On 28.04.2022 19:43, Siva Mullati wrote:
>
> On 14/04/22 17:41, Balasubramani Vivekanandan wrote:
> > On 04.04.2022 15:01, Mullati Siva wrote:
> >> From: Siva Mullati <siva.mull...@intel.com>
> >>
> >> Convert CT commands and descriptors to use iosys_map rather
> >> than plain pointer and save it in the intel_guc_ct_buffer struct.
> >> This will help with ct_write and ct_read for cmd send and receive
> >> after the initialization by abstracting the IO vs system memory.
> >>
> >> Signed-off-by: Siva Mullati <siva.mull...@intel.com>
> >> ---
> >> drivers/gpu/drm/i915/gt/uc/intel_guc_ct.c | 200 +++++++++++++---------
> >> drivers/gpu/drm/i915/gt/uc/intel_guc_ct.h | 9 +-
> >> 2 files changed, 127 insertions(+), 82 deletions(-)
> >>
> >> diff --git a/drivers/gpu/drm/i915/gt/uc/intel_guc_ct.c
> >> b/drivers/gpu/drm/i915/gt/uc/intel_guc_ct.c
> >> index f01325cd1b62..64568dc90b05 100644
> >> --- a/drivers/gpu/drm/i915/gt/uc/intel_guc_ct.c
> >> +++ b/drivers/gpu/drm/i915/gt/uc/intel_guc_ct.c
> >> @@ -44,6 +44,11 @@ static inline struct drm_device *ct_to_drm(struct
> >> intel_guc_ct *ct)
> >> #define CT_PROBE_ERROR(_ct, _fmt, ...) \
> >> i915_probe_error(ct_to_i915(ct), "CT: " _fmt, ##__VA_ARGS__)
> >>
> >> +#define ct_desc_read(desc_map_, field_) \
> >> + iosys_map_rd_field(desc_map_, 0, struct guc_ct_buffer_desc, field_)
> >> +#define ct_desc_write(desc_map_, field_, val_) \
> >> + iosys_map_wr_field(desc_map_, 0, struct guc_ct_buffer_desc, field_,
> >> val_)
> >> +
> > Did you try to make the change Lucas mentioned in his comment on rev0,
> > to pass `struct guc_ct_buffer_desc *` as first argument to the above
> > macros? Was it not feasible?
> It is not feasible.
> >> /**
> >> * DOC: CTB Blob
> >> *
> >> @@ -76,6 +81,11 @@ static inline struct drm_device *ct_to_drm(struct
> >> intel_guc_ct *ct)
> >> #define CTB_G2H_BUFFER_SIZE (4 * CTB_H2G_BUFFER_SIZE)
> >> #define G2H_ROOM_BUFFER_SIZE (CTB_G2H_BUFFER_SIZE / 4)
> >>
> >> +#define CTB_SEND_DESC_OFFSET 0u
> >> +#define CTB_RECV_DESC_OFFSET (CTB_DESC_SIZE)
> >> +#define CTB_SEND_CMDS_OFFSET (2 * CTB_DESC_SIZE)
> >> +#define CTB_RECV_CMDS_OFFSET (2 * CTB_DESC_SIZE +
> >> CTB_H2G_BUFFER_SIZE)
> >> +
> >> struct ct_request {
> >> struct list_head link;
> >> u32 fence;
> >> @@ -113,9 +123,9 @@ void intel_guc_ct_init_early(struct intel_guc_ct *ct)
> >> init_waitqueue_head(&ct->wq);
> >> }
> >>
> >> -static void guc_ct_buffer_desc_init(struct guc_ct_buffer_desc *desc)
> >> +static void guc_ct_buffer_desc_init(struct iosys_map *desc)
> >> {
> >> - memset(desc, 0, sizeof(*desc));
> >> + iosys_map_memset(desc, 0, 0, sizeof(struct guc_ct_buffer_desc));
> >> }
> >>
