On Wed, Nov 25, 2020 at 10:31:55AM -0500, Celejar wrote: > On Wed, 25 Nov 2020 17:43:00 +0300 > Reco <[email protected]> wrote: > > > On Wed, Nov 25, 2020 at 09:19:08AM -0500, Celejar wrote: > > > On Wed, 25 Nov 2020 10:42:59 +0300 > > > Reco <[email protected]> wrote: > > > > > > > On Tue, Nov 24, 2020 at 03:06:59PM -0800, [email protected] wrote: > > > > > This is the only oddity evident after ifup wlxa0f3c10a28f7. > > > > > > > > > > DHCPDISCOVER on wlxa0f3c10a28f7 to 255.255.255.255 port 67 interval > > > > > 3 > > > > > send_packet: No buffer space available > > > > > dhclient.c:2445: Failed to send 300 byte long packet over > > > > > wlxa0f3c10a28f7 interface. > > > > > > > > > > Nevertheless, after a few more lines from DHCP the link works. > > > > > Any ideas about buffer space? > > > > > > > > The kernel has no free RAM to queue a packet or that Tp-Link device > > > > you're is using low-quality kernel module. Happens with Tp-Link, but > > > > > > With TP-Link? Aren't they just using other manufacturers' chipsets? > > > > The hardware is definitely OEM'ed. The firmware is most likely theirs. > > Really? I thought that network drivers load firmware based solely on > the chipset.
It's more complicated than this. For instance, to identify a NIC chip you need something that responds to said identification. To put it simple, to load a firmware you need something to accept it. Some chips (ESP, nRF to name a few) allow you to write a firmware, but can hide lower-level functions like actual frame sending (i.e. - blobs in SDK). I.e. you provide a chip a tiny, yet complete OS, because otherwise all you have a chip that cannot even boot. From the host POV you just put some bits on a appropriate bus and hope that the other side is something you expect it to be. An example - [1]. Some chips (Broadcom, Realtek or Intel, for instance), have an OS part baked-in (or, very rarely - stored at EEPROM), and payload (i.e. application-level software) needs to be loaded by host. And that my "firmware" statement was referred to a part that's baked-in, not a blob from linux-firmware. After all, A0:F3:C1 in that NIC above is Tp-Link, not Realtek. Reco [1] https://github.com/espressif/esp-hosted
