Re: [PATCH 1/2] scsi: ufs: Add Multi-Circular Queue support

[Asutosh Das (asd)]

Hi Avri
Thanks for taking a look.

On 7/22/2022 12:31 AM, Avri Altman wrote:
> > +static int ufshcd_mcq_config_resource(struct ufs_hba *hba)
> > +{
> > +       struct platform_device *pdev = to_platform_device(hba->dev);
> > +       struct ufshcd_res_info_t *res;
> > +       struct resource *res_mem, *res_mcq;
> > +       int i, ret = 0;
> > +
> > +       memcpy(hba->res, ufshcd_res_info, sizeof(ufshcd_res_info));
> > +
> > +       for (i = 0; i < RES_MAX; i++) {
> > +               res = &hba->res[i];
> > +
> > +               res->resource = platform_get_resource_byname(pdev,
> > +                                                            IORESOURCE_MEM,
> > +                                                            res->name);
> > +               if (!res->resource) {
> > +                       dev_info(hba->dev, "Resource %s not provided\n", res-
> > > name);
> > +                       if (i == RES_MEM)
> > +                               return -ENOMEM;
> > +                       continue;
> > +               } else if (i == RES_MEM) {
> > +                       res_mem = res->resource;
> > +                       res->base = hba->mmio_base;
> > +                       continue;
> > +               }
> > +
> > +               res->base = devm_ioremap_resource(hba->dev, res->resource);
> > +               if (IS_ERR(res->base)) {
> > +                       dev_err(hba->dev, "Failed to map res %s, err = %d\n",
> > +                                        res->name, (int)PTR_ERR(res->base));
> > +                       res->base = NULL;
> > +                       ret = PTR_ERR(res->base);
> > +                       goto out_err;
> > +               }
> > +       }
> > +
> > +       res = &hba->res[RES_MCQ];
> > +       /* MCQ resource provided */
> > +       if (res->base)
> > +               goto out;
> > +
> > +       /* Manually allocate MCQ resource */
> Did you consider to force providing the MCQ configuration?
>
> > +       res_mcq = res->resource;
> > +       res_mcq = devm_kzalloc(hba->dev, sizeof(*res_mcq), GFP_KERNEL);
> > +       if (!res_mcq) {
> > +               dev_err(hba->dev, "Failed to alloate MCQ resource\n");
> > +               goto out_err;
> > +       }
> > +       res->is_alloc = true;
> > +
> > +       res_mcq->start = res_mem->start +
> > +                        mcq_sqattr_offset(hba->mcq_capabilities);
> > +       res_mcq->end = res_mcq->start + 32 * MCQ_QCFG_SIZE - 1;
> Shouldn't there can be MCQCap.MAXQ queues and no more than 32?
Yes correct. Will change it in the next version.
> > +int ufshcd_mcq_init(struct ufs_hba *hba)
> > +{
> > +       struct Scsi_Host *host = hba->host;
> > +       struct ufs_hw_queue *hwq;
> > +       int i, ret = 0;
> > +
> > +       if (!is_mcq_supported(hba))
> > +               return 0;
> > +
> > +       ret = ufshcd_mcq_config_resource(hba);
> > +       if (ret) {
> > +               dev_err(hba->dev, "Failed to config MCQ resource\n");
> > +               return ret;
> > +       }
> > +
> > +       ret = ufshcd_vops_config_mcq_rop(hba);
> > +       if (ret) {
> > +               dev_err(hba->dev, "MCQ Runtime Operation Pointers not
> > configured\n");
> > +               goto out_err;
> > +       }
> > +
> > +       hba->nr_queues[HCTX_TYPE_DEFAULT] = num_possible_cpus();
> > +       hba->nr_queues[HCTX_TYPE_READ] = 0;
> > +       hba->nr_queues[HCTX_TYPE_POLL] = 1;
> > +
> > +       for (i = 0; i < HCTX_MAX_TYPES; i++)
> > +               host->nr_hw_queues += hba->nr_queues[i];
> > +
> > +       host->can_queue = hba->nutrs;
> > +       host->cmd_per_lun = hba->nutrs;
> > +
> > +       /* One more reserved for dev_cmd_queue */
> > +       hba->nr_hw_queues = host->nr_hw_queues + 1;
> Is it possible, since MCQ memory space is *added* to the UTR & UTMR lists,
> That we'll keep using the legacy doorbell for query commands?
> Wouldn't it will simplify the hw_queue bookkeeping
Umm, I didn't understand this suggestion. Please can you elaborate a bit.
When MCQ mode is selected the Config.QT is set to 1.
So how would we keep using the legacy doorbell for query commands?

> > -#define ufshcd_hex_dump(prefix_str, buf, len) do {                       \
> > -       size_t __len = (len);                                            \
> > -       print_hex_dump(KERN_ERR, prefix_str,                             \
> > -                      __len > 4 ? DUMP_PREFIX_OFFSET : DUMP_PREFIX_NONE,\
> > -                      16, 4, buf, __len, false);                        \
> > +#define ufshcd_hex_dump(prefix_str, buf, len) do {                     \
> > +       size_t __len = (len);                                           \
> > +                                                                       \
> > +       print_hex_dump(KERN_ERR, prefix_str,                            \
> > +                      __len > 4 ? DUMP_PREFIX_OFFSET : DUMP_PREFIX_NONE, \
> > +                      16, 4, buf, __len, false);                       \
> > +                                                                       \
> >   } while (0)
> Should this be part of this patch?
No it shouldn't. Will remove this.
> > +#define UFSHCD_MCQ_IO_QUEUE_OFFSET     1
> Maybe add a comment above: "queue 0 is reserved for query commands" or something
> That is if the query commands don't use the  legacy doorbell
>
> > +static inline bool ufshcd_is_hwq_full(struct ufs_hw_queue *q)
> > +{
> > +       return (q->sq_hp_slot == ((q->sq_tp_slot + 1) %
> > +                                     q->max_entries));
> > +}
> Isn't sq_tp_slot is already % q->max_entries ?
This function is unused in this patchset and I will remove it in the 
next version.

> Thanks,
> Avri