Re: [PATCH 7/9] util/vfio-helpers: Have qemu_vfio_dma_map() propagate Error

["Philippe Mathieu-Daudé" <philmd@redhat.com>]

On 8/25/21 1:34 PM, Klaus Jensen wrote:
> On Aug 24 16:11, Philippe Mathieu-Daudé wrote:
>> Now that all qemu_vfio_dma_map() callers provide an Error* argument,
>> fill it with relevant / more descriptive message. Reduce 'ret'
>> (returned value) scope by returning errno directly to simplify
>> (removing the goto / 'out' label).
>>
>> Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com>
>> ---
>>  block/nvme.c        |  1 +
>>  util/vfio-helpers.c | 31 ++++++++++++++-----------------
>>  2 files changed, 15 insertions(+), 17 deletions(-)
>>
>> diff --git a/block/nvme.c b/block/nvme.c
>> index 663e5d918fa..80546b0babd 100644
>> --- a/block/nvme.c
>> +++ b/block/nvme.c
>> @@ -240,6 +240,7 @@ static NVMeQueuePair *nvme_create_queue_pair(BDRVNVMeState *s,
>>      r = qemu_vfio_dma_map(s->vfio, q->prp_list_pages, bytes,
>>                            false, &prp_list_iova, errp);
>>      if (r) {
>> +        error_prepend(errp, "Cannot map buffer for DMA: ");
> 
> Ah. Here is the missing error message that I noticed in patch 2 ;)

So I should use error_setg() in patch 2 and replace it with
error_prepend() here? I'm trying to address Eric's comments
from:

https://lore.kernel.org/qemu-devel/2cbd5471-4611-ae6b-d79f-db6ff19db5bf@redhat.com/
https://lore.kernel.org/qemu-devel/bd1cc1c9-3553-d252-3ca9-a23bc1035170@redhat.com/

> 
>>          goto fail;
>>      }
>>      q->free_req_head = -1;
>> diff --git a/util/vfio-helpers.c b/util/vfio-helpers.c
>> index 3e1a49bea15..f4c16e1dae5 100644
>> --- a/util/vfio-helpers.c
>> +++ b/util/vfio-helpers.c
>> @@ -729,7 +729,6 @@ qemu_vfio_find_temp_iova(QEMUVFIOState *s, size_t size, uint64_t *iova)
>>  int qemu_vfio_dma_map(QEMUVFIOState *s, void *host, size_t size,
>>                        bool temporary, uint64_t *iova, Error **errp)
>>  {
>> -    int ret = 0;
>>      int index;
>>      IOVAMapping *mapping;
>>      uint64_t iova0;
>> @@ -742,32 +741,34 @@ int qemu_vfio_dma_map(QEMUVFIOState *s, void *host, size_t size,
>>      if (mapping) {
>>          iova0 = mapping->iova + ((uint8_t *)host - (uint8_t *)mapping->host);
>>      } else {
>> +        int ret;
>> +
>>          if (s->high_water_mark - s->low_water_mark + 1 < size) {
>> -            ret = -ENOMEM;
>> -            goto out;
>> +            error_setg(errp, "iova exhausted (water mark reached)");
>> +            return -ENOMEM;
>>          }
>>          if (!temporary) {
>> -            if (qemu_vfio_find_fixed_iova(s, size, &iova0)) {
>> -                ret = -ENOMEM;
>> -                goto out;
>> +            if (qemu_vfio_find_fixed_iova(s, size, &iova0) < 0) {
>> +                error_setg(errp, "iova range not found");
>> +                return -ENOMEM;
> 
> Just curious.
> 
> Previously this did error_setg_errno in out. Why don't we want to do that here?

I didn't thought about it but indeed you are right.

Thanks for the review!

Phil.