diff mbox series

[01/16] nvme: fix pci doorbell size calculation

Message ID 20200415102445.564803-2-its@irrelevant.dk (mailing list archive)
State New, archived
Headers show
Series nvme: refactoring and cleanups | expand

Commit Message

Klaus Jensen April 15, 2020, 10:24 a.m. UTC
From: Klaus Jensen <k.jensen@samsung.com>

The size of the BAR is 0x1000 (main registers) + 8 bytes for each
queue. Currently, the size of the BAR is calculated like so:

    n->reg_size = pow2ceil(0x1004 + 2 * (n->params.num_queues + 1) * 4);

Since the 'num_queues' parameter already accounts for the admin queue,
this should in any case not need to be incremented by one. Also, the
size should be initialized to (0x1000).

    n->reg_size = pow2ceil(0x1000 + 2 * n->params.num_queues * 4);

This, with the default value of num_queues (64), we will set aside room
for 1 admin queue and 63 I/O queues (4 bytes per doorbell, 2 doorbells
per queue).

Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
---
 hw/block/nvme.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

Comments

Philippe Mathieu-Daudé April 15, 2020, 11:34 a.m. UTC | #1
On 4/15/20 12:24 PM, Klaus Jensen wrote:
> From: Klaus Jensen <k.jensen@samsung.com>
> 
> The size of the BAR is 0x1000 (main registers) + 8 bytes for each
> queue. Currently, the size of the BAR is calculated like so:
> 
>      n->reg_size = pow2ceil(0x1004 + 2 * (n->params.num_queues + 1) * 4);
> 
> Since the 'num_queues' parameter already accounts for the admin queue,
> this should in any case not need to be incremented by one. Also, the
> size should be initialized to (0x1000).
> 
>      n->reg_size = pow2ceil(0x1000 + 2 * n->params.num_queues * 4);
> 
> This, with the default value of num_queues (64), we will set aside room
> for 1 admin queue and 63 I/O queues (4 bytes per doorbell, 2 doorbells
> per queue).

It seems you are right, but I'd feel safer with a Ack-by from Keith.

Do you mind adding definitions such:

#define NVME_IO_SIZE (4 * KiB)
#define DOORBELL_SIZE 4
#define DOORBELL_COUNT 2

Looking at nvme_set_feature() I see better why you introduce 
max_ioqpairs in patch #7:

     case NVME_NUMBER_OF_QUEUES:
         trace_nvme_setfeat_numq((dw11 & 0xFFFF) + 1,
                                 ((dw11 >> 16) & 0xFFFF) + 1,
                                 n->num_queues - 1, n->num_queues - 1);
         req->cqe.result =
             cpu_to_le32((n->num_queues - 2) | ((n->num_queues - 2) << 16));
         break;

Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>

> 
> Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
> ---
>   hw/block/nvme.c | 2 +-
>   1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/hw/block/nvme.c b/hw/block/nvme.c
> index d28335cbf377..77f9c151a665 100644
> --- a/hw/block/nvme.c
> +++ b/hw/block/nvme.c
> @@ -1345,7 +1345,7 @@ static void nvme_realize(PCIDevice *pci_dev, Error **errp)
>       pcie_endpoint_cap_init(pci_dev, 0x80);
>   
>       n->num_namespaces = 1;
> -    n->reg_size = pow2ceil(0x1004 + 2 * (n->num_queues + 1) * 4);
> +    n->reg_size = pow2ceil(0x1000 + 2 * n->num_queues * 4);
>       n->ns_size = bs_size / (uint64_t)n->num_namespaces;
>   
>       n->namespaces = g_new0(NvmeNamespace, n->num_namespaces);
>
diff mbox series

Patch

diff --git a/hw/block/nvme.c b/hw/block/nvme.c
index d28335cbf377..77f9c151a665 100644
--- a/hw/block/nvme.c
+++ b/hw/block/nvme.c
@@ -1345,7 +1345,7 @@  static void nvme_realize(PCIDevice *pci_dev, Error **errp)
     pcie_endpoint_cap_init(pci_dev, 0x80);
 
     n->num_namespaces = 1;
-    n->reg_size = pow2ceil(0x1004 + 2 * (n->num_queues + 1) * 4);
+    n->reg_size = pow2ceil(0x1000 + 2 * n->num_queues * 4);
     n->ns_size = bs_size / (uint64_t)n->num_namespaces;
 
     n->namespaces = g_new0(NvmeNamespace, n->num_namespaces);