> >> static void guc_ct_buffer_reset(struct intel_guc_ct_buffer *ctb)
> >> @@ -128,17 +138,18 @@ static void guc_ct_buffer_reset(struct
> >> intel_guc_ct_buffer *ctb)
> >> space = CIRC_SPACE(ctb->tail, ctb->head, ctb->size) - ctb->resv_space;
> >> atomic_set(&ctb->space, space);
> >>
> >> - guc_ct_buffer_desc_init(ctb->desc);
> >> + guc_ct_buffer_desc_init(&ctb->desc_map);
> >> }
> >>
> >> static void guc_ct_buffer_init(struct intel_guc_ct_buffer *ctb,
> >> - struct guc_ct_buffer_desc *desc,
> >> - u32 *cmds, u32 size_in_bytes, u32 resv_space)
> >> + struct iosys_map *desc,
> >> + struct iosys_map *cmds,
> >> + u32 size_in_bytes, u32 resv_space)
> >> {
> >> GEM_BUG_ON(size_in_bytes % 4);
> >>
> >> - ctb->desc = desc;
> >> - ctb->cmds = cmds;
> >> + ctb->desc_map = *desc;
> >> + ctb->cmds_map = *cmds;
> >> ctb->size = size_in_bytes / 4;
> >> ctb->resv_space = resv_space / 4;
> >>
> >> @@ -218,12 +229,13 @@ static int ct_register_buffer(struct intel_guc_ct
> >> *ct, bool send,
> >> int intel_guc_ct_init(struct intel_guc_ct *ct)
> >> {
> >> struct intel_guc *guc = ct_to_guc(ct);
> >> - struct guc_ct_buffer_desc *desc;
> >> + struct iosys_map blob_map;
> >> + struct iosys_map desc_map;
> >> + struct iosys_map cmds_map;
> >> u32 blob_size;
> >> u32 cmds_size;
> >> u32 resv_space;
> >> void *blob;
> >> - u32 *cmds;
> >> int err;
> >>
> >> err = i915_inject_probe_error(guc_to_gt(guc)->i915, -ENXIO);
> >> @@ -242,27 +254,35 @@ int intel_guc_ct_init(struct intel_guc_ct *ct)
> >>
> >> CT_DEBUG(ct, "base=%#x size=%u\n", intel_guc_ggtt_offset(guc, ct->vma),
> >> blob_size);
> >>
> >> - /* store pointers to desc and cmds for send ctb */
> >> - desc = blob;
> >> - cmds = blob + 2 * CTB_DESC_SIZE;
> >> + if (i915_gem_object_is_lmem(ct->vma->obj))
> >> + iosys_map_set_vaddr_iomem(&blob_map,
> >> + (void __iomem *)blob);
> >> + else
> >> + iosys_map_set_vaddr(&blob_map, blob);
> >> +
> >> + /* store sysmap to desc_map and cmds_map for send ctb */
> >> + desc_map = IOSYS_MAP_INIT_OFFSET(&blob_map, CTB_SEND_DESC_OFFSET);
> >> + cmds_map = IOSYS_MAP_INIT_OFFSET(&blob_map, CTB_SEND_CMDS_OFFSET);
> >> cmds_size = CTB_H2G_BUFFER_SIZE;
> >> resv_space = 0;
> >> - CT_DEBUG(ct, "%s desc %#tx cmds %#tx size %u/%u\n", "send",
> >> - ptrdiff(desc, blob), ptrdiff(cmds, blob), cmds_size,
> >> - resv_space);
> >> + CT_DEBUG(ct, "%s desc %#x cmds %#x size %u/%u\n", "send",
> >> + CTB_SEND_DESC_OFFSET, (u32)CTB_SEND_CMDS_OFFSET,
> >> + cmds_size, resv_space);
> >>
> >> - guc_ct_buffer_init(&ct->ctbs.send, desc, cmds, cmds_size, resv_space);
> >> + guc_ct_buffer_init(&ct->ctbs.send,
> >> + &desc_map, &cmds_map, cmds_size, resv_space);
> >>
> >> - /* store pointers to desc and cmds for recv ctb */
