Re: [PATCH 11/15] habanalabs: add command submission module

From: Oded Gabbay
Date: Mon Jan 28 2019 - 08:49:47 EST


On Sun, Jan 27, 2019 at 5:11 PM Mike Rapoport <rppt@xxxxxxxxxxxxx> wrote:
>
> On Wed, Jan 23, 2019 at 02:00:53AM +0200, Oded Gabbay wrote:
> > This patch adds the main flow for the user to submit work to the device.
> >
> > Each work is described by a command submission object (CS). The CS contains
> > 3 arrays of command buffers: One for execution, and two for context-switch
> > (store and restore).
> >
> > For each CB, the user specifies on which queue to put that CB. In case of
> > an internal queue, the entry doesn't contain a pointer to the CB but the
> > address in the on-chip memory that the CB resides at.
> >
> > The driver parses some of the CBs to enforce security restrictions.
> >
> > The user receives a sequence number that represents the CS object. The user
> > can then query the driver regarding the status of the CS, using that
> > sequence number.
> >
> > In case the CS doesn't finish before the timeout expires, the driver will
> > perform a soft-reset of the device.
> >
> > Signed-off-by: Oded Gabbay <oded.gabbay@xxxxxxxxx>
> > ---
> > drivers/misc/habanalabs/Makefile | 3 +-
> > drivers/misc/habanalabs/command_submission.c | 787 +++++++++++++
> > drivers/misc/habanalabs/context.c | 52 +-
> > drivers/misc/habanalabs/device.c | 16 +
> > drivers/misc/habanalabs/goya/goya.c | 1082 ++++++++++++++++++
> > drivers/misc/habanalabs/habanalabs.h | 274 +++++
> > drivers/misc/habanalabs/habanalabs_drv.c | 23 +
> > drivers/misc/habanalabs/habanalabs_ioctl.c | 4 +-
> > drivers/misc/habanalabs/hw_queue.c | 250 ++++
> > drivers/misc/habanalabs/memory.c | 200 ++++
> > include/uapi/misc/habanalabs.h | 158 ++-
> > 11 files changed, 2842 insertions(+), 7 deletions(-)
> > create mode 100644 drivers/misc/habanalabs/command_submission.c
> > create mode 100644 drivers/misc/habanalabs/memory.c
> >
> > diff --git a/drivers/misc/habanalabs/Makefile b/drivers/misc/habanalabs/Makefile
> > index b5607233d216..d2fd0e18b1eb 100644
> > --- a/drivers/misc/habanalabs/Makefile
> > +++ b/drivers/misc/habanalabs/Makefile
> > @@ -5,7 +5,8 @@
> > obj-m := habanalabs.o
> >
> > habanalabs-y := habanalabs_drv.o device.o context.o asid.o habanalabs_ioctl.o \
> > - command_buffer.o hw_queue.o irq.o sysfs.o hwmon.o
> > + command_buffer.o hw_queue.o irq.o sysfs.o hwmon.o memory.o \
> > + command_submission.o
> >
> > include $(src)/goya/Makefile
> > habanalabs-y += $(HL_GOYA_FILES)
> > diff --git a/drivers/misc/habanalabs/command_submission.c b/drivers/misc/habanalabs/command_submission.c
> > new file mode 100644
> > index 000000000000..0116c2262f17
> > --- /dev/null
> > +++ b/drivers/misc/habanalabs/command_submission.c
> > @@ -0,0 +1,787 @@
> > +// SPDX-License-Identifier: GPL-2.0
> > +
> > +/*
> > + * Copyright 2016-2018 HabanaLabs, Ltd.
> > + * All Rights Reserved.
