As my understanding Below can be the case of using SRS:
If we consider CQI -> This conveys info about how UE is sensing eNodeB's DL power. Similarly, SRS should be conveying how eNodeB can sense the UL power. now eNB has two sources of estimating UL power characteristics -> indirectly by PHR and SRS.
With this I can deduce below usage of SRS (may be):
Based on PHR eNB can decide to enable SRS, Here PHR gives instantaneous path loss info experienced by DL signal. if eNB detects that there path loss is high (above some threshold), It can trigger SRS reports, to get continuous/more granular info about path loss.
SRS ideally this path loss will consider only losses due to fast fading. (Slow fading would have been already compensated by P0). Now eNB can keep track of SRS values (may be again filtered using suitable filter coefficient enough to ignore spurious spikes and get smooth curve) and use these filtered values to input to TPC commands and also decide on modulation sachems.
Further deterioration in SRS values (which cannot be compensated may trigger eNB to configure few additional E-ARFCN's or PCI to measure (which would had not been configured earlier to save UE's power) and report in the view of potential future HO.
Moreover, it can also be input for frequency selective UL scheduling.
Since it uses Zadoff-Chu sequences with cyclic shifts, Even SRS can be used for Timing alignment adjustments (First TA adjustment would have occurred when eNB had received for same UE the RACH preamble sequence which also uses ZC sequences).
I am not 100% sure that above usage are for SRS. But looking the the SRS characteristics, i assume aboveshould be the usage of SRS.