> >> - desc = blob + CTB_DESC_SIZE;
> >> - cmds = blob + 2 * CTB_DESC_SIZE + CTB_H2G_BUFFER_SIZE;
> >> + /* store sysmap to desc_map and cmds_map for recv ctb */
> >> + desc_map = IOSYS_MAP_INIT_OFFSET(&blob_map, CTB_RECV_DESC_OFFSET);
> >> + cmds_map = IOSYS_MAP_INIT_OFFSET(&blob_map, CTB_RECV_CMDS_OFFSET);
> >> cmds_size = CTB_G2H_BUFFER_SIZE;
> >> resv_space = G2H_ROOM_BUFFER_SIZE;
> >> - CT_DEBUG(ct, "%s desc %#tx cmds %#tx size %u/%u\n", "recv",
> >> - ptrdiff(desc, blob), ptrdiff(cmds, blob), cmds_size,
> >> - resv_space);
> >> + CT_DEBUG(ct, "%s desc %#x cmds %#x size %u/%u\n", "recv",
> >> + (u32)CTB_RECV_DESC_OFFSET, (u32)CTB_RECV_CMDS_OFFSET,
> >> + cmds_size, resv_space);
> >>
> >> - guc_ct_buffer_init(&ct->ctbs.recv, desc, cmds, cmds_size, resv_space);
> >> + guc_ct_buffer_init(&ct->ctbs.recv,
> >> + &desc_map, &cmds_map, cmds_size, resv_space);
> >>
> >> return 0;
> >> }
> >> @@ -279,6 +299,10 @@ void intel_guc_ct_fini(struct intel_guc_ct *ct)
> >>
> >> tasklet_kill(&ct->receive_tasklet);
> >> i915_vma_unpin_and_release(&ct->vma, I915_VMA_RELEASE_MAP);
> >> + iosys_map_clear(&ct->ctbs.send.desc_map);
> >> + iosys_map_clear(&ct->ctbs.send.cmds_map);
> >> + iosys_map_clear(&ct->ctbs.recv.desc_map);
> >> + iosys_map_clear(&ct->ctbs.recv.cmds_map);
> > I think there is no need of clearing the iosys_map when were are zeroing
> > the complete structure area below.
> Yes, I have updated this in latest revision.
> >> memset(ct, 0, sizeof(*ct));
> >> }
> >>
> >> @@ -291,8 +315,8 @@ void intel_guc_ct_fini(struct intel_guc_ct *ct)
> >> int intel_guc_ct_enable(struct intel_guc_ct *ct)
> >> {
> >> struct intel_guc *guc = ct_to_guc(ct);
> >> + struct iosys_map blob_map;
> >> u32 base, desc, cmds, size;
> >> - void *blob;
> >> int err;
> >>
> >> GEM_BUG_ON(ct->enabled);
> >> @@ -302,9 +326,14 @@ int intel_guc_ct_enable(struct intel_guc_ct *ct)
> >> GEM_BUG_ON(!i915_gem_object_has_pinned_pages(ct->vma->obj));
> >> base = intel_guc_ggtt_offset(guc, ct->vma);
> >>
> >> + if (i915_gem_object_is_lmem(ct->vma->obj))
> >> + iosys_map_set_vaddr_iomem(&blob_map, (void __iomem *)
> >> + __px_vaddr(ct->vma->obj));
> >> + else
> >> + iosys_map_set_vaddr(&blob_map, __px_vaddr(ct->vma->obj));
> >> +
> >> /* blob should start with send descriptor */
> >> - blob = __px_vaddr(ct->vma->obj);
> >> - GEM_BUG_ON(blob != ct->ctbs.send.desc);
> >> + GEM_BUG_ON(!iosys_map_is_equal(&blob_map, &ct->ctbs.send.desc_map));
> >>
> >> /* (re)initialize descriptors */
> >> guc_ct_buffer_reset(&ct->ctbs.send);
> >> @@ -314,15 +343,15 @@ int intel_guc_ct_enable(struct intel_guc_ct *ct)
> >> * Register both CT buffers starting with RECV buffer.