> > + */
> > +
> > +#include <uapi/misc/habanalabs.h>
> > +#include "habanalabs.h"
> > +
> > +#include <linux/sched/mm.h>
> > +#include <linux/sched/task.h>
> > +#include <linux/sched/signal.h>
> > +#include <linux/wait.h>
> > +#include <linux/mm.h>
> > +#include <linux/highmem.h>
>
> [ ... ]
>
> > +static void cs_do_release(struct kref *ref)
> > +{
> > + struct hl_cs *cs = container_of(ref, struct hl_cs,
> > + refcount);
> > + struct hl_device *hdev = cs->ctx->hdev;
> > + struct hl_cs_job *job, *tmp;
> > +
> > + cs->completed = true;
> > +
> > + /*
> > + * Although if we reached here it means that all external jobs have
> > + * finished, because each one of them took refcnt to CS, we still
> > + * need to go over the internal jobs and free them. Otherwise, we
> > + * will have leaked memory and what's worse, the CS object (and
> > + * potentially the CTX object) could be released, while the JOB
> > + * still holds a pointer to them (but no reference).
> > + */
> > + list_for_each_entry_safe(job, tmp, &cs->job_list, cs_node)
> > + free_job(hdev, job);
> > +
> > + /* We also need to update CI for internal queues */
> > + if (cs->submitted) {
> > + hl_int_hw_queue_update_ci(cs);
> > +
> > + spin_lock(&hdev->hw_queues_mirror_lock);
> > + /* remove CS from hw_queues mirror list */
> > + list_del_init(&cs->mirror_node);
> > + spin_unlock(&hdev->hw_queues_mirror_lock);
> > +
> > + /*
> > + * Don't cancel TDR in case this CS was timedout because we
> > + * might be running from the TDR context
> > + */
> > + if ((!cs->timedout) &&
> > + (hdev->timeout_jiffies != MAX_SCHEDULE_TIMEOUT)) {
> > + struct hl_cs *next;
> > +
> > + if (cs->tdr_active)
> > + cancel_delayed_work_sync(&cs->work_tdr);
> > +
> > + spin_lock(&hdev->hw_queues_mirror_lock);
> > + /* queue TDR for next CS */
> > + next = list_first_entry_or_null(
> > + &hdev->hw_queues_mirror_list,
> > + struct hl_cs, mirror_node);
> > + if ((next) && (!next->tdr_active)) {
> > + next->tdr_active = true;
> > + schedule_delayed_work(&next->work_tdr,
> > + hdev->timeout_jiffies);
> > + spin_unlock(&hdev->hw_queues_mirror_lock);
> > + } else {
> > + spin_unlock(&hdev->hw_queues_mirror_lock);
> > + }
>
> 'else' can be dropped, just move spin_unlock() outside the 'if'
>
Fixed
> > + }
> > + }
> > +
> > + hl_ctx_put(cs->ctx);
> > +
> > + if (cs->timedout)
> > + dma_fence_set_error(cs->fence, -ETIMEDOUT);
> > + else if (cs->aborted)
> > + dma_fence_set_error(cs->fence, -EIO);
> > +
> > + dma_fence_signal(cs->fence);
> > + dma_fence_put(cs->fence);
> > +
> > + kfree(cs);
> > +}
>
> [ ... ]
>
> > +static int allocate_cs(struct hl_device *hdev, struct hl_ctx *ctx,
> > + struct hl_cs **cs_new)
> > +{
> > + struct hl_dma_fence *fence;
> > + struct dma_fence *other = NULL;
> > + struct hl_cs *cs;
> > + int rc;
> > +
> > + cs = kzalloc(sizeof(*cs), GFP_ATOMIC);
> > + if (!cs)
> > + return -ENOMEM;
>
> Does this ever run from a context that cannot use GFP_KERNEL?
> This applies to other allocations below.
>
It *always* runs from a context that cannot use GFP_KERNEL, because we
mustn't sleep during command submission due to low latency
requirements.

> > +
> > + cs->ctx = ctx;
> > + cs->submitted = false;
> > + cs->completed = false;
> > + INIT_LIST_HEAD(&cs->job_list);
> > + INIT_DELAYED_WORK(&cs->work_tdr, cs_timedout);
> > + kref_init(&cs->refcount);
> > + spin_lock_init(&cs->job_lock);
> > +
> > + fence = kmalloc(sizeof(*fence), GFP_ATOMIC);
>
> kzalloc?
