+		m = m_getcl(M_WAITOK, MT_DATA, M_PKTHDR);
+		xferq->bulkxfer[i].mbuf = m;
+		m->m_len = m->m_pkthdr.len = m->m_ext.ext_size;
+		STAILQ_INSERT_TAIL(&xferq->stfree, &xferq->bulkxfer[i], link);
+	}
+}
+
+static __inline void
+fw_iso_free_chunks(struct fw_xferq *xferq, struct malloc_type *mtype)
+{
+	int i;
+
+	for (i = 0; i < xferq->bnchunk; i++) {
+		if (xferq->bulkxfer[i].mbuf != NULL)
+			m_freem(xferq->bulkxfer[i].mbuf);
+	}
+	free(xferq->bulkxfer, mtype);
+	xferq->bulkxfer = NULL;
+}
+
+/*
+ * Dequeue an ISO receive mbuf, replace with a fresh one.
+ * Returns the consumed mbuf, or NULL on error/allocation failure.
+ */
+static __inline struct mbuf *
+fw_iso_dequeue(struct fw_xferq *xferq, struct fw_bulkxfer *sxfer,
+    struct firewire_comm *fc)
+{
+	struct fw_pkt *fp;
+	struct mbuf *m, *m0;
+
+	fp = mtod(sxfer->mbuf, struct fw_pkt *);
+	if (fc->irx_post != NULL)
+		fc->irx_post(fc, fp->mode.ld);
+
+	m = sxfer->mbuf;
+
+	m0 = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
+	if (m0 == NULL) {
+		/* Allocation failed; recycle the original mbuf. */
+		STAILQ_INSERT_TAIL(&xferq->stfree, sxfer, link);
+		return (NULL);
+	}
+
+	m0->m_len = m0->m_pkthdr.len = m0->m_ext.ext_size;
+	sxfer->mbuf = m0;
+	STAILQ_INSERT_TAIL(&xferq->stfree, sxfer, link);
+
+	if (sxfer->resp != 0) {
+		m_freem(m);
+		return (NULL);
+	}
+
+	return (m);
+}
+
+static __inline void
+fw_iso_rearm(struct fw_xferq *xferq, struct firewire_comm *fc, int dma_ch)
+{
+
+	if (STAILQ_FIRST(&xferq->stfree) != NULL && dma_ch >= 0)
+		fc->irx_enable(fc, dma_ch);
+}
+
+static __inline void
+fw_iso_wait_inactive_locked(struct mtx *mtx, int *active, const char *wmesg)
+{
+
+	mtx_assert(mtx, MA_OWNED);
+	while (*active)
+		msleep(active, mtx, PWAIT, wmesg, hz);
+}
+
+static __inline void
+fw_iso_rearm_done(struct fw_xferq *xferq, struct firewire_comm *fc,
+    struct mtx *mtx, int *active, int *cur_dma_ch, int dma_ch)
+{
+
+	mtx_lock(mtx);
+	if (*cur_dma_ch == dma_ch)
+		fw_iso_rearm(xferq, fc, dma_ch);
+	*active = 0;
+	wakeup(active);
+	mtx_unlock(mtx);
+}
+
+#endif /* _DEV_FIREWIRE_FW_HELPERS_H_ */