Re: [U-Boot] cadence_qspi_apb: cache issues on mach-socfpga
Rush, Jason A wrote;:
I don't believe the patchset I submitted for DT bindings were merged in.
I can confirm that. I'd strongly vote for them to get in as cadence_qspi is otherwise not usable on mach socfpga.
How can I ensure a tested-by from me gets related to that patch set? I can't reply to it as I was not subscribed to the list at that time.
Regarding the bounce buffer change: isn't it wrong to use the generic bounce buffer here? Given the dcache calls in it, it seems like this one is for DMA but we're doing cpu copy in the cadence qspi driver.
I tend to agree for the socfpga architecture; however, I'm not very familiar with the bounce buffer and dcache. From what I recall, it looked like the data is copied from the QSPI by the CPU on the socfpga so the data itself might be in the dcache? Then the call to bounce_buffer_stop invalidates the dcache, so the QSPI data that was there is now invalidated, and whatever random data in higher level memory is copied over the QSPI data.
Right. That's what my results were, too. I only discovered this after starting this thread and sent a patch later today
Simon
On 11/15/2017 05:15 PM, Goldschmidt Simon wrote:
Rush, Jason A wrote;:
I don't believe the patchset I submitted for DT bindings were merged in.
I can confirm that. I'd strongly vote for them to get in as cadence_qspi is otherwise not usable on mach socfpga.
How can I ensure a tested-by from me gets related to that patch set? I can't reply to it as I was not subscribed to the list at that time.
You should rebase that patch anyway and then resend it.
Regarding the bounce buffer change: isn't it wrong to use the generic bounce buffer here? Given the dcache calls in it, it seems like this one is for DMA but we're doing cpu copy in the cadence qspi driver.
I tend to agree for the socfpga architecture; however, I'm not very familiar with the bounce buffer and dcache. From what I recall, it looked like the data is copied from the QSPI by the CPU on the socfpga so the data itself might be in the dcache? Then the call to bounce_buffer_stop invalidates the dcache, so the QSPI data that was there is now invalidated, and whatever random data in higher level memory is copied over the QSPI data.
Right. That's what my results were, too. I only discovered this after starting this thread and sent a patch later today
Simon
participants (2)
-
Goldschmidt Simon -
Marek Vasut