> >> * Descriptors are in first half of the blob.
> >> */
> >> - desc = base + ptrdiff(ct->ctbs.recv.desc, blob);
> >> - cmds = base + ptrdiff(ct->ctbs.recv.cmds, blob);
> >> + desc = base + CTB_RECV_DESC_OFFSET;
> >> + cmds = base + CTB_RECV_CMDS_OFFSET;
> >> size = ct->ctbs.recv.size * 4;
> >> err = ct_register_buffer(ct, false, desc, cmds, size);
> >> if (unlikely(err))
> >> goto err_out;
> >>
> >> - desc = base + ptrdiff(ct->ctbs.send.desc, blob);
> >> - cmds = base + ptrdiff(ct->ctbs.send.cmds, blob);
> >> + desc = base + CTB_SEND_DESC_OFFSET;
> >> + cmds = base + CTB_SEND_CMDS_OFFSET;
> >> size = ct->ctbs.send.size * 4;
> >> err = ct_register_buffer(ct, true, desc, cmds, size);
> >> if (unlikely(err))
> >> @@ -371,31 +400,33 @@ static int ct_write(struct intel_guc_ct *ct,
> >> u32 fence, u32 flags)
> >> {
> >> struct intel_guc_ct_buffer *ctb = &ct->ctbs.send;
> >> - struct guc_ct_buffer_desc *desc = ctb->desc;
> >> u32 tail = ctb->tail;
> >> u32 size = ctb->size;
> >> u32 header;
> >> u32 hxg;
> >> u32 type;
> >> - u32 *cmds = ctb->cmds;
> >> + u32 status = ct_desc_read(&ctb->desc_map, status);
> >> unsigned int i;
> >>
> >> - if (unlikely(desc->status))
> >> + if (unlikely(status))
> >> goto corrupted;
> >>
> >> GEM_BUG_ON(tail > size);
> >>
> >> #ifdef CONFIG_DRM_I915_DEBUG_GUC
> >> - if (unlikely(tail != READ_ONCE(desc->tail))) {
> >> + if (unlikely(tail != ct_desc_read(&ctb->desc_map, tail))) {
> >> CT_ERROR(ct, "Tail was modified %u != %u\n",
> >> - desc->tail, tail);
> >> - desc->status |= GUC_CTB_STATUS_MISMATCH;
> >> + ct_desc_read(&ctb->desc_map, tail), tail);
> >> + status |= GUC_CTB_STATUS_MISMATCH;
> >> + ct_desc_write(&ctb->desc_map, status, status);
> >> goto corrupted;
> >> }
> >> - if (unlikely(READ_ONCE(desc->head) >= size)) {
> >> + if (unlikely(ct_desc_read(&ctb->desc_map, head) >= size)) {
> >> CT_ERROR(ct, "Invalid head offset %u >= %u)\n",
> >> - desc->head, size);
> >> - desc->status |= GUC_CTB_STATUS_OVERFLOW;
> >> + ct_desc_read(&ctb->desc_map, head), size);
> > Can we avoid reading the head from ctb->desc_map twice? Since each
> > iosys_map read translates to memcpy, it is better to avoid repeating if
> > we can cache from the earlier read.
> Do you see real need of having a variable to cache value locally, considering
> the path is very unlikely to hit.
At first I thought reading the head twice just affects performance, but
after our discussion, I realized it can potentially mislead debugging.
As it is possible that head can be updated concurrently, we might print
a different value than what was really used in the if condition.
Printing a different value can mislead debugging. So we should print the
same value used in the condition.
Even in the original code, READ_ONCE is used only in the condition not
while printing.