Can't waste time on memset

>
> > + if (!fence) {
> > + rc = -ENOMEM;
> > + goto free_cs;
> > + }
> > +
> > + fence->hdev = hdev;
> > + spin_lock_init(&fence->lock);
> > + cs->fence = &fence->base_fence;
> > +
> > + spin_lock(&ctx->cs_lock);
> > +
> > + fence->cs_seq = ctx->cs_sequence;
> > + other = ctx->cs_pending[fence->cs_seq & (HL_MAX_PENDING_CS - 1)];
> > + if ((other) && (!dma_fence_is_signaled(other))) {
> > + spin_unlock(&ctx->cs_lock);
> > + rc = -EAGAIN;
> > + goto free_fence;
> > + }
> > +
> > + dma_fence_init(&fence->base_fence, &hl_fence_ops, &fence->lock,
> > + ctx->asid, ctx->cs_sequence);
> > +
> > + cs->sequence = fence->cs_seq;
> > +
> > + ctx->cs_pending[fence->cs_seq & (HL_MAX_PENDING_CS - 1)] =
> > + &fence->base_fence;
> > + ctx->cs_sequence++;
> > +
> > + dma_fence_get(&fence->base_fence);
> > +
> > + dma_fence_put(other);
> > +
> > + spin_unlock(&ctx->cs_lock);
> > +
> > + *cs_new = cs;
> > +
> > + return 0;
> > +
> > +free_fence:
> > + kfree(fence);
> > +free_cs:
> > + kfree(cs);
> > + return rc;
> > +}
> > +
>
> [ ... ]
>
> > +
> > +static int goya_validate_cb(struct hl_device *hdev,
> > + struct hl_cs_parser *parser, bool is_mmu)
> > +{
> > + u32 cb_parsed_length = 0;
> > + int rc = 0;
> > +
> > + parser->patched_cb_size = 0;
> > +
> > + /* cb_user_size is more than 0 so loop will always be executed */
> > + while ((cb_parsed_length < parser->user_cb_size) && (!rc)) {
> > + enum packet_id pkt_id;
> > + u16 pkt_size;
> > + void *user_pkt;
> > +
> > + user_pkt = (void *) (parser->user_cb->kernel_address +
> > + cb_parsed_length);
> > +
> > + pkt_id = (enum packet_id) (((*(u64 *) user_pkt) &
> > + PACKET_HEADER_PACKET_ID_MASK) >>
> > + PACKET_HEADER_PACKET_ID_SHIFT);
> > +
> > + pkt_size = goya_packet_sizes[pkt_id];
> > + cb_parsed_length += pkt_size;
> > + if (cb_parsed_length > parser->user_cb_size) {
> > + dev_err(hdev->dev,
> > + "packet 0x%x is out of CB boundary\n", pkt_id);
> > + rc = -EINVAL;
> > + continue;
>
> For me !rc in the while statement was blind. Please consider break here and
>
> if (!rc)
> break;
>
> after the switch
Fixed
>
> > + }
> > +
> > + switch (pkt_id) {
> > + case PACKET_WREG_32:
> > + /*
> > + * Although it is validated after copy in patch_cb(),
> > + * need to validate here as well because patch_cb() is
> > + * not called in MMU path while this function is called
> > + */
> > + rc = goya_validate_wreg32(hdev, parser, user_pkt);
> > + break;
> > +
> > + case PACKET_WREG_BULK:
> > + dev_err(hdev->dev,
> > + "User not allowed to use WREG_BULK\n");
> > + rc = -EPERM;
> > + break;
> > +
> > + case PACKET_MSG_PROT:
> > + dev_err(hdev->dev,
> > + "User not allowed to use MSG_PROT\n");
