Re: [Patch v7 1/8] memory: tegra: add interconnect support for DRAM scaling in Tegra234

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 





On 09/05/23 19:06, Krzysztof Kozlowski wrote:
External email: Use caution opening links or attachments


On 09/05/2023 15:17, Sumit Gupta wrote:
+                     /*
+                      * MC driver probe can't get BPMP reference as
it gets probed
+                      * earlier than BPMP. So, save the BPMP ref got
from the EMC
+                      * DT node in the mc->bpmp and use it in MC's
icc_set hook.
+                      */
+                     mc->bpmp = emc->bpmp;

This (and ()) are called without any locking. You register first the
interconnect, so set() callback can be used, right? Then set() could be
called anytime between tegra_emc_interconnect_init() and assignment
above. How do you synchronize these?

Best regards,
Krzysztof


Currently, the tegra234_mc_icc_set() has NULL check. So, it will give
this error.
    if (!mc->bpmp) {

How does it solve concurrent accesses and re-ordering of instructions by
compiler or CPU?


Now, the "mc->bpmp" is set before tegra_emc_interconnect_init().
So, until the EMC interconnect initializes, set() won't be
called as the devm_of_icc_get() call will fail.

What if compiler puts "mc->bpmp" assignment after
tegra_emc_interconnect_init()?

What if CPU executes above assignment also after
tegra_emc_interconnect_init()?

Considering amount of code inside tegra_emc_interconnect_init() second
case is rather unlikely, but first possible, right?

Best regards,
Krzysztof


Yes, for protection from compiler re-ordering I can add
barrier() after the assignment.

  mc->bpmp = emc->bpmp;
  barrier();

Thank you,
Sumit Gupta



[Index of Archives]     [DMA Engine]     [Linux Coverity]     [Linux USB]     [Video for Linux]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]     [Greybus]

  Powered by Linux