> >> + status = ct_desc_read(&ctb->desc_map, status) |
> >> + GUC_CTB_STATUS_OVERFLOW;
> > Same here, can we avoid reading status since we already have it read. We
> > can do the same like how it is done for the GUC_CTB_STATUS_MISMATCH case.
> Addressed in latest patch
> >> + ct_desc_write(&ctb->desc_map, status, status);
> >> goto corrupted;
> >> }
> >> #endif
> >> @@ -418,14 +449,14 @@ static int ct_write(struct intel_guc_ct *ct,
> >> CT_DEBUG(ct, "writing (tail %u) %*ph %*ph %*ph\n",
> >> tail, 4, &header, 4, &hxg, 4 * (len - 1), &action[1]);
> >>
> >> - cmds[tail] = header;
> >> + iosys_map_wr(&ctb->cmds_map, 4 * tail, u32, header);
> > I prefer not to use the MAGIC number. My suggestion is to use
> > sizeof(tail) since we need the offset in bytes.
> I tried avoiding it, can't use sizeof(tail). Here tail is just an index which
> might be a misleading if we use it in this case, cmds would have been the
> right one but this has been changed to iosys map. Currently updated with
> sizeof(u32).
> >> tail = (tail + 1) % size;
> >>
> >> - cmds[tail] = hxg;
> >> + iosys_map_wr(&ctb->cmds_map, 4 * tail, u32, hxg);
> > MAGIC number again. There are numerous places in the patch, where
> > numbers are used. I prefer to remove it.
> >> tail = (tail + 1) % size;
> >>
> >> for (i = 1; i < len; i++) {
> >> - cmds[tail] = action[i];
> >> + iosys_map_wr(&ctb->cmds_map, 4 * tail, u32, action[i]);
> >> tail = (tail + 1) % size;
> >> }
> >> GEM_BUG_ON(tail > size);
> >> @@ -442,13 +473,14 @@ static int ct_write(struct intel_guc_ct *ct,
> >> atomic_sub(len + GUC_CTB_HDR_LEN, &ctb->space);
> >>
> >> /* now update descriptor */
> >> - WRITE_ONCE(desc->tail, tail);
> >> + ct_desc_write(&ctb->desc_map, tail, tail);
> >>
> >> return 0;
> >>
> >> corrupted:
> >> CT_ERROR(ct, "Corrupted descriptor head=%u tail=%u status=%#x\n",
> >> - desc->head, desc->tail, desc->status);
> >> + ct_desc_read(&ctb->desc_map, head),
> >> ct_desc_read(&ctb->desc_map, tail),
> >> + ct_desc_read(&ctb->desc_map, status));
> >> ctb->broken = true;
> >> return -EPIPE;
> >> }
> >> @@ -499,20 +531,22 @@ static inline bool ct_deadlocked(struct intel_guc_ct
> >> *ct)
> >> bool ret = ktime_ms_delta(ktime_get(), ct->stall_time) > timeout;
> >>
> >> if (unlikely(ret)) {
> >> - struct guc_ct_buffer_desc *send = ct->ctbs.send.desc;
> >> - struct guc_ct_buffer_desc *recv = ct->ctbs.send.desc;
> >> -
> >> CT_ERROR(ct, "Communication stalled for %lld ms, desc
> >> status=%#x,%#x\n",
> >> ktime_ms_delta(ktime_get(), ct->stall_time),
> >> - send->status, recv->status);
> >> + ct_desc_read(&ct->ctbs.send.desc_map, status),
> >> + ct_desc_read(&ct->ctbs.recv.desc_map, status));
> >> CT_ERROR(ct, "H2G Space: %u (Bytes)\n",
> >> atomic_read(&ct->ctbs.send.space) * 4);
> >> - CT_ERROR(ct, "Head: %u (Dwords)\n", ct->ctbs.send.desc->head);
