Operating Systems

Question 1

How many Data-TLB entries for 4KB pages in Intel icelake x86_64? What about for 2MB? For 1GB?

Same question for Instr-TLB.

Answer 1

Data TLB
- 4KB : 64
- 2MB : 32
- 1GB : 4

Instr TLB
- 4KB : 128
- 2MB : 8
- 1GB : 8

Question 2

What does 4-way, 8-way, 16-way, fully-associative TLB signify?

Trade-offs with more/less “ways”?

Answer 2

• gives the number of sets
• set index is pulled from VPN
• everything in the set is searched in parallel

Higher (e.g., 8-way, 16-way):
- Fewer collisions, better hit rate
- More complex, slightly slower
Lower (e.g., 2-way, 4-way)
- Faster lookup
- More collisions (conflict misses)

Question 3

What is STLB in Intel Ice Lake?

Answer 3

L2 TLB, larger, slower

Question 4

fetch time for L1 TLB vs L2 TLB

Answer 4

0.5-1 cycle vs 5-10 cycles

Question 5

What are some free space management policies?
When/How to use them?

Answer 5

• free space management policies are used in memory allocators like in malloc, or address sanitizers
• api is malloc(int size), free(void* ptr)
• we use splitting and coalescing
• The Header block is used to store the size of an allocated memory block

Question 6

How can you get infinite TLB misses? How to prevent that?

Answer 6

• When the TLB miss handler’s pages are not in TLB you get infinite…
• The translations for TLB-miss handler are pinned in the TLB

Question 7

Metadata in a TLB entry

Answer 7

• Address space identifier / Global identifier
• valid
• dirty
• caching info

Question 8

Why is TLB so fast

Answer 8

• spatial etc etc
• but looking for “parallel search”

Question 9

What are inverted page tables

Answer 9

mapping from physical page frame to virtual page, instead of the other way around

but who cares, not how intel x86_64 does it

Question 10

What is the difference between swap space and page cache

Answer 10

• swap space is for anonymous memory
• page cache is for file backed pages

Question 11

Give an example of when LRU misbehaves

Answer 11

• the memory region you’re referencing is one more than the cache size in a loop, so every access is a miss

Question 12

in linux when does a swap daemon run? what is a swap daemon, what does it do, why is it needed?

Answer 12

• kswapd: kernel swap daemon
• runs when the memory is under pressure
• kswapd picks a page using LRU, avoiding active/pinned ones.
• Dirty CPU cache lines for that page are flushed to RAM by hardware (cache coherence).
• TLB entry is invalidated, and the page is marked non-present in the page tables.
• Page is written to swap (if dirty) and removed from physical memory.
• Future access triggers a page fault → kernel reloads it from swap, updates page tables & TLB.

Question 13

How does linux’s page cache work

Answer 13

• 2Q policy