> > + rc = -EPERM;
> > + break;
> > +
> > + case PACKET_CP_DMA:
> > + dev_err(hdev->dev, "User not allowed to use CP_DMA\n");
> > + rc = -EPERM;
> > + break;
> > +
> > + case PACKET_STOP:
> > + dev_err(hdev->dev, "User not allowed to use STOP\n");
> > + rc = -EPERM;
> > + break;
> > +
> > + case PACKET_LIN_DMA:
> > + if (is_mmu)
> > + rc = goya_validate_dma_pkt_mmu(hdev, parser,
> > + user_pkt);
> > + else
> > + rc = goya_validate_dma_pkt_no_mmu(hdev, parser,
> > + user_pkt);
> > + break;
> > +
> > + case PACKET_MSG_LONG:
> > + case PACKET_MSG_SHORT:
> > + case PACKET_FENCE:
> > + case PACKET_NOP:
> > + parser->patched_cb_size += pkt_size;
> > + break;
> > +
> > + default:
> > + dev_err(hdev->dev, "Invalid packet header 0x%x\n",
> > + pkt_id);
> > + rc = -EINVAL;
> > + break;
> > + }
> > + }
> > +
> > + /*
> > + * The new CB should have space at the end for two MSG_PROT packets:
> > + * 1. A packet that will act as a completion packet
> > + * 2. A packet that will generate MSI-X interrupt
> > + */
> > + parser->patched_cb_size += sizeof(struct packet_msg_prot) * 2;
> > +
> > + return rc;
> > +}
>
> [ ... ]
>
> > +static int goya_patch_cb(struct hl_device *hdev,
> > + struct hl_cs_parser *parser)
> > +{
> > + u32 cb_parsed_length = 0;
> > + u32 cb_patched_cur_length = 0;
> > + int rc = 0;
> > +
> > + /* cb_user_size is more than 0 so loop will always be executed */
> > + while ((cb_parsed_length < parser->user_cb_size) && (!rc)) {
> > + enum packet_id pkt_id;
> > + u16 pkt_size;
> > + u32 new_pkt_size = 0;
> > + void *user_pkt, *kernel_pkt;
> > +
> > + user_pkt = (void *) (parser->user_cb->kernel_address +
> > + cb_parsed_length);
> > + kernel_pkt = (void *) (parser->patched_cb->kernel_address +
> > + cb_patched_cur_length);
> > +
> > + pkt_id = (enum packet_id) (((*(u64 *) user_pkt) &
> > + PACKET_HEADER_PACKET_ID_MASK) >>
> > + PACKET_HEADER_PACKET_ID_SHIFT);
> > +
> > + pkt_size = goya_packet_sizes[pkt_id];
> > + cb_parsed_length += pkt_size;
> > + if (cb_parsed_length > parser->user_cb_size) {
> > + dev_err(hdev->dev,
> > + "packet 0x%x is out of CB boundary\n", pkt_id);
> > + rc = -EINVAL;
> > + continue;
>
> Ditto
Fixed
>
> > + }
> > +
> > + switch (pkt_id) {
> > + case PACKET_LIN_DMA:
> > + rc = goya_patch_dma_packet(hdev, parser, user_pkt,
> > + kernel_pkt, &new_pkt_size);
> > + cb_patched_cur_length += new_pkt_size;
> > + break;
> > +
> > + case PACKET_WREG_32:
> > + memcpy(kernel_pkt, user_pkt, pkt_size);
> > + cb_patched_cur_length += pkt_size;
> > + rc = goya_validate_wreg32(hdev, parser, kernel_pkt);
> > + break;
> > +
> > + case PACKET_WREG_BULK:
> > + dev_err(hdev->dev,
> > + "User not allowed to use WREG_BULK\n");