> >> - CT_ERROR(ct, "Tail: %u (Dwords)\n", ct->ctbs.send.desc->tail);
> >> + CT_ERROR(ct, "Head: %u (Dwords)\n",
> >> + ct_desc_read(&ct->ctbs.send.desc_map, head));
> >> + CT_ERROR(ct, "Tail: %u (Dwords)\n",
> >> + ct_desc_read(&ct->ctbs.send.desc_map, tail));
> >> CT_ERROR(ct, "G2H Space: %u (Bytes)\n",
> >> atomic_read(&ct->ctbs.recv.space) * 4);
> >> - CT_ERROR(ct, "Head: %u\n (Dwords)", ct->ctbs.recv.desc->head);
> >> - CT_ERROR(ct, "Tail: %u\n (Dwords)", ct->ctbs.recv.desc->tail);
> >> + CT_ERROR(ct, "Head: %u\n (Dwords)",
> >> + ct_desc_read(&ct->ctbs.recv.desc_map, head));
> >> + CT_ERROR(ct, "Tail: %u\n (Dwords)",
> >> + ct_desc_read(&ct->ctbs.recv.desc_map, tail));
> >>
> >> ct->ctbs.send.broken = true;
> >> }
> >> @@ -549,18 +583,19 @@ static inline void g2h_release_space(struct
> >> intel_guc_ct *ct, u32 g2h_len_dw)
> >> static inline bool h2g_has_room(struct intel_guc_ct *ct, u32 len_dw)
> >> {
> >> struct intel_guc_ct_buffer *ctb = &ct->ctbs.send;
> >> - struct guc_ct_buffer_desc *desc = ctb->desc;
> >> u32 head;
> >> u32 space;
> >> + u32 status = ct_desc_read(&ctb->desc_map, status);
> >>
> >> if (atomic_read(&ctb->space) >= len_dw)
> >> return true;
> >>
> >> - head = READ_ONCE(desc->head);
> >> + head = ct_desc_read(&ctb->desc_map, head);
> >> if (unlikely(head > ctb->size)) {
> >> CT_ERROR(ct, "Invalid head offset %u >= %u)\n",
> >> head, ctb->size);
> >> - desc->status |= GUC_CTB_STATUS_OVERFLOW;
> >> + status |= GUC_CTB_STATUS_OVERFLOW;
> >> + ct_desc_write(&ctb->desc_map, status, status);
> >> ctb->broken = true;
> >> return false;
> >> }
> >> @@ -803,11 +838,10 @@ static void ct_free_msg(struct ct_incoming_msg *msg)
> >> static int ct_read(struct intel_guc_ct *ct, struct ct_incoming_msg **msg)
> >> {
> >> struct intel_guc_ct_buffer *ctb = &ct->ctbs.recv;
> >> - struct guc_ct_buffer_desc *desc = ctb->desc;
> >> u32 head = ctb->head;
> >> - u32 tail = READ_ONCE(desc->tail);
> >> + u32 tail = ct_desc_read(&ctb->desc_map, tail);
> >> u32 size = ctb->size;
> >> - u32 *cmds = ctb->cmds;
> >> + u32 status = ct_desc_read(&ctb->desc_map, status);
> >> s32 available;
> >> unsigned int len;
> >> unsigned int i;
> >> @@ -816,23 +850,26 @@ static int ct_read(struct intel_guc_ct *ct, struct
> >> ct_incoming_msg **msg)
> >> if (unlikely(ctb->broken))
> >> return -EPIPE;
> >>
> >> - if (unlikely(desc->status))
> >> + if (unlikely(status))
> >> goto corrupted;
> >>
> >> GEM_BUG_ON(head > size);
> >>
> >> #ifdef CONFIG_DRM_I915_DEBUG_GUC
> >> - if (unlikely(head != READ_ONCE(desc->head))) {
> >> + if (unlikely(head != ct_desc_read(&ctb->desc_map, head))) {
> >> CT_ERROR(ct, "Head was modified %u != %u\n",
> >> - desc->head, head);
> >> - desc->status |= GUC_CTB_STATUS_MISMATCH;
> >> + ct_desc_read(&ctb->desc_map, head), head);
> > head read twice here as well.
