[Info-vax] VMS Software Q1 '23 Update

[Arne Vajhøj]

On 2/3/2023 6:19 PM, Dave Froble wrote:
> On 2/3/2023 1:57 PM, Arne Vajhøj wrote:
>> My understanding of the problem based on the PPT linked to
>> in previous discussions then the core of the problem
>> is that the API between process code and kernel code
>> need to be byte centric not block centric.
>>
>> Both SYS$QIO(W) and SYS$IO_PERFORM(W) specify
>> block number not byte number.
>>
>> Which can cause concurrency problems for IO's with
>> no overlapping byte range but overlapping blocks.
> 
> The simple solution to that is to determine the starting block and 
> ending block, then lock the entire range of blocks.
> 
> I'm betting that that happens on any OS, since disk devices are block 
> oriented.

I don't think it matters that disks are accessed by block.

The problem is a user mode buffer per process vs a
single kernel mode buffer. As long as a single
kernel mode buffer get updated then it does not
matter how it gets written back to disk.

(as I read the before mentioned PPT)

Arne