> > + rc = -EPERM;
> > + break;
> > +
> > + case PACKET_MSG_PROT:
> > + dev_err(hdev->dev,
> > + "User not allowed to use MSG_PROT\n");
> > + rc = -EPERM;
> > + break;
> > +
> > + case PACKET_CP_DMA:
> > + dev_err(hdev->dev, "User not allowed to use CP_DMA\n");
> > + rc = -EPERM;
> > + break;
> > +
> > + case PACKET_STOP:
> > + dev_err(hdev->dev, "User not allowed to use STOP\n");
> > + rc = -EPERM;
> > + break;
> > +
> > + case PACKET_MSG_LONG:
> > + case PACKET_MSG_SHORT:
> > + case PACKET_FENCE:
> > + case PACKET_NOP:
> > + memcpy(kernel_pkt, user_pkt, pkt_size);
> > + cb_patched_cur_length += pkt_size;
> > + break;
> > +
> > + default:
> > + dev_err(hdev->dev, "Invalid packet header 0x%x\n",
> > + pkt_id);
> > + rc = -EINVAL;
> > + break;
> > + }
> > + }
> > +
> > + return rc;
> > +}
>
> [ ... ]
>
> > static void goya_get_axi_name(struct hl_device *hdev, u32 agent_id,
> > u16 event_type, char *axi_name, int len)
> > {
> > @@ -4645,6 +5677,48 @@ static void goya_disable_clock_gating(struct hl_device *hdev)
> >
> > }
> >
> > +static bool goya_is_device_idle(struct hl_device *hdev)
> > +{
> > + u64 offset, dma_qm_reg, tpc_qm_reg, tpc_cmdq_reg, tpc_cfg_reg;
> > + bool val = true;
> > + int i;
> > +
> > + offset = mmDMA_QM_1_GLBL_STS0 - mmDMA_QM_0_GLBL_STS0;
> > +
> > + for (i = 0 ; i < DMA_MAX_NUM ; i++) {
> > + dma_qm_reg = mmDMA_QM_0_GLBL_STS0 + i * offset;
> > +
> > + val = val && ((RREG32(dma_qm_reg) & DMA_QM_IDLE_MASK) ==
> > + DMA_QM_IDLE_MASK);
> > + }
> > +
> > + offset = mmTPC1_QM_GLBL_STS0 - mmTPC0_QM_GLBL_STS0;
> > +
> > + for (i = 0 ; i < TPC_MAX_NUM ; i++) {
> > + tpc_qm_reg = mmTPC0_QM_GLBL_STS0 + i * offset;
> > + tpc_cmdq_reg = mmTPC0_CMDQ_GLBL_STS0 + i * offset;
> > + tpc_cfg_reg = mmTPC0_CFG_STATUS + i * offset;
> > +
> > + val = val && ((RREG32(tpc_qm_reg) & TPC_QM_IDLE_MASK) ==
> > + TPC_QM_IDLE_MASK);
> > + val = val && ((RREG32(tpc_cmdq_reg) & TPC_CMDQ_IDLE_MASK) ==
> > + TPC_CMDQ_IDLE_MASK);
> > + val = val && ((RREG32(tpc_cfg_reg) & TPC_CFG_IDLE_MASK) ==
> > + TPC_CFG_IDLE_MASK);
> > + }
> > +
> > + val = val && ((RREG32(mmMME_QM_GLBL_STS0) & MME_QM_IDLE_MASK) ==
> > + MME_QM_IDLE_MASK);
> > + val = val && ((RREG32(mmMME_CMDQ_GLBL_STS0) & MME_CMDQ_IDLE_MASK) ==
> > + MME_CMDQ_IDLE_MASK);
> > + val = val && ((RREG32(mmMME_ARCH_STATUS) & MME_ARCH_IDLE_MASK) ==
> > + MME_ARCH_IDLE_MASK);
> > + val = val && ((RREG32(mmMME_SHADOW_0_STATUS) & MME_SHADOW_IDLE_MASK) ==
> > + 0);
>
> Huh, these are neat, but IMHO plain
>
> if ((RREG(reg) & mask) != mask)
> return false;
>
> are more readable...
>
Fixed
> > +
> > + return val;
> > +}
> > +
>
> [ ... ]
>
> --
> Sincerely yours,
> Mike.
>