> Same answer.
> >> + status |= GUC_CTB_STATUS_MISMATCH;
> >> + ct_desc_write(&ctb->desc_map, status, status);
> >> goto corrupted;
> >> }
> >> #endif
> >> if (unlikely(tail >= size)) {
> >> CT_ERROR(ct, "Invalid tail offset %u >= %u)\n",
> >> tail, size);
> >> - desc->status |= GUC_CTB_STATUS_OVERFLOW;
> >> + status = ct_desc_read(&ctb->desc_map, status) |
> >> + GUC_CTB_STATUS_OVERFLOW;
> > No need to read the status again as we already have it cached.
> addressed.
> >> + ct_desc_write(&ctb->desc_map, status, status);
> >> goto corrupted;
> >> }
> >>
> >> @@ -849,7 +886,7 @@ static int ct_read(struct intel_guc_ct *ct, struct
> >> ct_incoming_msg **msg)
> >> CT_DEBUG(ct, "available %d (%u:%u:%u)\n", available, head, tail, size);
> >> GEM_BUG_ON(available < 0);
> >>
> >> - header = cmds[head];
> >> + header = iosys_map_rd(&ctb->cmds_map, (4 * head), u32);
> > MAGIC numbers
> >> head = (head + 1) % size;
> >>
> >> /* message len with header */
> >> @@ -857,11 +894,15 @@ static int ct_read(struct intel_guc_ct *ct, struct
> >> ct_incoming_msg **msg)
> >> if (unlikely(len > (u32)available)) {
> >> CT_ERROR(ct, "Incomplete message %*ph %*ph %*ph\n",
> >> 4, &header,
> >> + 4 * (head + available - 1 > size ? size - head :
> >> + available - 1), ((__px_vaddr(ct->vma->obj) +
> >> + CTB_RECV_CMDS_OFFSET) + (4 * head)),
> >> 4 * (head + available - 1 > size ?
> >> - size - head : available - 1), &cmds[head],
> >> - 4 * (head + available - 1 > size ?
> >> - available - 1 - size + head : 0), &cmds[0]);
> >> - desc->status |= GUC_CTB_STATUS_UNDERFLOW;
> >> + available - 1 - size + head : 0),
> >> + (__px_vaddr(ct->vma->obj) +
> >> CTB_RECV_CMDS_OFFSET));
> >> + status = ct_desc_read(&ctb->desc_map, status) |
> >> + GUC_CTB_STATUS_UNDERFLOW;
> > No need to read status again
> >
> > Regards,
> > Bala
> >> + ct_desc_write(&ctb->desc_map, status, status);
> >> goto corrupted;
> >> }
> >>
> >> @@ -869,17 +910,18 @@ static int ct_read(struct intel_guc_ct *ct, struct
> >> ct_incoming_msg **msg)
> >> if (!*msg) {
> >> CT_ERROR(ct, "No memory for message %*ph %*ph %*ph\n",
> >> 4, &header,
> >> + 4 * (head + available - 1 > size ? size - head :
> >> + available - 1), (ctb->cmds_map.vaddr + (4 *
> >> head)),
> >> 4 * (head + available - 1 > size ?
> >> - size - head : available - 1), &cmds[head],
> >> - 4 * (head + available - 1 > size ?
> >> - available - 1 - size + head : 0), &cmds[0]);
> >> + available - 1 - size + head : 0),
> >> ctb->cmds_map.vaddr);
> >> return available;
> >> }
> >>
> >> (*msg)->msg[0] = header;
> >>
> >> for (i = 1; i < len; i++) {
> >> - (*msg)->msg[i] = cmds[head];
> >> + (*msg)->msg[i] = iosys_map_rd(&ctb->cmds_map,
> >> + (4 * head), u32);
> >> head = (head + 1) % size;
> >> }
> >> CT_DEBUG(ct, "received %*ph\n", 4 * len, (*msg)->msg);
> >> @@ -888,13 +930,15 @@ static int ct_read(struct intel_guc_ct *ct, struct
> >> ct_incoming_msg **msg)
> >> ctb->head = head;
> >>
> >> /* now update descriptor */
> >> - WRITE_ONCE(desc->head, head);
> >> + ct_desc_write(&ctb->desc_map, head, head);
> >>
> >> return available - len;
> >>
> >> corrupted:
> >> CT_ERROR(ct, "Corrupted descriptor head=%u tail=%u status=%#x\n",
> >> - desc->head, desc->tail, desc->status);
> >> + ct_desc_read(&ctb->desc_map, head),
> >> + ct_desc_read(&ctb->desc_map, tail),
> >> + ct_desc_read(&ctb->desc_map, status));
> >> ctb->broken = true;
> >> return -EPIPE;
> >> }
> >> @@ -1211,13 +1255,13 @@ void intel_guc_ct_print_info(struct intel_guc_ct
> >> *ct,
> >> drm_printf(p, "H2G Space: %u\n",
> >> atomic_read(&ct->ctbs.send.space) * 4);
> >> drm_printf(p, "Head: %u\n",
> >> - ct->ctbs.send.desc->head);
> >> + ct_desc_read(&ct->ctbs.send.desc_map, head));
> >> drm_printf(p, "Tail: %u\n",
> >> - ct->ctbs.send.desc->tail);
> >> + ct_desc_read(&ct->ctbs.send.desc_map, tail));
> >> drm_printf(p, "G2H Space: %u\n",
> >> atomic_read(&ct->ctbs.recv.space) * 4);
> >> drm_printf(p, "Head: %u\n",
> >> - ct->ctbs.recv.desc->head);
> >> + ct_desc_read(&ct->ctbs.recv.desc_map, head));
> >> drm_printf(p, "Tail: %u\n",
> >> - ct->ctbs.recv.desc->tail);
> >> + ct_desc_read(&ct->ctbs.recv.desc_map, tail));
> >> }
> >> diff --git a/drivers/gpu/drm/i915/gt/uc/intel_guc_ct.h
> >> b/drivers/gpu/drm/i915/gt/uc/intel_guc_ct.h
> >> index f709a19c7e21..867fe13fb47d 100644
> >> --- a/drivers/gpu/drm/i915/gt/uc/intel_guc_ct.h
> >> +++ b/drivers/gpu/drm/i915/gt/uc/intel_guc_ct.h
> >> @@ -7,6 +7,7 @@
> >> #define _INTEL_GUC_CT_H_
> >>
> >> #include <linux/interrupt.h>
> >> +#include <linux/iosys-map.h>
> >> #include <linux/spinlock.h>
> >> #include <linux/workqueue.h>
> >> #include <linux/ktime.h>
> >> @@ -32,8 +33,8 @@ struct drm_printer;
> >> * holds the commands.
> >> *
> >> * @lock: protects access to the commands buffer and buffer descriptor
> >> - * @desc: pointer to the buffer descriptor
> >> - * @cmds: pointer to the commands buffer
> >> + * @desc: iosys map to the buffer descriptor
> >> + * @cmds: iosys map to the commands buffer
> >> * @size: size of the commands buffer in dwords
> >> * @resv_space: reserved space in buffer in dwords
> >> * @head: local shadow copy of head in dwords
> >> @@ -43,8 +44,8 @@ struct drm_printer;
> >> */
> >> struct intel_guc_ct_buffer {
> >> spinlock_t lock;
> >> - struct guc_ct_buffer_desc *desc;
> >> - u32 *cmds;
> >> + struct iosys_map desc_map;
> >> + struct iosys_map cmds_map;
> >> u32 size;
> >> u32 resv_space;
> >> u32 tail;
> >> --
> >> 2.33.0
> >>
